Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Catastrophizing, fear and avoidance in the development of chronic pain McMurtry, Bruce Wiliam 2004

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2004-931552.pdf [ 13.19MB ]
Metadata
JSON: 831-1.0091918.json
JSON-LD: 831-1.0091918-ld.json
RDF/XML (Pretty): 831-1.0091918-rdf.xml
RDF/JSON: 831-1.0091918-rdf.json
Turtle: 831-1.0091918-turtle.txt
N-Triples: 831-1.0091918-rdf-ntriples.txt
Original Record: 831-1.0091918-source.json
Full Text
831-1.0091918-fulltext.txt
Citation
831-1.0091918.ris

Full Text

Catastrophizing, fear and avoidance in the development of chronic by BRUCE WILLIAM MCMURTRY B.A., (Hons.), The University of Calgary, 1992 M . A . , The University.of British Columbia, 1997 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T OF T H E R E Q U I R E M E N T S FOR T H E D E G R E E OF  DOCTOR OF PHILOSOPHY  in T H E F A C U L T Y OF G R A D U A T E STUDIES (Department of Psychology) We accept this thesis as conforming to the required standard  T H E UNIVERSITY OF BRITISH C O L U M B I A August 2004 ©Bruce W. McMurtry, 2004  Catastrophizing, Fear  ii  Abstract A self-perpetuating cycle of catastrophizing, pain related fear (PRF), and avoidance following tissue stress or damage has been proposed to underlie the development of chronic pain in some individuals (Vlaeyen, Kole-Snijders, Rotteveel, et al., 1995; Vlaeyen & Linton, 2000). A number of instruments have been designed to measure aspects of PRF and catastrophizing, and research utilizing them has provided support for many aspects of the pain related fear/avoidance cycle (Vlaeyen & Linton, 2000). However, there has been relatively little research on PRF or catastrophizing early in the transition to chronic pain and disability. The current study examined the structure of PRF and catastrophizing early in the course of a low-back pain episode, and the prediction of three-month outcome from those variables. Work disabled individuals with lowback pain completed questionnaires, including measures of PRF (i.e.. Pain Anxiety Symptoms Scale (McCracken, Zayfert & Gross, 1992), Fear Avoidance Beliefs Questionnaire, (Waddell, Newton, Henderson, Somerville & Main, 1993), Tampa Scale of Kinesiophobia (Vlaeyen, KoleSnijders, Boeren & van Eek, 1995)) and catastrophizing (Pain Catastrophizing Scale (Sullivan, Bishop & Pivik, 1995)), by mail at less than seven weeks post-pain onset and a second set of questionnaires, including measures of pain, perceived disability, depressive symptoms, and return to work at 3 months post-onset. Confirmatory factor analyses indicated that in this sample, all of the instruments except the T S K conformed to the factor structure intended by the scale designers. Structural Equation Modeling provided support for the relationships between pain level, PRF, catastrophizing and avoidance proposed by Vlaeyen and Linton to characterize fear-avoidance cycle. Regression analyses demonstrated that at least some aspects of PRF and/or catastrophizing contributed significantly to the prediction of all aspects of three-month outcome after controlling for demographics and initial pain level. These results provide support for the  Catastrophizing, Fear role of catastrophizing and PRF in the development of chronic pain and for the mechanisms proposed by Vlaeyen and Linton (2000) to underlie that effect, but suggest there may be utility further refinement of PRF measures. Limitations to the study and broader theoretical and clinical implications are discussed.  Catastrophizing, Fear  iv  T A B L E OF CONTENTS Abstract  i  Introduction  1  Literature Review  8  Theories of Chronic Pain Defining Chronic Pain and Disability  8 8  The Biomedical Model, of Pain  10  The Gate Control Theory  12  Behavioral/Learning Theory  13  Cognitive Behavioral Theories  15  The Role of Fear in Chronic Pain  16  Letham and Colleagues' Model of Pain Related Fear  17  Vlaeyen & Colleagues'Model of Pain Related Fear/  20  Empirical Evidence for the Role of Pain Related Fear in the Development of Chronic Pain  26  Tne Role of Catastrophizing in the Fear Avoidance Model of Chronic Pain  36  Measures of Pain Related Fear and Catastrophizing  37  Pain Catastrophizing Scale  38  The Pain Anxiety Symptom Scale  41  The Fear Avoidance Beliefs Questionnaire  46  Tampa Scale for Kinesiophobia  47  What is Being Measured? Fear Versus Anxiety  51  The Relationship Between Aspects of Pain Related Fear and Catastrophizing  54  A Proposed Model of the Relationship Between Aspects of Pain Related Fear  57  The Cognitive Model of Panic Disorder with Agoraphobia  59  The Cognitive Theory of Health Anxiety  61  Implications For a Model of Pain Related Fear  64  Summary of the Proposed Model  65  Summary of Review and Rationale for Research Questions Method Structural Equation Modeling  68 70 70  Catastrophizing, Fear  v  General Procedure for Structural Equation Modeling  70  Assessment of Model Fit  72  Comparison of Alternate Models  74  Model modification  76  Participants  79  Measures  83  Pain Related Fear and Catastrophizing  85  PASS  85  FABQ  86  PCS  87  TSK  87  Measures Specific to Structural Model Analyses  88  Pain Intensity  88  Avoidance Behaviour  89  Outcome Variables  90  Average Pain  90  Depressive Symptoms  90  Perceived Disability  91  Return to Work  91  Procedure  91  Results  93  C F A Results  93  PCS  93 Model 1 (one factor)  93  Model 2 (Two factor)  96  Model 3 (three factor)  96  Summary of PCS C F A  96  PASS  99  Model 1 (single factor)  99  Model 2 (4 factor original; McCracken et a l , 1993; Osman et al., 1994)  99'  Model 3 (5 factor; Larsen, Taylor & Asmundson, 1997)  99  Model 4 (4 factor; Larsen, Taylor & Asmundson, 1997)  103  Catastrophizing, Fear  vi  Model 5 (PASS-20 4 factor; McCracken & Dhingra, 2002)  103  Summary of PASS C F A Analyses  106  FABQ  107  Model 1  111  Model 2  111  Model 3  111  Model 4  115  Summary of F A B Q C F A analyses.  115  TSK  119 Model 1  119  Model 2  119  Model 3  124  Model 4  124  Summary of T S K C F A analyses  124  Test of theoretical structure  125  Measurement Model  125  Description of measurement for each construct  128  Catastrophizing  128  Avoidance  128  Fear of Pain  128  Fear of Activity  129  Pain Intensity  129  Testing and Modification of Measurement Model  129  Structural Analysis  133  Alternative Models  133  Model 1  .  1  3  3  Model 2  136  Model 3  136  Model 4 Model 5 Assessment of Model Fit Model 1  '  136 136 136 136  Catastrophizing, Fear  vii  Model 2  141  Model 3  141  Model 4  141  Model 5  142  Modification of Best Fitting Model Prediction of Outcome  142 145  Correlations between Independent and Dependent Variables  146  Prediction of Time Two Depressive symptoms  163  A l l Pain Related Fear Variables  163  Individual Pain Related Fear Scales  163  Prediction of Time Two Perceived Disability  170  A l l Pain Related Fear Variables:  170  Individual Pain Related Fear Scales  172  Prediction of Time Two Average Pain level  176  A l l Pain Related Fear Variables  176  Individual Pain Related Fear Scales  180  Prediction of Return to Work  184  A l l Pain Related Fear Variables  184  Individual Pain Related Fear Scales  186  Discussion Summary of Findings  193 193  Confirmatory Factor analyses  194  Structural Model  200  Pain Related Fear in the prediction of three month outcome  207  Overall Summary  213  Clinical Relevance of the findings  219  Limitations of the Current work  229  Implications for Future Research  239  References  242  Appendices  265  Catastrophizing, Fear  viii  LIST OF FIGURES Figure 1. Fear-avoidance model (Vlaeyen Kole-Snijders, Boeren, & van Eek, 1995)  21  Figure 2. Revised fear-avoidance model (Vlaeyen & Linton, 2000).  21  Figure 3. Taylor and Rachman (1992) models of fear and agoraphobic avoidance  62  Figure 4. Taylor & Rachman (1992) model of fear and avoidance of sadness  62  Figure 5. Model of Anxiety Sensitivity and Fear of Pain (Asmundson & Taylor, 1996)  63  Figure 6. Proposed structural model of pain, catastrophizing and pain related fear and avoidance Figure 7. Single factor model of PCS  66 ,  94  Figure 8. Two factor model of PCS  94  Figure 9. Three factor model of PCS  95  Figure 10. Single factor model of PASS  100  Figure 11. Four factor model of PASS (McCracken et al., 1993)  101  Figure 12. Five factor model of PASS (Larsen, et al., 1997)  102  Figure 13. Four factor model of the PASS (Larsen et al., 1997)  104  Figure 14. Four factor model of PASS-20 (abbreviated item set; McCracken & Dhingra, 2001) Figure 15. One Factor Model of F A B Q (Waddell, et al., 1993); 16 items minus item 6  105 112  Figure 16. Two factor model of full item set (Waddell, et al., 1993); 16 items minus item 6  113  Figure 17. One factor model of abbreviated F A B Q item set (Waddell et al., 1993)  114  Figure 18. Two factor mode! of abbreviated F A B Q item set (Waddell et al., 1993)  116  Figure 19. T S K model 1; single factor model of T S K (Vlaeyen, Kole-Snijders, Rotteveel, e t a l , 1995)  120  Catastrophizing, Fear  ix  Figure 20. T S K model 2; Four factor model of T S K (minus items 5,7,8,15 & 16; Vlaeyen, Kole-Snijders, Rotteveel, et al., 1995)  121  Figure 21. T S K model 3; two factor model of TSK (Clark, et al., 1996; Geisser, et al.., 2000; Goubert et al., 2004)  122  Figure 22. T S K model 4; Two factor model of TSK (Swinkel-Meewisse, Swinkel, et al.,2003) Figure 23. Measurement model for Pain Related Fear Constructs  123 130  Figure 24. Measurement model for all variables including pain intensity and activity avoidance  132  Figure 25. Structural model 1 (first alternative model)  135  Figure 26. Structural model 2 (second alternative model)  137  Figure 27. Structural model 3 (third alternative model)  138  Figure 28. Structural model 4 (fourth alternative model)  139  Figure 29. Best fitting structural model with standardized parameter estimates  144  Catastrophizing, Fear  x  LIST OF T A B L E S  Table 1 Descriptive statistics for S E M and regression samples (time one data)  84  Table 2 Goodness of Fit for the Confirmatory Factor Analyses of the PCS  97  Table 3 PCS Three-factor solution: loadings and factor labels  98  Table 5 Goodness of Fit for the Confirmatory Factor Analyses of the PASS  108  Table 6 Four-factor solution on abbreviated item pool (PASS 20): loadings and factor labels  109  Table 7 Factor intercorrelations for the four factor solution on abbreviated item pool (PASS 20)  110  Table 8 Goodness of Fit for the Confirmatory Factor Analyses of the F A B Q  117  Table 9 Four-factor solution of F A B Q on abbreviated item pool (Waddell et al., 1993) minus item 6: loadings and factor labels  118  Table lO.Goodness of Fit for the Confirmatory Factor Analyses of the T S K  124  Table 11 Goodness of Fit for Measurement Model  134  Table 12 Goodness of Fit for Structural Model  140  Table 13 Intercorrelations Between Full Scale Scores and Time Two Depressive (CES-D) Symptoms  147  Table 14 Intercorrelations Between PCS Subscale Scores and Time Two Depressive (CES-D) Symptoms  148  Table 15 Intercorrelations Between PASS Subscale Scores and Time Two Depressive (CES-D) Symptoms  149  Table 16 Intercorrelations Between F A B Q Subscale Scores and Time Two Depressive (CES-D) Symptoms  150  Table 17 Intercorrelations Between pain related fear Full Scale Scores and Time Two Perceived Disability (PDI)  151  Table 18 Intercorrelations Between PCS Subscale Scores and Time Two Perceived Disability (PDI)  152  Catastrophizing, Fear  xi  Table 19 Intercorrelations Between PASS Subscale Scores and Time Two Perceived Disability (PDI)  153  Table 20.1ntercorrelations Between F A B Q Scale Scores and Time Two Perceived Disability (PDI).  154  Table 21 Intercorrelations Between pain related fear Full Scale Scores and Time Two Average pain  155  Table 22 Intercorrelations Between PCS Subscale Scores and Time Two Average pain  156  Table 23 Intercorrelations Between PASS Subscale Scores and Time Two Average pain  157  Table 24 Intercorrelations Between F A B Q Subscale Scores and Time Two Average pain Table 25 Intercorrelations Between pain related fear Full Scale Scores and Return to Work Status at Time Two  158 159  Table 26. Intercorrelations Between PCS Subscale Scores and Return to Work Status at Time Two  160  Table27 Intercorrelations Between PASS Subscale Scores and Time Two Return to Work Status  161  Table 28 Intercorrelations Between F A B Q Subscale Scores and Time Two Return to Work Status  162  Table 29 Pain Related Fear Scales in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  164  Table 30 Pain Anxiety Symptom Subscales in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  165  Table 31 Pain Catastrophizing in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  166  Table 32 Fear Avoidance Beliefs in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  167  Table 33 Kinesiophobia in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  168  Table 34 Pain Related Fear Scales in the Prediction of Time Two Perceived Disability: Hierarchical Regression  171  Catastrophizing, Fear  xii  Table 35 Pain Anxiety Symptom Subscales in the Prediction of Time Two Perceived Disability: Hierarchical Regression  173  Table 36 Pain Catastrophizing Subscales in the Prediction of Time Two Perceived Disability (PDI): Hierarchical Regression  174  Table 37 Fear Avoidance Beliefs Subscales in the Prediction of Time Two Perceived Disability: Hierarchical Regression  175  Table 38 Kinesiophobia in the prediction of Time Two Perceived Disability: Hierarchical Regression  177  Table 39 Pain Related Fear in the Prediction of Time Two Average Pain: Hierarchical Regression  178  Table 40 Pain Catastrophizing Subscales in the Prediction of Time Two Average pain: Hierarchical Regression  179  Table 41 Pain Catastrophizing Subscales in the Prediction of Time Two Average pain: Hierarchical Regression  181  Table 42 Fear Avoidance Behaviour Subscales in the Prediction of Time Two Average pain: Hierarchical Regression  182  Table 43 Kinesiophobia in the Prediction of Time Two Average Pain: Hierarchical Regression  183  Table 44 Pain Related Fear in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  185  Table 45 Pain Anxiety Symptoms Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  187  Table 46 Pain Catastrophizing Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  188  Table 47 Fear Avoidance Beliefs Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  190  Table 48 Fear of movement/(re)injury in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  192  Catastrophizing, Fear  xiii  Acknowledgements A number of people were instrumental in the completion of this work and I am grateful to all of them. I would like to specially thank: Ken Craig, my supervisor, you are a source of inspiration, both academically and personally. Your reassurance and encouragement helped to tame my self-doubts and your advice helped me to create a project I am proud of. I hope that I can emulate, in some small measure, your dedication to good science, to your students, and to your family. Renee, Erinna and Ryan, my dear family, because your love and support kept me going through many long sleepless nights and frustrating days completing this "Dimitation". You were always there for me even when I had little left to give back, and I will always remember that. Mom, for your unending belief in me and reminding me that things will indeed work out. Mike my friend who generously tolerated my own catastrophizing and provided invaluable advice and a couch to sleep on even as your own deadlines crept closer. You are one of the few people who can truly understand what it took to finish and I am grateful to have gotten to know you. Izabela Schultz my secondary supervisor. Thank you for sharing your thoughts, which were the impetus for this work, for assisting me to gain access to the W C B database, and for providing invaluable advice and support at every step along the way. Buno (Bruon) Zumbo who helped me to navigate the dark world of statistics and demonstrated a knack for providing just the right the amount of information to answer my questions and reassure me about what I did know, without increasing my anxiety over how much I didn't know. Wolfgang Linden and Geoff Hall, the other members of my committee, who so generously gave of their time and their knowledge to improve the final document. David Wong, our lab manager who responded quickly and efficiently to so many last minute requests without complaint and without whose assistance I would have had no data. To the staff at FIT whose unwavering support and words of encouragement helped me to see the light at the end of the tunnel. I would especially like to thank Carol Brown who understood the importance of my completing this work and provided me the time and resources to do so and John Marshal for his invaluable advice and his willingness to discuss even the most esoteric topics at the most inconvenient times. Thank you as well to Annalyn, Gregg and Marc for your support and for sharing your own experiences, so that I knew what to expect. And finally, to the rest of the WCB'research group, Gregg Meloche, Ruth Milner, Joan Crook, Ooghna Zuberbier, Jonathon Berkowitz and Wendy Meloche, for allowing me to attend your meetings where I learned a great deal about back-pain research with injured workers. A special thanks to Wendy for cheerfully providing advice and practical support throughout the project.  Catastrophizing, Fear  1  Introduction Low-back pain is one of the most common reasons for seeking medical attention (Von Korff Dworkin, LeResche & Kruger, 1988) and it has been estimated to affect 80% of the population at some point in their lives. The back is a mechanically complex structure innervated with pain receptors at many points and therefore acute pain may arise from any of a variety of compromises to that structure. In the majority of cases, the source of acute low-back pain is identified as soft-tissue injury, and the pain is expected to resolve as the injured tissue heals, typically within approximately three months. However, for some individuals, pain, which began with an acute injury to the back, does not resolve in the expected three months and can be considered chronic (Merskey & Bogduk, 1994). Chronic low-back pain is frequently associated with extensive psychological distress (Craig, 1999), disability (Fordyce, 1995) and associated financial costs (Nachemson, 1992; Turk, 1997). Understanding the process underlying the transition from acute to chronic pain is central to reducing these impacts and promises to aid in reducing the incidence of chronic back pain. While pain has frequently been described in terms of its physiological and anatomical aspects, current conceptualizations of pain recognize that psychological and social factors also play an important role (Block, Kremer & Fernandez, 1999). There is considerable evidence demonstrating that psychological and social processes are implicated in the persistence of pain, distress and disability (Schultz, et al., 2004; Turk, 1997; Waddell, Newton, Henderson, Somerville & Main, 1993). Two psychological constructs, which have gained considerable recent attention, are fear and catastrophizing (e.g., Asmundson, Norton & Allerdings, 1997; Letham, Slade, Troup & Bentley, 1983; Vlaeyen & Linton, 2000, Waddell et al., 1993).  Catastrophizing, Fear  2  While fear and pain have long been known to be closely related, the exact nature of that relationship is unclear (Craig, 1999). However, several authors have argued that fear may interfere with recovery from acute pain, thereby causing chronic pain to develop (Vlaeyen & Linton, 2000). Vlaeyen and Linton (2000) synthesized a growing body of empirical and theoretical work in this area to develop a comprehensive model of the process underlying that transition. They proposed that individuals who are prone to catastrophize, that is individuals who expect negative outcomes and doubt their ability to control their pain, are likely to become fearful of pain or of activities expected to cause pain. According to Vlaeyen and Linton, this fearfulness leads to hypervigilence for signs of pain or injury and/or avoidance of activities, which in turn prevents a normal return to activity necessary for healing, thus prolonging physical healing and exacerbating emotional distress. While allowing that this cycle typically begins with a physical injury, Vlaeyen and Linton argue that it can become self-perpetuating, such that it persists even as the original injury heals, resulting in chronic pain. Vlaeyen and Linton further argue that if pain related fear can be identified early and addressed through psychological intervention, the likelihood of chronicity developing might be reduced. In the broader literature on fear and anxiety, there is abundant evidence of the effectiveness of such interventions in reducing levels of fear and anxiety and the impact they have on an individual's life (McLean & Woody, 2001). That research also points to the importance of identifying exactly what it is the person fears, in order to most efficiently address that fear. The success of intervention in preventing chronic pain via treatment of pain related fear, hinges on the ability to accurately identify individuals experiencing, or at risk for experiencing pain related fear (Vlaeyen & Linton, 2000). It also depends on developing clear  Catastrophizing, Fear  3  understanding of the various aspects of pain-related fear and which might be most profitably addressed prior to pain and disability becoming chronic. A number of psychometric instruments have been developed to assess pain related fear, for example, the Fear-Avoidance Beliefs Questionnaire (FABQ; Waddell et al., 1993), the Tampa Scale for Kinesiophobia (TSK; Vlaeyen, Kole-Snijders, Rotteveel, Ruesink & Heutts, 1995) and the Pain Anxiety Symptom Scale (PASS; McCracken, Zayfert & Gross, 1992). The F A B Q was designed to assess fear of work and other activities and the belief they should be avoided, while the T S K assesses the degree to which an individual fears movement/(re)injury. The PASS (McCracken, Zayfert & Gross, 1992) assesses self-reported cognitive, physiological and behavioral responses to pain. The Pain Catastrophizing Scale (PCS; Sullivan, Bishop & Pivik, 1995) was recently developed specifically to measure pain catastrophizing. Research employing one or more of these instruments has confirmed many elements of the fear avoidance model (e.g. see reviews by; Asmundson, Norton, & Norton, 1999; Vlaeyen and Linton. 2002) and there is evidence that pain related fear and catastrophizing may have utility at the early stages following pain onset, in predicting which individuals will go on to suffer chronic pain (e.g., Burton, Tillotson, Main & Hollis, 1995; Fritz, George, & Delitto, 2001). However, a number of issues remain to be resolved i f a full understanding of the mechanism of this effect is to be achieved. One issue relates to the limited evidence to date for many causal aspects of Vlaeyen's theory. The prediction that pain related fear leads to the persistence of pain, suffering and disability, a process that characterizes the transition from acute to chronic pain, is the cornerstone of the theory. If that prediction holds true, the measurement of pain related fear early in that transition promises to aid in the prediction and prevention of chronic pain.  Catastrophizing, Fear  4  However, studies using the current instruments (Crombez, Vlaeyen, Heuts & Lysens, 1999; McCracken, Gross, Aikens, & Carnrike, 1996) have primarily employed cross sectional designs and chronic pain samples. There have been very few longitudinal studies, even fewer beginning early after pain onset, and even less comparing different pain related fear measures. If measures of pain related fear are to be used to predict and prevent the development of chronic pain, it will be important to demonstrate that the instruments are appropriate in individuals whose pain is not yet chronic and that they are indeed predictive of persistent pain, distress and disability in such a sample. While there is some evidence for the factorial validity of all of the instruments in chronic pain samples, and in some cases in community samples (e.g., Van Dame, Crombez, Bijttebier, Goubert, & Houdenhove, 2002; Osman, Barrios, Osman, Schneekloth, and Troutman, 1994; Waddell, Newton, Henderson, Sommervile, & Main, 1993; Goubert et al. 2004), only the factor structure of the T S K has been examined in a sample of individuals with recent onset pain (Swinkels-Meewise, Roelofs, Verbeek, Oostendorp & Vlaeyen, 2003), and that study indicated a different structure than what had been found in chronic pain samples. Another related issue is the relationship between the instruments and the constructs they measure. In a 1996 paper, McCracken et al. (1996) pointed out that "while there appears to be a clear rationale for assessing fear and anxiety in patients with chronic pain, it is not clear how these variables should be defined and measured". Given the causal implications of Vlaeyen's theory, a similar and perhaps more compelling argument applies to the measurement of those variables in patients whose pain has not yet become chronic. More recently, McNeil and Vowles (in press) pointed out that test selection remains an issue in measuring pain related fear and that to date there is limited evidence on how the instruments compare.  Catastrophizing, Fear  5  Empirical evidence suggests that the constructs tapped by the various pain related fear instruments are closely related, but intercorrelations between them are not so high as to indicate that they are entirely redundant (e.g., McCracken et al., 1996; Crombez et al., 1999). The instruments also show different patterns of correlation with other pain related variables (e.g., McCracken et al., 1996; Crombez et al., 1999), suggesting that the measures, and the constructs they assess are not redundant. Importantly, these findings were based on chronic pain samples and the relationship between instruments and the constructs they measure may be different for individuals whose pain has not yet become chronic. Examination of the item content of the instruments also suggests that they are tapping different domains. For example, items on the Pain Anxiety Symptom Scale (PASS; McCracken, Zayfert & Gross, 1992), focus on assessing fear and avoidance of pain, while other instruments such as the F A B Q (FABQ), and the Tampa Scale of Kinesiophobia (TSK) focus on fear of activities expected to cause pain (McNeil & Vowles, in press). The Pain Catastrophizing Scale, as the name implies, addresses catastrophic cognitions about the meaning of pain. However, there are complexities, which are not captured in these descriptions. For instance, the PASS (McCracken, Zayfert & Gross, 1992) attempts to assess physiological, behavioral and cognitive aspects of fear, while the F A B Q , the T S K and the PCS focus exclusively on cognitive aspects. The PCS is specifically designed to assess pain catastrophizing and treats it as a separate construct, while the PASS includes a subscale, which appears to tap into catastrophizing, but includes it as a component of Pain Anxiety. Vlaeyen and Linton's (2000) model provides a framework for understanding some aspects of these constructs and their interrelationships, for example the postulation that catastrophizing causally precedes other aspects of pain related fear and therefore that other  Catastrophizing, Fear  6  aspects of pain related fear should mediate the relationship between catastrophizing and avoidance. However, to date there does not appear to have been empirical verification of this postulate. In addition, the other pain related fear constructs, for example fear of pain and fear of activity are generally recognized as distinct constructs, but in Vlaeyen and Linton's model they are grouped together as pain related fear. A better understanding of the interrelationships of the instruments, and by extension, of the constructs they assess, would inform theory and also be useful in targeting intervention. For example, i f catastrophizing precedes pain related fear in the transition to chronic pain, early intervention to address catastrophizing might prevent the development of the aspects of pain related fear and consequently reduce the likelihood of pain becoming chronic. Conversely, at later stages when fear of pain and/or activity has already developed and generalized, it may be necessary to directly address that fear. The present study addressed three specific goals. The first was to examine the factor structures of the F A B Q , PASS, PCS and TSK in a subacute (pain of less than 7 weeks duration) back pain population. The second goal was to develop and empirically test a theoretical model of relationships between aspects of pain related fear tapped by those instruments at the subacute stage. The third goal was to examine and compare the relative utility of the F A B Q , PASS, PCS and TSK at the subacute stage of pain in predicting pain, disability and emotional distress longitudinally. Questionnaire data were collected by mail from a group of Workers' Compensation Claimants with subacute low-back pain, a population at high risk for the development of chronic pain (e.g., Reesor & Craig, 1988). It would be expected that injury severity might confound the relationship between pain related fear and outcome, for that reason, it was  Catastrophizing, Fear  7  considered important to control for injury severity in some manner. Statistical control of injury severity would require quantifying it in some way. However, injury severity is very difficult to quantify in terms of the expected effect on recovery, so it was decided to attempt to achieve a relatively homogeneous sample with regard to that variable. In order to obtain achieve that homogeneity with regard to injury severity, the sample was restricted to individuals with relatively uncomplicated soft-tissue injuries of the low-back. Participants were asked to complete two sets of questionnaires, one within the first seven weeks post injury (subacute) and another at three months post injury (chronic). Questionnaire packages included the F A B Q , PASS, PCS and T S K as well as measures of pain, disability and emotional distress. The factor structure of each instrument was tested using confirmatory factor analysis through structural equation modeling (SEM). A model of interrelationships between the various aspects of pain related fear was developed rationally based on the literature review and tested empirically using S E M analyses. The utility of the various instruments in predicting prospective pain, disability, emotional distress and return to work was evaluated through regression analyses.  Catastrophizing, Fear  8  Literature Review The review includes empirical and theoretical work in the area of pain related fear (pain related fear) and relevant work on more general fear and anxiety to provide a background for the development of a conceptual model of pain related fear and anxiety. For the sake of parsimony, in the following discussion the term pain related fear is used as a blanket term to include all aspects of fear/anxiety regarding pain, activity, work, and (re)injury. While pain catastrophizing is frequently considered separately from the other constructs (e.g., Vlaeyen et al. 1995, Vlaeyen & Linton., 2000), some (e.g., McNeil & Vowles, In press) discuss measures of catastrophic thinking in the context of pain related fear. Likewise, the authors of the PASS, McCracken, Zayfert, & Gross (1992) include aspects of catastrophic thinking in their measure of pain related fear. However, in the interest of clarity, catastrophizing will be differentiated from the other aspects of pain related fear in the following discussion. The following review begins with a broad description and discussion of the biopsychosocial approach to chronic pain as a context for review of the literature on pain related fear. The next section will provide detailed analysis of theoretical and empirical work in the area of pain related fear and anxiety. This analysis focuses on the psychometric instruments most commonly used in the area, and concludes with an examination of the aspects of pain related fear, which appear to be tapped by each instrument. In order to provide a framework for understanding how these aspects of pain related fear and anxiety might interrelate, relevant work on more general aspects of fear and anxiety will then be presented. Finally, a theoretical model of pain related fears as measured by existing instruments and informed by the preceding review will be presented. Theories of Chronic Pain Defining Chronic Pain and Disability There has been some variability in the application of the term chronic pain. For example, until recently, the term was reserved for pain of greater than six months duration. However, the International Association for the study of pain has defined chronic pain as pain lasting beyond  Catastrophizing, Fear  9  the expected time for tissue healing, which is typically three months or less. Research suggests that three months is a pivotal point in time, as persisting pain, distress and disability at that point predict poor long-term outcome (Burton Tillotson, Main & Hollis, 1995; Phillips & Grant, 1991). Pain which has not become chronic is frequently considered acute, but Spitzer (1987) in a report from the Quebec Task Force on Spinal Disorders has argued that the term acute pain should be reserved for the first week following injury and that between 7 days and 7 weeks post onset, pain should be labeled subacute. As stated in the introduction, back pain is frequently associated with significant distress and disability. From the perspective of occupational disability, Krause & Ragland (1994) have suggested that since pain and disability do not always follow the same trajectory (Waddell, 1987), and because a variety of factors other than pain, influence disability, categorizations based on duration of pain are too broad to reflect the phases of functional disability. They argued for consideration of changing systemic (e.g. medical and work related) factors over the course of disability in addition to duration of pain in defining stages of disability. They proposed eight phases of disability based on pain duration combined with work status, from phase one, the period before an injury is reported, to phase eight, permanent disability. Phases one (nondisabling pain), two (formally reported injury) and three (short term disability) occur over the first six days of missed work following injury and therefore roughly corresponds to the acute stage of pain as defined by Spitzer. Phase four of Krause and Ragland's model reflects work disability of 1-7 weeks and therefore roughly corresponds to the subacute category of pain. Disability of 7-12 weeks duration falls in phase five and is considered long term disability. Krause and Ragland argue that this phase is important in terms of treatment decisions, because it is during this phase that hopes of recovery appear to fade (Hendler, 1984) and therefore interventions should shift from primary care to surgery or multidisciplinary treatment as appropriate.  Catastrophizing, Fear  10  The Krause and Ragland (1994) model is a useful adjunct to the clinical categorization of stages of pain in considering occupational disability issues. However, the existing psychological literature typically reports sample characteristics in terms of pain stage, likely because samples often include non-employed individuals to whom disability phases do not apply. The following discussion, being focused on psychological variables, and outcomes beyond work disability, will likewise primarily focus on stages of pain, however, since the sample employed in the current work consisted of Workers' Compensation Claimants, and because workplace disability was considered one aspect of outcome, phases of disability were discussed in the literature review where relevant. In the current work, the Quebec Task Force (Spitzer, 1987) definitions of acute, subacute and chronic pain were utilized, pain of less than 7 weeks will be considered subacute and pain of more than three months will be considered chronic. The existing psychology research varies in use of the acute, subacute and chronic (and in some instances sub-chronic) to describe stages of pain. Where study authors utilized different definitions or terminology than those laid out here, specific durations of pain were provided in the current review, but the authors' own terminology was retained. Regardless of the particular categorization of stages, it is important to recognize that factors affecting pain and disability are likely to change over the course of the problem and therefore research on individuals at one stage does not necessarily apply to individuals at other stages (Krause & Ragland). . The Biomedical Model of Pain Throughout the 19th and early 20th centuries, attempts to treat and understand chronic pain were guided by a biomedical model of pain (Robinson & Riley, 1999a), which focused on the anatomical and physiological substrates of pain transmission and perception (Turk & Flor, 1999). While variants of this model exist, they share the common assumption that pain results from unidirectional neural transmission of a nociceptive signal from peripheral sensory neurons to the central nervous system. Despite the fact that such models have contributed to considerable advances in our understanding of the neural and biochemical mechanisms of pain, a growing  Catastrophizing, Fear  11  body of clinical observation and research findings highlight inadequacies of a purely biomedical account. One difficulty with a purely biomedical model of pain is that it does not appear to account well for the often noted discrepancy between the intensity of a pain stimulus and the subjective experience and behavioral expression it elicits. Physician Henry Beecher (1956) was one of the first to highlight the degree to which observable injury does not correspond to subjectively experienced pain. He noted that soldiers, who had suffered severe injuries in battle, often reported less pain and exhibited less distress than did similarly injured patients from his general practice. Beecher attributed this effect to the fact that for the soldiers their injury represented an opportunity to avoid the battlefield, while, for civilian patients, the pain likely represented something more threatening. More recently, it has been noted that blocking or even cutting the pathways known to be responsible for transmitting nociceptive signals often has little or no effect on pain sensation (Turk & Flor, 1999). A n extreme example of this phenomenon is phantom limb pain, where an individual reports pain in a limb that has been amputated (Melzack & Wall, 1965). Chronic back pain is particularly difficult to explain from a purely biomedical perspective. From a biomedical perspective, pain should be associated with the presence of a nociceptive stimulus, and should resolve when that stimulus is removed. In the case of injury, the subjective experience and behavioral expression of pain should correlate with the degree of tissue damage and should therefore decline as that damage heals. B y definition, chronic pain has persisted beyond the normally expected healing time for tissue damage (Merskey & Bogduk, 1994). In some cases, the persistence of back pain can be attributed to ongoing tissue pathology such as arthritic or malignant changes in the spine. However, in many cases no such pathology can be detected. Radiographic investigation often fails to identify any apparent pathology and paradoxically many individuals who report no subjective pain do show significant tissue pathology on radiographic examination (Flor & Turk, 1999). While diagnostic testing is not entirely reliable and may not be sensitive to all underlying pathology, these measurement issues  Catastrophizing, Fear  12  should decrease as technology improves. Despite dramatic technological advancements, the ability to reliably identify causes of persistent back pain and disability through medical testing is limited (Waddell et al., 2002). Even in cases of chronic pain where pathology is clearly identifiable, the degree of pathology generally bears little relation to the level of pain, disability and distress experienced by the patient (Waddell, 1987). As might be expected from the foregoing discussion, treatments for chronic pain derived from a biomedical understanding have been of limited benefit, particularly in the case of back pain. In an address to the North American Spine Society, Haldeman (1990), a physician, lamented that the lack of evidence that "the proliferation of new technology and advanced clinical skills for the investigation and treatment of spinal pain has influenced the overall incidence, morbidity, cost, or disability related to spinal disorders." (p. 718). As the inadequacy of the traditional biomedical approach for understanding and treating chronic pain became more apparent in the latter half of the 20th century, a variety of broader, more inclusive models including environmental, psychological and behavioral aspects of pain perception and experience has been developed. These models focus on different aspects of pain, and assign different causal priorities among them, but they are largely compatible and form a natural confluence toward what has been called the bio-psycho-social (Robinson & Riley, 1999a) or bio-behavioral (Turk & Flor, 1999) model. Central to the development of these more satisfactory models of chronic pain is the gate control theory of pain developed by Melzack and Wall (1965). The Gate Control Theory Melzack & Wall (1965) proposed the existence of a spinal gating mechanism located in the dorsal horn of the spinal cord. They hypothesized that this spinal gate allows neural input, including descending signals from the central nervous system, to moderate ascending nociceptive signals. They further proposed that in addition to sensory features, pain also includes affectivemotivational and cognitive-evaluative components each subserved by different brain structures (Melzack & Casey, 1968).  Catastrophizing, Fear  13  While there has been some debate about specific components of the theory, and it has required some revision, there is a great deal of support for the basic tenets (Wall, 1996). The theory has served as the impetus for further research into psychological facets of pain (Robinson & Riley, 1999a), because it acknowledged the multifaceted nature of pain and provided a putative mechanism through which psychosocial variables might influence pain perception. However, the gate control theory was not formulated specifically to account for chronic pain (Robinson & Riley, 1999a) and it does not provide a detailed account of the psychological processes involved (Turk & Flor, 1999). Other models, largely compatible with the gate control theory but more specific to chronic pain, have been developed and do provide a more comprehensive account. Probably the most widely employed and best supported are the behavioral/learning theory (e.g., Fordyce, 1976; Keefe & Williams, 1989) and the cognitive behavioral theory (Turk & Rudy, 1989; Turk & Flor, 1999). As the names imply, these theories share many common tenets, and are largely compatible, differing chiefly in their focus. Behavioral/Learning Theory Fordyce has been the pre-eminent advocate for the behavioral approach to chronic pain (Turk & Flor, 1999). Drawing on the behaviorist tradition in psychology, his account of chronic pain focused on the observable behaviors which express pain, and the application of learning principles to understanding and changing those behaviors (Fordyce, 1976). Fordyce defined chronic pain as the persistence of pain behaviors due to learned associations and contingencies. The most immediate and perhaps most apparent behavioral responses to pain are escape and avoidance. Fordyce proposed that these responses, while initially triggered by a pain stimulus, could persist through a combination of classical and operant conditioning. According to Fordyce, behaviors such as escape and avoidance are initially selfreinforcing because they reduce pain, which is a negative stimulus. In addition, he noted that pain behaviors including avoidance, frequently elicit favorable outcomes such as sympathy and assistance from others and relief from disliked responsibilities. Fordyce argued that through  Catastrophizing, Fear  14  these outcomes, we come to learn contingencies, which elicit pain behaviors such as avoidance out of proportion to the pain stimulus or even in the absence of such a stimulus. The demonstrated effectiveness of treatment based on the behavioral theory of chronic pain (e.g., see review by Keefe & Williams, 1989) supports the basic tenets of the theory. Keefe and Williams reviewed a body of research indicating that altering the contingencies between pain behavior and environmental consequences, for example reducing positive reinforcement, can reduce those behaviors. However, it has been argued that the direct reinforcement is not adequate to explain chronic pain, as the negative consequences, in terms of emotional distress and lost function, would seem to far outweigh the degree positive reinforcement most patients receive for their pain behavior (e.g., Phillips, 1987). In addition, concerns have been raised about the narrow focus of operant models in particular on observable behaviour at the expense of considering other, subjective experiences (e.g. Turk, 1996). Social learning is another mechanism which may affect pain behavior, but without direct experience (Craig, 1986). Bandura (1969) proposed that social information may be as important in shaping behavior as conditioning in that it provides information on potential consequences, • which might not be available through direct experience. The social context provides considerable information about the nature and consequences of pain and this information appears to be integrated into our own responses (Craig, 1986). Numerous laboratory studies have demonstrated the extent to which pain responses are subject to modeling influences, (e.g., Craig & Weiss, 1971; Craig & Patrick, 1985; Prkatchin & Craig, 1986), to the extent that pain reports can be elicited by non- noxious stimuli, through the influence of modeling (Craig, Best & Reith, 1974). It is conceivable that such mechanism could also contribute to the persistence of pain behavior, which characterizes chronic pain. While the social learning theory retains a focus on observable behavior, it allows for the role of cognitive processes such as appraisal and expectancies in shaping that behavior.  Catastrophizing, Fear  15  Cognitive Behavioral Theories There are a number of somewhat divergent cognitive behavioral accounts of how chronic pain may come to develop, but they share some common principles and there have been efforts to integrate them (e.g., Novy, Nelson, Francis & Turk, 1995). Cognitive behavioral theory integrates aspects of behavioral and learning theory, but is considerably broader in scope. In cognitive behavioral accounts of chronic pain, the focus shifts from observable behavior to cognitive process, and the subjective experience. This position emphasizes the role of cognitive processes, such as appraisals, interpretations, and beliefs as central to not only the expression, but to the experience of chronic pain (Turk & Rudy, 1989). Like the gate control theory, cognitive behavioral theories attempt to integrate affective, behavioral, cognitive, and biomedical aspects of pain (Novy et al., 1995), emphasizing the interactive nature of the relationships among these domains (Turk & Rudy, 1989). The determinants of behavior are viewed to be an interaction of individual and environment. (Turk & Rudy, 1989). While the model encompasses pain behavior and learning processes, cognitive processes, as opposed to behavior, assume a more central role. Human beings are seen as active information processors and this information processing or cognition can influence behavior and emotion (Turk & Rudy, 1989). Because of the breadth of constructs subsumed under cognitive behavior theory, research has generally focused on narrow areas of interest at the expense of integration (Turk, 1997). Nevertheless, a considerable body of research indicates that cognitive behavioral mechanisms including coping strategies (Snpw-Turek, Norris, & Chan, 1996), beliefs (Jensen, Turner, & Romano, 1991) and appraisals (Jensen & Karoly, 1991; Sullivan & D'eon 1990), influence how chronic pain is manifest. Reesor & Craig (1988) showed that maladaptive cognitions differentiated patients exhibiting medically incongruent pain symptoms from patients demonstrating symptoms consistent with medical data. Cognitions including appraisals and beliefs have been shown to play a role in the persistence of chronic pain (Williams & Thorn, 1989, Lacroix & Barbee, 1992). Further evidence for the importance of cognition in chronic pain is available in the treatment literature. For example, a meta-analysis of laboratory studies  Catastrophizing, Fear  16  indicated that cognitive behavioral interventions can increase tolerance to various pain stimuli (Fernandez & Turk, 1989) and treatments based upon cognitive behavioral principles have been demonstrated to reduce self-reported pain, depression, and disability in chronic pain patients (Turner & Jensen, 1993). The Role of Fear in Chronic Pain The relationship between fear and pain is likely quite complex, and it may be reciprocal (Craig, 1999). From a cognitive behavioral perspective, fear, like pain, can be seen as a multidimensional phenomenon including affective, cognitive and behavioral aspects (Craig, 1999). Fear is one of the onset salient responses to a painful stimulus, and in the case of acute pain it serves an important purpose. Pain often signals danger of injury and, if we fear the stimuli that trigger it, we are motivated to escape it if possible and to avoid it in the future. However, many authors have argued that fear can also play a less adaptive role in pain, exacerbating the inherent aversiveness of pain and/or prolonging it. The primary mechanism of this effect is seen to be avoidance behaviour (Phillips, 1987, Letham, Slade, Troup & Bentley, 1983; Vlaeyen et al. 1995, Vlaeyen & Linton, 2000). Theoretical work on pain related fear and its effects has been influenced by both the behavioral/learning and cognitive-behavioral accounts of chronic pain. Phillips (1987) applied cognitive behavioral principles to the phenomenon of pain avoidance as highlighted by Fordyce (1976). Drawing on the work of Fordyce, Phillips emphasized the importance of avoidance behavior in the development of chronic pain, but applied the cognitive constructs of beliefs and expectations to understanding the phenomenon. In contrast to Fordyce, Phillips emphasized the negative consequences of avoidance behavior and argued that they far outweigh the positive reinforcement that might be derived. Contrary to the intended prevention of pain, there is evidence that avoidance behavior can actually sensitize individuals to pain (Phillips & Jahanshahi, 1985) and avoidance does not appear to reduce ongoing levels of pain (Phillips & Jahanshahi, 1985; 1986). Phillips, therefore, argued that basic conditioning principles are inadequate to explain why avoidance behavior would persist.  Catastrophizing, Fear  17  In providing her explanation, Phillips (1987) emphasized the role that beliefs and expectations such as the amount of pain anticipated, might play in motivating avoidance behavior. Phillips argued that through pain experience, an individual begins to develop expectations that certain behaviors or situations will cause or exacerbate pain and a belief that these behaviors or situations should be avoided. These fearful expectations may be inaccurate from the beginning, or may become so over time, as tissue healing occurs and activities no longer pose real threat of pain or injury. Phillips argued that these fearful beliefs and expectations will be resistant to change because avoidance prevents the kinds of experience that would disconfirm them. Research has shown that through experience individuals will correct their inaccurate expectations of pain, such that predictions become more accurate with exposure (Rachman & Arntz, 1991). However, avoidance precludes such learning from occurring. Phillips argued that through this mechanism, the cycle of expected pain and avoidance can become self-perpetuating even in the absence of nociceptive stimuli. Letham and Colleagues' Model of Pain Related Fear A more specific discussion of the role of fear, in pain related avoidance behavior was provided by Letham et al. (1983).  Like Phillips (1987), Letham and colleagues argued that  avoidance is central in chronic pain and they emphasized the role of fear in motivating avoidance. Letham et al., drew on the differentiation between the motivational-affective and sensory-discriminatory aspects of pain first proposed by Melzack and Wall (1965) to explain how fear might contribute to the persistence of pain. Letham and colleagues proposed that under normal circumstances, the affective and sensory components of pain are concordant, such that the affective distress elicited by painful stimuli is directly related to its sensory strength. According to Letham et al. (1983), some individuals are more predisposed to respond to pain in a fearful way. When they experience pain, they also experience significant fear, which magnifies the motivational-affective components of the pain experience. This causes the pain stimulus to be more aversive and more likely to be avoided. In another similarity to the description provided by Phillips (Letham and  Catastrophizing, Fear  18  colleagues suggested that avoidance prevents the kind of experiences which would correct or recalibrate the affective component of pain perception. In Letham and colleague's account, without this corrective learning experience, fear, and, therefore, avoidance of pain stimuli, increases, allowing the discordance between the sensory and affective aspects of pain to increase. Thus, pain perception, pain behavior, and/or physiological responses to pain "come to be out of all proportion to demonstrable organic pathology or current levels of nociceptive stimulation" (pp. 402). Letham and colleagues, also highlight the physical consequences of avoidance, and the role these play in prolonging or exacerbating pain. They described the loss of muscle tone and reduced flexibility, which result from inactivity, as potentially creating new sources of pain or exacerbating existing sources. Through this mechanism, avoidance exacerbates the sensory aspects of pain, further feeding into the self-perpetuating cycle of exaggerated responses ). Letham and colleagues (1983) described a group of variables, including history of painful events, current stress, personality and pain coping strategies, which provide the context where an individual will experience an excessive affective response to pain and therefore predispose that individual to chronic pain. However, these contextual variables are not clearly specified and their role is not clearly defined. In the text of their paper, Letham et al., suggest that the contextual variables contribute to fear. However, the diagram representing the theory in the same article suggests that the psychosocial context created by these variables mediates between fear and avoidance. This lack of clarity also affects later empirical work based upon the model. Research has demonstrated that contextual variables such as those proposed by Letham et al., do play a role in the persistence and manifestation of chronic pain, (Klenerman et al., 1995), but these studies did not measure fear itself, so it is not clear that the mechanism suggested by Letham et al (1983) was responsible or that fear or avoidance played a role in the effects observed. The accounts of Phillips and Letham share many common themes, particularly the emphasis on avoidance as a central factor in the development of chronic pain and the role of fear in motivating that avoidance, but they also diverge on some points (Phillips & Jahanshahi, 1985).  Catastrophizing, Fear  19  Phillips & Jahanshahi (1985) proposed that fear of pain as described by Letham et al., may not be the important factor in predicting avoidance behavior. They found that over a series of trials, headache sufferer's tolerance for a painful laboratory task was better predicted by pain experienced in prior exposures to the task than by the pain experienced during a particular exposure. Based in this finding, they suggested that the "individual's beliefs about the nociceptive power of various stimuli and situations" are more important than the affective component of the pain response itself. In essence Phillip's proposes that avoidance and not an exaggeration of the fear component of the pain response, is the primary phenomenon and that avoidance is motivated by expectations of pain. The primary focus of Phillip's arguments appears to be the assertion that anticipated aversiveness rather than the actual aversiveness is the primary motivation for avoidance behaviour, and therefore that cognitive processes play a more important role than a conditioned fear response. However, it is possible that both processes occur, since there is evidence that fear of pain specifically predicts avoidance behaviour in chronic pain patients (e.g., McCracken, Gross, Zayfert, Sorg, & Edmunds, 1992; Burns, Mullen, Higdon, Wei, & Lansky, 2000). Specifically, McCracken et al. and Burns et al. found that scores on the PASS predicted performance on a physically demanding leg raise task. This divergence from Phillips & Jahanshahi's (1985) findings, may be due to different pain populations being used; chronic headache (Philips & Jahanshahi, 1985) versus chronic musculoskeletal pain (e.g., McCracken, Gross et al., 1992; Burns, Higdon, Wei, & Lansky, 2000). However the difference in outcome may also have been related to the way in which fear of pain was conceptualized in the various studies. Phillips and Jahanshahi conceptualization of fear of pain as corresponding to the affective component of the pain response itself (i.e. Affective scale of the M c G i l l Pain Questionnaire; MPQ). However, while the M P Q affective scale does include items reflecting fear (e.g., category 13 consists of the words "fearful", "frightful", and "terrifying"), the scale also assesses five other affective components (Melzack & Torgerson, 1971). Thus the affective scale of the M P Q may not be exclusively assessing fear of pain. In addition, the PASS, does attempt to measure cognitions reflecting beliefs, as described by  Catastrophizing, Fear  20  Phillips and Jahanshahi, but beliefs specifically about pain itself (e.g., "Even i f I do an activity which causes pain, I know it will decrease later"; reverse scored and item 5, "when I feel pain, I am afraid that something terrible will happen) rather than beliefs about the amount of pain a particular stimuli will elicit. Thus it may be that two different types of fear are operative in pain related avoidance, fear of pain and fear of stimuli expected to cause pain and both may include cognitive anticipatory components. Vlaeyen & Colleagues' Model of Pain Related Fear Vlaeyen and colleagues (Vlaeyen Kole-Snijders, Boeren, & van Eek, 1995; Vlaeyen & Linton, 2000) have presented a more clearly delineated model of pain related fear and avoidance in the development of chronic pain. These models are displayed graphically, in Figures 1 and 2. Drawing on the work of Letham et al. (1983), Phillips (1987), and Waddell et al. (1993), in addition to more recent empirical work, this model provides a relatively well specified description of how an acute injury can lead into a self perpetuating cycle of fear and avoidance resulting in chronic pain. In line with Phillips and Letham et al., Vlaeyen and Linton proposed that individuals who experience pain related fear will avoid activity, leading to disuse, depressed mood, and disability, which prolong or exacerbate the pain experience further feeding the cycle. As can also be seen in the model, the authors include pain catastrophizing, which they define as "an exaggerated negative orientation toward noxious stimuli" (p.320) as a precursor to pain related fear. This element is absent in the Letham et al. (1983) model, but it corresponds to some degree with the type of beliefs outlined by Phillips and Jahanshahi (1985) in reflecting a tendency to expect a poor outcome. The later version of the model (Vlaeyen & Linton, 2000) contains several revisions to the earlier form (Vlaeyen, Kole-Snijders, Boeren, et al., 1995). One change is the inclusion of negative affectivity and threatening illness information as contributing to pain catastrophizing.  Catastrophizing, Fear  DISUSE DEPRESSION DISABILITY  RECOVERY  INJURY  AVOIDANCE  tFEAR OF  21  CONFRONTATION  PAIN EXPERIENCE  MOVEMENT/REINJURY  PAINCATASTROPHIZING  NO-FEAR  Figure 1 Fear-avoidance model (Vlaeyen Kole-Snijders, Boeren, & van Eek, 1995). Reprinted from Pain, 62, The role of fear of movement/(re)injury in chronic low-back pain and its relation to behavioural performance, Pages 363-372, Copyright (1995), with permissionfromElsevier.  DISUSE DEPRESSION DISABILITY  INJURY  RECOVERY  PAIN EXPERIENCE  CONFRONTATION  AVOIDANCE HYPERVIGILENCE  t  PAIN RELATED FEAR  PAINCATASTROPHIZING  NO-FEAR  T NEGATIVE AFFECTIVITY THREATENING ILLNESS INFORMATION  Figure 2 Revised fear-avoidance model (Vlaeyen & Linton, 2000). Reprinted from Pain, 85, Fearavoidance and its consequences in chronic musculoskeletal pain: A state of the art. Pages 317-332, Copyright 2000, with permission from International Association for the Study of Pain.  Catastrophizing, Fear  22  Negative affectivity reflects a relatively stable tendency to' experience negative affect such as worry, anxiety and self-criticism (Clark & Watson, 1991). The hypervigilence to threat which characterizes negative affectivity (Watson & Pennebaker, 1989) increases the likelihood that an individual will develop a fear response when confronted with pain. Highlighting the complexity of the relationships between these constructs, other writers have considered fear of pain and fear of catastrophizing as lower order constructs, both reflecting the higher order construct of negative affectivity (Keogh & Asmundson, in press). In another change from the earlier model, Vlaeyen and Linton (2000) use the term "pain related fear" where the previous model (Vlaeyen, Kole-Snijders, Boeren, et a l , 1995) referred specifically to "fear of movement/re-injury". The term "pain related fear" encompasses fear of pain and fear of activities expected to cause pain, in addition to fear of movement/(re)injury (Vlaeyen & Linton, 2000). These different descriptors were derived from the instruments, which were used to measure pain related fear in the studies Vlaeyen and colleagues reviewed in developing their model. For example the original model was based on work using the T S K (Vlaeyen, KoleSnijders, Rotteveel, et al., 1995), which was designed to measure fear of movement/(re)injury, while the latter model was based on work which also used other measures o f pain related fear which apparently assess different types of fear, the F A B Q (Waddell et al., 1993) which taps into fear or anxiety of activity and work, and the PASS, which was designed to measure fear/anxiety about pain itself. The model was apparently broadened in order to accommodate and assimilate the broader research using instruments other than the TSK. While the earlier version of the model (Vlaeyen Kole-Snijders, Boeren, et al., 1995) focused exclusively on avoidance as the mediating factor between fear and negative outcome, Vlaeyen & Linton (2000) extended the model to include hypervigilence as another mediating variable. Increased attention to pain might exacerbate the aversiveness of pain itself (McCracken, Faber, & Janeck, 1998), or it might lead to avoidance by increasing anticipation of pain. Evidence suggests that both processes may occur.  Catastrophizing, Fear  23  In has been hypothesized that pain related fear might directly exacerbate pain symptoms through increasing general autonomic arousal (McCracken et al., 1998). However, the effect of fear and anxiety on pain experience appears to be mediated by attention (Arntz et al., 1991). Arntz, Dreesen, & Merckelback demonstrated that anxiety did not directly increase subjective perceptions and physiological responses during a painful dental procedure, but attention to pain did. Waddell et al. (1993) found that pain related fear correlated significantly with a tendency to focus on somatic cues. Further evidence that pain related fear may cause individuals to divert attention toward nociceptive stimuli, comes from work utilizing laboratory attention tasks (e.g., Crombez, Eccleston, Bayenns, Van Houdenove, & Van Den Brock, 1999; Crombez, Eccleston, Bayens & Eelen, 1998). Crombez, Eccleston et al. (1999) found that in a sample of individuals with chronic pain, levels of pain related fear interacted with pain levels to predict the degree of attentional interference in a numerical task. Similarly, Crombez et al. (1998) found that catastrophic thinking in the presence of a pain stimulus increased interference with a tone discrimination task. There is some evidence that the difficulty attending to such tasks results from a shift of attention toward the threatening pain stimulus, comes from Asmundson and Taylor (1996). They demonstrated that individuals with chronic pain who were high in anxiety sensitivity, had greater difficulty shifting attention away from pain related stimuli, than were those who were lower in anxiety sensitivity. The role of autonomic arousal, hypervigilence and attention to pain clearly needs further elucidation (Asmundson, Norton & Vlaeyen, in press), but those relationships were not directly addressed in the current work, where the focus was on the effects of pain related fears and catastrophizing via their impact on avoidance. Fear appears to increase expectancies of pain as well (Arntz, Van Eck, & Heijmans, 1990), possibly through increased arousal (Crombez et al., 1998) and the anticipation of pain may be important in motivating avoidance than pain experience (e.g., Phillips, 1995). Arntz et al. (1990) found that anticipatory anxiety increased the amount of anxiety experienced by dental patients during a dental procedure but did not predict the amount of pain experienced. They also found that experienced anxiety led to greater anticipated pain on subsequent procedures.  Catastrophizing, Fear  24  McCracken, Gross, Sorg, & Edmands (1993) likewise found that pain related fear was related to higher ratings of anxiety but not to higher pain report during a leg raise task in chronic pain patients. Mirroring Arntz et al.'s, results, McCracken et al. found that over a series of trials, high anxiety patients were more likely to over predict pain levels on subsequent trials. Crombez, Vervaet, Bayens, Lysens and Eelen (1996), found that higher pain anticipation was associated with greater pain related fear and predicted reduced effort on a leg raise task, which is often considered a proxy for avoidance. Interestingly, Crombez et al. (1996) found that pain overexpectancies were relatively quickly corrected with experience, but that this effect did not tend to generalize completely to new tasks. A study by Sullivan, Rodgers, & Kirsch (2001), suggests a similar effect for pain catastrophizing. They found that in an undergraduate sample, pain expectancies mediated the relationship between pain catastrophizing and pain experience. While it is unclear exactly what the causal relationships between pain related fear, affective distress and pain, there is considerable evidence that pain related fear is associated with affective distress. Pain related fear has been shown to correlate highly with measures of general anxiety and depressed mood (e.g. McCracken et al., 1992; Crombez, Vlaeyen, et al., 1999; Sullivan & D'Eon, 1990). Pain related fear (PASS; McCracken, Zayfert, & Gross, 1992) has also been demonstrated to correlate with a measure of non-specific physical symptoms usually associated with emotional distress, even after controlling for depression and pain level (McCracken etal., 1998). Further elaborations of Vlaeyen's (Vlaeyen, Kole-Snijders, Boeren, et al., 1995; Vlaeyen & Linton, 2000) model have been recently been proposed by Asmundson et al. (in press). The former expands the concept of pain related fear as a tripartite entity consisting of behavioral (avoidance), cognitive and physiological components, which are reciprocally related. They add pain appraisals to the pain experience portion of Vlaeyen's model and add anxiety sensitivity to the negative affectivity and threatening illness information segment. The most central amendment is to suggest that physiological arousal directly contributes to pain appraisals and thereby pain experience, and also to catastrophizing (Norton & Asmundson, 2003).  Catastrophizing, Fear  25  More recent work by Asmundson et al. (in press), elaborates the model even further to differentiate between fear and anxiety and propose that each will have different behavioral consequences, escape for fear and avoidance for anxiety. They argue that such a model is helpful in accounting for pain that is incongruous with actual injury by including a feedback loop between fear of pain and pain perception, and for pain in the absence of identifiable injury or pain that persists after injury has apparently healed. This differentiation is helpful from a theoretical perspective and is addressed in a subsequent section of this review titled Anxiety versus fear. However, as Asmundson et al. (in press) acknowledge the differentiation between fear and anxiety is a subtle one, particularly in the context of a third, and often overlapping condition, pain and that existing measures and theoretical work largely confound the two states. While there has been some evolution toward a more elaborate but clearly delineated model of pain related fear and avoidance, the basic premise remains the same over the various models, namely that individuals who are fearful and/or anxious about pain and/or activities expected to cause pain will be more likely to avoid activity, and this avoidance will prolong their recovery and exacerbate their fears. As evidenced by the publication of a recent book (Asmundson, et al. in press) entirely dedicated to the topic, interest in the role of pain related fear in chronic pain has gained considerable momentum over the last decade. Recent reviews (e.g., Asmundson, Norton, & Norton, 1999; Keogh & Asmundson, in press; Vlaeyen and Linton, 2000) provide substantial evidence for many aspects of Vlaeyen and Linton's model. However, a number of issues remain unresolved (Vlaeyen & Linton, 2000). From a clinical perspective, one of the biggest promises inherent in the theory of pain related fear, is the potential for the early identification of individuals with acute or subacute pain who are at risk of developing chronic pain (e.g., Linton, & Boersma, in press). If prediction and prevention of chronicity is the goal, it will be important to study the phenomenon of pain related fear in populations who have not yet become chronic but are at risk. Another, related issue, if prediction and prevention are goals, is the choice of measures. As indicated by the expansion of the fear component in the revised version of  Catastrophizing, Fear  26  Vlaeyen's model (Vlaeyen & Linton, 2000), there are a variety of different measures for pain related fear, each derived from a slightly different conceptualization of the phenomenon. The following sections will review some of the evidence for the causal aspects of Vlaeyen's model, particularly focusing on pain related fear and avoidance, and discuss the most common ways in which pain related fear is conceptualized. Implications for the current work will be provided in each section. Empirical Evidence for the Role of Pain Related Fear in the Development of Chronic Pain A number of instruments have been used to examine the role of pain related fear in chronic pain, but three instruments specifically designed to measure pain related fear appear most frequently in the current literature. Those are the PASS (McCracken, Zayfert & Gross, 1992) which was devised to measure fear of pain, The F A B Q (Waddell et al., 1993), which was devised to measure fearful beliefs about the effects of work and more general activities, and the TSK (Vlaeyen, Kole-Snijders, Rotteveel, et al., 1995), intended to measure fear of exercise/(re)injury. Each has been used in a number of contexts to test various components of the general fear/avoidance model of chronic pain (Asmundson et al., 1999; Vlaeyen & Linton, 2000). There are also a number of instruments designed to measure other constructs which Vlaeyen's model suggests to be associated with pain related fear (e.g., Anxiety Sensitivity; ASI, Peterson & Reiss, 1992; PCS, Sullivan, et al., 1995). The current work is focused primarily on the central concepts in the Vlaeyen model, pain related fear, avoidance and catastrophizing and therefore the following review will focus on those constructs and the primary measure used to assess them. The instruments utilized in particular studies will be identified, but a more detailed discussion of the differentiating characteristics of the instruments and what they measure is left until the subsequent section. There is considerable evidence that pain related fear does predict avoidance behavior at least in chronic pain patients. In experimental studies of pain, reduced effort expended in a pain provoking task can be viewed as a proxy measure for pain avoidance (e.g. Burns, et al., 2002; Crombez, Vlaeyen, et al., 1999). The inherent assumption is that poor performance reflects avoidance of painful stimuli. A number of authors have found pain related fear to affect effort or performance on physical tasks by chronic pain patients. Burns et al. (2002) found that pain  Catastrophizing, Fear  27  related fear as measured with the PASS predicted performance on a physical capacity examination. Similarly, Vlaeyen, Kole-Snijders, Rotteveel, et al. (1995) found a negative association between T S K scores and performance on a static-lifting task. Utilizing yet another modality, a trunk extension-flexion task, Crombez, Vlaeyen et al. (1999) found that pain related fear as measured with the T S K was the best predictor of performance even after controlling for the effects of pain intensity. Geisser, Haig, & Theisen (2000), similarly found that pain related fear scores (activity avoidance items from the TSK) predicted the degree to which individuals with disabling pain performed up to their maximum estimated capacity in a lifting task, controlling for demographic variables, depressive symptoms and physiological exertion (maximal heart rate). It has been questioned the extent to which low performance on physical tasks actually represents avoidance behaviour, and other factors such as motivation (Burns et al. 2002), or physical capacity may play some role. In a more direct test of the association between pain related fear and avoidance, Bussman, van de Laar, Neleman, and Stam (1998) found associations between pain related fear and active postures during a typical day, in chronic pain patients, as measured with an automated activity monitor. Reanalysis of the Bussman et al. (1998) data by Vlaeyen & Linton (2000) revealed that fearful patients avoided standing during the day, as compared to patients with lower levels of pain related fear. Perhaps the most immediate subjective consequence of fear motivated avoidance behavior would be an increase in self-perceived disability, a sense that one is unable to perform normal daily activities. A number of cross sectional studies have found support for the relationship between pain related fear and self-reported disability in chronic pain patients. Waddell et al. (1993) demonstrated that F A B Q scores were more closely related to self-rated disability and work loss in the preceding year than were biomedical variables or pain severity. Vlaeyen, Kole-Snijders, Rotteveel, et al. (1995) similarly found that TSK scores were a better predictor of self rated disability than were biomedical findings or pain intensity levels. McCracken et al. (1996) found that self rated disability was more closely associated with Pain related fear (PASS; McCracken, Zayfert, & Gross, 1992 & F A B Q ) than with general anxiety as  Catastrophizing, Fear  28  measured by the State section of the State-Trait Anxiety Inventory (Spielberger, 1983). Crombez, Vlaeyen et al. (1999) reported that the scores on the F A B Q , PASS and TSK all predicted concurrent self-reported disability in chronic back pain patients. Likewise, the PCS has been demonstrated to predict perceived disability after controlling for pain intensity. A prospective study of the effects of fear on disability in chronic pain patients did not support the causal effect of pain related fear (Mullen, 1998). Mullen conducted a treatment outcome study to test the mediational effects of pain related fear (PASS; McCracken, Zayfert, & Gross, 1992) on a work hardening program. Work hardening programs are based upon operant learning principles and focus on teaching the individual that hurt (pain) does not equal harm (Fordyce, 1995). While not specifically designed for the purpose, such programs utilize a graduate return to activity, which might be conceptualized as exposure treatment, and therefore helpful in reducing fear. Mullen found that initial PASS scores or change in PASS scores did not predict improvement in functional capacity or self-rated disability after such a treatment program. There are methodological concerns with this study including a small sample size (n = 43) and a very high rate of attrition (> 60%), with those who failed to complete the protocol scoring higher on initial pain severity.- While the author does not report a comparison of PASS scores for dropouts and those who completed the study, the mean initial PASS score of those retained (67.17) was closer to that of a non clinical population (e.g., Osman, et al., 1994; M = 65.04) than to that of a chronic pain sample (e.g., McCracken et a l , 1992; M = 94.24). This suggests that either Mullen's original sample was less fearful than typical chronic back pain samples or that only less fearful individuals remained at follow-up. In either case it does not appear that the sample used in the analyses was representative of the general population of chronic back pain patients and it is possible that this factor attenuated the effects of pain related fear. Together, these results show strong support for some aspects of fear based models of chronic pain, namely that fear is associated with greater avoidance behavior in the context of pain and is related to self-reported disability, in chronic pain patients. It would also appear that  Catastrophizing, Fear  29  fear might directly affect pain perception among those who have chronic pain and those who do not, through increased attention to pain signals, or possibly through increased autonomic arousal. However, longitudinal evidence for the role of pain related fear in the persistence of pain is lacking as is evidence for the effects of pain related fear in the earlier stages of pain. Only a subset of patients who suffer acute and subacute pain will go on to suffer chronic pain and disability. Pain and disability change over time and different factors may predict their persistence at different points in time (Krause, Dasinger, Deegan, Rudolph & Brand, 2001). Similarly, the relationships between various predictors may change over time. Ceiling phenomenon may come into effect over time as fear, avoidance and pain rise to their maximal levels, obscuring any causal relationships. It is also possible that anxiety and fear merge with, or transform into more complex perceptions of disability over time (Schultz et al., 2004) and thus become less amenable to treatment. Therefore, it is important to study the impact of pain related fear and catastrophizing early after pain onset. A limited number of studies have addressed the role of pain related fear in individuals with more recent onset of pain. In one study directly assessing the role of pain related fear in the transition to chronic pain status, Linton & Hallden (1998) constructed a brief questionnaire of items reflecting various constructs thought to play a role in that transition. They included three items intended to address pain related fear. Two of the items were selected from the F A B Q ("Physical activity makes my pain worse" & "I should not do my normal work with my present pain") and one item was chosen from another instrument ("An increase in pain is a signal that I should stop what I am doing until the pain decreases"). Discriminate function analyses revealed that fear-avoidance beliefs significantly contributed to the prediction of pain and self reported impairment of activity as measured six months later. The fear-avoidance beliefs also contributed significantly to the prediction of sick days over the six month period. However, although these authors labeled their initial sample as acute and subacute, it contained individuals with up to 4 months pain duration at baseline, which, according to current guidelines from the International Association for the Study of Pain (IASP; Merskey & Bogduk, 1994), would make at least some  Catastrophizing, Fear  30  of them chronic pain patients. Therefore, this study was not a clear demonstration of the effects of fear avoidance beliefs at the acute or subacute stage.  A n additional limitation was the use of  only three items to assess pain related fear. The reliability of findings based on such a limited measure of pain related fear is questionable. Klenerman et al. (1995) also provided evidence suggesting a causal role for pain related fear in the development of chronic pain. They assessed individuals presenting with a first episode of "benign, musculoskeletal low-back pain" on four variables hypothesized by Letham et al. (1983) to reflect components of the fear-avoidance model, including stressful life events, personality, personal pain history and pain coping strategies. Klenerman et al. (1995) found that these variables, measured at the acute stage of pain (within one week of pain onset) predicted self-ratings of pain severity, disability, and days of sick leave at two months post injury, and again at one year post-injury. However, Klenerman et al. (1995) did not directly assess pain related fear. The assumption that the variables chosen addressed pain related fear and avoidance seems somewhat tenuous. As stated earlier, Letham's model was not clear as to the role of fear in the relationship between the contextual variables and chronic pain, but it appeared to imply a mediational role. Klenerman et al. (1995) assumed this relationship and therefore argued that their results provided evidence for the effect of pain related fear on the development of chronic pain. However, given that no direct measure of pain related fear was included in the study, it does not seem appropriate to assume such a relationship exists. It is quite possible to derive pathways not involving fear or avoidance in the demonstrated effects. The variables hypothesized to operate through fear avoidance included a broad range of constructs, which may or may not be mediated by fear and avoidance. For example coping strategies may affect chronic pain through their effect directly on self efficacy and/or depression, as opposed to fear. A final limitation to this study was the high drop-out rate, with only 41% of subjects being assessed at all three points. Burton et al. (1995) conducted another follow-up study with a similar sample, but had a much lower drop out rate (26%). Their subjects had pain of varying duration ranging from less  Catastrophizing, Fear  31  than three weeks to several years. The authors split the sample into three groups based on duration of current pain complaint, acute (defined as < 3 weeks), sub-chronic (defined as > 3 weeks & < 52 weeks) and chronic (defined as > 52 weeks). Burton et al. (1995) found that fear avoidance beliefs were not predictive of self-rated disability for any of the patient groups. However, among the acute patients, catastrophizing as measured with the Coping Strategies Questionnaire (Rosensteil & Keefe, 1983), was the strongest predictor of one year outcome, accounting for a full 47% of the variance in self rated disability. It may be that the effect of fear was obscured by the variables included and the analysis chosen in this study. Burton et al. (1995) were primarily interested in comparing the predictive power of psychosocial variables versus medical diagnostic variables and therefore their analysis did not provide the best test of the effect of pain related fear. They utilized stepwise multiple regression, which is useful for achieving maximally predictive models but is less valuable in identifying the importance of individual variables. (Tabachnick & Fidell, 2001). It is possible that fear did not load significantly because it shared variance with other variables, which had slightly higher correlations with the outcome variable and therefore entered the equation first. For example, somatic perceptions and coping strategies were both included in the final model. Both of these variables were conceptualized as indicators of fear-avoidance in the Klenerman et al. (1995) study where they were also shown to predict outcome. Therefore, it is quite possible that these variables would share a considerable amount of predicted variance in the prediction of outcome with fear. Burton et al. (1995) do not report the correlations between the independent variables so it is impossible to quantify the degree of shared variance. The effects of fear may also have been obscured by high correlation with catastrophizing, although again, without access to first order correlations, it is not possible to assess this possibility. Burton et al. (1995) used only the activity avoidance scale of the F A B Q , because the other subscale, work beliefs, pertains only to individuals who have been working, and some members of their sample were unemployed prior to onset of pain. This choice may also have  Catastrophizing, Fear  32  obscured the effects of pain related fear as the F A B Q work subscale has been found to correlate more highly with disability than does the work subscale (e.g., Waddell et al., 1993). Burton, Waddell, Tillotson, & Summerton (1999) conducted a treatment trial which also bears upon the question of whether pain related fear affects the persistence of acute pain. They created an experimental educational booklet, specifically designed to change fearful beliefs and behavior. The sample included patients presenting for primary care for a new episode of back pain, which had started in the last three months. The dropout rate was quite low (22% at oneyear follow-up), but as in the Burton et al. (1995) study, only the F A B Q activity scale was used because some patients were not employed before the onset of pain. Again this sample mixed subjects from various stages of pain and disability. Based on relative risk ratios, Burton et al. (1999) found that patients receiving the experimental booklet were more likely to have clinically significant improvement in the F A B Q at 2 weeks, 3 months, and 1 year than patients receiving the control booklet were. Clinically important changes in F A B Q scores at two weeks and 3 months were significantly related with clinically important improvement in self-rated disability at 3 months. Those patients with initially high F A B Q who received the experimental booklet were more likely to show improvement in self perceived disability at two week follow-up than those who received the control booklet. This may reflect a mediational effect of fear reduction on effectiveness of the booklet, but the mediational hypothesis was not directly tested. Another study (Fritz, et a l , 2001) examined the effect of pain related fear (FABQ) at the acute stage (defined in this case as less than 3 weeks duration) of work-related low-back pain on disability and work status following a four week course of physiotherapy treatment. They found that neither subscale of the F A B Q predicted concurrent disability but the work subscale predicted four week disability and work status even after controlling for initial pain intensity, physical impairment and disability and type of therapy received (all had one of two types of physiotherapy).  Catastrophizing, Fear  33  Pain related fear and catastrophizing are only two of a number of biopsychosocial factors implicated in the development of chronic pain and long-term disability and it is important to place fear and catastrophizing in the context of those other predictors. There are numerous studies examining broader classes of predictors, but methodological problems complicate interpretation of that work (Turk, 1997; Crook, Milner, Schultz & Stringer, 2002). Turk (1997) in his review of research on demographic and psychosocial predictors, concluded that very few variables have been demonstrated to consistently predict which individuals suffering acute or subacute pain will go on to become chronic, and that interpretation of research in this area is hampered by several methodological inconsistencies within and across studies. Likewise, Crook, Milner, Schultz and Stringer (2002) highlighted methodological problems with the existing literature, including the lack of a theoretical direction, variability across studies in methodology, and background systemic issues such as differences in health care and compensation systems. Crook et al. concluded that only 19 of 2170 studies examining prediction of low-back disability screened in their review met the criteria for methodological adequacy. Both of these reviews suggest that psychosocial variables related to catastrophizing and pain related fear play a role in the development of chronic pain and disability. In his review of this literature, Turk (1997) found support for the role of several cognitive variables in the development of chronic pain. The majority of these variables appear closely related to catastrophizing and/or fear. Heightened somatic concerns (Klenerman et al., 1995), the conviction that one has an incurable disease (Cats-Baril & Frymoyer, 1991; Dworkin & Whitney., 1992; Hazard, Haugh, Reid, Preble, & McDonell, 1996), general perceptions of poor health (Biering-Sorenson, 1983; Deyo & Diehl, 1988; Linton et al., 1994; Linton & Buer, 1995), and perceived severity of illness and disability (Tait, Chibnall & Richards, 1988; Lacroix et al., 1990; Fuerstein & Thebarge, 1991; Hazard et al., 1996) have all been implicated in the development of chronic pain and each of these variables appears to capture some aspect of catastrophic thinking. Somatic concerns represent a tendency to be overly concerned with physical symptoms and their meaning as does disease conviction. Both perceived illness severity  Catastrophizing, Fear  34  and disability are hypothesized to result from fear (Letham et al., 1983; Vlaeyen, Kole-Snijders, Rotteveel et al., 1995). Turk (1997) pointed out in his review that much of the work on psychosocial predictors has utilized different measures and that frequently the choice of measures is not based on theory, and that both of these factors complicate comparison and interpretation of the results. He also highlighted the tendency of studies to rely on narrow measures of outcome, which, given the multidimensional nature of chronic pain, is problematic. A broader and more recent, systematic review focusing specifically on occupational disability, (Crook, Milner, Schultz & Stringer, 2002) reported that biopsychosocial factors across a range of domains have been demonstrated to be predictive of outcome. These authors considered studies examining a broad range of predictors, including psychosocial, medical, pain related, and occupational variables measured in the first six months following onset of back pain to predict outcome at one year or more post onset. They divided outcomes into five categories, time off work, recurrence of pain, return to work by a specified point in time, health care costs and the persistence of pain and disability. Crook et al. (2002) found support for the role of some medical/physical variables. Clinical indicators such as no pain, no sprains, good back flexion, straight leg raise, radiating pain, number of x-rays, and greater mobility, arthritis and widespread pain, were all implicated in one or more aspects of outcome. Medical history variables including previous hospitalization and previous episodes of back pain were also predictive. Demographic variables including age, sex, and number of children were also found to be prognostic. Workplace variables identified as important in predicting aspects of disability included availability of a modified job, control over work, problems with the job or coworkers, and worker's compensation status. Crook et al. (2002) also found support for the role of psychological variables including depression, locus of control and pain related fear in predicting outcome. The wide variety of variables indicated as potentially important in this review suggests that i f the highest degree of prediction is desired, it is important to consider a range of potential classes of biopsychosocial  Catastrophizing, Fear  35  variables. Indeed, in a recent attempt to identify the best predictive model of occupational disability among a group of Worker's Compensation Claimants, Schultz et al, (2002) utilizing stepwise regression procedures, found that the best model for predicting return to work included physical/medical, psychosocial, pain behaviour-guarding, and workplace factors. Psychosocial factors included in the final model included perceptions of current health and physical status and expectations of recovery. Both of these variables represent cognitions that may reflect fear and/or catastrophizing. The preceding work indicates pain related fear and catastrophizing and related cognitive and emotional distress variables are likely important in the prediction of pain and disability, it also suggests that a number of other biopsychosocial factors also play a role in the persistence of pain and disability. It also important to acknowledge that the role of pain related fear and catastrophizing may change with the progression of pain and disability as other factors in the biological, psychological and social domains change and evolve. The current work was focused specifically on assessing the construct validity of pain related fear and catastrophizing and as part of that undertaking whether those variables predicted outcome rather than on maximizing the predictive power of an actuarial model Given the broad range of predictors that have been implicated in previous work (e.g., Crook et a l , 2002) and the lack of an encompassing theoretical framework for understanding the relationship between those broader classes of variables, it was considered important to retain a focus on pain related fear and catastrophizing. Until the relationships among pain related fear and catastrophizing variables are clarified, as was the goal of the current study, it was considered useful to restrict the predictive analyses to those variables. Once those relationships are more clearly understood, it will be important to integrate them into a broader biopsychosocial framework in order to gain a better understanding of their relative importance and relationship to the many other relevant factors, such as those in the workplace and systemic domains. It is expected that pain related fear and catastrophizing might be closely related to and/or interact with a number of these other constructs.  Catastrophizing, Fear  36  The Role of Catastrophizing in the Fear Avoidance Model of Chronic Pain The role of catastrophizing as measured with the CSQ (Rosensteil, & Keefe, 1983) and the PCS (Sullivan, Bishop & Pivik, 1995) in chronic pain is well demonstrated. There is clinical and experimental evidence that catastrophizing exacerbates the pain experience (Rosensteil & Keefe, 1983; Sullivan et a l , 1995; Sullivan, Stanish, Waite, Sullivan & Tripp, 1998) and contributes to disability (Martin et. a l , 1996; Sullivan et al. 1998) in chronic pain and non patient samples. Sullivan et al. (1998) demonstrated that catastrophizing predicted disability after controlling for pain intensity, depression and anxiety in chronic pain patients. In a prospective study, Linton, Buer, Vlaeyen and Hellsing (2000), found that both pain related fear (FABQ) and pain catastrophizing predicted the likelihood of back pain in the general population. Burton et al. (1995) similarly found that the level of catastrophizing in acute pain patients predicted their selfreported disability one year later. In non-clinical samples, catastrophizing has been demonstrated to predict the level of pain and emotional distress reported by participants in laboratory pain studies (Sullivan et al., 1995). There is however, relatively little evidence speaking to the causal relationship between pain related fear and catastrophizing proposed by Vlaeyen & Linton (2000) and theoretical alternatives have been proposed. While Vlaeyen et al., separate the constructs and propose that catastrophizing causally precedes fear of pain, others have argued (Keogh & Asmundson, in press), that the two constructs represent lower order factors in a hierarchy of negative emotional states (i.e. negative emotionality, trait Anxiety). The inclusion of a subscale consisting of items which appear to reflect catastrophic thinking (fearful appraisals) in the PASS implies that the authors of that scale (McCracken, Zayfert & Gross, 1992) viewed catastrophizing as an aspect of pain related fear constructs. Sullivan et al., 2001 proposed a number of theoretical mechanisms through which catastrophizing may affect pain, but in line with Vlaeyen et al., they propose that the fear/avoidance model provides a plausible account for the persistence of catastrophizing and that fear of pain may mediate the relationship between catastrophizing and disability. Pain catastrophizing has been demonstrated to correlate highly with pain related fear in both chronic pain (McCracken & Gross, 1993; McCracken, Zayfert & Gross, 1992; Crombez,  Catastrophizing, Fear  37  Vlaeyen, Heuts, & Lysens, 1999) and non-clinical samples (Sullivan et al., 1995) utilizing different measures of pain related fear and catastrophizing. There is also evidence that individuals who catastrophize, demonstrate pain related fear when threatened with intense pain in laboratory situations (Crombez, Eccleston, Van den Broeck, Van Houdenhove, & Goubert, 2002; Eccleston, Baeyens, & Eelen, 1998). Tentative support for the contention the effect of catastrophizing on disability is mediated by pain related fear can be found in the work of Linton et al. (2000). In their study predicting onset of back pain and reduced function in the general population, Linton et al. found that pain related fear (FABQ) was a stronger predictor than catastrophizing, suggesting that pain related fear may play a more direct role in pain and loss of function than does catastrophizing. Research by Vlaeyen, Kole-Snijders, Rotteveel et al. (1995) is also informative. They found that pain related fear (TSK scores) predicted self perceived disability in a group of chronic back pain patients after controlling for biomedical findings, pain intensity and catastrophizing, and that catastrophizing predicted pain related fear after controlling for biomedical findings and pain intensity. This implies additional support for the mediation of pain related fear in the effect of catastrophizing proposed by Vlaeyen and colleagues (Vlaeyen, Kole-Snijders, Boeren, et a l , 1995; Vlaeyen & Linton, 2000), but the analysis employed by (Vlaeyen, Kole-Snijders, Rotteveel, et al., 1995) did not specifically test the mediational hypothesis and so that support must be viewed as tentative. In addition, the evidence for these relationships is derived from chronic and non clinical samples, and have not been examined in individuals with pain of recent onset. Given the apparent importance of this relationship and the limited evidence available to date, Keogh and Asmundson's (in press) call for further research that "addresses the causal relationship between these variables is required." (pp. 100). In addition, given the potential importance of the early stages of pain and disability, it is important that these relationships be studied in individuals with recent onset pain. Measures of Pain Related Fear and Catastrophizing Several self-report scales tap into aspects of pain related fear. For example, the Harm subscale of the Survey of Pain Attitudes Scale (SOP A ; Jensen, Karoly, & Huger, 1987), and the Pain and Impairment Rating Scale (PAIRS; Riley, Ahem, & Follick, 1988), include some beliefs or appraisals, which appear to reflect fear. However, only recently have measures specifically  Catastrophizing, Fear  38  designed to tap into pain related fear been developed. The most commonly utilized instruments in the current literature are the PASS, F A B Q and TSK. Recently, Sullivan et al. (1995), developed the Pain Catastrophizing Scale (PCS), an instrument specifically designed to measure multiple dimensions of pain catastrophizing, which had been previously described in the literature. As stated earlier, the manner in which various instruments measure pain related fear speaks to the different ways in which pain related fear is conceptualized. The most basic assumptions about the nature of pain related fear as measured by each instrument are revealed in the factor structure of the instrument. The factor structure of an instrument speaks to the way in which items on the questionnaire cluster into specific conceptual groupings and those conceptual groupings suggest what the basic elements of pain related fear might be. The following sections will review research on the factor structures of each of the instruments most commonly used to measure pain related fear and catastrophizing. A subsequent section will discuss the theoretical implications of those factor structures for understanding the constituent aspects of pain related fear and how they might relate to one another. Pain Catastrophizing Scale. Until recently, pain catastrophizing was primarily measured with the catastrophizing subscale of the Coping Strategies Questionnaire (CSQ; Rosensteil & Keefe, 1983). However, this scale appears to measure only one of multiple aspects of pain catastrophizing, a tendency to experience helplessness and doubt about one's ability to cope with pain (Sullivan et al., 2001). In an attempt to provide a broader assessment of pain catastrophizing, Sullivan et al. (1995) developed the Pain Catastrophizing Scale (PCS). The PCS items (see Appendix A) were derived from previous work, including the CSQ catastrophizing scale (Sullivan et al., 1995) to measure a broader range of pain related catastrophic thinking. The scale is comprised of 13 items, which assess a general tendency to respond to pain with negative cognitions. There is a considerable body of research on the factor structure of the PCS.  Catastrophizing, Fear  39  A n initial exploratory factor analysis (EFA) of the 13 items administered to a college undergraduate sample yielded a three factor solution (Sullivan et al., 1995). The three factors were labeled helplessness, rumination and magnification. Rumination was described as a tendency to dwell on negative aspects of pain (e.g., item 10, "I keep thinking about how much it hurts"). Magnification was seen by the authors to reflect exaggeration of the threat value of the pain stimulus (e.g., item 13, "I wonder whether something serious may happen.") and the helplessness subscale items were seen to pertain to perceptions that one camiot cope with pain (e.g., item 11, "There's nothing I can do to reduce the intensity of the pain"). Sullivan et al., found the magnification subscale to have relatively low internal reliability (alpha = .60), which they argued might reflect the low number of items (3) on that scale. However, they also suggest that the low reliability may have reflected variability in the pain experiences recalled by their study participants, since this being a non-clinical sample, these individuals had to rely on memory for past experiences. The particular experiences recalled may therefore have varied. Another E F A conducted by Osman, et al. (1997) on data collected from a college student sample failed to support the three factor solution. Instead, utilizing the same procedure as Sullivan et al. (1995), Osman et al. found a two factor solution to be supported. One factor was comprised of six items mostly reflecting rumination, while the other factor consisted of seven items, reflecting helplessness and magnification. However, in a subsequent confirmatory factor analysis (CFA) reported in the same paper, they found that a three factor solution corresponding to the original scale construction provided significantly better fit than the two factor model. They subsequently tested the three factor model with one cross loading (helplessness item 5, "I feel I can't stand it anymore", also loading on helplessness), and found that the cross loading provided further significant improvement in the model fit. A second C F A by Osman et al. (2000) on data from a randomly selected community sample, also supported the three correlated factor model with modifications (correlated error terms for items nine and ten) over a single factor model and an orthogonal three factor model. It is unclear why the authors did not test the two factor model in this study. They also tested a second order factor model with the three first order factors loading on a higher order "catastrophizing" factor as opposed to having them simply correlate with one another as in the first order model. The fit of this model was very close to that of the three correlated factors  Catastrophizing, Fear  40  model, but the authors argue that it was superior based on relative parsimony (i.e., it accounted for the data with more degrees of freedom). However, examination of their results suggests this interpretation may be unwarranted. First, a higher order model with three first order factors does not statistically differ from a simple three factor model in the number of parameters estimated or the relative contribution to model fit. The only way to compare those two models would be to impose some extra constraint on the higher order model such as constraining two paths to be equal. While the difference in degrees of freedom between the two models (Adf. = 2) would suggest that such a constraint was imposed, the authors do not report on what that constraint might have been or the logic of imposing it. Sullivan, Tripp, Rodgers, and Stanish (2000) conducted another confirmatory factor analysis of the PCS in two samples, a group of sedentary individuals and a group of individuals who regularly engaged in sporting activities. In both samples a three factor model provided good fit to the data, but only after allowing for correlated error terms between items suggesting that some factor beyond the three specified was operative. Multi-group testing-further revealed that the factor structure was stable across the two samples. In the most recent factor analysis of the PCS, Van Damme, et al. (2002) tested a Dutch version of the scale. They found that the three factor model provided the best fit to the data in three separate samples, students, chronic low-back pain patients, and Fibromyalgia patients. However, it should be noted that the fit of the three factor model was marginal in the student sample and did not meet the author's stated criteria for adequate fit (CFI > .9; R M S E A < 0.08) in the other two samples. There has been some debate about the discriminant validity of pain catastrophizing (Sullivan et al., 2001). Pain catastrophizing is highly associated with depression and it has been argued that catastrophizing may represent the cognitive aspect of depression (Sullivan & D'Eon, 1990). However, in a study, comparing the PCS to measures of depression, state anxiety and fear of pain, Sullivan et al. (1995) found that PCS scores accounted for unique variance to the prediction of self reported pain levels beyond that accounted for by depressive symptoms and trait anxiety during a cold pressor task, indicating that catastrophizing can be considered a separate construct from more general aspects of negative emotion such as depression. Likewise,  Catastrophizing, Fear  41  Osman et al. (1997) demonstrated, using structural equation modeling (SEM), that the three PCS constructs represent a separate construct from depression and anxiety. While catastrophizing is generally viewed as a predisposing factor for poor response to pain, there is tentative evidence that, as predicted by the Vlaeyen model, pain or its persistence may trigger or exacerbate pain catastrophizing (Sullivan et al., 1998). Sullivan et al. found that scores on the PCS were higher in a sample of chronic pain patients than in a non-clinical sample. While the factor structure of the PCS is reasonably well demonstrated, that structure has not yet been tested in a subacute pain population. If the instrument is to be used in subacute pain populations to predict and/or prevent chronic pain, distress and disability, it will be useful to establish that the factor structure holds in that population. In addition, there has been limited research examining pain related fear and catastrophizing together and it is not clear if the relationship proposed by Vlaeyen et al holds (Vlaeyen et al., 1995, Vlaeyen & Linton, 2000). The Pain Anxiety Symptom Scale. The PASS (see Appendix A) is a forty-item scale devised to measure fear of pain (McCracken, Zayfert, & Gross, 1992). It was designed to measure four aspects of fear of pain. The authors assumed that fearful responses to pain could be divided into three domains, cognitive, physiological, and behavioral. The cognitive domain was, further divided into fearful appraisals and cognitive anxiety symptoms. Items were devised to assess each of these concepts and selected based on their psychometric properties. Each PASS item asks the respondent to rate the extent they experience particular symptoms in response to pain on a 5 point scale from "never" to "always". Cognitive anxiety items on the PASS include symptoms such as racing thoughts and disorientation. Somatic anxiety items include items assessing physiological arousal, while the learning responses include escape and avoidance behaviors such as avoidance of activity, taking medication or seeking reassurance. The fearful appraisal subscale was intended to measure "fearful thoughts related to the experience of pain or anticipated negative consequences of pain" (McCracken, Zayfert, & Gross, 1993; pp. 183). Three separate factor analytic studies have been conducted on the PASS. A study applying principal components analysis (McCracken, Gross, Hexum, & Semenchuk, 1993) to PASS scores from 180 patients with chronic pain indicated a four factor solution with factors  Catastrophizing, Fear  42  corresponding to the original scales and accounting for 49% of the variance. Interscale correlations were in the same range (0.49 to 0.76) as those found by McCracken, Zayfert, & Gross (1992), indicating that the scales are closely related but likely not redundant. Another examination of the PASS factor structure (Osman, et al., 1994) was conducted using a non-pain sample. Results of a confirmatory factor analysis supported the original 4 factor structure, with factors corresponding to the original scales. In addition, a one-factor solution was supported, suggesting the validity of the essential construct embedded in the instrument. However, Osman et al. found that three items were poor measures of their respective scales. Those items were, item 11 "go to emergency room" and item 31 "I do it even if it causes pain" which loaded poorly on avoidance (<.340) and item 16 "Even i f I do an activity which causes pain, I know it will decrease later" which loaded poorly on the fearful appraisal subscale. Osman et al., also found the PASS total scale and all of the subscales to have high internal consistency, comparable to the levels found by McCracken, Zayfert & Gross (1992). Also, similar to what was found by McCracken, Zayfert and Gross, the interscale correlations were moderately high (0.55 to 0.73), suggesting they are related but did not overlap in this sample. The preceding findings have been criticized by Larsen, Taylor and Asmundson (1997), because in the case of McCracken, Gross, et al. (1993), decision criteria used to decide the number of factors retained were not reported and in the case of Osman et al., confirmatory factor analyses may not have been appropriate at such an early stage in scale development. This latter point is debatable, since confirmatory factor analyses are generally considered appropriate i f there is theoretical reason to assert a particular factor structure and that procedure represents a more stringent test than exploratory factor analysis (Ullman, 2001). Larsen, et al. (1997) conducted an exploratory factor analysis of the PASS, utilizing principal components analysis with an oblique rotation. To determine the appropriate number of factors, they utilized a technique called parallel analysis (Horn, 1965). Their sample was 259 patients referred to a multidisciplinary treatment program. In contrast to McCracken, Zayfert and Gross (1992), Larsen, Taylor, & Asmundson (1997) found that a five-factor solution  Catastrophizing, Fear  43  accounting for 44% of the variance was indicated. The solution provided partial support for the factorial validity of the originally specified factor structure (McCracken, Gross, Hexum, et al., 1993), but there were some significant differences. The Larsen et al. (1997) Catastrophic Thoughts subscale contained most of the items from the original fearful appraisal subscale, the physiological anxiety subscale and the escape/avoidance subscales contained items from the corresponding original subscales of the same name and the Cognitive interference subscale contained some items form the original cognitive anxiety subscale, but other Cognitive Anxiety symptoms loaded on the Catastrophic thoughts scale or the Coping Strategies subscale. The Coping strategies subscale also included items from the original Escape/Avoidance subscale and one item from the original Fearful Appraisals subscale. The new scales demonstrated good to excellent internal consistency and showed the expected correlations with other pain related variables. Correlations between the derived factors tended to be smaller (ie.-0.22 to +0.27) than those found by McCracken, Zayfert and Gross. Interestingly, Larsen et al. (1997) retested the factor structure after covarying for medication use, on the assumption that it might alter the factor structure, but they found that the structure did not change significantly. Due to the deviation of their solution from the proposed factor structure, Larsen, et al. (1997) concluded that the PASS was in need of revision and further evaluation. Based on that conclusion, McWilliams and Asmundson (1999) subsequently developed a revised and expanded Pain Anxiety Symptom Scale (PASS-R). Mc Williams and Asmundson added a number of items intended to increase the stability and validity of the new subscales indicated in their previous research. Utilizing Principal Components Analysis with oblique rotation, they determined that the new scale with 72 items did in fact reflect a similar five factor structure. The new subscales demonstrated good to excellent internal reliability and the expected correlations with other pain related variables. However, since the item pool for this instrument deviates substantially from the original item set, and the latter was the focus of the current work, the McWilliams and Asmundson (1999) PASS revision was not further examined here.  Catastrophizing, Fear  44  McCracken and Dhingra (2002) have proposed another revision of the PASS. Because the original 40 item scale was considered too long for use in clinical settings, McCracken and Dhingra set out to create an abbreviated version of the PASS while retaining the psychometric properties of the original. They derived the 20 item abbreviated scale (PASS-20) by selecting the five items from each of the four subscales which showed the highest item-total correlations with their respective subscales and contributed reasonably to scale variability (they dropped two of the items with the highest correlations because they had low means and reduced scale variability). They achieved subscales with satisfactory reliability and variance (average alpha = 0.81, range 0.75 to 0.87). As might be expected, given the item selection criteria, the shortened subscales correlated highly with the corresponding original subscales (average r=0.95) and showed considerably lower correlation with the non-corresponding subscales of the full instrument (average r= 0.57). The abbreviated subscales also demonstrated a very similar pattern of correlations with outside variables including pain, depression, and disability. Internal consistencies for all subscales were high (mean alpha = 0.81) as was the item total consistency for the entire scale (alpha == 0.91). The latter was high enough for the authors to suggest the possibility of a single factor rather than the four factor structure but they did not report the interscale correlations for the abbreviated subscales. The authors concluded that the results of their work would need to be replicated in an independent sample, but that does not appear to have been done and to date there does not appear to be any reported factor analytic examination of the PASS-20 scale. There are indications that the PASS subscales assess more than one construct. Correlations among the PASS subscales have been found to be moderately high, ranging from 0.49 between the escape/avoidance subscale and the fearful appraisals subscale to 0.72 between the cognitive interference and fearful appraisals subscale, but low enough to suggest that they are somewhat independent (McCracken, Zayfert & Gross, 1992). Interestingly, given the apparent divergence in the concepts tapped by the subscales, most research to date has employed a total scale score (sum of subscale scores). The limited work assessing the subscales  Catastrophizing, Fear  45  independently indicates that the subscales show differential correlations with other pain related variables, such as depressive symptoms, pain severity, and use of medications (McCracken, Zayfert, & Gross, 1992, McCracken & Dhingra, 2002). The inclusion of the fearful appraisals subscale which appears closely related to the construct of pain catastrophizing, suggests that if Vlaeyen and colleagues (Vlaeyen, Kole-Snijders, Boeren, et al., 1995; Vlaeyen & Linton, 2000) contention that catastrophizing is separate and precedes pain related fear, holds, the PASS is actually assessing at least two separate constructs. Several questions remain unanswered with regard to the PASS. Perhaps most significant, is the debate about the most appropriate factor structure. Secondly, is the degree to which shortening the scale is warranted. As McCracken and Dhingra argue, a 20 item scale is certainly more efficient that one with 40 items, particularly for clinical purposes. However, as those authors also point out, the shortened scale requires cross validation on new samples. No factor analysis of the shortened scale has apparently been conducted to date and the instrument does not appear to have been used in an acute or subacute pain population. Similarly, the factor structure for the full instrument has not yet been examined in an acute or subacute pajn population, As with the PCS, if the PASS is to be used in for prediction and/or prevention at the early stages of pain complaint, it will be important to determine how it behaves in that population. The high degree of internal consistency of the total scale PASS score based on both the 40 item version (McCracken, Zayfert, & Gross, 1992) and 20 item version (McCracken & Dhingra, 2002) suggests that both versions might be better conceptualized as representing a single factor, but there is also reason to believe that the four factors may represent divergent but related constructs. One goal of the present work was to develop a model of relationships between pain related fear constructs. For the purposes of developing that model in subsequent sections of this review, the four-factor solution corresponding to the PASS original subscales was assumed. However, confirmatory factor analyses of the scale were conducted prior to the structural analysis and the results informed the final selection of PASS items to include in the structural model and how to assign them to latent variables in the model.  Catastrophizing, Fear  46  The Fear Avoidance Beliefs Questionnaire. The F A B Q (see Appendix A) has also been employed frequently in the study of pain related fear. As the name suggests, the F A B Q assesses beliefs about the appropriateness and consequences of various activities for the individual. In designing the F A B Q , Waddell and colleagues (1993) were interested in accounting for the weak correlation between injury severity and disability, which is commonly found in chronic pain. Like Phillips (1987), they proposed that beliefs, specifically fear/avoidance beliefs about work and activity, might account for the discrepancy between disability and injury severity. The authors of the F A B Q specifically drew upon the existing literature in the area of beliefs and coping strategies, which contained examples of appraisals and beliefs potentially related to fearavoidance. They argued that existing measures, (i.e., SOPA; Jensen et a l , 1987; CSQ; Rosensteil & Keefe, 1983, and PAIRS; Riley et al., 1988) were not specific enough to adequately explain low-back disability. In addition to pre-existing instruments, Waddell and colleagues also drew upon the concept of disease conviction (Pillowsky & Spence, 1975; 1983), which includes beliefs about the seriousness of the illness and its effect on the patient's life, and the concept of somatic focus and increased somatic awareness discussed by Main (1983). Items on the F A B Q are rated on a seven point Likert scale from "strongly disagree to strongly agree". Beginning with an initial 16 items Waddell et al. (1993) selected 11 items, based on psychometric characteristics. They found the final 11 item F A B Q to have adequate test-retest reliability over a period of two days Principal components analysis indicated two factors, one involving beliefs about the relationship between pain and work and the second beliefs about physical activity in general (Waddell et al, 1993). Despite their derivation through orthogonal rotation, the two subscales correlated moderately and a single factor solution showed high internal reliability suggesting the two subscales tap a single construct. Crombez, Vlaeyen, et al. (1999) further supported that contention, finding the two subscales to correlate highly. However, the two subscales have been demonstrated to have discriminant validity (e.g., Waddell et al., 1993; Fritz, et al., 2001).  Catastrophizing, Fear  47  The constituent subscales of the F A B Q have been examined independently (e.g., Waddell et al., 1993; Fritz, et al., 2001). Interestingly, in their initial work Waddell et al. (1993) found that the activity subscale of the F A B Q was more highly associated with the tendency to focus on somatic cues than was the work subscale, but the work scale was a stronger predictor of concurrent depression, disability, retrospective time off work and current work status, than was the activities scale in their chronic pain sample. Subsequent work (Fritz, et al.) similarly found the work subscale to be more highly predictive of return to work and disability in a prospective study among those with more recent onset pain. Other work has demonstrated the F A B Q activity subscale to be more predictive of performance on physical tasks (Crombez, Vlaeyen et al., 1999). Although the F A B Q is the most commonly used instrument to date in prospective studies and has demonstrated significant promise in that regard (Fritz, et al., 2001, Burton et al., 1999, Burton et al., 1995; Linton et al., 2002) there has been very limited examination of its psychometric characteristics. The only apparent examination of the factor structure appears to be the initial exploratory analysis conducted by Waddell, et al. (1993). Despite its use in recent onset pain samples for the purpose of predicting outcome, the factor structure of the F A B Q has never been examined in an acute or subacute pain sample. In addition, despite Waddell et al. having published the full 16 item set from which they derived the final 11 items, there has been no examination of the full item set, beyond that conducted by Waddell et al. themselves. Tampa Scale for Kinesiophobia. A final measure, which appears to overlap conceptually with the F A B Q , the PCS and the PASS, is the TSK. Kinesiophobia has been defined as "the fear of movement and physical activity that is (wrongfully) assumed to cause (re)injury." (Vlaeyen, Kole-Snijders, Rotteveel, et a l , 1995, pp. 364). Each of the 17 items constituting the TSK is rated on a four point Likert scale ranging from one (strongly disagree) to four (strongly agree). It would appear the majority of work examining the factor structure of the T S K is based on a Dutch (TSK-DV) translation developed by Vlaeyen, Kole-Snijders, Boeren, et al. (1995). Those authors, utilizing E F A (principal components analysis), found a four correlated factors model to  Catastrophizing, Fear  48  best fit the data. They dropped five items from this four-factor solution (items 5, 7, 8, 16 & 17) because they failed to load above .40 on any factor, and they did not report the loadings for these items. Vlaeyen and colleagues labeled the four factors, harm, fear of reinjury, importance of exercise, and avoidance of activity. However, despite relatively low interscale correlations (.02 to -.31) the authors argued that the scales were not totally independent and because the total score had higher internal consistency (Cronbachs alpha = 0.77) than the individual subscales (0.53 to 0.71), use of the total score has been recommended (Vlaeyen & Linton, 2000). Most of the subsequent work employing the T S K and/or examining its factorial structure also utilized the Dutch translation. One study using the English version of the scale (Clark, Kori, & Brockel, 1996) used E F A on data collected from male veterans admitted to a chronic pain program. Item analysis revealed that all four of the reverse scored items (items 4, 8, 12 and 16) were weakly correlated with the TSK total score and they were excluded from further analysis on that basis. It should be noted that only two of these items were in the group dropped by Vlaeyen Kole-Snijder, Boeren, et al. (1995). Principal components analysis with oblique rotation indicated a two factor solution for the remaining 13 items. The two factors were labeled "activity avoidance" (the belief that activity will result in injury or worsen pain), and "pathological somatic focus", a belief that pain reflects an underlying and serious medical problem. Geisser et al. (2000) found a very similar solution to hold in a chronic pain population using the same statistical procedure as Clark et al. (1996). Twelve of 13 items loaded on the same factor as that indicated by Clark et al. (1996). The one item which did not load as expected, showed high loadings on both scales and the authors therefore argued that the Clark, et al. structure should be retained. Goubert et al. (2004) attempted to replicate these results utilizing confirmatory factor analysis. They compared the two factor structure of Clark, et al. (1996) and Geisser et al. (2000) to a single factor model including all items, the four factor model proposed by Vlaeyen, KoleSnijders, Boeren et al. (1995), and a single factor model with the reverse scored items deleted.  Catastrophizing, Fear  49  They argued that the two factor structure held and was invariant across two separate samples, a chronic pain sample and a Fibromyalgia sample. However, examination of their results suggests that this interpretation may have been somewhat optimistic. The fit indices for the two factor model as applied to the Fibromyalgia sample did meet the criteria stated apriori by the authors. However, when applied to the chronic pain sample, the fit of the two factor model by one of the reported indices (CFI= 0.086) was below the apriori stated threshold for that index, despite the fact that the threshold chosen by the authors for that index (0.09) was by some standards somewhat liberal (e.g., H u & Bentler (1999) recommend a cutoff off in the range of 0.95 or greater for the CFI). Goubert et al. (2004) also tested a second order model of the TSK, where the two first order factors were specified to load on a higher order "fear of movement/(re)injury factor. They reported adequate fit for that model, although it is not clear how they specified this model in a manner, which would permit such a conclusion. Statistical identification, a necessary condition for confirmatory factor analysis with S E M (Ullman, 2001) requires a minimum of at least three lower order factors in the case of one higher order factor (Byrne, 1994). Even with three first order factors the higher order model does not differ statistically from the first order model where the factors are permitted to correlate, and the fit values should be identical. Therefore, those two models cannot be compared, unless specific constraints, such as setting two factor loadings to be equal are imposed. The higher order model and first order model should have identical fit. The imposition of constraints such as equal loadings requires a strong rationale for why the two loadings would be equal. In the case of two first order factors it is not even clear what constraints could be imposed to compare the model to the first order model. Vlaeyen and colleagues do not provide a discussion of the procedure they used to achieve identification of the higher order model, so it is not possible to evaluate that procedure. Swinkels-Meewise, Roelofs, Verbeek, Oostendorp, & Vlaeyen (2003) conducted another examination of the factor structure of the T S K - D V , using both C F A and E F A . They first conducted a C F A , similar to that of Goubert et al., comparing a single factor model of all items, a single factor model without reverse scored items^ the two factor model of Clark, et al. (1996),  Catastrophizing, Fear  50  and the four factor model of Vlaeyen, Kole-Snijders, Rotteveel, et al. (1995). They found that none of these models provided adequate fit to the data for their sample of individuals with acute (defined as < 4 weeks duration) low-back pain seeking primary care. Interestingly, they argued that the Clarke model, in particular, fit poorly because of the CFI index value of 0.86, which is actually slightly better than the value found by Goubert et al., who argued that the model fit adequately. Swinkels-Meewise et al. then proceeded to randomly split their sample and run an exploratory factor analysis on one half. Using P C A with oblique rotation, they found a two factor solution differing from that of Clark et al., to best fit the data. The first factor items reflected "danger and injury" (p.374), and the second factor items reflected "activity and avoidance of activity" (p. 374) and the two factors showed small intercorrelations. Subsequently a C F A was run on the second half of the sample using this two factor model. The fit indices for this model were poor, and all reverse scored items were found to load poorly on their respective scales. These items were dropped and the model rerun, providing very good fit to the data. They labeled the two factors, "harm", the belief that something is seriously wrong with the body, and "activity avoidance", beliefs that" avoiding exercise or activity might protect against pain" and reported that the two factors were moderately intercdrrelated (0.55). Consistent evidence for a single appropriate T S K factor structure can only be considered tentative. The best evidence suggests a two-factor structure, but that structure could only be achieved by dropping four items. Even that evidence does not appear to be strong and the entire 17 item scale continues to be used in most studies. Finally, although the T S K is alone in having had the factor structure examined in a sample with acute (defined as < 4 weeks duration) pain (Swinkells-Meewise, Roeloffs, et al., 2003) that work found the structure tentatively supported in previous studies with chronic patients, did not hold for the recent onset pain sample. The supported structure was similar to that found for chronic patients in having two factors after dropping the four negative scored items, but the differences between the two solutions was substantial enough that the derived factors were given different meanings than the two factors  Catastrophizing, Fear  51  found with chronic samples. Finally, there has been very little work to date utilizing the English version of the TSK, despite the fact that the original instrument was written in English. What is Being Measured? Fear versus anxiety. The apparent confounding of fear and anxiety in both theoretical and empirical work in the pain related fear literature is another issue that warrants discussion in considering how the pain related fear constructs may interrelate. Asmundson et al. (in press) also highlighted this issue and it was one of the primary motivations behind their most recent elaboration of the Vlaeyen and Linton (2000) model. In the literature on pain related fear, the terms anxiety and fear have often been used interchangeably, (e.g., Asmundson et al., 1999; Vlaeyen & Linton, 2000). This practice even extends to the way instruments are named. For example, McCracken, Zayfert and Gross (1993), described their instrument, the PASS, as measuring "four aspects of anxiety associated with clinical pain symptoms" one of which is "Fear of Pain" (p. 183). The lack of clear delineation between anxiety and fear is not specific to pain related fear/anxiety. It characterizes much of the work in the broader field of fear and anxiety and it probably arises because in practice it can be very difficult to differentiate the two states (Rachman, 1998). Despite this, it has been argued (e.g., Marks, 1987; Rachman, 1998) that the two states can be differentiated on the basis of temporal and phenomenological characteristics, causes, and maintaining factors. According to Rachman (1998) fear is an intense and brief "emergency reaction", with a specific triggering stimulus, while anxiety tends to be an ongoing state of heightened vigilance with a vague cause and no clear onset or ending. Therefore, fear tends to be episodic and circumscribed, while anxiety is more persistent and pervasive (Rachman, 1998). Fear is a mobilization for avoidance/escape of a potentially dangerous situation, so it involves the type of changes that provide for immediate, organized action (Marks, 1987). The reaction pattern associated with anxiety is more diffuse and less closely associated with immediate activity (Marks, 1987). Despite these differences, Rachman (1998) has argued that avoidance behavior is often associated with both fear and anxiety. However, Asmundson, Norton and Vlaeyen (in  Catastrophizing, Fear  52  press) argued that because it is more related to immediate threat, fear is more likely to motivate defensive behaviors, such as escape, rather than avoidance. Anxiety, being more future oriented is more likely to motivate defensive behaviors such as avoidance according to Asmundson et al. They further argue that the fear-escape and anxiety avoidance dyads are mutually reinforcing, for example, the increased vigilance created by anxiety may lead to an increased likelihood that threat will be perceived and fear experienced. Rachman (1998), has argued that this reciprocal relationship is precisely the reason that fear and anxiety are difficult to differentiate in practice. The situation becomes immensely more complex when a third variable pain, which is associated with both fear and anxiety and shares many features, is considered. It is not surprising, given the complexity of the relationship between fear and anxiety, that the two are somewhat confused in the field of pain related fear. The issue applies no less to the measurement instruments themselves than it does to the literature. For instance, in addition to the reference to fear of pain in the PASS, the physiological symptoms assessed by the PASS physiological Symptoms Scale reflect the type of intense arousal (e.g., more associated with fear than with anxiety). Conversely, the items on the F A B Q appear to assess beliefs more akin to anxiety than fear in that they refer to future oriented anticipation (e.g., "physical activity might harm my back"), despite the use of the word "Fear" in the instrument title. Adding to the complexity of this issue, pain as a threat stimulus can be discrete and identifiable, such as would be experienced in acute pain, or it can be ongoing and more diffuse such as can be the case in pain of longer duration. Likewise, activity as a threat stimulus, can be either discrete and easily identifiable such as would be the case in a specific movement in the moment before an individual is tp engage in it, or it can be more diffuse and future oriented, such as would be the case for individuals experiencing anxiety over the perception that any movement or anticipate that hypothetical activities might cause pain. Similarly, the response to pain as a feared stimulus, may relate to the immediate aversiveness, or to immediate consequences such as injury, or it may be related to longer-term, less well defined consequences, such as incapacitation  Catastrophizing, Fear  53  or failure to recover. Most items from the F A B Q refer to "pain" as a consequence but in stating the consequence, do not differentiate between long term exacerbation, future flare-up or immediate exacerbation (e.g., the word "harm" which is stated as a feared consequence might refer to immediately apparent damage or unnoticed long term damage. Likewise, the items of the PASS refer to both long-term (e.g., "I am afraid of dying") and more specific immediate consequences (e.g., "I will become paralyzed"). It may be possible to differentiate fear and anxiety responses in the context of pain and Asmundson, et al. (in press) make a strong argument for the value of doing so, but that is beyond the scope of the current project. Measuring internal states is difficult, particularly with selfreport measures, and it is one goal of the current work to examine the existing self-report measures. Although it is acknowledged that the current instruments tend to confound fear and anxiety, the degree to which fear and anxiety overlap and interact, and the imprecision inherent in self-report measures is likely to obscure any effects of the difference between fear and anxiety as measured by self-report. The PASS, F A B Q , TSK and PCS appear to tap into elements of both fear and anxiety, and it may reasonably be expected that both play a role in avoidance behavior. Since the intent of this project is to compare the instruments as they exist for their theoretical significance, the issue of whether they tap fear and/or anxiety will not be addressed further and the terminology used by the scale authors will be used here without further comment as to its accuracy. It is important to note however, that all of the pain related fear measures tap into a more specific concept than the general anxiety assessed with instruments such as the State-Trait , Anxiety Inventory (STAI; Spielberger, 1983), which is often used clinically and in pain research. Pain related fear contributes significantly to the prediction of concurrent disability even after controlling for general anxiety (McCracken et al., 1996) among chronic pain patients. It may be that individuals who are highly anxious in general are more likely to develop more specific pain related fears, or that pain related fear, in the context of ongoing pain, leads to elevated general  Catastrophizing, Fear  54  anxiety. The former is reflected in the proposition that pain related fear and catastrophizing represent lower order manifestations of trait anxiety (Keogh & Asmundson, in press). The Relationship Between Aspects of Pain Related Fear and Catastrophizing. Beyond the relationship between catastrophizing and pain related fear, there has been limited discussion of how different aspects of pain related fear might relate to one another. As noted previously in this review, there is research demonstrating that the various measures of pain related fear are moderately to highly correlated, but there is also some evidence that there are discernable differences between the constructs they measure. The correlations between instruments and even between subscales of instruments are not so high as to suggest they overlap entirely or that the scales are redundant. In their review of the various measures of pain related fear, McNeil and Vowles (in press) stated that, "the issue of test selection is not one of "which measure is best? but is a matter of which measure or measures best taps the population or concept under investigation" (p. 205). The latter part of this question is difficult to answer unless there is more clarity on exactly what concept(s) each of the measures is tapping. Factor analyses are one way of deriving a more clear understanding of the concepts captured by a particular instrument. While factor analysis examines the relationships between items within a scale, examination of the relationship between scales can provide further information, by placing the instruments and by extension the concepts they measure into a nomological network. The relationships between instruments can be theoretically informative, but also clinically relevant, as the quote from McNeil and Vowles (in press) suggests. To use an analogy, if given a choice between measuring height and weight it is useful to know which is most important for a particular situation, for example height is more relevant for determining required headroom in a car, while weight is more important for determining required seat width. However, if one is presented with a choice of measuring either height or weight, in order to gain the most benefit from that choice, it is useful to know which measure is more predictive of the other. In the case of height and weight, height is likely to be more predictive of weight than vice versa, because in the human body, there is a minimal mass for each centimeter of height. There  Catastrophizing, Fear  55  is no minimal gradation of height for each gram of weight. A tall person will generally be heavier than a shorter person will, because each centimeter of height requires a certain amount of tissue. It would be rare to gain height and not gain weight, but unfortunately the reverse is not true! Therefore, height is likely to be more predictive of weight than weight will be predictive of height. If given a choice between those two measures, height provides more information, because it is predictive of weight. The same may hold for pain related fear and catastrophizing. If aspects of these constructs can be differentiated and demonstrated to be predictive of other aspects, the aspects that are most predictive may provide a more efficient measure. More concretely, i f catastrophizing is predictive of pain related fear it would suggest that at least early in the progression to chronic pain, catastrophizing might prove the more efficient construct to measure. As McNeil and Vowles (in press) highlight, there have been few studies comparing pain related fear instruments. They review two recent relevant studies (McCracken et al, 1996; Crombez, Vlaeyen et al., 1999). McCracken et al. (1996) compared pain related fear as measured by the F A B Q , PASS and a third pain related fear measure, the Fear of Pain Questionnaire - III, with trait anxiety as measured by the State -Trait Anxiety Inventory (Spielberger, 1983), a more general form of anxiety, in the prediction of pain severity, perceived disability, and pain behaviors in a sample of chronic pain patients. They found that the pain related fear measures were more highly correlated with the other pain related dimensions than was the more general anxiety scale. Crombez, Vlaeyen, et al. (1999) utilized the PCS, F A B Q , TSK, and/or PASS in various combinations across three separate studies examining the effects of pain related fear on perceived disability and performance on physical tasks in chronic pain patients. The pattern of results suggested that the T S K and F A B Q are more strongly associated with self-reported disability and poor behavioral performance and the PASS and PCS more closely associated with negative affect. More recently, Vlaeyen, de Jong, Geilen, Heuts, and van Breukelen (2001) and Vlaeyen, de Jong, Onghena, Kerkhoffs-Hanssen and Kole-Snijders (2002), utilized items selected from the PASS, T S K and PCS and F A B Q in a daily diary completed by  Catastrophizing, Fear  56  patients in a treatment study aimed specifically at reducing perceived disability through treatment of pain related fear in chronic pain patients. While it is difficult to generalize from a small subset of items, they found some indication that while for the most part, the different aspects of pain related fear and catastrophizing were highly associated, they behaved somewhat differently in one of the studies (Vlaeyen et al., 2002). It is difficult to interpret these results, particularly with regard to the PASS, since the total score combines subscales that appear to tap somewhat divergent domains. As indicated above, the PASS has been shown to have an impact on self reported disability and on poor performance on behavioral tasks. In addition, as Vowles and McNeil (in press) point out, the PASS escape/avoidance subscale is specifically designed to measure avoidant tendencies and therefore it would be expected to be highly predictive of avoidance behavior. The use of the PASS total scale score could obscure this effect, however, since other aspects of pain related fear may be more related to affective constructs. The PASS avoidance subscale does not appear to have been tested independently of the other subscales in predicting disability. This situation is not particular to the avoidance subscale, however, as none of the PASS subscales has been examined individually nor have they been compared. This leaves questions as to the construct validity of the subscales. For example it has been argued that the physiological anxiety subscale may simply be assessing arousal symptoms (e.g., sweating), which are direct consequences of pain rather than fear or anxiety (Larsen, et al, 1997). A final concern with regard to the PASS is the status of fearful appraisals. As stated earlier in this review, there is considerable evidence that the PASS fearful appraisals subscale is measuring catastrophizing. If that is the case, is it reasonable to consider catastrophizing as part of fear of pain, or is it best conceptualized as a separate, but related construct? While it is clear the instruments and by extension the constructs they measure are closely related, the nature of those relationships is not clear. A better understanding of that relationship would help to clarify the nature of pain related fear, providing a framework for further research and for attempts to predict and prevent the transition to chronic pain.  Catastrophizing, Fear  57  A Proposed Model of the Relationship Between Aspects of Pain Related Fear. In order to advance knowledge on the nature of pain related fear and catastrophizing constructs and their relationship to each other in the development of chronic pain, a theoretical model was developed and tested through Structural Equation Modeling (SEM). S E M is unique in that it allows simultaneous evaluation of the construct validity of the model and testing of hypothesized causal relationships between the constructs (Byrne, 1994). The development of this model was based on the foregoing discussion with regard to; 1) content of the various questionnaires; 2) Vlaeyen & Linton's (2000) model and; 3) reference to selected theoretical work in other aspects of fear and anxiety (e.g., cognitive behavioural models of Panic and Agoraphobia and Health Anxiety and Taylor and Rachman's 1992 models of fear and avoidance of aversive emotional states). The starting point for the proposed model was the model proposed by Vlaeyen and Linton (2000), which itself contains many untested assertions. Based on the Vlaeyen and Linton's model it is expected that pain intensity will contribute to catastrophic thinking, catastrophic thinking will contribute to pain related fear and pain related fear will contribute to avoidance behavior. While Vlaeyen and Linton, specify catastrophizing as causally preceding pain related fear, they do not specify relationships among the various pain related fear constructs. As noted above, reviews of the literature on pain related fear (e.g., Asmundson, et al., 1999; Vlaeyen & Linton, 2000), are also largely uninformative in this regard, and there is little guidance in the limited empirical work available to date. The findings of studies utilizing the various measures are typically discussed cumulatively as reflecting the importance of pain related fear in general. To develop an elaboration of the Vlaeyen model including specified relationships between aspects of pain related fear, it was necessary to draw on sources which are less obviously relevant. One is an issue raised earlier in this review, in the context of discussing a point made by Phillips and Jahanshahi (1987). In reviewing the argument regarding the motivation for avoidance put forward by Phillips and Jahanshahi (1987), it was in passing proposed that two different types of fear may be  Catastrophizing, Fear  58  operative in pain related avoidance, fear of pain and a fear of stimuli expected to cause pain. Examination of the pain related fear instruments and reading of various reviews of the literature employing them (Vlaeyen & Linton, 2000, Asmundson et al., 1999, McNeil & Vowles, in press) suggests that the constructs assessed by existing instruments do fall into these categories, with the PASS assessing primarily fear of pain and the TSK and F A B Q assessing fear of stimuli expected to cause pain. In the case of the TSK the objects of fear are movement and (re)injury, . while in the case of the F A B Q , the objects of fear are work and general activity. Categorization of the instruments in this manner is perhaps a first step in understanding how they interrelate, however, it still does not imply any causal relationship. This relative lack of understanding regarding the theoretical structure of pain related fear and catastrophizing, raises several issues. Perhaps most basic is that it complicates comparison across studies that use different measures. Second, as stated earlier, if the different instruments are tapping different constructs which are causally related in some manner, those causal relationships may inform clinical practice in helping clinicians to better focus assessment and intervention. Explication of the relationships between the instruments and the constructs they tap will advance theory in the area of pain related fear and will help to provide a framework for understanding the literature on pain related fears, A better understanding of the relationship between the instruments will also inform decisions about which is most appropriate for particular uses. One source, which appears to inform the understanding of pain, related fear is the relatively better developed literature on other types of fear/anxiety. Theoretical work on two other conditions, Panic Disorder with Agoraphobia and Health Anxiety, hold promise in this regard, and this will be briefly reviewed below with attention to implications for understanding pain related fear. As the purpose of the current work is not to provide a detailed exploration of these phenomena, the review is not comprehensive, and the interested reader is referred to recent work by Barlow (2002) and Warwick, (1989).  Catastrophizing, Fear  59  The cognitive model of Panic Disorder with Agoraphobia. According to the Diagnostic and Statistical Manual of Mental Disorders- Fourth edition (DSM-IV; American Psychiatric Association, 1994) "the essential feature of Panic Disorder is the presence of recurrent, unexpected Panic Attacks" (p. 397), followed by a period of persistent i  concern about future attacks or behavioral change due to the attacks. A panic attack is an isolated, brief period of intense fear or discomfort (Rachman, 1988). Agoraphobia is anxiety about being in public places, which typically leads to pervasive avoidance of such places (DSMIV; American Psychiatric Association, 1994). Agoraphobia often occurs in the context of Panic Disorder and the agoraphobic anxiety and avoidance relates to a fear of having a panic attack (White & Barlow, 2002). Current cognitive behavioral theories of Panic Disorder with Agoraphobia (e.g., Clark, 1986; Barlow, 2002) propose that Panic Disorder results from anxiety over physiological symptoms of panic, and Agoraphobia serves as a strategy for coping with Panic Attacks, by avoiding situations which might trigger panic (White & Barlow, 2002). Indeed, most agoraphobic patients attribute their avoidance to episodes of panic (Rachman, 1998). Clark (1986) emphasized the role of catastrophic misinterpretation of bodily symptoms in the initiation and prolongation of panic attacks. There is a substantial body of evidence supporting cognitive- behavioral models of Panic Disorder with Agoraphobia, and in particular it has been demonstrated that those who experience panic are more sensitive to bodily sensations and prone to interpret those sensations in a catastrophic manner (see reviews by Clark et al., 1996 and Barlow, 2002). The bodily sensations typically misinterpreted by individuals with panic are those produced by the autonomic nervous system in response to fear and anxiety, for example dizziness, breathlessness, or increased heart rate. In those predisposed to catastrophic misinterpretation of those sensations, they are taken to indicate an imminent disaster such as a heart attack. These interpretations lead to further bodily sensations creating a vicious circle, which culminates in a panic attack. Therefore, panic may be considered to result from a fear of  Catastrophizing, Fear  60  fear itself. A variety of internal (e.g., hyperventilation) or external (e.g., feared situation) stimuli, which are perceived as threatening, can trigger these sensations. Taylor and Rachman (1992) conducted a study examining the relationship between fear of anxiety symptoms, agoraphobic fear (e.g., fear of public places) and agoraphobic avoidance (see Figure 3 for graphical depiction) using structural equation modeling. They found support for a model where both fear of anxiety and agoraphobic fear predicted agoraphobic avoidance but the relationship between fear of anxiety and agoraphobic fear was partially mediated by agoraphobic fear (see Figure 3). Interestingly, in the same study, Taylor and Rachman also found that a similar set of relationships held for another aversive state, sadness. They demonstrated that fear of sadness and fear of cues to sadness (e.g., sad movies) both predicted avoidance of saddening stimuli but the relationship between fear of sadness and fear of sadness cues was partially mediated by fear of sadness cues (see Figure 4 for graphical depiction). Both of these models demonstrate that a more basic fear of an internal state (fear or sadness) leads to avoidance behavior. However, it also generalizes into a fear of stimuli, which are expected to trigger it (e.g., public spaces or sad movies). This has implications for understanding pain related fear, i f one considers that the PASS appears to measure fear of pain while the TSK and F A B Q appear to measure fear of stimuli or situations expected to cause pain. Reiss and McNalley (1985) refer to the fear of anxiety symptoms as Anxiety Sensitivity (AS) and propose that it represents a basic fear, central to the development of many anxiety disorders including, but not restricted to, panic attacks. According to their theory, the individual high in AS responds to anxiety symptoms with further anxiety and, therefore, AS amplifies any fear reactions. In an individual with high AS exposure to a minimally anxiety provoking event, this amplification can lead to the development of clinically significant levels of fear. Anxiety sensitivity also appears to be closely connected to fear of pain. Using Structural Equation Modeling, Asmundson & Taylor (1996) found evidence that AS might play a role in the development of fear of pain (see Figure 5). Using subscales of the PASS and other indicators, they demonstrated that anxiety sensitivity predicted fear of pain, which in turn predicted  Catastrophizing, Fear  61  avoidance behavior. Interestingly, in testing the same model, they found that the relationship between pain severity and avoidance was completely mediated by fear of pain. The Cognitive Theory of Health Anxiety The concept of health anxiety bears similarities to pain related fear and the cognitive theory of health anxiety shares many similarities with the cognitive theory of panic. The concept of health anxiety provides a cognitive learning explanation for Hypochondriasis (Warwick & Salkovskis, 1990), a clinical disorder characterized by an overwhelming conviction that one has a serious physical illness. Warwick and Salkovskis proposed that Hypochondriasis represents an extreme example of a more general tendency to attend to illness related information and to misinterpret the information in a catastrophic and personally threatening manner. The information may be internal bodily sensations or health information gleaned from the media or other forms of communication. This information becomes the basis for the formation of specific beliefs or assumptions such as "bodily changes are usually a sign of serious disease, because every symptom has to have an identifiable physical cause" (Warwick & Salkovskis, 1990; pp. 110). The cognitive model of health anxiety is similar to the cognitive model of panic disorder in that these fearful cognitive biases are proposed to lead to anxiety about any indication of ill health and this anxiety produces physiological symptoms which are interpreted as further evidence of illness or behaviors (rubbing or repeated checking) which actually create new physical symptoms (Warwick, 1989). One response to such fear of bodily sensation leads to avoidance of illness cues. Since such cues are not entirely avoidable, the result is sometimes excessive reassurance seeking. In its extreme form this results in the frequent use of medical care, which characterizes Hypochondriasis. The model also accounts for less extreme forms of health anxiety (Hadjistavropoulos, Craig & Hadjistavropoulos, 1998). Warwick, Clark, Cobb & Salkovskis (1996) demonstrated that patients diagnosed with Hypochondriasis benefited from cognitive behavioural treatment based on treatments for panic was effective in reducing hypochondriachal symptoms. The therapy included identifying  Catastrophizing, Fear  62  Figure 3. Taylor & Rachman (1992) model of fear and agoraphobic avoidance . 1  Figure 4. Taylor & Rachman (1992) model of fear and avoidance of sadness . 1  Reprinted from Journal of Anxiety Disorders, 6, Fear and avoidance of aversive affective states: Dimensions and Causal relations, Pages 15-25, Copyright (1992), with permission from Elsevier.  Catastrophizing, Fear  63  Figure 5. Model of Anxiety Sensitivity and Fear of Pain (Asmundson & Taylor, 1996).  Cf. Asmundson & Taylor, 1996; ellipses represent latent variables (constructs) while boxes represent observed indicators of the construct. A hypothesized direct arrowfrompain severity to avoidance was not supported. ReprintedfromJournal of Behavioral Medicine, 19, 577-586. Role of anxiety sensitivity in pain related fear and avoidance.Copyright (1996), with permission from Springer and the first author.  Catastrophizing, Fear  64  and challenging evidence for misinterpretations of symptoms and signs and helping the patients to construct more realistic interpretations. It also included Behavioural experiments, to assist the patients in testing these new interpretations. These experiments included exposure to both feared innocuous bodily symptoms by deliberate inducement of those symptoms and focusing the patient's attention on them. In addition, graded exposure to previously avoided illnessrelated situations was included. This supports an approach to health anxiety as a set of separate but related fears, fear of illness symptoms and fear of illness related situations, a similar construal as applied by Taylor and Rachman (1992) to panic and agoraphobia and fear of sadness. The cognitive theory of health anxiety bears obvious relevance to understanding pain related fear. It is possible that pain related fear represents a specific type of health anxious response to painful injury. A n individual predisposed to fearful misinterpretation of physical symptoms, when faced with an injury would seem likely to engage in the kind of catastrophic thinking measured by the fearful appraisals subscale of the PASS or the PCS. The PASS item "When I feel pain, I think that I might be seriously i l l " seems to directly reflect health anxiety. Further evidence for the connection between health anxiety and pain related fear and avoidance comes from work by Hadjistavropoulos, et al. (1998), who found that individuals high on health anxiety reported more catastrophic cognitions during a painful procedure and withdrew from the procedure earlier than less health anxious individuals. Implications for a Model of Pain Related Fear The foregoing discussion has several implications for a model of fear of pain. The cognitive theory of panic and health anxiety support the proposition derived from Vlaeyen's model, that catastrophic thinking will mediate between the pain stimulus and pain related fear. The PCS and one subscale from the PASS (fearful appraisals) assess catastrophic appraisals. The cognitive theories of panic and health anxiety also suggests specific relationships between aspects of pain related fear. Specifically, based on the Taylor and Rachman (1992) model, it is proposed that fear of pain may contribute to fear of activity and that both may predict avoidance behavior. If the fearful appraisal subscale is taken to measure catastrophizing, and avoidance is  Catastrophizing, Fear  65  considered to be a separate construct, the two remaining subscales from the PASS (cognitive interference and physiological anxiety) appear to tap into fear of pain. As stated earlier, the F A B Q and TSK appear to assess fear about stimuli, which might be expected to cause pain. Following the reasoning of Taylor and Rachman (1992), it seems reasonable that individuals who are afraid of pain are likely to develop a progressive fear of situations and stimuli which are expected to cause pain, similar to the way in which individuals with panic often come to experience agoraphobia. It is important to acknowledge that the pain related fear measures assess fear in a somewhat different way than the type of fear measures used in Taylor and Rachman's (1992) study. In that study, respondents were specifically asked to rate their fear of particular stimuli on a Likert Scale. The stimuli included a range of intensity, for example, fear of sadness was assessed by having subjects rate their fear of "feeling mildly sad", "feeling moderately sad" and "feeling extremely sad". The pain related fear scales do not assess fear in such a direct manner. The PASS cognitive interference and physiological anxiety subscales ask the individual to report on the occurrence of particular symptoms in the presence of pain o fun specified intensity, which ostensibly reflect fear or anxiety. The F A B Q specifically assesses particular beliefs and appraisals, which again ostensibly reflect fear. Some of the F A B Q items do refer specifically to a feared outcome, namely an increase in pain, (e.g., "I should not do physical activities which (might) make my pain worse") while others are less direct (e.g., " M y work is too heavy for me"). Summary of the Proposed Model The proposed model will consist of a set of hypothetical relationships between pain intensity and various subscales of the PCS, PASS, TSK and F A B Q . The model is provided graphically in Figure 6. The model begins with pain intensity and ends with avoidance behavior. Arrows labeled with lower case Roman numerals indicate the hypothesized relationships between variables. The first hypothesized relationship illustrated (arrow i), is that increased pain levels are expected to predict increased levels of Catastrophizing. This path is derived directly from  Catastrophizing, Fear  Figure 6. Proposed structural model of pain, catastrophizing and pain related fear and avoidance  66  Catastrophizing, Fear  67  Vlaeyen's model. The second hypothesized relationship (arrow ii) is that Catastrophizing is expected to predict fear of pain. The third hypothesized relationship (arrow iii) indicated is that Catastrophizing will predict fear of activity. Stated another way, Catastrophizing is expected to mediate between Pain Intensity and pain related fear (Fear of Pain and Fear of Activity). The fourth hypothesized relationship (arrow iv) is that Fear of Pain will predict Avoidance behavior and the fifth hypothesized relationship (arrow v) is that Fear of Activity will predict Avoidance. No direct effect of Catastrophizing on Avoidance is predicted in this model. Stated another way, pain related fear (Fear of Pain and Fear of Activity) is expected to mediate between Catastrophizing and Avoidance Behavior. The sixth relationship (arrow vi) indicates the hypothesis that Fear of Pain will predict Fear of Activity. This relationship (arrows vi) represent the hypothesis that Fear of Pain represents a more central fear predicting Fear of Activity, or stated another way, that the relationship between Fear of Pain and Avoidance behaviour will be partly mediated by Fear of Activity. Although it is generally advisable to utilize item level data for structural equation modeling and typically several indicators per latent variable are recommended, there is an upper limit on the number of observed variables. As the number of estimated parameters approaches the sample size (in the current case N = 210), the stability of the parameter estimates comes into question. In order to achieve a balance between these demands, the current study involved selecting components of various instruments to use in the structural model, rather than the complete instruments. It was anticipated that the PASS cognitive anxiety and physiological anxiety subscales would represent fear of pain, that fear of activity would be represented by the activity subscale of the F A B Q and/or the TSK, and that catastrophizing would be represented by the PCS. However, the choice of specific items was to be informed by the C F A ' s of the  Catastrophizing, Fear  68  respective scales and therefore specific item selection will be discussed following the C F A results. Summary of Review and Rationale for Research Questions Pain related fear and catastrophizing appear to hold great promise for the prediction and prevention of chronic pain. However, there are obstacles to using the existing measures efficiently for that purpose. One is the limited use to date of any of the instruments in early after pain onset. Since the assessment of pain related fear early in the development of pain and disability holds promise in predicting and preventing chronicity, it seems important to ascertain the utility of pain related fear in the early stages of pain and disability. One important aspect, which needs to be examined, is the factorial validity of the existing measures in a subacute population. Although one instrument, the T S K has been tested in a sample with recent onset pain, that context, the results are not convincing. The current study attempted to address this issue through confirmatory factor analyses comparing various factor structures of each instrument reported in the literature. A second issue is the theoretical relationship between aspects of pain related fear and catastrophizing. Vlaeyen and Linton have proposed that catastrophizing causally precedes pain related fear in their model, but support for that proposition is limited. In addition, although the pain related fear measures appear to tap into different concepts, there has not yet been an attempt to provide a framework for understanding how those constructs might interrelate. The current work addressed this issue through the development of a model of the pain related fear constructs and testing of that model using structural equation modeling. A third issue addressed is the utility of the various pain related fear measures and by extension the constructs they assess in the prediction of outcome in subacute pain patients. While the prospective studies conducted to date suggest that pain related fears may play a role  Catastrophizing, Fear  69  in pain, suffering and disability that persists after back injury, the research is limited. There are few prospective studies examining the effect of pain related fear, and even fewer utilizing recent onset pain samples. The predictive measures utilized have been limited (3 studies employing the F A B Q and one study employing the PCS) and only one study has employed more than one measure of pain related fear and catastrophizing. The outcome variables examined have been limited to perceived disability and return to work. Turk (1997) has argued for the used of multiple indicators of outcome to reflect the multidimensional nature of pain. These three issues were addressed through three separate sets of analysis on data provided by the same sample of participants. Data were collected at two time points, one within the first seven weeks of onset of back pain, and then at three months (12 weeks + 3 weeks) post injury. First, confirmatory factor analyses through structural equation modeling were conducted on time one data to determine the best fitting factor structure for each of the pain related fear measures (PASS, F B Q & TSK) and the PCS. Second, the structural hypotheses stated above were tested using structural equation modeling on the same data. Finally, regression analyses were conducted using time one pain related fear and catastrophizing variables to predict time two outcome (depressive symptoms, perceived disability, pain level and return to work). Structural equation modeling is a complex statistical procedure and therefore a brief overview of general considerations for its use are provided in the following section.  Catastrophizing, Fear  70  Method A l l procedures for the study were approved by both the University of British Columbia Behavioral Ethics Research Board and the W C B Freedom of Information and Privacy Office. Data were collected from a single sample at two time points, but three distinct sets of analyses were employed, 1) structural equation modeling (SEM) confirmatory factor analyses (CFA), 2) S E M test of structural model, and 3) regression analyses. As each of these sets of analyses involved different subsets of data and of participants the methods and results are presented in three separate sections. The discussion will review all three sets of results and serve to integrate them. Structural Equation Modeling In the present work, S E M was utilized to address questions relating to the factor structure of the various instruments and for testing the structural model. S E M is generally an iterative process where a hypothesized model is tested and ideally compared to one or more alternatives. Because of its flexibility and the promise of simultaneously testing a number of hypotheses, and because it lends itself well to easily understood graphical representations, S E M has become a popular tool for analyzing complex data in psychology (Ullman, 2001). However, there are many potential pitfalls in the process and many published papers either apply the technique inappropriately or fail to provide the reader with logical explanation for decision processes involved in the analysis (Breckler, 1990), leaving the validity of the findings in question. Therefore, space will be dedicated here to a general outline of the S E M procedures utilized in the current work. The following briefly outlines the modeling and analysis of covariance structures used to test the hypothesized model. General Procedure for Structural Equation Modeling The broad goal of structural equation modeling is to determine the degree to which the  Catastrophizing, Fear  71  parameters implied by a specified model "fit" the covariance structure obtained from a set of data. S E M can be utilized to conduct factor analysis or regression style path analysis, but its greatest strength lies in the ability to combine both. S E M can be utilized to simultaneously test complex relationships between variables, allowing for the inclusion of constructs that are not directly observable (i.e., latent) and therefore imperfectly measurable. Two separate types of variables are specified in a structural equation model, observed (measured) variables, and latent variables, which are the constructs assumed to underlie and cause the observed variables. The factor analytic or "measurement" aspect of a structural equation model consists of hypothesized relationships between the observed and latent variables. In its simplest form, this type of model corresponds to a typical factor analysis, with factors represented by latent variables and items specified as observed variables. However, because structures are hypothesized a priori and S E M output provides tests of overall fit, the procedure can be considered confirmatory rather than exploratory. In addition, for structural models, that is models with specific directional relationships specified between latent variables, the inevitable error in measurement, which would otherwise bias the parameter estimates, can be estimated within the model. Since psychological constructs are generally conceptualized as being latent, that is not directly observable, the ability to acknowledge and model measurement error is an important advantage of S E M over traditional path analytic or regression procedures (Hoyle, 1995). However, S E M procedures are complex and each analysis involves a number of decisions, some of them based on subjective judgment. Structural equation modeling is based on the assumption of multivariate normality (Ullman, 2001). Deviations from multivariate non-normality may lead to conservatively biased estimates, and increase the likelihood of incorrectly rejecting a true model (West, Finch, & Curran, 1995). The EQS program (Bentler, 1995), which was utilized in the current study,  Catastrophizing, Fear  72  provides a statistic, Mardia's Kappa, which assesses the degree of multivariate kurtosis. In each of the structural equation modeling analyses conducted for the present study, the distributional characteristics of the data (e.g., univariate and multivariate normality) are reported. Distributional non-normality of data can be corrected in a number of ways, including transformation of variables, alternative estimation procedures (e.g., Asymptotically Distribution Free Estimator, Browne (1984), bootstrapping and corrections to the statistical estimates (e.g., Satorra Bentler % ; Satorra, 1990) but each of these has limitations (West, Finch & Curran, 2  1995). Bootstrapping, alternate estimation procedures and the Satorra Bentler correction all require larger data sets than the one available for this study (Chou, Bentler, & Satorra, 1991) and therefore none of these procedures was considered appropriate. Transformation of individual items can correct univariate distributions, but will not necessarily alter multivariate distributional characteristics (West et al.). In addition, as West et al., point out, transformation alters the meaning of the particular responses to the particular transformed item, and, in the context of several other items on the same questionnaire, it becomes impossible to interpret such changes. Fortunately, if non-normality is less than severe, the effect is only a minimal bias toward more conservative estimates (West et al.) that is a slightly greater likelihood of rejecting a true model. Therefore, in the current work, distributional characteristics of the data utilized in each analysis were reported, but no adjustments were made. The result of any existing non-normality, was that tests may have been somewhat conservative and therefore parameter estimates somewhat attenuated. Assessment of Model Fit S E M involves postulating a hypothetical set of relationships between variables and then comparing the covariance structure implied in that model to the actual covariance matrix of an observed data set. The degree of correspondence between observed and hypothesized covariance  Catastrophizing, Fear  73  matrices is often assessed using the x likelihood ratio test. Since the distribution of % is known, a probability level can be calculated as in traditional hypothesis testing procedures. In contrast to traditional decision rules for hypothesis testing utilizing % , in S E M , a non-significant % value 2  2  supports the hypothesized model, indicating a good match between the hypothesized and actual covariance matrices. Despite the advantage of allowing for statistical hypothesis testing, the %  2  test is rarely used as the sole measure of fit in structural equation modeling, because it is overly sensitive to sample size. With large samples, such as that required for accurate estimation in S E M the x statistic is likely to be significant, even for a correctly specified model (Marsh, 2  Balla, & McDonald, 1988). Due to this limitation in the % test, a number of alternative indices 2  for assessing closeness of fit have been developed (Marsh et. al., 1988; Satorra & Saris, 1985; Tanaka 1993). Interpretation of the close fit indices is based upon "rules of thumb" derived from simulation studies (e.g., Browne & Cudeck, 1993). Based on results of Monte Carlo simulation studies, Hun and Bentler (1999) recommended using a combination of close fit indices in the assessment of overall model fit, because characteristics of the data, such as distribution and sample size, and characteristics of the specified model, such as the number of parameters estimated, affect each of the indices in different ways and to different degrees. Hun and Bentler specifically recommend examining the Standardized Root Mean Residual (SRMR), along with at least one other index, such as the Comparative Fit index (CFI; Bentler, 1988) or the Root Mean Square Error of Approximation (RMSEA; Browne & Cudeck, 1993). The S R M R is the standardized average difference between the sample variances and covariances and those estimated in the population based on the proposed model. Smaller values of S R M R indicate better fit. Hu and Bentler (1999) reported that S R M R values in the range of 0.08 or lower represent a good fit when the index is used in conjunction with other indices.  Catastrophizing, Fear  74  The CFI compares the fit of the hypothesized model to that of a model where all variables are specified to be uncorrected. A larger CFI value indicates better fit and a value close to 0.95 indicates good fit when accompanied by a S R M R value in the range of 0.08 or lower (Hu & Bentler, 1999). The R M S E A is an index comparing the fit of the hypothesized model to the fit of a model where all variables are allowed to correlate. A smaller R M S E A value indicates better fit. While Hu and Bentler, suggest that SRMR and CFI alone are adequate, Cheung and Rensvold (2002) demonstrated that R M S E A is less affected by model complexity than are other fit indices, which will tend to over-reject more complex models. Therefore, the R M S E A will also be examined in the current study. Hu and Bentler recommend a R M S E A nearing 0.06 as indicating good fit when the S R M R is in the range of 0.08.  However, those authors also  acknowledge that, when sample sizes are relatively small in S E M terms ( N < 250), a cutoff of R M S E A < 0.06 leads to over rejection of true models. Earlier work by Browne and Cudeck (1993) indicated that while a R M S E A value of 0.05 indicates very good fit, values up to 0.08 represent acceptable fit and 1.0 suggests the model is unfeasible. The more liberal cutoff was considered appropriate for the current study where n was less than 250. To summarize for the present purposes, a model was considered as adequately fitting the data i f S R M R approached 0.08, CFI approached 0.95, and R M S E A was < 0.08. % values are presented, but were not used 2  to inform the assessment of model adequacy because of the aforementioned sensitivity to sample size. Comparison of Alternative Models Demonstrating that a particular model provides adequate fit to the data does not rule out the possibility of alternative, but equally well fitting models. Therefore, it is generally recommended that alternative models also be developed and compared to the proposed model (e.g., Breckler, 1990). For the C F A analyses in the present work, a number of alternative factor  Catastrophizing, Fear  75  structures, derived from previous research, were tested for each instrument. Likewise, a number of alternative models for the relationships between the pain related fear and catastrophizing constructs were developed. A benefit of testing alternative structural models is that specific segments of the hypothesized model can be tested by devising a model, which is identical to the hypothesized model in every regard save the particular parameter of interest. For example if one is interested in testing the importance of a mediating path in the hypothesized model, an alternative model without that path can be compared to the hypothesized model. If the alternative model does not provide worse fit than the hypothesized model that provides evidence that the path was not necessary. In the current work a series of models culminating in the hypothesized structural model, were specified such that each model in the series differed from the preceding one by the addition or subtraction of particular paths to test a specific hypothesis. The alternative models are described in detail in the methods section. Models, which are nested, which are to say that they contain the same latent variables, but differ in the number of paths between those latents, can be compared with a statistic test using the obtained % values for the two models. The difference in % values between two nested models is itself be distributed as a % statistic on degrees of freedom equal to the difference in degrees of freedom between the two models. Therefore, the probability of the % difference can be derived, providing a statistical test of the difference in fit. The interpretation of a significant difference is such that given two models the one with fewer estimated parameters (higher df) fits better, if the difference in % between the models is not statistically significant. Conversely, a 2  statistically significant difference in % values indicates that the model with more specified 2  parameters (fewer df) is superior. In other words, a statistically significant drop in % with the 2  Catastrophizing, Fear  76  addition of (a) path(s) to a model indicates that the addition was warranted and that the path(s) are important to the model. A failure to find a significant drop in the % value with the addition 2  of (a) path(s) indicates that the added path(s) are not warranted. Of course, such comparisons only apply if both models provide adequate fit and reasonable parameter estimates. The comparison of two ill-fitting models would be meaningless, since neither adequately accounts for the data covariances. In the present study, a % difference 2  test was utilized wherever two or more alternative and nested models were found to adequately explain the data. The various structural models to be tested did meet criteria for nesting, since they contained the same latent constructs, with only the relationships between the constructs differing across models. Therefore, in the structured analyses, % difference tests were utilized to ascertain 2  which model best accounted for the data. However, in the C F A analyses many of the alternative factor structures postulated a different number of latent variables and therefore were not formally nested and a % difference test was not appropriate. In the absence of nesting, or if two models 2  have equal degrees of freedom, and both demonstrated reasonable fit to the data, the model with fewest unspecified (estimated) parameters was considered superior on the basis of parsimony. Model modification It is common practice to modify ill-fitting models to achieve better fit (Byrne, 1994). Typically, various aspects of the statistical output are examined, changes to the model specification made and the model retested. In practice, changes are often made based purely on statistical grounds. Since the modification procedure is post hoc, it is exploratory and risks capitalizing on sample specific characteristics of the data, thereby reducing the degree to which the results generalize to other samples. Therefore, it is generally recommended that model modification be approached in a judicious manner with any changes being theoretically  Catastrophizing, Fear  77  justifiable (Byrne, 1994). Recourse to theory in addition to statistical considerations reduces the risk that modifications are simply capitalizing on chance fluctuation in population parameters within the specific sample. In the present study, if none of the proposed models provided adequate fit to the data, but one or more approached close fit, those models were judiciously modified through consideration of statistical and theoretical issues. A l l modifications were tested through reanalysis of the model and modification was stopped once acceptable fit was achieved. If statistical considerations suggested model improving modifications, but none appeared theoretically justified, they were not made, in order to avoid capitalizing on chance. The EQS program (Bentler, 1995) was utilized to conduct all S E M in the current work. EQS output provides many clues as to the source of model misfit, and thereby provides guidance as to what modifications might improve fit. S E M involves a complex process of matrix algebraic calculations based on the original data file and the specified model. If a model is severely mis-specified, for example i f severe multicolinearity is present, or the number of latent variables specified is far from explaining the data, the necessary matrices cannot be calculated and the program will not run properly. In such cases, error messages are provided which can point to the source of misfit. Occasionally, a model will provide adequate fit, but provide parameter estimates, which are unreasonable. The parameter estimates provided by S E M analyses correspond to: 1) the loading of each item on its associated factor for the measurement part of the model 2) the regression weights for paths between latent variables 3) error terms for each dependent variable and 4) variance for each independent variable. In an EQS model, a dependent variable is a variable predicted by any other variable in the model. Independent variables are those, which serve as predictors but are not predicted by any other. Parameter estimates for any of these values outside of the expected range can point to specification problems. For example  Catastrophizing, Fear  78  standardized values for any parameter outside the range of 1 to 0, called "Heywood cases", often indicate identification or multicolinearity issues. Similarly, values, which are in the opposite direction to what would be predicted from the proposed model, suggest that the data does not conform to the hypothesized theory, even if the model provides adequate fit to the data. The derived parameter estimates may also deviate substantially from what would be expected theoretically, for example, regression weights of the opposite valence to that expected (i.e., a negative value where a positive relationship is expected) suggesting a problem with the structure of the model or coding of the data. EQS calculates a z score for each parameter. A non-significant z score (i.e., z < 1.96) indicates that the estimated path value does not differ significantly from zero. Such non-significant paths can often be dropped from the model, improving parsimony with minimal decrement in model fit, indicating that they are not necessary. Other valuable pieces of information provided in the EQS output useful for suggesting potential model modifications include the standardized residuals for each covariance and the Lagrange Multiplier (LM) Test. The multivariate Lagrange Multiplier (LM) Test indicates which changes to model parameters will improve model fit and the significance of that improvement using a process similar to stepwise multiple regression. The standardized residuals provide a measure of the degree to which the relationship between two variables in the data set is not accounted for by the model that is the error in predicting for each covariance in the data from the specified model. A high standardized residual associated with the covariance of a pair of variables, suggests that some specification involving the two variables is problematic. In the current study, modifications were made to models approximating close fit utilizing the above information, but only if they could also be justified theoretically. A new test was conducted after each modification to determine i f it significantly improved model fit and in no  Catastrophizing, Fear  79  case were more than two iterations of modification made. The degree of care in conducting modifications utilized in the current study minimized the degree to which the analyses capitalized on chance sampling fluctuations and should lend credence to the findings. Participants Potential participants were identified from W C B claimant lists. The initial criteria for inclusion was subacute (< 7 weeks) soft tissue injuries of the low-back. In order to provide a homogeneous sample of individuals with uncomplicated soft tissue injury to the low-back, instructions in the mailing included a list of exclusionary diagnoses and potential participants were asked to self exclude i f any of the diagnoses applied to them. Participants were asked to self exclude if: 1) their English reading skills were not adequate to read a newspaper (approximately grade eight level) 2) they were pregnant 3) they had a history of previous back surgery 4) they had been diagnosed with malignant pain (e.g., cancer pain) 5) they had been diagnosed with an arthritic condition of the spine 6) had been diagnosed with spondylolisthesis 7) they had been diagnosed with disk herniation or 8) they had been diagnosed with a vertebral fracture. Participants were informed in their time one questionnaire package that a second questionnaire would be sent to them for completion in approximately six weeks. Participants were offered an honorarium of 20 dollars for completion of both sets of questionnaires. Three hundred and three individuals responded to the time one mailing. Based on review of the data provided, some individuals who responded did, met the exclusion criteria listed above and were therefore inappropriate for the study and others responded outside of the appropriate time window. A series of procedures was used at this stage to restrict the sample to useable data. At the initial stage of data cleaning, participants were dropped if: a) they responded to the time one questionnaire too late (> 7 weeks post injury), b) they did not meet diagnostic criteria, c) they reported their most severe pain as anywhere other than their low-back, or d) they could be  Catastrophizing, Fear  80  considered to have recovered. The criteria for the latter, was worst pain over the preceding week of less than 30mm on a 100 mm scale (anchors; 0 = "no pain" and 100 = "worst pain possible") and reported having returned to work. Thirty millimetres on a 100-millimetre pain scale has been demonstrated to represent a clinically significant level of pain. The choice of cutoff for minimal pain level was based on research by Stratford and Spadoni (2001) who demonstrated that the 90% confidence interval for one time ratings of pain by back pain patients on an 11 point numerical rating scale was + 2.13 points. In the same sample the 90% confidence level for change in ratings over a brief period was + 2.87 points. Taken together, these findings suggest that ratings below three might not be reliably different from 0. Since this was the worst pain over the preceding week, individuals who provided ratings of 30 or less and had returned to work, were unlikely to be experiencing a significant level of pain. The fact that the excluded individuals had returned to work provides further indication that their pain was not a significant concern. Of the 303 individuals potential participants who completed and returned time one questionnaires, 19 returned their questionnaires more than seven weeks post-injury and so were dropped from further analysis. Of the 284 remaining participants, 18 reported that their main locus of pain was other than their low-back and they were dropped, leaving 266 participants. Of these, 20 more were dropped because they were judged to have recovered because they had returned to work and had worst pain ratings less than 30mm, leaving 246 participants. Twentyone more potential participants reported diagnoses, which excluded them, leaving 225 participants. Of the remaining subjects, the vast majority (N= 185) reported diagnoses of sprain and/or strain. Further attenuation of sample size occurred due to missing data. As variables used in the S E M analyses (CFA and structural model analyses) and regression analyses were different, the  Catastrophizing, Fear  81  actual participants dropped due to missing data varied slightly. It was considered necessary to scan for missing data at the item level, because the covariances utilized in the S E M involved item scores. Scanning for subjects missing more than ten percent of item level data on any of the questionnaires included in the S E M analyses (FABQ, TSK, PASS, PCS, Pain Behaviour Checklist, and M c G i l l Pain Questionnaire - Short Form) led to 15 participants being dropped, leaving N = 210 for the S E M analyses. Descriptive data for this sample, including means and standard deviations on all subscales used in the C F A , are displayed in Table 1. The sample included 93 women and 117 men. With regard to marital status, 71 percent of the sample reported being married or living common-law, 17 percent reported being single and never married, and 11 percent reported being divorced or widowed. Eighty-two percent of the sample reported their ethnic origin as Caucasian, 7.6 percent described themselves as Asian-Canadian, 3.8 percent as Indo-Canadian, 2.9 percent as First Nations, 1 percent as African-Canadian and 1.4 percent as other. Two percent of the sample reported having worked less than 20 hours per week prior to their injury, 73 percent worked between 20 and 40 hours per week and twenty five percent worked more than 40 hours per week. Approximately 30 percent of the sample reported having been off work with back pain at least once prior to the current injury, but fifty percent of the sample did not respond to this question. Approximately 46 percent of participants reported they provide more than 75 percent of their family's income. Another 31 percent earned between 50 percent and 75 percent of the family income approximately 16 percent earned between 25 percent and 50 percent of the family income and the remainder earned less than 25 percent of the family income. With regard to education level, approximately 4 percent had a maximum of grade school education, 47 percent had high school and the remaining 49 percent had post secondary education.  Catastrophizing, Fear  82  Since the regression analyses were longitudinal, completed time two data were required for those analyses. Of the 225 subjects meeting inclusion criteria at time one, 158 returned their time two questionnaires. After dropping subjects who completed the questionnaires late, who had more than ten percent of the relevant data missing or who reported that they had not returned to work for a reason other than their identified back pain, the final sample for the regression analyses consisted of 137 participants. Demographically, this sample was very similar to the larger sample utilized in the S E M analyses. It consisted of 80 men and 57 women. The mean age of individuals in the sample was 41 years (SD = 10.7 years). With regard to marital status, 71 percent of the sample reported being married or living common-law, 17 percent reported being single and never married, and 12 percent reported being divorced or widowed. Eighty-three percent of the sample reported their ethnic origin as Caucasian, 6.6 percent described themselves as Asian-Canadian, 5.1 percent as Indo-Canadian, 2.2 percent as First Nations, 1.5 percent as African-Canadian and 1.5 percent as other. Approximately four percent of the sample reported having worked less than 20 hours per week prior to their injury, 71 percent worked between 20 and 40 hours per week and twenty five percent worked more than 40 hours per week. With regard to proportion of family income provided by the participant, 43 percent reported that they provided more than 75 percent of the family income, 37 percent provided between 50 and 75 percent of the family income and 14 percent provided 14 percent of the family income, while the remaining 6 percent provided less than 25 percent of the family income. Educationally, 1.5 percent of this subsample had grade eight or less education, 45 percent had high school, and 53 percent had post-secondary education. Approximately 30 percent of the sample reported having been off work with back pain at least once prior to the current injury, but fifty-five percent of the sample did not respond to this question. For this sample, the mean duration of injury at completion at time two was 105 days. Individual t-tests comparing the 137 participants included  Catastrophizing, Fear  83  in the final sample with the 210 used in the S E M analyses on time one scores for the pain related fear variables, and time two outcome variables (see Table 1) indicated no significant difference in scores on any of the measures. Similarly, comparison of the two groups on demographic characteristics indicated no significant differences For the analyses predicting time two return to work, participants who had returned to work at time one were dropped from the analysis. Of the 137 individuals included in the regression sample, 59 had returned to work by time two, leaving a sample size for that analysis of 84. The rate of return to work for his attenuated sample was relatively high (70 percent), but consistent with what might be expected in a sample composed of individuals with relatively uncomplicated pain complaints (i.e., no indication of serious physical injury) surveyed at the subacute stage. For example, Fritz, et al., 2001, found a similar return to work rate of 72 percent at a follow-up of four weeks in a group of acute/subacute pain patients. Statistical analyses indicated that none of the demographic variables from the return to work prediction sample differed significantly from the larger regression sample. A l l three subsamples were very similar demographically to that of Schultz et al. (2002) who also utilized a British Columbia Workers Compensation Board sample, although their sample included chronic pain patients in addition to those with subacute pain. The only substantive difference from the sample in that study was that the Schultz et al. sample consisted of more than 2/3 men, while the current samples were more evenly split between men and women. Measures The questionnaire package for time one data collection, including all instruments listed below are included in Appendix A . The time two questionnaires were identical to the first set, with some redundant demographic questions (e.g., age and sex) deleted. The time two demographic questions are included in Appendix B. Time two data were employed for the regression analyses. Time one data from the PASS, TSK, F A B Q and PCS as described above  Catastrophizing, Fear Table 1 Descriptive statistics for S E M and regression samples (time one data)  Variable  S E M sample (N = 210) Standard Mean Deviation  Regression sample (N= 137) Standard Mean Deviatio n  PCS Rumination  7.55  4.23  7.52  4.28  PCS Magnification  3.44  2.82  3.49  2.79  PCS Helplessness  7.36  5.42  7.23  5.22  PCS Total  18.35  11.42  18.24  11.28  PASS Fearful Appraisals  13.80  9.16  13.70  8.95  PASS Cognitive Interference  20.84  8.99  20.53  8.75  PASS Avoidance  24.25  8.57  24.48  8.48  PASS Physiological Symptoms  12.88  9.69  12.77  9.79  PASS Total  71.77  31.39  71.77  31.39  F A B Q Activity  21.26  6.42  21.70  6.48  F A B Q Work  40.39  10:02  39.84  9.7  F A B Q Total  61.65  14.07  61.59  13.87  TSK  39.73  8.68  39.60  8.39  McGill Pain Questionnaire - Total  13.43  7.88  13.68  7.59  McGill Pain Questionnaire - Sensory  10.75  6.00  10.73  6.02  McGill Pain Questionnaire Affective  2.68  2.57  2.70  2.43  Pain Behaviour Checklist Total  0.55  0.24  0.53  0.26  Pain Disability Index  35.93  14.87  37.19  13.78  CES-Depression  17.58  10.92  17.62  10.85  Highest Level of Pain  71.80  19.87  71.77  20.61  84  Catastrophizing, Fear  85  were utilized in all three sets of analyses. As the instruments are discussed in some detail above, they will be only briefly summarized here. The structural model analyses employed time one data for avoidance behaviour and pain intensity. For the regression analyses, time two data for outcome variables, depressive symptoms, perceived disability, average pain level and return to work status, as described below was utilized. Pain Related Fear and Catastrophizing (Used in all Three Sets of Analyses) PASS. The 40-item version of the PASS (McCracken et a l , 1992) was administered, since it subsumes the shorter version. Therefore, both the full and abbreviated (PASS-20; McCracken & Dhingra, 2002) versions could be analyzed. Both versions were designed to assess the same four factors, fearful appraisals, cognitive anxiety, escape/avoidance and physiological anxiety. Each subscale comprises 10 items in the original version and five items in the abbreviated (PASS-20) version. Items are answered on a scale of 0 = "never", to 5 = "always". Four items on the full version are reverse scored and the items in each scale are summed to achieve a score for that subscale. The reverse scored items were dropped from the abbreviated version. Each item consists of a specific pain anxiety symptoms and respondents are asked to rate how often the engage in the thought or activity. Internal consistency has been demonstrated to be moderate to high for the total score (full scale alpha = .94) and subscale scores (alpha = .81 to .89) in a chronic pain sample. Similar levels of internal consistency have been demonstrated for the instrument in a community sample with alphas ranging from .75 to .94 for the full and subscale scores. The validity of the PASS has been demonstrated in chronic pain samples through positive correlations with pain and self-perceived disability (McCracken and Gross, 1995; Crombez et al., 1999). It has also been demonstrated to predict performance on physical tasks in chronic pain patients (Burns et al., 2000). The abbreviated version of the Scale has been demonstrated to correlate highly with the original version at the full scale and subscale  Catastrophizing, Fear  86  level (.93 to .97), and the abbreviated instrument has also been demonstrated to have high internal consistency at the full scale (alpha = .91) and subscale (alphas = .75 to .86) levels. Analysis of reading level indicated a Flesch-Kinkaid Grade Level score of 4.8, indicating that the subjects in this study should have been capable of understanding the PASS instructions and items. FABQ. A l l 16 items from the original F A B Q item pool (Waddell et al., 1993) were administered to allow comparison of the full item pool with the items selected by Waddell et al. The two subscales of the F A B Q assess fear/avoidance beliefs about work and about general activity. Items are rated on a seven point Likert scale ranging from "strongly disagree" to "strongly agree". The 11-item F A B Q has been demonstrated to have adequate internal consistency at the subscale level with alphas for the work and activity subscales being .88 and .77 respectively in a chronic pain sample (Waddell et al. 1993). Test-retest reliability over 48 hours in that same sample was adequate (alpha = .74). However, Crombez et al. in a study using two chronic pain samples, (1999) found only the work subscale to have adequate internal consistency (alpha = .84 to .92), while the alphas (.52 and .57) for the activity subscale were much lower. Evidence for the validity of the scale comes from findings that both subscale scores were predictive of disability and work loss in a chronic pain sample, although the work subscale was a consistently better predictor (Waddell et al., 1993). Further evidence for the validity of the scale comes from findings that both the activity subscale (Burton et al., 1999) and work subscale (Fritz et al., 2001) were predictive of later self-perceived disability when measured early after pain onset and that the work subscale predicted work status (Fritz et al., 2001). Analysis of reading level indicated that the F A B Q should be understandable by individuals with a grade 4 to 5 reading ability, indicating it to be appropriate for this sample.  Catastrophizing, Fear  87  PCS. The PCS is comprised of 13 items designed to assess three components of exaggerated negative thoughts about pain, helplessness, magnification and rumination. Respondents are asked to provide a numerical rating from 0 = "not at all" to 4 "all the time", in response to questions each prefaced by "When I am in pain...". The rumination scale comprises four items (e.g., I keep thinking about how much it will hurt) while the magnification subscale contains three items (e.g., I keep thinking about other serious events", and six items make up the helplessness subscale (e.g., "There's nothing I can do to reduce the intensity of the pain"). A l l items are positively scored and subscale scores are derived by summing the items on that scale. A total catastrophizing scale is computed by summing the scale scores. The pain catastrophizing scale has been demonstrated to have moderate test-retest reliability ((Bentler, 1995) (Bentler, 1995) alpha = .70) over 10 weeks and high internal consistency (alpha = .87) in an undergraduate student sample (Sullivan et al. 1995). The 3 subscales have also been demonstrated to have moderate to high levels of internal consistency, with alphas of .91, .75 and .87 in another undergraduate sample (Osman et al., 1997). The validity of the scale has been demonstrated through prediction of pain ratings in chronic pain (Sullivan et al., 1998), and sport participant (Sullivan et al., 2000) samples. It has also been demonstrated to predict emotional distress in response to aversive stimulation (Sullivan et al., 1995). Analysis of required reading level indicated that the PCS should be understandable to individuals with a grade 5 to 6 reading level. TSK. The T S K is comprised of 17 items intended to measure fear of movement/(re)injury. Each item is rated on a numerical scale, ranging from 1 "strongly disagree" to 4 "strongly agree". Wording for each item is self contained (e.g., I'm afraid I might injure myself i f I exercise.") and respondents are asked to circle the number "that corresponds to how you feel". Four items (items 4,8,12 & 16) are reverse scored. As reported in the preceding discussion, a variety of factor structures have been indicated in previous studies, but the total  Catastrophizing, Fear  88  scale score achieved by summing the item scores is used in most of the published research. The scale has been demonstrated to have moderate reliability in an acute (< 4 weeks duration) pain sample (Swinkels-Meewise, Swinkels, et al., 2003), as indicated by internal consistency (a = .70) and test-retest reliability (a_= .76). Evidence for the validity of the scale in a similar sample comes from evidence that T S K scores predicted concurrently measured disability scores and indirectly predicted concurrently measured participation in activities. Analysis of reading level indicated that the T S K should be understandable to individuals with a grade 7 to 8 reading level, indicating that participants should have had no difficulty comprehending the items or instructions. Measures Specific to the Structural Model Pain intensity. Items from the Sensory subscale of the short-form M c G i l l Pain Questionnaire (SF-MPQ; Melzack, 1987; see Appendix A) were used as indicators for the pain intensity construct. The SF-MPQ was developed as an abbreviated version of the McGill Pain Questionnaire, for use in research settings where time is limited, but more information than that provided by single item scales is desired Melzack and Katz (1992). Consistent with the multidimensional view of pain (Melzack & Wall, 1965), the SF-MPQ consists of 15 items representing Sensory ( N = 11) and affective £N= 4) dimensions of pain. Each item is a word reflecting an aspect of the sensory or affective dimension, and the respondent is asked to rate the degree to which the word describes their pain on a 4 point scale ranging from 0) "none" to 3) "severe". The subscale scores of the SF-MPQ have been demonstrated to correlate highly with the major indices of the long-form MPQ. Further evidence for the validity of the scale comes from studies demonstrating it to be sensitive to changes in pain level following physical pain treatment interventions (Melzack, 1987). It has been employed in chronic pain populations (Gronblad,  Catastrophizing, Fear  89  Lukinmaa, & Konttinen, 1990; Guieu, Tady-Gervet, & Roll, 1991) and Melzack (1987) demonstrated that the instrument might be capable of discriminating between different pain syndromes. Avoidance behavior. As noted earlier, the PASS includes an escape/avoidance subscale. For the reasons discussed in the preceding section, it was considered important to have a measure of avoidance behavior other than the PASS avoidance/escape subscale in order to maximize the discriminant validity between the latent variables in this analysis. The Pain Behavior Checklist (PBC; Phillips and Jahanshahi, 1986) is a self-report questionnaire designed to assess multiple dimensions of pain behavior including avoidance, complaint, and help seeking (Phillips and Jahanshahi). The instrument consists of an introductory statement asking respondents if they "have done any of the following activities in relation to your present pain condition" and a series of statements (e.g., "avoid or minimize heavy lifting"), each referring to a particular pain behavior. Respondents check "yes", "no" or "not applicable". Phillips and Jahanshahi found the instrument to demonstrate adequate test-retest reliability over a one-week period, in a group of chronic headache sufferers (e.g., alpha = 0.72 for the total score and alpha = 0.77 for the avoidance subscale). The P B C has been used to study self-reported pain behavior in chronic headache (Phillips and Jahanshahi, 1986) and acute neck and back pain (Phillips, et al., 1991). The 33 items on the six subscales reflecting avoidance behavior in Phillips and Jahanshahi's study were included in the current questionnaire package. Pilot testing of the instrument indicated that respondents were confused by the wording of the introductory statement and so the phrase "For example, for question 1 i f you DO avoid night-clubs and dances, put a tick in the " Y E S " column". The questionnaire was scored by summing the yes responses in each category and dividing by the number of items endorsed with "yes" or "no" (total items minus not applicable). Consequently, each subscale score is the proportion of applicable activities in that  Catastrophizing, Fear  90  category, which were avoided. Outcome variables (regression analyses) Average pain over the preceding week. A 100 mm visual analogue scale (VAS) anchored with No Pain and Worst Possible Pain was employed as the index of pain level. Patients were asked to rate their average pain over the preceding week by marking the line with an X at the appropriate point between anchors. Average pain over the preceding week was utilized in case the participant was experiencing a temporary spike or lull I their pain at the moment of filling in the questionnaire. There is considerable evidence for the validity of V A S s in the measurement of pain intensity . Responses to V A S scales have been demonstrated to correlate with other selfreport measures of pain intensity (Paice & Cohen, 1997) and observed pain behaviour (Gramling & Elliot, 1992). V A S scales have been demonstrated to be sensitive to the effects of clinical intervention (Turner, 1982). Because they provide a large number of response categories (i.e. 101 when measured to the millimeter) the resulting data can be considered continuous and over group comparisons, there is even indication that V A S s provide ratio level data (Price & Harkins, 1987). Depressive Symptoms. Depressive symptoms were assessed with the Center for Epidemiologic Studies-Depression Scale (CES-D; Radloff, 1977). The CES-D has 20 items reflecting depressive symptoms. Respondents rate how often during the past week they experienced each symptoms on a 4 point numerical scale ranging from 0 = "rarely/none (less than one day)": to 3 = "Most/all of the time (5-7 days per week)". The scale has been demonstrated to be highly reliable, with Alpha coefficients ranging from .85 in a community population to .90 in a psychiatric population (Radloff, 1977). The scale has been used in pain research and Geisser et al. (1997) demonstrated that the CES-D discriminated between chronic pain patients who had major depression and those who did not. The CES-D has also been  Catastrophizing, Fear  91  demonstrated to be sensitive to changes in depressive symptoms (Santor, Zuroff, Ramsey, Cervantes, & Pacios, 1995). The CES-D contains fewer items reflecting somatic disturbance than other self-report measures of depression and therefore it may be more appropriate for use in pain patients Bradley & McKendree-Smith (2001). Perceived Disability. Perceived disability was assessed with the Pain Disability Index (PDI; Pollard, 1984). This measure consists of seven items each describing a particular category of life activities (e.g. family/home responsibilities) and the respondent is asked to rate the level of disability they experience in that category. Ratings are made on a 10 point numerical rating scale anchored with 0 = "no disability', and 10 = "total disability". Tait, et al., 1990) found that the instrument had high internal consistency (alpha = .86) in a sample of chronic pain patients. With regard to validity, Tait et al. found the authors also found scores on the instrument to predict significant differences in observed pain behaviours. The instrument has also been shown to discriminate between pain patients who are receiving financial compensation and those who are not, (Tait, et al.). The instrument is also face valid, which makes it simple for respondents to complete. Return to work. Participants were asked if they had returned to work at the time of completing the questionnaire and i f yes, the date. Procedure A list of claimants having low-back pain as a primary complaint was extracted from the W C B database on a weekly basis over the course of approximately 12 months. A total of 1477 claimants were mailed a set of time one questionnaires, including a cover letter, instruction sheet, a consent form and information on informed consent (see Appendix A), and asked to return the questionnaires and consent form in a prepaid envelope. The questionnaires were designed to take less than one hour to complete and claimants were offered twenty dollars compensation for  Catastrophizing, Fear  92  completing both sets of questionnaires. Those individuals who returned the time one questionnaire were mailed a set of time two questionnaires. The choice of three months post injury for time two data collection was based on the IASP guidelines which suggest that pain can be considered chronic when it has lasted longer than the expected time for tissue healing and in the case of uncomplicated soft-tissue injury to the back, three months is the typical maximum time for such healing. In addition, there is some indication in the literature that the three month point is pivotal in predicting longer term outcome (Phillips & Grant, 1987). Likewise, Krause (1994), argued that by the three to six months post injury period (phase 6 in his model), clinically, pain can be considered chronic and treatment focus should shift from acute pain intervention to reactivation. In order to maximize the reliability of the data, several steps were taken. First, the cover letter and instructions for the questionnaire packages specified that participants were to complete the questionnaires in a quiet space on their own. Participants were provided with a telephone number to contact the researcher should they have any questions or concerns. Telephone contact was attempted by a member of the research team approximately one week following the mailing to answer any questions participants might have, to assure that questionnaires had been received, and to encourage participants who chose to participate to complete the questionnaires in an appropriate environment and timely manner. As a check that the procedures were followed, participants were also asked five questions at the end of the questionnaire relating to the environment, their comprehension and the procedure they had followed in completing the questionnaire. The questions are included in Appendix A , as part of the questionnaire package. After subjects were dropped for excessive missing data, any missing data remaining was replaced with the mean score for that item.  Catastrophizing, Fear  93  Results CFA Results Each of the pain related fear and catastrophizing measures was subjected to confirmatory factor analysis, utilizing the EQS S E M program. For each measure, several potential factor structures were modeled and assessed. These included the factor structure implied in the recommended scoring procedures for that instrument (i.e. the subscales specified by the scale author(s)), in addition to any other factor structure supported by previous literature and a single factor model to test for unidimensionality. PCS Variances and covariances for the PCS items are displayed in Appendix C for the reader interested in replicating the following analyses. Three models of the PCS were tested and are displayed graphically in Figures 7 to 9. These were derived from previous research, and included a single factor model (Figure7) and two factor models (Figure 8; Osman et al., 1997), in addition to the original three-factor model (Figure 9; Osman et al., 1997; Osman et a l , 2000; Sullivan et a l , 1995; Sullivan et al., 2000; Van Damme et al., 2002). A l l items were found to be adequately distributed (skewness = -0.76 to 1.50; kurtosis 1.02 to 1.32). However, moderate multivariate kurtosis was indicated (normalized Mardia's coefficient = 10.59). Model 1 (single factor; see Figure 7 for graphical representation). The % value for the one factor model of 269.04 (df = 65, p_ < 0.001) was statistically significant and the model showed poor approximate fit by all indices (CFI = 0.89; S R M R = 0.55; R M S E A = 0.12 (90%CI; 0.107 to 0.137), except S R M R (0.55).  Catastrophizing, Fear  Catastrophizing, Fear Figure 9. Three factor model of PCS.  95  Catastrophizing, Fear  96  Model 2 (two factor; see Figure 8 for graphical representation). The two-factor model (X (64, N = 210)= 199.1 l , p < 0.001) was somewhat more appropriate, achieving good fit 2  according to the S R M R index (SRMR = 0.054) and the CFI value (0.932) was marginal, but the R M S E A value of 0.100 (90%CI; 0.083 to 0.115) indicated poor fit. Model 3 (three factor; see Figure 9 for graphical representation). The three factor model x (62, N = 210) = 131.44, p_ < 0.001 provided a good fit to the data based on all indices 2  (CFI = 0.96; S R M R = 0.042; R M S E A = 0.073 (90%CI; 0.56 to 0.090)). Summary of PCS CFA. A comparison of fit for the three models tested is presented in Table 2. The analyses support the contention that catastrophizing, as measured by the PCS, can best be conceptualized as consisting of three separate but correlated dimensions (Sullivan et al., 1995; Osman et a l , 1997; Osman et al., 2000; Sullivan et al., 2000; Van Damme et al., 2002) and that this factor structure holds for individuals with subacute low-back pain. The standardized item factor loadings are displayed in Table 3 and show that all items load highly on their hypothesized factor. The correlations between factors are displayed in Table 4. As can be seen, all the three subscales are highly correlated (r = .81 to r = .90). Correlations of this magnitude suggest that the three components might best be conceptualized as dimensions of a higher order catastrophizing factor. S E M can and has been used to test the viability of such higher order factor models. However, because of statistical identification issues, testing a higher order model with only three first order factors requires the imposition of specific model constraints such as fixing two of the higher order factor loadings to be equal (Byrne, 1989). The current author was not aware of a rationale for any such constraint in this case.  Catastrophizing, Fear  97  Table 2 Goodness of Fit for the Confirmatory Factor Analyses of the PCS Model  Df  CFI  SRMR  RMSEA  269.04  65  0.89  0.055  0.120  199.11  64  0.93  0.054  0.100  131.44  62  0. 96  0.040  0.073  MLY  1. One factor  3  2. Two factor  b  3. Three factor  2  Notes. M L , maximum-likelihood; CFI, comparative fit index; SRMR, standardized root mean residual; R M S E A , Root Mean Square Error of Residual. Osman e t a l , 1997 Sullivan, Tripp, Rodgers, & Stanish, 2000; Sullivan, Bishop & Pivik, 1995; Van Damme, Crombez, Bjittebier, Goubert, & Van Houdenhove, 2002; Osman et al., 1997; Osman et al 2000 a  b  98  Catastrophizing, Fear  Table 3 PCS Three-factor solution: loadings and factor labels Factor 1 2 3 Rumination Magnificati Helplessness on  Abbreviated item  9.1 can't seem to keep it out of my mind.  0.89  10.1 keep thinking about how much it hurts.  0.87  11.1 keep thinking about how much I want the pain to end  0.79  8.1 anxiously want the pain to go away.  0.71  6.1 become afraid that the pain may get worse. 13.1 wonder whether something serious may  0.88 0.75  7.1 keep thinking of other painful events. 3 It's terrible and I think it's never going to get better  0.63 0.86  4. It's awful and I feel it overwhelms me.  0.85  5.1 feel I can't stand it any more.  0.82  2.1 feel I can't go on.  0.77  1.1 worry all the time about whether the pain will end 12. There's nothing I can do to reduce the intensity of the pain.  Table 4 Factor intercorrelations for the three factor solution of the PCS Factor Rumination Magnification Magnification  .82  Helplessness  .81  .90  0.73  0.59  Catastrophizing, Fear  99  PASS. Variances and covariances for the PASS items are displayed in appendix D for the reader who is interested in replicating the following analyses. Five models of the factor structure of the PASS were tested. Along with the four factor structure proposed by McCraken et al. (1993), other models tested included a single factor model, the four factor model of the abbreviated version (PASS-20) of the instrument proposed by McCracken and Dhingra (2002) and an alternative four factor model and five factor model both reported by Larsen et al.(1997). A l l items were found to have acceptable univariate distributions (skewness = 0.63 to 1.47; kurtosis = 0.97 to 0.44). However, a moderate degree of multivariate kurtosis was also indicated (normalized Mardia's coefficient = 16.25. Model 1 (single factor: See Figure 10 for graphical representation). The single factor model provided a significant % value of 1780.64 (df =740, N = 210, p_ < 0.001). The S R M R 2  (0.073) suggested adequate fit, but the CFI (0.73) and R M S E A (0.082; 90%CI = 0.077 to 0.087) did not. Model 2 (4 Factor Original; McCracken et al., 1993; Osman et al., 1994; See Figure 11 for Graphical Representation). The original 4-factor structure provided a significant %  2  value (1486.49, df=734, N = 210, p < 0.001), but demonstrated reasonable fit to the data according to the S R M R (0.68) and R M S E A = 0.070 (90%CI; 0.66 to 0.76). However, the CFI (0.81) indicated substantial misfit. Model 3 (five factor; Larsen et al., 1997; See Figure 12 for graphical representation). The five factor model proposed by Asmundson also provided a reasonable fit for the data (x (730, N = 210) = 1473.07 , p_ < 0.001), according to the S R M R (0.68) and R M S E A = 0.070 (90%CI; 0.064 to 0.075), but the CFI (0.81) indicated substantial misfit.  Catastrophizing, Fear ;ure 10. Single factor model of PASS.  100  Catastrophizing, Fear  Figure 11. Four factor model of PASS (McCracken et al., 1993).  E36_ E38_  Catastrophizing, Fear  Figure 12. Five factor model of PASS (Larsen, et al., 1997).  E4 . E5 . E6 . E7 . E8 .  E39_ E40_  Catastrophizing, Fear  103  Model 4 (4 factor; Larsen et al., 1997; for graphical representation see Figure 13). The four factor model also provided a significant  (1532.22 , df = 164 g < 0.001). It provided a  marginally satisfactory fit, based on the based on the R M S E A (0.072; CI, 0.067 to 0.077) and S R M R (0.070), but the CFI (0.79) indicated substantial misfit Model 5 (PASS-20 four factor; McCracken & Dhingra, 2002; see Figure 14 for graphical representation). The four factor model of the PASS-20 (% = 334.74, df = 164, p < 2  0.001), provided a good fit based on the R M S E A (0.071; CI, 0.060 to 0.081) and S R M R (0.058) and approached the cutoff for an adequate CFI value (0.91). Given the relatively better initial fit, for this model as compared to the preceding ones, it was decided to attempt modification of this model. Examination of Lagrange Multiplier Test statistics indicated that the greatest improvement in fit would be achieved by allowing item 37 ("I worry when I am in pain") of the cognitive symptoms subscale to cross load on the fearful appraisals subscale. Theoretically, there are strong grounds for allowing this item to load on the fearful appraisal subscale. As argued earlier in this work, items on the fearful appraisal subscale of the PASS appear to be assessing catastrophizing. Although it was placed by the instrument authors in the cognitive symptoms subscale, "I worry when I am in pain" also appears to measure catastrophizing. Indeed, it is very similar to PCS item 1, which reads "I worry all the time about whether the pain will end." Additional support for allowing item 37 to load on fearful appraisals subscale comes from Larsen et al. (1997) who also found item 37 to load together with fearful appraisal items in their factor analysis of the PASS. The model was rerun allowing item 37 ("I worry when I am in pain") to load on both factor 1 in addition to factor 2. The resulting improvement in fit provided a CFI very close to the  Catastrophizing, Fear  Figure 13. Four factor model of the PASS (Larsen et al., 1997).  E13E16_  ^|  Item 35  E21— E25E29E33E 2 _ E 6 _  EIOE14-  E26_ E 3 _ E 3 _  E37E M -  E U E l _ E l _  E27E31E35-  Item 39  E39-  Item 7 E12E17E20E24E28_  Catastrophizing, Fear  105  Figure 14. Four factor model of PASS-20 (abbreviated item set; McCracken & Dhingra, 2001).  Catastrophizing, Fear  106  cutoff (.94), and slight improvements in all other indices (% (163, N = 210) = 282.67, p < 0.001), 2  R M S E A (0.059; CI, 0.047 to 0.070) and S R M R (0.05). With item 37 allowed to load on both factor 1 and factor 2 the loading on factor 2 was too low to achieve statistical significance (standardized estimate = 0.11, z-score =1.41), and loaded highly on factor 1 (standardized estimate = .70, z-score = 7.37). Therefore, the four factor model of the PASS-20 was run once more, specifying item 37 to load on factor 1 (fearful appraisals) but not on factor 2 (cognitive symptoms). This model resulted in significant % (278.00, df = 164, p_ < 0.001), acceptable S R M R = 0.050, R M S E A = 2  0.060 (90% CI = 0.047, 0.070) and improved but still marginal CFI (0.94). The modification predicted to have the next largest effect on fit according to the L M test was a correlation between the measurement errors for item 32 ("Pain makes me nauseous") and item 24 ("when I have pain, I feel dizzy or faint"). Both of these items represent vestibular symptoms, with nausea often accompanying dizziness. Therefore, it was considered appropriate to allow their error terms to correlate. One final modification allowing correlated errors for variable 12 (PASS item 24) and variable 15 (PASS item 32) provided adequate fit by all indices (% = 266.22, df = 163, p < 0.001), CFI = 0.95, S R M R = 0.048, R M S E A = 0.055 2  90% CI (0.043 to 0.067). Summary of PASS CFA analyses. A comparison of all models tested is presented in Table 5. Of the five models tested, a four factor model conforming to the original factor structure of the 40 item PASS (McCraken et al., 1993) applied to the 20 item version of the test (McCracken & Dhingra, 2002) provided the only adequate fit to the data. The modifications necessary to achieve adequate fit were specification of item 37 to load on the fearful appraisals  Catastrophizing, Fear  107  subscale rather than the cognitive symptoms subscale and specification of correlated error terms for item 24 and item 32. The standardized factor loadings are displayed in Table 6 and show that all items load highly on their hypothesized factor (all factor loadings were significant at g < 0.05). The correlations between factors is displayed in Table 7. As can be seen, the four subscales are highly correlated. It is not possible to compare these correlations to those found in previous work with the PASS-20, as the authors (McCracken & Dhingra, 2002) of the only published work using that instrument, do not report the interscale correlations. However, the range of correlations (e.g. r = 0.65 between fearful appraisals and avoidance, to r = 0.88 between fearful appraisals and physiological symptoms) and their pattern was quite similar to those found in community (Osman et al., 1997) and chronic pain (e.g. McCracken et al., 1993) samples utilizing the full PASS. FABQ Variances and covariances for the F A B Q items are displayed in Appendix E for the reader interested in replicating the results. It was anticipated that item 6, " M y pain was caused by my work or by an accident at work", would be irrelevant in the current sample, because selection criteria (i.e., W C B compensable injury) imply a work related injury, and the vast majority of participants would answer in the affirmative. Examination of the distribution of this item indicated that it was severely skewed (-5.5) and kurtotic (32.77) with 95 percent of subjects answering "completely agree" and no participant answering "completely disagree". Therefore, this item was dropped from the factor analyses.  Catastrophizing, Fear  108  Table 5 Goodness of Fit for the Confirmatory Factor Analyses of the PASS Model  MLy  df  CFI  SRMR  RMSEA  1780.64  740  0.73  0.073  0.082  1486.49 1473.07  734  0.81  0.068  0.070  730  0.81  0.068  0.070  1532.22  734  0.79  0.070  0.072  334:74  164  0.91  0.058  0.070  282.67  163  0.94  0.050  0.060  278.00  164  0.94  0.050  0.060  266.22  163  0.95  0.048  0.055  2  Original (40 Items) 1. One factor 2. Four factor 3. Five factor  b  4. Four factor  3  2  Abbreviated (PASS-20, 20 items) 5. Four factor 6. Four factor modification l 7. Four factor modification 2  d  e  0  8. Four factor modification 3  f  Note. M L , maximum-likelihood, CFI, comparative fit index, SRMR, standardized root mean residual, R M S E A , Root Mean Square Error of Residual. ' Larsen, Taylor, & Asmundson, 1997, McCracken, Gross, Hexum, & Semenchuck, 1993, . McCracken & Dhingra, 2002. ' Item 37 loading on fearful appraisal and cognitive symptoms, Item 37 loading only on fearful appraisal, . Item 37 loading only on fearful appraisal with correlated error terms for items 24 & 32. a  b  c  d e f  Catastrophizing, Fear  109  Table 6 Four-factor solution on abbreviated item pool (PASS 20): loadings and factor labels Abbreviated item  Factor Fearful Cognitive Escape/ Appraisals Anxiety Avoidance  PAS37 PAS5 PAS33 PAS25 PAS21 PAS1 PAS22 PAS26 PAS34 PAS 14 PAS23 PAS 15 PAS39 PAS7 PAS 19 PAS36 PAS 12 PAS 17 PAS24 PAS32  I worry when I am in pain. When I feel pain, I am afraid that something terrible will happen. When pain comes on strong, I think that I might become paralyzed or more disabled. Pain sensations are terrifying. When I feel pain, I think that I might be seriously ill. I think that if my pain gets too severe, it will never decrease. During painful episodes, it is difficult for me to think of anything besides the pain. When I hurt, I think about the pain constantly. I find it hard to concentrate when I hurt. I can't think straight when in pain. I avoid important activities when I hurt. I will stop any activity as soon as I sense pain coming on. I try to avoid activities, which cause pain. I go immediately to bed when I feel severe pain. As soon as pain comes on I take medication to reduce it. I find it difficult to calm my body down after periods of pain. I begin trembling when engaged in an activity, which increased pain. Pain seems to cause my heart to pound or race.. When I sense pain, I feel dizzy or faint. Pain makes me nauseous.  PhysioLogical  0.78 0.77 0.74 0.72 0.69 0.58 0.83 0.81 0.79 0.71 0.70 0.68 0.64 0.50 0.50 0.73 0.71 0.66 0.58 0.58  Catastrophizing, Fear  Table 7 Factor intercorrelations for the four factor solution on abbreviated item pool (PASS 20) Subscale Fearful Appraisals Fearful Appraisals  3  Cognitive Anxiety  3  Cognitive Anxiety 3  3  Escape/ Avoidance  0.75  Escape/Avoidance  0.65  .73  Physiological  0.88  .80  .67  Note. item 37 is included in the fearful appraisals subscale rather than the cognitive symptoms subscale as originally specified (McCracken & Dhingra, 2001) 3  110  Catastrophizing, Fear  111  Four models of the F A B Q factor structure were tested, 1) a single factor model applied to the full item set minus item 6 (15 items), a two factor model applied to the full item set minus item 6, 3) a single factor model of the attenuated item set, selected by Waddell et al. (1993), minus item 6, and 4) a two factor model of the attenuated item set minus item 6. The models are displayed graphically in Figures 15 to 18. The majority of the F A B Q items demonstrated approximately normal distributions, with only item 15 demonstrating moderately high skew (2.85) and kurtosis (7.86). The data were found to deviate moderately from multivariate normality (normalized Mardia's coefficient = 15.73). As with the preceding analyses, normal theory M L estimation with non-adjusted fit indices was employed. Model 1 (single factor model of full item set; for graphical representation, see Figure 15). Utilizing the complete item pool, the one factor model provided a poor fit to the data ( x = 666.19, df = 90, p < 0.001), CFI = 0.50, S R M R = 0.118, R M S E A = 0.175 90% 2  CI 0.162 to 0.187). Model 2 (two factor model offull item set; for graphical representation, see Figure 16) Likewise, the two factor model of the complete item pool provided better, but still poor fit to the data ( x = 491.59, df = 90, p < 0.001), CFI = 0.65, S R M R = 0.105, R M S E A = 2  0.147 90% CI(0.134 to 0.159). Model 3 ( one factor model of abbreviated item set; for graphical representation, see Figure 17). The fit of the single factor model on the abbreviated F A B Q item set also provided a poor fit to the data ( x = 270.51, df = 35, p < 0.001), CFI = 0.67, S R M R = 0.115, 2  R M S E A = 0.179, 90% CI = 0.159 to 0.199).  Catastrophizing, Fear  112  Figure 15. One Factor Model of F A B Q (Waddell, et al., 1993); 16 items minus item 6.  Catastrophizing, Fear Figure 16. Two factor model of full item set (Waddell, et al., 1993); 16 items minus item 6.  El  -^1  Item 1  E2  -T\  Item 2  E3  ->|  Item 3  E4  Item 4  E5  Item 5  E6  s/  Item 8  E7 E8  N  ?  Item 9 Item 10  E9 E10  Item 7  >  Item 11  Ell  Item 12  E12  Item 13  E13  Item 14  E14  Item 15  E15  Item 16  &  Catastrophizing, Fear  114 -  Figure 17. One factor model of abbreviated F A B Q item set (Waddell et al., 1993).  El E2 E3 E4 E5  Fear/Avoidance Beliefs Fl  E6 E7 E8 E9 E10  Note: Item number in brackets refers to the number of the item on the original questionnaire  >> )  Catastrophizing, Fear  115  Model 4 (two factor model of abbreviated item set; for graphical representation, see Figure 18). The two factor model provided a good fit based on S R M R (0.062) and it approached adequate fit, based on the R M S E A (0.097, 90% CI = 0.074 to 0.118) and CFI = 0.91. The Chi square statistic was significant (% = 97.30, df = 34, p < 0.001). Because the fit 2  values for this model were marginal, it was decided to attempt modification. Inspection of the L M statistics for model 4 indicated that allowing correlated errors for items 4 ("I should not do physical activities which (might) make my pain worse") and 5 ("I cannot do physical activities which (might) make my pain worse") would significantly improve model fit. Given that the wording is nearly identical for these two items, it appears reasonable that they might share variance beyond that accounted for by their common factor (the activity subscale) The model was rerun with this path freed and resulted in adequate fit (% = 64.44, df 2  = 32, CFI = .96, S R M R = 0.060, R M S E A = 0.073, CI = 0.045 -0.094). The correlation between the two error terms in the final model (0.45), was statistically significant at p < 0.05. Summary of FABQ CFA analyses. A comparison of fit for all factor models of the F A B Q tested is presented in Table 8. The two factor model of the F A B Q items selected by Waddell et al., 1993 (minus item 6: " M y pain was caused by my work or by an accident at work") was the only model to provide adequate fit to the data. Factor loadings for items on the two scales in this model are displayed in Table 9. As can be seen in Table 9, some of the loadings were quite low, particularly for factor 2, but all were statistically different from zero at p < 0.05. Item 7 and item 15 achieved loadings below .40. For item 15, the low loading may have been due to the degree to which responses to that item were not normally distributed. Item seven demonstrated relatively better distributional characteristics but also deviated from normality. Two (skew = -2.25, kurtosis = 4.71). The correlation between the two F A B Q factors for this model was moderate (0.52) but statistically significant (p < 0.05).  Catastrophizing, Fear Figure 18. Two factor model of abbreviated F A B Q item set (Waddell et al., 1993).  Note: Item number in brackets refers to the number of the item on the original questionnaire  Catastrophizing, Fear  117  Table 8 Goodness of Fit for the Confirmatory Factor Analyses of the FABQ l Original 16 items (minus item 6) 1. One factor  M  o  d  MLx  e  CFI  SRMR  RMSEA  666.39  90  0.50  0.118  0.175  491.39  89  0.65  0.105  0.147  207.51  35  0.67  0.115  0.179  97.3  34  0.91  0.062  0.094  a  2. Two factor  3  Recommended 11 items (minus item 6) 3. One factor  M  a  4. Two factor" 5. Two factor  with 1 correlated error 64.44 33 036 0.060 0.073 Note. ML, maximum-likelihood; CFI, comparative fit index, SRMR, standardized root mean residual, RMSEA, Root Mean Square Error of Residual. ' Waddell etal 1993 b Correlation between error terms for items 3 & 4 freely estimated b  a  Catastrophizing, Fear  118  Table 9 Four-factor solution of FABQ on abbreviated item pool (Waddell et al., 1993) minus item 6: loadings and factor labels Abbreviated item  Factor Activity  FABQ 3 Physical activity might harm my back.  0.87  FABQ 2 Physical activity makes my pain worse.  0.71  FABQ 5 I cannot do physical activities, which (might) make my pain worse.  0.60  FABQ 4 I cannot do physical activities which (might) make my pain worse.  0.58  Work  FABQ 11 My work might harm my back.  0.88  FABQ 10 My work makes or could make my pain worse.  0.87  FABQ 12 1 should not do my normal work with my present pain.  0.54  FABQ 9 My work is too heavy for me.  0.46  FABQ 7 My work aggravated my pain.  0.32  FABQ 15 I do not think that I will ever be able to go back to work.  0.25  Catastrophizing, Fear  119  TSK For the reader interested in replicating the following analyses, variances and covariances for the TSK items are displayed in Appendix F. Models of the T S K tested included 1) a single factor model of all 17 items, 2) the four factor model of 12 items reported by Vlaeyen, Kole-Snijders, Rotteveel, et al. (1995), 3) the two factor model of 13 items reported by Clarke et al (1996), Geisser et al (2000), and Goubert et al (2004), and 4) the two factor model reported by SwinkelsMeewise, Swinkels, et al (2003). The various factor models are displayed graphically in Figures 19 to 22. Model 1 (single factor). A model with all items loading on a single factor, provided a significant % (119, N = 210) = 306.86, p < 0.001, a marginally acceptable fit based on the 2  SRMR (0.74), but a poor fit based on the CFI = 0.76 and R M S E A = 0.087 90% CI = 0.075 to 0.099). Examination of the residuals and L M test statistics revealed no theoretically plausible modifications, which would have substantially improved this fit. Model 2 (fourfactor, Vlaeyen, Kole-Snijders, Rotteveel, et al, 1995; 13 items). Because Vlaeyen, Kole-Snijder, Rotteveel, et al. (1995), did not report the factor loadings for items loading less than .40 on their four factor model (items 1, 5, 7, 8, 16 & 17), those items could not be assigned to a specific subscales in the current study and therefore were dropped from the analysis (for graphic representation of this model see Figure 20). The four factor model of the remaining 12 items provided a significant % (47, N = 210) = 107.69, p < 0.001. The SRMR 2  (.066) was somewhat lower than that of the single factor model and fell in the acceptable range, but the CFI value (0.85), indicated poor fit. Likewise, the R M S E A (0.093, 90% CI = 0.074 to 0.111) indicated poor fit. Again, the modification indices and residuals suggested no modifications, which appeared theoretically justifiable.  Catastrophizing, Fear  120  Figure 19. T S K model 1; single factor model of T S K (Vlaeyen, Kole-Snijders, Rotteveel, et al., 1995).  tem 1 tem 2 FT  tem 3  Pd  tem 4  FS  tem 5  Ffi  tem 6  F7  tem 7  FX  tem 8  FQ  tem 9  E10"  tem 10  C1 i  tem 11  F.l?  tem 12  PU  tem 13  F14  tem 14  Fl S  tem 15  F1fi  tem 16  F17  tem 17  K  Fear of movement/ (re)injury Fl  C atastrophizing, Fear  121  Catastrophizing, Fear  122  Figure 21. T S K model 3; two factor model of T S K (Clark, et al., 1996; Geisser, et al.., 2000; Goubert et al., 2004).  Ell E3 E6 E9  Item 1 Item 2 Item 9 Item 10  El  Item 13  E14  Item 14  E4  Item 15  E12  Item 17  E2  Item 3  E13  Item 5  E15  Item 6  E10  Item 7  E10  Item 11  Catastrophizing, Fear  123  Catastrophizing, Fear  124  TSK Model 3 (Two factor; Clark et al., 1996; Geisser et al, 2000; Goubert et al, 2004; 12 items). The two factor model of 13 items (reverse scored items 4, 8, 12, 16 not included) is displayed graphically in Figure 21. This model (x = 168.12, df = 64) produced a SRMR 2  (0.066) value indicating adequate fit, but other indices (CFI = .83, R M S E A = 0.088, CI = 0.072 to 0.104) indicated poor fit. The modification indices did not suggest any theoretically justifiable modifications to improve this model. Model 4 (Two factor; Swinkel-Meewisse, Swinkel et al., 2003; 13 items). The two factor model reported by Swinkel-Meewise et al. (2003) to fit well in a sample of acute/subacute pain patients failed to approach adequate fit in this study (x = 170.08, df = 64, 2  CFI = .83, S R M R = .065, R M S E A = 0.089, 90% CI = 0.073 - 1.05). Summary of modelfitfor TSK CFA analyses. The fit statistics for all four models of the factor structure of the T S K are presented in Table 10. As indicated there, none of the previously reported factor structures approached fit values indicating adequate fit to the TSK data in the current analyses. Correlations and factor loadings are not reported here because such values are not reliable when derived from a poor fitting model. Table 10. Goodness of Fit for the Confirmatory Factor Analyses of the T S K ML Model Original 16 items (minus item 6) 1. One factor ' 306.86 2  X  Df  CFI  SRMR  RMSEA  119 47 64 64  0.76 0.85 0.83 0.83  0.074 0.066 0.066 0.065  0.087 0.093 0.088 0.089  a  a  2. Four factor  b  b  3. Two factor 4. Two factor  0  d  107.69 168.12 170.08  Note. M L , maximum-likelihood; CFI, comparative fit index; SRMR, standardized root mean residual; R M S E A , Root Mean Square Error of Residual. Vlaeyen, Kole-Snijders, Rotteveel, Roesink, and Heuts, 1995 Vlaeyen, Kole-Snijders, Rotteveel, Roesink, and Heuts, 1995 Clark et al., 1996; Geisser et al.„ 2000; Goubert et al., 2004 Swinkel-Meewisse et a l , 2003 a  b  0  d  Catastrophizing, Fear  125  Test of theoretical structure Based on the proposed model, indicators were selected to represent catastrophizing, fear of pain and fear of activity constructs. Discussion of the process for selecting the final set of indicators is provided below. In addition, two other scales were used as indicators of pain intensity and avoidance. Structural equation modeling with EQS was also utilized to test the five hypothesized models of pain related fear including the target (expected) model. The first step in this analysis was to derive an empirically viable measurement model. Measurement Model The measurement model contained all latent variables of interest and their indicators, but no directional relationships were specified between the latents and they were allowed to correlate freely. Adequate fit of the measurement model was a necessary condition for testing the more constrained models containing directional hypotheses.  Several factors were  considered in this process, including statistical and theoretical considerations and post hoc examination of the confirmatory factor analyses. With regard to statistical considerations, there is value in having multiple indicators for each latent. Having a number of observed indicators for each construct allows for one of the most beneficial aspects of S E M , that is the ability to account for and model imprecision in the measurement of the latent constructs (Hoyle, 1995). Secondly, a necessary condition for identification of structural equation models is that there are enough data points (covariances between observed variables) to estimate all specified paths with residual degrees of freedom (Ullman, 2001). Identification means that there is at least one unique solution for fitting the data to the model. In order to test a model, it is necessary that it be over-identified, that is the  Catastrophizing, Fear  126  number of relationships between indicators must exceed the number of parameters estimated. As the ratio increases the power to detect misspecified models increases. However, there are also reasons to avoid too many indicators. As the ratio of observations (N) to indicators decreases, the reliability of the derived parameter estimates is reduced. At a minimum, the number of indicators cannot exceed the number of observations (Ullman, 2001). The recommended minimum ratio of observations to indicators is generally in the range of five to one (Hoyle, 1995). Given the above considerations and the sample size of 210 for the current analysis, the goal was to select approximately 40 items. The main goal of the analysis was to test the hypothesized conceptualization of pain related fear as three separate but related constructs: fear of pain, fear of activity and catastrophizing, and secondly to test the hypothesized relationships between those constructs within the context of a larger model predicting avoidance from pain. Since the proposition that pain related fear can be divided into catastrophizing, fear of pain, and fear of activity was based partly on the content of existing instruments used to measure pain related fear, and the overall goal of this work was to speak to theoretical issues in pain related fear as it is currently measured, it was considered desirable to keep these constructs as true to the original instruments (PCS, PASS, F A B Q & TSK) as possible. However, it was not possible to utilize all four of the instruments in their entirety, because this would inflate the number of items well beyond 40. Therefore, it was necessary to select a subset of items. Apriori inspection of the instruments and existing research suggested that there would be overlap and redundancy in some of the instruments. For example it was apparent that the PASS fearful appraisal subscale tapped into pain-catastrophizing and therefore might  Catastrophizing, Fear  127  reasonably be dropped. Inspection of the relationship between the PCS total score and the PASS fearful appraisal subscale, consistent with previous findings (Crombez, Vlaeyen et al., 1999) confirmed the strength of this relationship. The correlation between the PCS total score and PASS fearful appraisal scale score (including item 37) in the current sample, was r = 0.75. Another benefit of dropping the PASS fearful appraisal subscale was that it would maximize discriminant validity between the constructs by not having items from the PASS represent two different constructs (fear of pain and catastrophizing). Utilizing the PASS fearful appraisals subscale to indicate catastrophizing and other scales to indicate fear of pain would confound covariance between the two constructs and covariance due to common measurement. A final reason to not include the PASS fearful appraisal items as indicators of catastrophizing is that it would complicate the structure of the catastrophizing latent, by introducing items with as yet unknown loadings. It is possible that PASS items would load across the three existing factors or they might create a fourth factor. While this issue is interesting, it was beyond the scope of the current work to address it. The final consideration for the selection of items was examination of the empirical results from the preceding CFAs. Since the fit of individual components of the measurement model (i.e. each latent and its indicators) constrained the maximal fit achievable for the total measurement model, it was considered important to utilize indicators which had been found to adequately measure the constructs in the CFAs. For example, since no adequate factor structure was found to describe the TSK, including T S K items, it would be unlikely that T S K items could be included in an adequately fitting overall measurement model.  Catastrophizing, Fear  128  Description of Measurement Models for Each Construct Catastrophizing. For catastrophizing, the potential items included all items from the PCS and items from the PASS fearful appraisals subscale. As stated above, since other PASS subscales were to be utilized as measuring fear of pain, it was considered appropriate to drop the PASS fearful appraisal items in order to maximize discriminant validity. Further rationale for this decision was the finding that the three factor model of pain catastrophizing was well validated both in the preceding C F A in the current work and in previous work. The three factors of catastrophizing appear to be measured with reasonable efficiency by 13 items of the PCS. Avoidance. A similar rationale for maximizing discriminant validity led to the decision to use a measure of avoidance behavior other than the PASS avoidance subscale, namely the Pain Behavior Checklist. Therefore, the PASS avoidance subscale was also dropped at this stage. Each of the six Pain Behavior Checklist subscale scores was specified as an indicator of avoidance behavior. Fear of pain. After dropping the fearful appraisal and avoidance items of the PASS, the remaining items fell on two subscales, the cognitive anxiety and physiological anxiety subscales. The items from the cognitive symptoms and physiological symptoms subscales were specified to load on their respective subscale, and both of the subscales were specified to load on a higher order Fear of Pain factor. Although item 37 was intended by the scale authors (McCracken & Dhingra, 2002) to assess cognitive anxiety, in the current work that item was found to load more highly on the fearful appraisal subscale. Therefore, it was not considered appropriate to include item 37 as an indicator of cognitive symptoms and it was dropped.  Catastrophizing, Fear  129  Fear of Activity. With regard to the fear of activity, both the T S K and F A B Q contained subscales, which were expected to measure this construct. However, confirmatory factor analysis of the TSK did not yield support for any of the specified factor structures. In contrast, the F A B Q avoidance scale items were indicated to load together on a stable and meaningful factor, which appeared to reflect avoidance beliefs about activity. Hence, the items from that subscale were selected as indicators for the fear of activity construct. Pain intensity. The 11 items from the M P Q sensory subscale were specified to load on a factor representing that construct. Testing and Modification of Measurement Model Testing of the overall measurement model was conducted in two steps in order that sources of misfit could be more easily isolated. First, a measurement model of the three pain related fear constructs was tested. Based on the foregoing discussion, the measurement model portrayed in Figure 23 was constructed and subjected to S E M . As indicated in Figure 23, the 3-factor model of the PCS was imposed on the PCS items, with all three subscales loading on a single higher order factor labeled "catastrophizing". The Items from the cognitive and physiological subscales of the PASS-20 were specified to load on their respective PASS factor, with both of those factors loading on a single higher order factor labeled "fear of pain". The five activity items from the F A B Q were specified to load on a single factor labeled "fear of activity". Correlated error terms were carried over from the C F A analyses resulting in two such specifications, the error terms for items 24 and 32 from the PASS, and the error terms for F A B Q items 4 and 5. The rationale for these correlated errors is discussed in the preceding C F A results. The three main latent variables fear of pain, fear of acting and catastrophizing were permitted to freely correlate.  Catastrophizing, Fear  131  Testing of the model depicted in Figure 23 resulted in a significant % , (427.08, df= 2  289, p< .001) but other indices suggesting good fit (CFI = 0.96, S R M R = 0.051, R M S E A 0.048, CI 0.038 to 0.057). The final step in deriving the measurement model was to add items representing avoidance, and pain intensity. The full measurement model is portrayed in Figure 24. Each of the six PBC subscales score was specified to load on a single factor representing activity avoidance. The 11 items from the Sensory subscale of the M P Q were specified to load on a single factor representing pain intensity. This model provided a significant % (1208.42, df = 843, p< 0.001). The R M S E A 2  (0.046, 90% CI= 0.040 to 0.051) and SRMR (0.058) indicated adequate fit, and the CFI value (0.91) approached a value indicating good fit. Although the fit at this point was marginally acceptable, modification was pursued to improve fit in order allow for testing of the structural model, which being more constrained than the measurement model, was expected to result in poorer fit. Examination of the residuals indicated that covariances relating to items 2, 3 and 4 of the M P Q were producing the largest residuals. The Lagrange multiplier test indicated that allowing correlations between the error terms from items 2 & 4 and between the error terms for items 3 & 4 were the modifications most likely to improve model fit. When one error term correlates with two others, such as item 4 did in this case, it is considered most appropriate to allow correlation between all three error terms (Ullman, 2001). Examination of item 2 ("shooting"), item 3 ("stabbing") and item 4 ("sharp") suggested a commonality, namely sensations which might be characterized as piercing and brief in duration. Descriptors for the other M P Q sensory items might best be characterized as describing more dull, protracted sensations (e.g., "tender" & "heavy"). This commonality  Catastrophizing, Fear  132  Catastrophizing, Fear  133  between items 2, 3 and 4 was likely contributing to shared variance between the items, beyond that accounted for by their common loading on the M P Q sensory scale and therefore it was considered appropriate to allow their error terms to correlate in the model. Fit indices and comparison of the initial and respecified measurement models is displayed in Table 11. Respecification of the measurement model allowing three correlated errors (MPQ2 & MPQ3, MPQ3 & MPQ4, MPQ2 & MPQ4) provided a significant % (1119.88, df = 840, but 2  good fit by S R M R (0.058) and R M S E A (0.041, CI = 0.035 to 0.047), and a marginal value for the CFI (0.94). The improvement in  ( y£ = 88.74,  df= 3) achieved by allowing the  correlated errors was substantial and statistically significant (g < 0.001). Although this model did not quite meet the threshold for adequate fit specified apriori due to the CFI value of 0.94, recall that Hu & Bentler recommended values "approaching .95" for the CFI. In addition, examination of the residuals and modification indices did not suggest further modifications, which would have been theoretically justifiable. Therefore, the measurement models for each latent variable from this model were utilized to test the structural models. Structural analysis Alternative Models As indicated earlier, a series of alternative models were devised for comparison to the hypothesized model. Each model in the series is designed to test the importance of a particular hypothesis implied in the hypothesized model. Model 1. The first alternative model is depicted in Figure 25. This model included direct paths from pain to catastrophizing, fear of pain fear of activity and avoidance and direct paths from catastrophizing fear of pain, fear of activity to avoidance.  Catastrophizing, Fear  Table 11 Goodness of Fit for Measurement Model  Model Adf  A%  2  E of A x  2  Model 1 3 FOP Model 2 Full Model 2 Full Revised  145.29  3  < 0.01  MLx  2  df  CFI S R M R  RMSEA  90% CI RMSEA  427.08  289  0.96  0.051  0.048  0.038-0.057  1208.42  843  0.91  0.065  0.046  0.040-0.051  1119.68  840  0.94  0.058  0.041  0.034-0.047  Note:A% , change in % from preceding model; Adf, change in df from preceding model; M L , maximum-likelihood; CFI, comparative fit index; SRMR, standardized root mean residual; R M S E A , Root Mean Square Error of Residual; 90% CI R M S E A , ninety percent confidence interval endpoints for R M S E A value. 2  2  134  Catastrophizing, Fear Figure 25. Structural model 1 (first alternative model).  135  Catastrophizing, Fear  136  Model 2. The second alternative model is displayed in Figure 26. This model differs from model 1, in that the direct path from pain to avoidance is not specified. This model therefore implies that the relationship between pain and avoidance is fully mediated by catastrophizing and pain related fear. Model 3. The third alternative model is depicted in Figure 27. This model differs from model 2 in that the direct path from catastrophizing to avoidance is not specified. Therefore, this model implies that the effect of catastrophizing on avoidance is totally mediated by the pain related fear variables. Model 4. The fourth hypothesized model, is depicted in Figure 28. It differs from model 3 in that the direct paths from pain to the two pain related fear variables are not specified. This model therefore implies that the relationship between pain and the pain related fear variables is fully mediated by catastrophizing. Model 5. is the hypothesized model (depicted in Figure 29) and differs from model 4 in that a path from fear of pain to fear of activity is specified. This model as compared to model 4 implies that the path from catastrophizing to fear of activity is partially mediated by fear of pain. Assessment of Model Fit A comparison of all structural models tested is reported in Table 12. The Chi square value and fit index values used to assess the models (CFI,  SRMR & RMSEA), in addition to the  1 change statistic between adjacent models are reported there. Model 1. The first alternative model tested (see Figure 22) provided a statistically significant  x  2  value of  1242.36, df = 843 (p < 0.001). The RMSEA (0.049, 90% CI = 0.043 -  0.054) indicated adequate fit and the CFI value 0.91 SRMR (0.100) indicated poor fit.  approached an acceptable level, but the  Catastrophizing, Fear Figure 26. Structural model 2 (second alternative model).  137  Catastrophizing, Fear Figure 27. Structural model 3 (third alternative model).  138  Catastrophizing, Fear  Figure 28. Structural model 4 (fourth alternative model).  139  Catastrophizing, Fear  140  Table 12 Goodness of Fit for Structural Model  Model  Ax  Adf  2  p of  Ax  2  ML  2 X  df  CFI  SRMR  RMSEA  90% CI RMSEA  Model 1  1242.36  843  0.91  0.101  0.048  0.042-0.053  Model 2  1243.68  844  0.91  0.101  0.048  0.042-0.053  Model 3  1122.25  843  0.94  0.058  0.039  0.032-0.045  Model 4 Model 5 Ho: Target Model  12.86  2  < 0.001  1135.11  845  0.94  0.063  0.041  0.034-0.046  0.34  -1  N.S.  1134.77  844  0.94  0.063  0.041  0.034-0.046  df  CFI  SRMR  RMSEA  90% CI RMSEA  Post Hoc revisions to model 3 : Model  Ax  2  Model 3 Revision 1  . vs Model 3 +1.87  Model 3 Revision 2  vs Rev.l +2.17  Adf p of  Ax  2  ML  2 X  N.S.  1124.12  844  0.94  0.059  0.040  0.033-0.046  N.S.  1126.29  845  0.94  0.060  0.040  0.033-0.045  Note: Significance of change only presented for adequately fitting models; Ax , change in x from preceding model; Adf, change in df from preceding model; ML, maximum-likelihood; CFI, comparative fit index; SRMR, standardized root mean residual; RMSEA, Root Mean Square Error of Residual; 90% CI RMSEA, ninety percent confidence interval endpoints for RMSEA value. 2  2  Catastrophizing, Fear  141  Model 2. The second alternative model, displayed in Figure 23 was tested next. As indicated in the Figure that model as compared to model one involved dropping the direct path from pain intensity to avoidance. That model provided a % of 1243.68 (df = 844, p < 0.001). 2  Again, the R M S E A (0.049, 90% CI = 0.043 - 0.054), suggested adequate fit, the CFI (0.91) was marginal but the S R M R (0.99) indicated substantial misfit. Despite the additional degree of freedom achieved by dropping the direct path between sensory pain and avoidance, the fit did not significantly worsen (Ax = 0.68, df = 1 p > 0.05), suggesting that the path was not 2  necessary. Model 3. The third alternative model was tested next (see Figure 24). This model differed from model two in the absence of a direct path from catastrophizing to avoidance and the addition of direct paths from catastrophizing to fear of pain and fear of activity. In essence, these changes test the hypothesis that the relationship between catastrophizing and avoidance will be fully mediated by fear of pain and/or fear of activity. For this model, although x was 2  significant (1122.25, df = 843, p_ < 0.001), the fit indices all indicated reasonable fit (CFI = 0.94, SRMR = 0.058, R M S E A = 0.039 (90% CI = 0~.032 - 0.045). The finding of adequate fit for this model and poor fit for model two, indicated that dropping the path from catastrophizing to avoidance and replacing it with paths to fear of pain and fear of activity was warranted. This suggests that the prediction of avoidance from catastrophizing is completely mediated by fear of pain and/or fear of activity. Model 4. The fourth alternative model is portrayed in Figure 25. This model differed from model three in that model four had no direct paths from sensory pain to either fear of pain or fear of activity. Again the x was significant (11.35.11, df = 845, p < 0.001), but the fit 2  indices while slightly worse than those of the previous model, all indicated reasonable fit (CFI =  Catastrophizing, Fear  142  0.94, SRMR = 0.063, R M S E A = 0.041 (90% CI = 0.034 - 0.046). The A% between this model 2  and the preceding one was 12.86 on 2 df, which is statistically significant at p < 0.001, indicating that model three provided better fit to the data, despite two fewer degrees of freedom. Therefore, the direct path between pain intensity and fear of pain and/or the direct path from pain intensity to fear of activity improved model fit. Model 5. As previously indicated in Figure 7, model 5, the hypothesized model,' differs from model 4 in that a path from fear of pain to fear of activity is specified. According to the fit indices (CFI = 0.934, S R M R = 0.063, R M S E A = 0.041, 90% CI = 0.034- 0.046), the model provided adequate fit to the data, despite the significant % value (1122.25, df = 843, p_< 0.001). 2  However, comparison of this model to model 4 indicated that the improvement in model fit achieved by adding the path from fear of pain to fear of activity did not significantly improve model fit (Ax = 0.34, Adf = 1, p>0.05), indicating that the path from fear of pain to fear of 2  activity was not warranted. Given that the fit of model five was inferior to that of model four, and that the fit of model four was inferior to model three, the fit of model 5 is obviously significantly worse than the fit of model three. Comparison of model five to model three, the best fitting model to this point, revealed that with the loss of one degree of freedom, model three provided a significantly better fit to the data (Ax = 12.52, Adf = 1, p_ < 0.001). 2  Modification of Best Fitting Model. Given the superior fit of model three, it was decided to examine the parameter estimates for that model to determine the importance of individual paths. Examination of the parameter estimates for model 3 indicated that all fell within the acceptable range, and most values were different from 0 at a statistically significant level (z > 1.96, p < 0.05). However, z-scores for two paths were not significant, the path from pain intensity to fear of pain (z =1.40, p > 0.05), and the path from fear of activity to avoidance (z =  Catastrophizing, Fear  143  1.36, p > 0.05). Since these paths did not appear to be significantly contributing to the model, it was decided to drop them one at a time to determine if model fit could be retained while making the model more parsimonious. As a first step, the path from Pain Intensity to Fear of Pain was removed, as that path was not specified in the hypothesized model. Fit indices for that model are reported in Table 12. The x value was statistically significant, but the other indices, (CFI = 0.94, SRMR = 0.059, 2  R M S E A = 0.040 (90% CI = 0.033 to 0.045) deteriorated only slightly from the original model 3. As indicated in Table 12, the A x (df = 1) value of 1.87 was not statistically significant 2  despite the increase in df, indicating that removing the path did not significantly worsen fit. After dropping the path from pain intensity to fear of pain the path from fear of activity to avoidance, although marginally increased, did not differ significantly from zero (z = 1.49, p > 0.05). Therefore, the path from activity to avoidance was dropped and the model re-tested. The fit indices were quite similar to those from the first revision (X = 1126.29, df= 845, p_< 2  0.001, CFI = 0.94, S R M R = 0.060, R M S E A = 0.040, (90% CI = 0.033 to 0.045). Again, the Ax value from the first revision of 2.17 (Adf = 1, p > 0.05) was not statistically significant, 2  indicating that this model accounted for the data as well as the first revised model with fewer estimated parameters. Given the relative increase in parsimony achieved by dropping two paths, this model was considered superior to the original model three. Parameter estimates for this final model are reported in Figure 26. A l l parameter estimates were statistically significant at a p of less than 0.05. The R (0.19) for the prediction of avoidance in the model indicated 2  that 19 percent of the variance in avoidance behaviour was accounted for by the model.  Catastrophizing, Fear Figure 29. Best fitting structural model with standardized parameter estimates.  Note: Measurement level of model is not displayed, but corresponds to that displayed in Figure 24.  Catastrophizing, Fear  145  Prediction of Outcome Separate regressions were used to compare the time one Pain Related Fear variables in the prediction of time two average pain (VAS), perceived disability (PDI), depressive symptoms (CES-D) and return to work.  Two sets of analyses were used to predict each dependent  variable. 1) To determine the effect of all pain related fear total scale scores, the total scores for time one PASS, PCS, F A B Q and T S K were entered as a group in a separate equation for each of the four dependent variables (4 separate analyses), and 2) to determine whether the component subscales of each pain related fear instrument contributed to each of the dependent variables, all subscales for each of the instruments were entered as a block in four separate analyses for each of the four dependent variables (16 separate analyses). For the TSK, as there is no replicated factor structure for the scale in a sample with recent onset pain, the full scale score was entered in a separate predictive analysis, independent of the other scales (PCS, PASS & FABQ). The regression analyses were informed by the findings from the confirmatory factor analyses. Specifically, the PASS scores were calculated in accordance with the best fitting factor model for that instrument, the 20 item set with item 37 dropped because of the cross loading. Likewise, F A B Q item 6 was dropped in calculating the scores for that instrument. PCS scores and T S K scores were calculated in accordance with the authors' specifications. In the case of the T S K this was because no adequate factor structure could be determined in the current work and for the PCS, it was because the author's (Sullivan et al, 1995) proposed factor structure was found to hold in this sample. A hierarchical procedure was used for all regression analyses with five control variables entered as an initial block followed by a second block containing all of the variables of interest for that particular analysis. The variables entered in the control block were age, sex, and duration of pain at time one response, duration of pain at time two response and time one pain  Catastrophizing, Fear  146  level. Age and sex were included because each has been demonstrated to be related to both pain related fear (Sullivan et al., 2001) and persistence of chronic pain (Turk, 1997). Because the analyses were longitudinal, duration of pain was considered relevant. Therefore, the duration of pain at time one and duration of pain at time two were included in the analysis. Finally, since pain intensity might be expected to be related to both pain related fear and outcome, pain level, average pain over the preceding week (time one) was also included in the control block. The second block containing the particular pain related fear variable(s) of interest for each analysis were entered with no specified order of entry within the block. Therefore, the results indicated the amount of variance each variable contributed to the prediction of the dependent variable, with all shared variance removed. Analyses predicting return to work utilized binary logistic regression with the same, procedure for entering the independent variables as utilized for other regressions. Correlations Between Independent and Dependent Variables Tables 13 to 28 show the zero order correlation coefficients (Pearsons r) between the independent and dependent variables grouped according the 20 regression analyses. There is no separate correlation table for prediction of outcome from T S K scores, since the correlation between the T S K and each dependent variable is displayed in the tables indicating pain related fear full-scale scores with the dependent variables. As indicated in those tables, most of the pain related fear and Catastrophizing variables were significantly correlated with each other and with all dependent variables. Of the control variables, only time one average pain over the preceding week was significantly associated with any of the dependent variables. Specifically, time one average pain over the preceding week showed significant correlation with time two depressive symptoms, time two perceived disability and time two average pain over the preceding week. Of all the independent variables, only the work subscale (minus item 6) of the F A B Q showed significant correlation with time two return to work status.  Catastrophizing, Fear  CN O  # #  CN CN O  o  d  * * O  * * OO  *CN i/o  PH  o  1  1  in CN o  0.12  UO  PH  * *  O o  -0.04  *CN  -0.04  oo  CN  o•  o  o  VO  ro  PH  O  * * *  ca i-  o  OO  o i s <N  Q o Q  ©  ©  o  ©  ©  ©  1  CN O  o  CN <  oo  oo  o  p  ©  CN  I  O  1  o  1  VO  I/O  i—i  O  o  o  ©  o  1  * * #  o  * *  o  VO o  * *  CN  p o  o  * *  co CN  ©  d  * #  * **  v> o  o  0.09  *4  CN  oo ro d  co  o  ©  o  •  ©  co O  ©  •  o d  1  co  p  OO  60  d  ©  CN d  - H  co  CN  '-4—»  cd  OO  w O  (U  oo  >i 3  Q  CN  co ca 3  Q  ca  a  UO  o  o o  60  ca  <u >  oo PH  <: PH  p d  00  V V  PH  #  U  o  ^ H  p. a  p d  V  a  * * * *  147  Catastrophizing, Fear  148  Table 14 Intercorrelations Between PCS Subscale Scores and Time Two Depressive (CES-D) Sympt. Age CES-D t2  -0.12  Age Sex Duration tl Duration t2 Average pain tl PCS Rumination tl PCS Magnification tl * p<0.05 ** p<0.01 ***p<0.001  Sex  Duration Duration Average PCS PCS PCS tl t2 pain tl Rumination Magnification Helplessne tl tl RSt1  0.03  0.12  0.12  0.26*"  0.49"*  0.45***  0.50***  0.11  0.06  -0.08  -0.10  -0.15  -0.14  -0.15  -0.05  -0.08  -0.09  -0.00  0.06  -0.05  0.07  -0.04  0.23*  0.22*  0.23*  0.02  0.06  0.09  0.05  0.19**  0.19*  0.18*  Catastrophizing, Fear  °2 <3 « OO  12  ftl  oo-S  <  &~ O  Pi  ON  o d  co  ^ £  CJ O  oo S oo ca  d  ca o  oo  '5b  ,2 ,  < .2 '  CO  *  ** *  ON T f  d  o d  p  •  ** * p  * *  d  d  d  o. d  co p  CN .  o  d  d  d  d  •  ON  o  d  oo  O  i  o  UO  *CN  co vO  d  d  d  d  d  *ON  *ON  4—1 1—1  p d  i  O  OO  **  **  * *  149  * * *  ** * vq d  o  d  ** *  * *  CN CN  IT)  ©  d  **  uo p  o  uo CN  d  d  d  p  CN p  d  d  1  PH OT PH U  ^ H  ca  .s  >  -4—» CN  !3 Q  o 3  Q oo CJ  60  vo CN  o  ON  d  d  d  •  o  -  ca  <o ft  c3  **  4->  oo  CN  00  d  d  CN  VO O  o  d  d  d  p  i  ' do •  p d  uo  co O  CN  S3  ca o  O  v.  CN  T3  fc! o o (-1  o  ca  OO  W U  60  X oo  13 Q  (3  u 60  !3 Q  <  ca  o  rx  O '-*-»  ca  >  CJ  '5b  ca  oo oo  < PH  PH  >  o  oo oo  < PH  " '32  3  oo a oo o < PH  O  U0  ^H  O  o p p o d d  V V V ft ft p .  *  * * * * *  Catastrophizing, Fear  oo ro  o  r--  10 O  p  d  d  *  *  ro d  d  o  UH  O ro  cy'  -H  o <N  © 1  d  d  CN  i  i  ON  o <  tN  > Si  "cd  o  ©  * (U  CO  o  I  O d  * *VD CN  ©  ©  o  CN o  1  d  1  oo o  o  o  d  Q  CM  VO o  d  o  d  O 00  d  CN  co  CN  60 cS  • iH  W  l-l <  oo  3  Q  &,  l-c  3 Q  VH  1  S d  ©  £ d  V V V  ci, a a, *  * * * * *  Catastrophizing, Fear  oo —  •3o  IN CN  11  d  * * # oo ro d  PH  00  <H PH  I<:  ©  cal  H  * * o ro d  (3 o  1  al  ro O d  i o •4-* cd  o d  ,1  _  d  o d  1  in p d  1  * * o ro d  PH  £  # * * uo ro  vo  X  o  OO  d  CD  o o d  Q PH  d  1  1  CN O d  o  d  d  d  CN ©  VO o d  o CN  OO p d  CN O  ON o  * * #  ro CN ©  uo  * #  ©  in d  o  oo  * * * o  d  * * *  oo CO ©  ro  ro O  •  OO o d •  OO o d  VO O d  uo o d  U OX)  d  m o d  o I  * * * m CN d  * * *  * * *  CJ  oo  o d  —'  CN  ca  o  c o  CJ  3 Q  ca  S3  PH  60 ca cj  3  X  oo oo < PH  o d 00  o PH  ° d  s °  d  V V V a  ft  P-  ^ * # * * * *  151  Catastrophizing, Fear  in  00  * *  M  *VO  ©  a  in o  ©  d  1  CO CN  co  O  o  vo  ON  CO  d  S3 o  '3  * #  oo o  CJ  IM  co CO  '  -3-  CN CN  o  d  d  d  ON ©  d  cd o  oo t3 O .S  so  1—1  o co .  • *  o  cd (3 O  ro  d  ro © d  S3 o  cd  © d  o d  CO  d  o  o  ON  d  d  i—<  •  oo o d  CN  i  oo  •  o d  vo O d  p d  O d  i  VO  ON  o d  d  CO  o  o d  ll o 00  60  d  ^  (  d  O  S3  o  S3 _o  CN  Q  <  00  cd (3  ft  cd  cd  Q  Q  iu  60 cd u  CN  > <  "fa  O  cd  HH  O  S3  CJ « PH  o  I 00 o  PH  cd  ^ ©^ S°  o o  d  d  V V V O, ft ft # #  C atastrophizing, Fear  03  .XI  <D cd UH .52  *  CN d  oo ft  ft  PH  -.g  ,  * # *  O ©  IS  o  co O  u o  2 1 Cd o  < .2 PH  o ©  T-H  ,—*  »—1  d  d  ON  CO d  CO  vo  d  •<  00 OO • g PH  CO  ON  oo  W PH  D  ™ *z ,> S o  ©  * * *  o  m d  # #  d  CN  o  d  d  o ©  p  o  ON  co  o ©  i—i  0\  o ©  d  •  CO  oo  d  d  •  d  o d  o  »—<  Os  T—1  o  u oo  d  p d  O d  d  o d  oo  i  VO  p  o  i  o  p  d  d  CN  vo  d  d  i  o  m  oo  o  c o  -4-»  CO  *  1—1  d  CO d  o  p  • 1-H  i  d  * * * CO CO  CN  © d  d  CN CN  CN  co  o d  o d  d  (  *-H  d  o o d  cd o , —  Q PH  so  I)  oo  H  CN  o  c o  cd  cd  Q  Q  o  o  cd ft  D  60  . cd lH  >  O  T3 00 00 < PH  43  oo  2  oo |) oo ©  ^ PH  PH  d  —< o O- o d  d  V V V ft ft ft #  ** **  153  Catastrophizing, Fear  154  Table 20. Intercorrelations Between F A B Q Scale Scores and Time Two Perceived Disability (PDI).  PDI time 2  Average FABQ Work tl pain tl  FABQ Activity tl  Age  Sex  0.00  0.10  0.07  -0.03  0.30**  0.18  0.34***  -0.11  0.06  -0.08  -0.10  -0.08  -0.11  0.05  0.08  -0.09  -0.14  0.07  0.07  -0.09  0.11  0.05  0.03  -0.03  0.07  0.26**  0.31***  Age Sex Duration tl Duration t2 Average pain tl FABQ Work tl * p<0.05 ** p<0.01 ***p<0.001  Duration t2  Duration tl  0.41***  Catastrophizing, Fear  oo  VO —'  VO CN  PH  OO OO  d1  T f  o  CN  <  o  *  1 1  OO 1—1  d  ©  *  #  o d  1  uo CN  o  CN  VO O  Ov O  co CN  d  **  **  d  d  o  *  uo  d ** *  o d  **  ON  d  00  Tf  d  0.60  H  CN CN di  0.20*  *  i/o CM  00  oo co  PH  CN  T f  o  VO CN  o  ©  PH  60  ca cj ca  fN  o d  Ov O  (N O  00  OO  CN  vo  PH  2^  o  d  o  CN  d  T f  co  d  CO  O  o  3  Q  ca  o d  d  S3 Q cj  oo  vo o d  <  p d  x  60  uo  o  ca -«-»  <  60  <  PH  X  cj OO  r-H  (N  c  o  ca  ca  S3 Q  3  Q  ca  PH CJ 60  . ©  ca  00 00  I-I CJ  >  PH  < PH  ©  ° © © oo O PH  1  a, y, y,  ^ a I*  *  P.  *  Catastrophizing, Fear  u  o  S3  CN  #  m  in  M  d•  o d  ro (N  -0.1  oo  vo o d  CN CN  in  VO  O  d  _o *  Cd  o  0.25  oo  *  ON  ro  ©  cd  _  00  o  • I—I  +->  cd  CN  .S ^ 6  O  d  *  *  |cs  o  ON  CN  ro  oo  oo  d  d•  d  d  i—<  o  vo o d  VD O  d  oo  i  CN  CN  <D  1  t  O  o  vo o  —  o d •  d  c o  CN  1  d  CN  > cd  d•  o o d  in  p  ON i—i  d  O  o  i in  o d  •  d  o d  00  <  oo CN ID C <  P.  <D  00  <  X ID  00  -H  CN  cd  c o  0  ID  3  3  o  Q  OH  00  o  c o  • T—I  cd  «H  C  S  td  o cd  cd  >-< (D >  oo U  00  PH  PH  O  m <-H o 0 o o odd V V V  a . D- a . *  1*  * * *#  isal  Catastrophizing, Fear  CO  1  Appi  cd i—i  ©  (D  o  o  oo • ~ ,  £  ON  a  co O  ON  0  O  d  vo  m  ©  /^i  0  O  -H  co  ON  d  f^J  vo ©  ©  O  PH  o  CD  o oo c3  ^  O-  S3  1  3  CO  d  o  CO ©  CN i—  1  ©  © 7—1 ©  * #  CN CN  i n i n  ©  CD  60 c3  l-i  CD  o  o &o 'Bp oo  -2 . < .2  p d  ON  © d  i n  * #  © ©  © ©  CO  i n  CN  ©  cn  PH  H CD  H  cu  PH  rt (3 tu rt  PH  >  CD  o o  CD  o 00 00 00  S3 O  rt  CN  IH  Q (3 O '-»3 rt  ©  ON ©  CN  CN  ©  ©  ©  CN  OO  00  CN  VO ©  © ©  d  Q X  CJ 00  o  PH  CD CD  CD  £ CJ  o d  CO  CN  <;  o ©  © ©  ©  ©  m ©  ©  Pi  m 53  -4-»  O  '3 CD  fc! o o lH  CD  rt  PH  CD 60  rt |H  CD  —i  CN  S3  S3  o CU  <  60  X CD  00  3  Q  O  rt >3  Q  rt o  rt  '5b  CD 60  ' «  rt |H >  <  CD O  o  PH  00  oo  < PH  43 PH  00 00  < PH  '>3  CD >  on §>  oo o PH  « S o•> — © © o d d V  o,  I*  V  V  * * *  *  ft  Catastrophizing, Fear  m co  31* o  60  ~  <D  -ft «i  >  o  ©  CN  o ©  o  o  ON  ©  ©  S3 CN  © ©  00  00  VO ©  UO  © ©  o  ro ©  cn  00  CN  l  o  uo  o  o o  o o  CN ©  CN ©  CO ©  ro  Q o I-I  S3  1 — 1  CN O  o o  O X  cu  00  (D  VO  © o © ©  CN •*->  S3 "3  CN +-»  OH  S3  CO  o  bO I—i Cu)  CL)  oO i  X  cu 00  '•a  Q  S3  o  VH  Q  PH  CD  00 l-H  3  O  uo ©  © V  a  *  *™H  O  o © o ©  v v a  * #  #  # #  a  Catastrophizing, Fear  r--  00  ©  H  oo •  o  PM  ©  00  o  oO < PM  -  >-H  ©  00  i—i ©  * CN CN  d• CO  d  ,_ CN  d  CN  d  i—l d *  ro  CN  d  d  ro  >-H  d  d  CN  T-H  in  ©  doi  o  ©  1  o d  CU  60  cS  >H  C  cu . H >  S i  OS CN  Q a o  ©  d  i  1= in  • r-H  o  1  o  o ©  1  Q  x ca oo  ro  •  o di  vo CN  d•  o  o  CO  CO  d  d  d  o  d  o  d  di  in  m  d  d  * * CN  d  o  VO  * # *  o  * # *  00  rd  ro  CN CN  o  di  #  d  VO  o  ON  d  CN  d  CN  1  *-  o d  *  o  **  * *  m  CN  *—<  d  o  in  ©  CO  60  o  '53 CN  H P4  *-*-» cu 60  X CU  00  'Q 2  CU  60  "is  c3  3  Q  00 00  )H CU  >H  CN  >  <  o  m  ^H  V  V V  o o o o d d  CH  c o  PH  < PM  00  U  PM  CH  I#  OH PH  *  #  *  #  #  159  Catastrophizing, Fear  160  w  CD  c  o  Tf  ^  CD  o  o d  d  o  Tf  CN  d  ©  O  m CN  o  >n  00  a  o  '•5 oo o rj  t c  vo O  •  ""I  d  o  1  00  00  —t  o o d  vo  CN  CN CN  d  d  CN i-H  m CN  d  d  o d  ON  CN  o  CN  d  cd  _o oo "S.  u .S PH  ON  O  d  1  PH  TT  S3  S <  ,£ s, i-i  \<  d  £  CD  cd  i  -  o d  d  a o  ON  o  cd <N  Q c o  ©  O d  o d  in  CO  CO  o d  '•si  i  O • O •  i  o d  d  CN  i  d  i  d  o d  m  o  <D  00  m  CD  o d  00 <3  S3 O  S3 T-H  CN  CN -»-»  o  S3  Is  Id  O H  5  CD  <  s>  00  I-I  CD  o  Q Q  CD >  <  u  s  oo cd  'tl  o  cd  00 3 u rt PH  t C  1 PH  m  ^H  o o o o d d  o  V V V O H  O H  O H *  I*  *  *  Catastrophizing, Fear  CD  cd  ro  ro  oo PH  i—<  d  OH  d  CN  CO CN  CN O  d  d  . d  o d  d  CN CN  oo  d  161  CO  llO  d  <H  CD  00 • ~ ,  ro  £  d  PH  a  CO  o  O  o  2  d  d  KO  vo  U  CD CD VO  *1  d  d  o d  00  o  o  Tt  d  o  o d  CO CN  CO  Tt  d  d  CN VO  cd  o  loo 'g> oo o , PH  d  >N  ,£3 PH  CD cd •H  Tt  S3 "3 '  o d  o d  CN  o  d  d  —>  CN  S3  o  UO  o  '"cd <N|  ON  O  o d  ro O  co  d  d  PS3 o  o  o  13 uo O  CD  00  d ro  CD  o  bfj  <3  CN  S3  CN  5  o  CD  <l  ^  00  'fa VH  3  S3 O  cd  S3 "c3  -  OH  CD  CD O  5b o  CD  cd  OO J3  CD  <  <  oo 'o oo >  00 8) oo o  PH  PH  3,  I-I  PH  PH  CD >  <  <  uo >—>  o o o o d d  O  V VV OH  I*  OH OH *  * *  * *  Catastrophizing, Fear  162  Table 28 Intercorrelations Between F A B Q Subscale Scores and Time Two Return to Work Status  Age  RTW t2 Age Sex Duration t l Duration t2 Average pain t l F A B Q Work t l * p<0.05 ** p O . O l ***p<0.001  Sex  -0.03 0.05  Duration Duration tl t2  Average pain tl  FABQ Work tl *  FABQ Activity tl  -0.05  -0.05  0.11  0.22  0.03  0.09  -0.04  -0.11  -0.07  0.03  0.01  -0.09  0.21*  -0.08  0.17  -0.12  0.08  0.00  -0.20  0.00  0.10  0.44  0.12  0.43 0.49  ***  Catastrophizing, Fear  163  Prediction of Time Two Depressive Symptoms All pain related fear variables. Table 29 displays the unstandardized regression coefficients (B) and intercept, the standardized regression coefficients (/?), the squared semipartial correlations ( s r  2 un  jq ), ue  and R, R , and adjusted R after entry of all IVs in the prediction of 2  2  time two depressive symptoms. In addition, the incremental change in R (sr ) after each block 2  2  inc  is displayed. R was significantly different from zero at the end of step 1 and step 2. With only the control variables in the equation, R = 0.13 ( F 2  inc  (5, 131) = 4.03, p <0.01) indicating that the  control variables contributed significantly to the prediction of time two CES-D scores. Addition of the four pain related fear variables (PASS, PCS, F A B Q & T S K total scores) in step two resulted in a significant increment in R to 0.36 (sr 2  2 inc  = 0.23, F  i n c  (4,127) = 10.79, p < 0.001). .  After step two, none of the individual scale scores contributed uniquely to the prediction of time two depressive symptoms. Therefore, together the pain related fear and catastrophizing variables when measured in the subacute stage, predicted 23 percent of the variance in depressive symptoms at three months post injury, after controlling for initial pain level and demographics. Individual pain relatedfear scales. The results of regressing time two depressive symptoms on the individual pain related fear scales in four separate analyses is reported in Tables 30 to 33. Because the total contribution of the control block to prediction of time two depressive symptoms is identical for all five analyses, and it was reported in the preceding section it will not be reported again. The results of regressing time two depressive symptom scores on the four PASS subscales is indicated in Table 30. The addition of the four PASS subscale scores in step two 2  2  following entry the control variable block, resulted in a significant increment in R to 0.33 (sr  m c  = 0.20, Fine (4,127) = 4.21, p_ < 0.001). None of the PASS subscale scores contributed unique variance to the prediction of time two depressive symptoms at this stage. As might be expected  Catastrophizing, Fear  164  Table 29 Pain Related Fear Scales in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  2  Variables  B  a  sr (unique)  SEB''  2  sr (incremental)'  3  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.05 1.43 0.04 0.06 0.07  0.08 1.65 0.14 0.05 0.04  -0.05 0.08 0.02 0.09 0.14  0.00 0.01 0.00 0.01 0.02  0.08  0.07  0.16  0.02  0.13  0.07  0.23  0.02  0.25  0.12  0.23  0.02  0.04  0.14  -0.01  0.00  Block 2  0.13 F A B Q total PASS total PCS total T S K total  0.23 Intercept = -8.56 R =0.36 Adjusted RI = 0.32 R = 0.60* 3  22  *p<.05 **p<.01 ***p<.001 values after step 2 (all variables in model) values after each step  3 b  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury At time two completion; Average pain 1, Average pain over week proceeding time one.  Catastrophizing, Fear  165  Table 30 Pain Anxiety Symptom Subscales in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  Variables  sr (unique)  SEB'  11  0.08 1.69 0.14 0.05 0.04  -0.09 0.09 0.07 0.09 0.19  0.01 0.01* 0.00 0.01 0.03*  sr (incremental)  3  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.09 -1.94 0.13 0.06 0.09  0.13  Block 2 PASS Physiological PASS Avoidance PASS Cognitive PASS Fear Appraisal  0.34  0.26  0.16  0.01  0.24  0.21  0.11  0.01  0.51  0.26  0.22  0.02  0.06  0.19  0.04  .0.00  0.20 Intercept = ••4.94 R = 0.33 Adjusted RI == 0.28 0.28 R = 0.57*' 2  a  *p<.05 a b  **p<.01  22  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  166  Table 31 Pain Catastrophizing in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression 2 V  a  n  • b ul i e  a  s  B  s  a  S E B  a  /?  a  2  r  (unique)  (incremental)  3  13  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.03 -1.66 0.04 0.07 0.10  0.08 1.68 0.14 0.05 0.04  -0.03 -0.07 0.02 0.10 0.20  0.00 0.01 0.00 0.01 0.04  0.54  0.31  0.21  0.02  0.11  0.50  0.03  0.00  0.58  0.28  0.27  0.02*  Block 2 PCS Rumination PCS Magnification PCS Helplessness  0.19 Intercept = -4.89 R =0.33 Adjusted R = 0.28 R = 0.57*** 2  a  a  b  2  *p<.05 **p<.01 ***p<.001 values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  167  Table 32 Fear Avoidance Beliefs in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression 2  Variables Block 1 Age Sex Duration 1 Duration 2 Average pain 1  B  a  a  -0.08 1.62 0.21 0.07 0.09  0.08 1.81 0.15 0.05 0.04  -0.07 0.07 0.11 0.11 0.19  0.01 0.01 0.01 0.01 0.03  0.21  0.15  0.13  0.01  0.47  0.17  0.25  0.05  Block 2 FABQ Work FABQ Activity  S E B  2  sr (unique)"  sr (incremental) " 1  *  Intercept = -12.40 Adjusted R = 0.18 « ._*** R = 0.47 a  2  *p<.05 **p<01 ***p<.001 values after step 2 (all variables in model) values after each step  a  b  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  168  Table 33 Kinesiophobia in the Prediction of Time Two Depressive Symptoms: Hierarchical Regression  Variables  B  Age Sex Duration 1 Duration 2 Average pain 1  -0.03 -0.75 0.14 0.08 0.10  0.09 1.78 0.15 0.05 0.04  -0.03 -0.03 0.08 0.11 0.20  0.44  0.11  0.34  a  SE5  3  /?  a  sr (unique)  sr 3  Block 1 0.00 0.00 0.01 0.01 0.19  Block 2 TSK  „  0.13  ^ „***  0.10  0.09  Intercept = -18.737 R =0.23 Adjusted R = 0.19 R = 0.48* 2  3  *p<.05  "p<.01  2  p<.001  values after step 2 (all variables in model) values after each step Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one. 3  b  Catastrophizing, Fear  169  given their high intercorrelations, the four PASS subscales contributed a significant amount of .17 in shared variance to the prediction of time two depressive symptoms. The results of regressing time two depressive symptom scores on the three PCS subscales is indicated in Table 31. Addition of the PCS subscale scores as a block to the equation in step two significantly increased R (sr inc = 0.19, F 2  2  inc  (3,128) = 12.14, p < 0.000) indicating that  together, the PCS subscales contributed significantly to the prediction of time two depressive symptoms. At step 2, the PCS helplessness subscale (sr j 2  un  que  = .02, p < .05) contributed  significant unique variance to the prediction of time two depressive symptoms. The three PCS subscales contributed a high level of shared variability (r = 0.17). Average pain over the week 2  preceding time one also contributed significant unique variance  (sr ique 2  un  = -03) to the equation  after step two of the regression analysis. The results of regressing time two depressive symptom scores on the two F A B Q subscales is indicated in Table 32. Addition of the F A B Q subscale scores in step two significantly increased R (sr j = 0.09, Fj (2,129) = 7.27, p < 0.01) indicating that together, the 2  2  nc  nc  F A B Q subscales contributed significantly to the prediction of time two depressive symptoms. At step 2, only the F A B Q activity subscale contributed significant unique variance to the equation. (sr unique 2  = 0.05, p_ < 0.01). The two F A B Q subscales together contributed only 0.04.  in shared variability indicating that they did not significantly overlap in the prediction of time two depressive symptoms The results of regressing time two depressive symptoms on the TSK total scores is reported in Table 25. Adding the T S K score to the equation after the control block resulted in a significant increase in R (sr j = 0.09, Fj (1,130), = 6.30, p < 0.001). At this stage, only 2  2  nc  nc  average pain over the week preceding time one (sr j u n  q u e  = 0.13) and the T S K continued to  contribute significant unique variance to the prediction of time two depressive symptoms.  Catastrophizing, Fear  170  Prediction of Time Two Perceived Disability Perceived disability at three months was regressed on the pain related fear total scores and subsequently on the constituent subscales of each instrument. Because the total contribution of the control block to prediction of time two disability symptoms is identical for all five analyses, it will be reported only once, in the context of the regression on pain related fear total scale scores. All pain relatedfear variables. Table 34 displays the unstandardized regression coefficients (B) and intercept, the standardized regression coefficients (j3), the squared semipartial correlations (sr nique), and R, R , and adjusted R after entry of all IVs in the prediction of 2  2  2  U  time two Perceived Disability scores. In addition, the incremental change in R {sr i ) after each 2  2  nc  block is displayed. R was significantly different from zero at the end of step 1 and step 2. With only the control variables in the equation, R = 0.11 (F 2  inc  (5, 131) = 3.31, p <0.001), indicating  that the control variables contributed significantly to the prediction of time two Perceived Disability. Addition of the pain related fear variables (PASS, PCS, F A B Q & TSK total scores) in step two resulted in a significant increment in R to 0.26 (sifj = .14, F j (4,127) = 5.68, p < 2  nc  nc  0.001). After step 2, none of the four pain related fear variables contributed uniquely to the prediction of time two disability scores. Together, the pain related fear variables contributed .09 in shared variance suggesting that they overlap significantly in predicting time two Perceived Disability and the failure of any to predict uniquely may be due to this overlap.  Catastrophizing, Fear  Table 34 Pain Related Fear Scales in the Prediction of Time Two Perceived Disability: Hierarchical Regression 2  2  sr Variables Block 1  R  a  SE5  a  /?  a  sr-  (unique)  3  Age Sex Duration 1 Duration 2 Avg. pain 1  0.14 3.00 -0.02 -0.04 0.14  0.13 2.70 0.22 0.08 0.06  0.09 0.09 -0.01 -0.04 0.19  0.01 0.01 0.00 0.00 0.03"  FABQ total PASS total PCS total TSK total  0.04 0.04 0.28 0.46  0.16 0.12 0.20 0.23  0.05 0.04 0.18 0.21  0.00 0.00 0.01 0.03*  (incremental)  b  0.14"* Intercept = -10.33 R = 0.25 Adjusted" R = 0.19 R = 0.50*" 2  2  *p<.05 a b  **p<01  ***p<001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  172  Individual pain related fear scales As indicated in Table 35, the addition of the PASS subscale scores at step two resulted in a significant increment in R (sr 2  2 inc  = 0.09, F  (4,127) = 3.50, p < 0.05). At step two, none of  i n c  the PASS subscales contributed significant unique variance to the prediction of time two perceived disability, but time one average pain over the preceding week, did (Sr i u n  q u e  = 0.04, p <  0.05). The four PASS subscales contributed .08 in shared variance to the prediction of time two perceived disability, indicating a significant degree of overlap and possibly explaining the failure of any to contribute significant unique variance to the equation. As indicated in Table 36, when entered as a block, following the control variables, the three PCS subscales significantly increased R (sr 2  2 inc  = .10, F j (3,128) = 5.49, p < 0.01) nc  indicating that together, the PCS subscales contributed significantly to the prediction of time two perceived disability (PDI). After entry of the PCS variables as a block, none contributed significant unique variance to the prediction of time two perceived disability. Average pain over the preceding week at time one, contributed significantly to the prediction of time two perceived disability (sr  2 uniqU  e = 0.05, p_ < 0.001) at this stage. Again the three PCS variables contributed a  large amount of shared variance (sr = 0.09), possibly explaining the failure of any to contribute 2  significant unique variance As displayed in Table 37, addition of the F A B Q work and activity subscale scores as a block in step 2 significantly improved the prediction of time two perceived disability scores (sr inc = 0.06, Fine (2,129) = 4.95, p < 0.01). The F A B Q activity subscale contributed significant 2  unique variance to the prediction of Perceived Disability (sr j 2  un  average pain (sr i e = 0.04). 2  un  qU  que  = 0.05, p_< 0.01), as did  Catastrophizing, Fear  173  Table 35 Pain Anxiety Symptom Subscales in the Prediction of Time Two Perceived Disability: Hierarchical Regression sr  1  Variables Block 1  Age Sex Duration 1 Duration 2 Average pain 1  PASS Physiological PASS Avoidance PASS Cognitive PASS Fear Appraisal  B  a  SES  a  sr  2  /T  (unique)  3  0.03 -4.30 0.14 -0.05 0.17  0.13 2.76 0.23 0.08 0.06  0.02 -0.13 0.05 -0.05 0.24  0.00 0.02 0.00 0.00 0.04**  0.34  0.43  0.11  0.00  0.50  0.35  0.16  0.01  0.37  0.43  0.11  0.00  -0.03  0.31  -0.01  0.00  Intercept = 4.42 R =0.20 Adjusted R = 0.14 R = 0.45** 2  2  *p<05 **p<.01 ***p<.001 values after step 2 (all variables in model) values after each step Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one. a  b  2  (incremental)  b  0.09*  Catastrophizing, Fear  174  Table 36 Pain Catastrophizing Subscales in the Prediction of Time Two Perceived Disability (PDI): Hierarchical Regression  V a r i a b l e s  B  a  SEff  sr  sr  2  (unique)  a  2  (incremental)"  3  Block 1 Age  0.10  0.13  0.07  0.00  Sex Duration 1 Duration 2 Average pain 1  4.05 0.03 -0.03 0.17  2.71 0.23 0.08 0.06  0.12 0.01 -0.03 0.24  0.01 0.00 0.00 0.05*  0.42  0.50  0.11  0.00  0.25  0.80  0.04  0.00  0.69  0.44  0.22  0.01  PCS Rumination PCS Magnification PCS Helplessness  0.10** Intercept = 5.23 R =0.21 Adjusted R = 0.16 R = 0.46*** 2  3  *p<.05 **p<.01 ***p<.001 values after step 2 (all variables in model) values after each step Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one. 3 b  2  Catastrophizing, Fear  175  Table 37 Fear Avoidance Beliefs Subscales in the Prediction of Time Two Perceived Disability: Hierarchical Regression  Variables Block 1  Age Sex Duration 1 Duration 2 Average pain 1  B  a  0.06 3.34 0.22 -0.04 0.16  SES  0.13 2.78 0.22 0.08 0.06  sr  (unique)  a  0.04 0.10 0.08 -0.04 0.22  (incremental)  3  0.00 0.01 0.01 0.00 0.04**  Block 2  0.11' FABQ Work FABQ Activity  0.05  0.23  -0.02  0.00  0.75  . 0.26  0.26  0.05** 0.06  Intercept = -2.77 R = 0.18 Adjusted RL = 0.13 R = 0.42* 2  3  *p<.05 3 b  *p<.01  *p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  22  Catastrophizing, Fear  176  Table 38 shows that adding the T S K score to the equation after the control block resulted in a significant increase in R to 0.22 (sr 2  2 n c  = 0.10, F  i n c  (1,130) = , 17.31, p < 0.001). At this  stage, average pain over the preceding week contributed significant unique variance ( s r j 2  un  que  =  0.04, p < 0.0) to the prediction of time two perceived disability as did the T S K (sr ique = 0.10, p 2  un  < 0.001). Prediction of Time Two Average Pain Level All Pain Related Fear Variables Table 39 displays the unstandardized regression coefficients (B) and intercept, the standardized regression coefficients (P), the squared semi-partial correlations (sr  2  ), and R, R , 2  unique  and adjusted R after entry of all IVs. In addition, the incremental change in (sr i ) after each 2  2  nc  block is displayed. At the end of step 1 with demographic variables (age and sex), duration variables, and time one pain level in the equation, R was significantly different from 0 (sr i = 0.13 2  nc  F i (5, 131) = 3.75, p < 0.01) indicating that they contributed significantly to the prediction of time nc  two pain level. After step two, with the pain related fear variables (PASS, PCS, F A B Q & TSK total scores) added to the equation, R . did not increase significantly (Sr j = .04, Fj (4, 127) = 2  2  nc  nc  1.65, p_ > .05), indicating that as a group they did not significantly improve prediction of time two pain. At this stage, average pain over the preceding week at time one (sr  2 unjque  = 0.11, p < 0.01)  independently contributed to the prediction of time two pain level but none of the pain related fear variables did so.  Catastrophizing, Fear  177  Table 38 Kinesiophobia in the Prediction of Time Two Perceived Disability: Hierarchical Regression  Variables  S E B  1  11  sr (unique)  sr (incremental)  3  Block! Age Sex Duration 1 Duration 2 Average pain 1  0.14 2.47 0.07 0.03 0.15  0.13 2.68 0.22 0.08 0.06  0.09 0.07 0.02 -0.03 0.21  0.01 0.01 0.00 0.00 0.04 0.11  Block 2 TSK  0.71  0.17  0.36  0.10 0.10*'  Intercept = -17.78 R =0.22 Adjusted RI = 0.18 R = 0.47* 2  3  *p<.05 3  b  **p<.01  2  *** p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  178  Table 39 Pain Related Fear in the Prediction of Time Two Average Pain: Hierarchical Regression  Sr (unique) 2  Variables  B  a  g  E  5  ^  a  sr (incremental) 2  3  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  0.10 4.49 0.44 0.03 0.28  0 21 4 43 0 36 0 13 0 10  0.04 0.09 0.10 0.02 0.25  F A B Q total PASS total PCS total T S K total  0.11 0.15 0.14 0.09  0 25 0 19 0 32 0 36  0.10 0.12 0.05 0.01  0.00 0.01 0.01 0.02 0.05 **  0.00 0.00 0.00 0.00 0.04  Intercept = -18.32 R = 0.17 Adjusted R = 0.11 R = 0.41 2  3  *p<.05 3  b  **p<.01  2  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  179  Table 40. Pain Anxiety Symptom Subscales in the Prediction of Time Two Average pain: Hierarchical Regression. Sr (unique) 1  Variables  11  S E B '  Block 1 -0.16 -4.73 0.76 -0.09 0.35  Age Sex Duration 1 Duration 2 Average pain 1  0.25 5.41 0.44 0.16 0.13  -0.06 -0.08 0.15 -0.05 0.24  sr (incremental)  3  0.00 0.01 0.02 0.00 0.05* 0.09  Block 2 PASS Physiological PASS Avoidance PASS Cognitive PASS Fear Appraisal  -0.48  0.44  -0.02  0.00  0.62  0.42  0.05  0.01  0.75  0.52  0.05  0.01  0.24  0.48  0.01  0.00  Intercept = -4.77  0.05  R = 0.13 Adjusted R = 0.03 R = 0.31 2  3  *p<.05 3  b  **p<.01  22  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Avg. pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  180  Individual Pain Related Fear Scales. PASS. The results of regressing time two average pain over the preceding week on the PASS variables is indicated in Table 40. The four PASS subscale scores did not contribute significantly to the prediction of average pain over the week preceding time two (sr j = 0.05, 2  nc  Fine (3,128) = 1.70, p > 0.05). Average pain over the preceding week at time one (sr.  u n i q u e  =  0.05, p < 0.05) contributed significant unique variance to the prediction of time two average pain over the preceding week but none of the four PASS subscale scores did. PCS. As can be seen from Table 41, addition of initial pain catastrophizing subscale scores as a block after the demographic variables, duration variables and initial pain intensity block did not significantly increase R (sr 2  2 inc  = 0.03, F  (3,128 ) = 1.47, p > 0.05).  i n c  Only average pain over the week preceding time one (sr ni 2  U  que  = 0.07, p_ < 0.01) contributed  significant unique variance to the prediction of average pain over the week preceding time two. FABQ. As indicated in Table 42, the F A B Q Work and Activity subscales, entered as a block following the control variables, contributed significantly to the prediction of average pain over the week preceding time two (sr i = 0.06, F 2  nc  inc  (2,129) = 4.31, p < 0.05). (see Table 34)  The F A B Q Activity subscale contributed significant unique variance to the equation (srf i un  0.05, p < 0.01), as did time one average pain (sr ni 2  U  que  que  =  = 0.03, p < 0.01). The amount of shared  variance in the prediction was less than 0.01, indicating that the two variables did not significantly overlap in prediction of average pain preceding time two. TSK. As indicated in Table 43, addition of the T S K total score to the equation predicting time 2 average pain over the preceding week did not result in a significant improvement in R (sfinc = 0.02, Fine (1,130) = 2.97, p > 0.05). Other than the TSK, the only variable to contribute  significant unique variance to the equation at this point was average pain over the week preceding time one (sr nique -07, p < 0.01). 2  =  U  Catastrophizing, Fear  181  Table 41 Pain Catastrophizing Subscales in the Prediction of Time Two Average pain: Hierarchical Regression  Sr (unique) 2  Variables  B  a  SE5  3  3  sr (incremental)  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  0.12 4.17 0.44 0.04 0.32  0.20 4.33 0.36 0.13 0.09  0.05 0.08 0.10 0.02 0.29  0.00 0.01 0.01 0.00 0.07*  Block 2  0.13 PCS Rumination PCS Magnification PCS Helplessness  0.39  0.80  0.037  0.00  0.63  1.28  0.07  0.00  0.33  0.71  0.07  0.00  .  0.03 R = 0.15 2  Intercept = 10.31  Adjusted R = 0.10 R = 0.39 2  *p<.05 a  b  **p<.01  ***p<.001  values after step 2(all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  182  Table 42 Fear Avoidance Behaviour Subscales in the Prediction of Time Two Average pain: Hierarchical Regression  Sr (unique)  sr (incremental)  2  Variables  B  a  S  E  B  *  ^  2  2  b  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -D.l 2 -2.94 0.79 -0.10 0.30  0.25 5.34 0.42 0.16 0.14  -0.04 -0.05 0.16 -0.06 0.20  0.00 0.00 0.02 0.00 0.03* 0.09*  FABQ Work FABQ Activity  -0.25  0.44  -0.05  0.00  1.11  0.50  0.21  0.03* 0.03  Intercept  2.53 R =0.12* Adjusted R = 0.08 R = 0.35 2  2  *p<05 a  b  **p<.01  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  Catastrophizing, Fear  183  Table 43 Kinesiophobia in the Prediction of Time Two Average Pain: Hierarchical Regression sr  Sr  1  Variables  B  Age Sex Duration 1 Duration 2 Average pain 1  -0.11 -4.09 0.74 -0.08 0.34  a  SE5  3  /?  a  (unique)  1  3  (incremental)  Block 1 0.25 5.40 0.43 0.16 0.13  -0.04 -0.07 0.15 -0.04 0.23  0.00 0.00 0.02 0.02 0.05*.  0.09*  Block 2 TSK  0.31  0.34  0.08  0.00 0.01  Intercept = -9.26 R =0.10 Adjusted R = 0.05 R = 0.31* 2  2  *p<.05 3  **p<.01  ***p<.001  values after step 2(all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  b  Catastrophizing, Fear  184  Prediction of return to work A series of binary logistic regressions were conducted to determine the degree to which each of the pain related fear and catastrophizing measures and their component subscales measured at time one contributed to prediction of return to work by three months post injury. Return to work was coded such that odds ratios greater than one indicated a decreased likelihood of returning to work. Of the 84 participants not returned to work at time one, 59 had returned to work at time two, while 25 had not, making the rate of return to work rate at time two, for individuals not working at time one 70.24 percent. All pain related fear variables The results of logistic regression of time two return to work on time one pain related fear full-scale scores are reported in Table 44. The Tables report x change after the addition of each 2  block, the regression coefficients (B), the Wald test statistic and the odds ratio for each variable after the addition of pain related fear variables to the model. As evident from Table 44, a regression model including only control variables as a group (X change = 2.73, p > .18) did not significantly improve the prediction of time two return to work beyond that afforded by the constant only model. Likewise, the addition of the pain related i  fear variables in step two did not add significantly to the equation (x change = 4.09, p > .05). At this stage, Wald test statistics indicated that none of the individual control or pain related fear variables made a significant unique contribution to the prediction of time two return to work. Inspection of the odds ratios indicated that only the value for sex (1.86) deviated from one by an appreciable magnitude. However, the 90% confidence interval (0.57 to 5.81) was very broad and spanned unity, indicating that sex did not reliably contribute to the prediction of failure to return to work. As the total contribution of the control variable block to the prediction of time two return to work is identical for all five analyses, it will not be repeated for the remaining analyses.  Catastrophizing, Fear  185  Table 44 Pain Related Fear in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  B  Variables Block 1 Age Sex Duration 1 Duration 2 Average pain 1  a  -0.01 0.62 -0.01 -0.01 0.01  SE B  0.03 0.58 0.04 0.02 0.01  a  Wald  a  AX  2  0.08 1.13 0.04 0.10 0.42  Odds Ratio  0.99 1.86 0.99 0.99 1.01 2.73  Block 2 F A B Q total PASS total PCS total T S K total  0.03 0.02 -0.05 0.02  0.03 0.02 0.04 0.05  1.03 1.02 0.95 1.02  1.16 0.77 1.40 0.19 3.66  Intercept = -2.46 Final model X = 6.39 Nagulkerke R = 0.10 2  *p<.05 a  b  **p<.01 ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  3  Catastrophizing, Fear  186  Individual Pain Related Fear Scales Results of the analysis predicting time two return to work from the four PASS subscales are reported in Table 45. The four PASS subscales when added to the equation as a group. Following entry of the control variables, added significantly to the prediction of return to work (X change = 11.45, p_ < 0.05). At this stage, the Wald test indicated that the Fearful Appraisal 2  subscale score was the only individual variable contributing significant unique variance to the prediction of time two return to work (p < 0.01). The beta weight (-.20) indicated that higher scores on fearful appraisals were associated with a lower risk of not returning to work. To state this another way, higher fearful appraisal scores increase the likelihood of returning to work. The odds ratio for Fearful appraisals was 0.82 (95%CI .71 to .94), indicating that an increase in the Fearful Appraisal score of one point corresponded with approximately an 18 percent increase in the likelihood of returning to work. The total model achieved a Nagulkerke R value of 0.19, 2  which indicates that, the demographics and PASS subscale scores as a group predicted 19 percent of the variance in likelihood of returning to work. Examination of the zero order correlations between PASS subscales and return to work indicate that the correlation between failure to return to work and fearful appraisals was small and not statistically significant, but negative, while the other PASS subscale scores correlated positively with failure to return to work. This suggests the possibility that the significant contribution of the fearful appraisals subscale score in the regression may be due to a suppression effect and therefore may represent a statistical artifact. Results of logistic regression of time two return to work on the three PCS subscales are reported in Table 46. The three PCS subscales added as a block did not significantly add to the prediction of time two return to work (% change = 6.99, p > 0.05). However, the PCS 2  Catastrophizing, Fear  187  Table 45  Pain Anxiety Symptoms Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  Variables  B*  SE B  a  Wald  a  A%  Odds ratio  2  1  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.01 -0.98 -0.01 0.01 0.02  0.03 0.61 0.05 0.02 0.01  0.13 2.53 0.06 0.10 1.68  PASS Physiological PASS Avoidance PASS Cognitive PASS Fearful Appr.  0.16  0.10  2.63  1.17  -0.06  0.07  0.57  1.06  0.08  0.09  0.82  1.09  -0.20  0.07  7.98**  0.82  0.99 2.65 0.99 0.99 1.02 2.73  11.45* Intercept = -2.10 Full Model x = 14.18 Nagulkerke R = .22 2  2  *p<.05  **p<.01  ***p<.001  Catastrophizing, Fear  188  Table 46  Pain Catastrophizing Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression Variables  B  a  SE B  a  Wald  3  Ay.  2  Odds ratio  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.01 0.87 -0.22 -0.08 0.02  0.03 0.61 0.04 . 0.02 0.02  0.15 2.18 0.25 0.16 1.49  0.18  0.08  3.57  1.20  -0.42  0.05  4.98*  0.65  -0.56  0.13  0.24  1.05  2.73  0.99 2.38 0.98 0.99 1.02  Block 2 PCS Rumination PCS Magnification PCS Helplessness  6.99* Intercept = 1.17 Final model % = 9.72 NagulkerkeR = 0.16 2  2  *p<.05 3  b  **p<.01 ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  3  Catastrophizing, Fear  189  magnification subscale did contribute significant unique variance to the prediction of time two return to work indicated by the Wald test (4.98, p_<.05). In a similar pattern to that found for Fearful appraisals, the Odds Ratio for the magnification (0.65, 95% CI = 0.45 to 0.95) indicated that it contributed significantly to the prediction of returning to work in the opposite direction to what was hypothesized. Specifically, an increase of one point in Magnification score corresponded to an approximately 34% lower likelihood of returning to work at three months. Inspection of the zero order correlations for PCS subscales and failure to return to work indicate a similar pattern to that observe in the PASS, namely, the PCS Magnification subscale demonstrated a small negative correlation with failure to return to work, while the other PCS subscales demonstrated positive correlations with failure to return to work. This, again suggests the operation of a suppressor effect. The combined model of all PCS subscales and control variables provided a Nagulkerke R of .16. 2  Values estimated for the logistic regression of time two return to work on the Activity and work subscales of the F A B Q are in reported in Table47. When entered as a single block, scores on the two subscales did not significantly contribute to the prediction of time two return to work but showed a trend toward statistical significance (x change = 5.25, p< 0.10). 2  Examination of individual Wald statistic values at this stage indicated that only the F A B Q work subscale contributed uniquely to the prediction of failure to return to work. The Odds ratio (1.08; 95% CI 1.01 to 1.16) for F A B Q Work subscale score indicates that an increase in increase of one point will increase the likelihood of returning to work of 11 percent. The Nagulkerke R statistic indicated that an increase in that score corresponded to a decreased likelihood of returning to work of approximately 8 percent. The complete model including all control variables and the two F A B Q subscales predicted 13 percent of the variance in likelihood of not returning to work.  Catastrophizing, Fear  190  Table 47 Fear Avoidance Beliefs Subscales in the Prediction of Time Two Return to Work: Hierarchical Logistic Regression  Variables B  a  SE B  a  Wald  a  Ax  Odds ratio  2  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  0.01 1.00 -0.02 -0.01 0.01  0.03 0.60 0.04 0.02 0.01  0.00 2.81 0.23 0.06 0.33  1.00 2.73 1.01 1.04 1.04 2.73  Block 2 FABQ Work FABQ Activity  0.12  0.05  5.36*  -0.05  0.05  1.07  1.13 6.17*5.25  Intercept =-2.96 X = 8.90 Nagulkerke R = 0.14 2  2  *p<.05 a b  **p<.01  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one.  0.95  3  Catastrophizing, Fear  191  TSK. The results for the prediction of time two return to work from T S K scores is indicated in Table 48. The addition of T S K scores to the equation after entry of the control variable block, did not significantly improve prediction of time two return to work (% change = 1.04, p > .05). Wald tests did not indicate that any variable was adding significant unique variance to the prediction of time two average pain over the preceding week.  Catastrophizing, Fear  192  Table 48 Fear of movement/(re)injury Logistic Regression  in the Prediction of Time Two Return to Work: Hierarchical  Odds ratio' Variables  SE B  a  Wald  A£  a  Block 1 Age Sex Duration 1 Duration 2 Average pain 1  -0.01 0.44 -0.02 -0.01 0.01  0.02 0.56 0.04 0.02 0.01  0.00 0.63 0.28 0.07 0.38  0.03  0.03  1.02  1.00 1.56 0.98 1.00 1.01  2.73 TSK  1.03 1.04  Intercept = -1.61 X =3.77 Nagulkerke R = 0.06 2  2  *p<.05 a b  **p<.01  ***p<.001  values after step 2 (all variables in model) values after each step  Note. Duration 1, duration of injury at time one completion; Duration 2, duration of injury at time two completion; Average pain 1, Average pain over week preceding time one; TSK, Tampa Scale of Kinesiophobia.  Catastrophizing, Fear  193  Discussion Summary of Findings and Relevance to the Literature Three related sets of analyses were conducted; 1) confirmatory factor analyses of each of the pain related fear and catastrophizing scales, 2) development and testing of a theoretical structural model relating various aspects of pain related fear and catastrophizing and 3) regression analyses predicting 3 month post injury outcome from pain related fear and catastrophizing measured in the subacute stage of low-back pain. The overall purpose of the study was to gain a better understanding of the various instruments used to measure pain related fear and catastrophizing and by extension, the constructs they measure in a subacute pain population. This is an important undertaking, because the assessment of pain related fear in the early stages following injury holds promise for the prediction and prevention of chronic pain. The CFAs of the instruments was intended to determine i f established or hypothesized factor structures held in this sample. The structural model built upon the findings from the C F A by placing several of the pain related fear constructs identified in the CFAs into a structural model hypothesizing specific relationships between them. The regression analyses built on the findings of the structural model analyses by determining the degree to which pain related fear and catastrophizing were predictive of three-month outcome. The discussion will begin with an overview of the findings in each of the three sets of analyses, placing them in the context of the existing literature and then synthesize the results with reference to the existing literature and clinical relevance. Finally, limitations of the current work and implications for future work will be discussed.  Catastrophizing, Fear  194  Confirmatory Factor Analyses The purpose of these analyses was to examine whether previously derived factor structures for the PCS, PASS, F A B Q and T S K could be replicated in an subacute pain sample. Individuals who have experienced painful injuries recently, represent a group at risk for the development of chronic pain, particularly if their injury is financially compensable. Since pain related fear and catastrophizing hold great promise for aiding in the prediction and prevention of chronic pain, it is important that the characteristics of the instruments currently used in assessing these states be understood in a subacute pain population. Factorial validity in a subacute population is one issue warranting attention. In order to assess the factorial validity of each of the instruments the data collected from individuals with recent onset low-back pain was subjected to confirmatory factor analysis to determine the degree to which the covariance matrix calculated from the data "fit" with the matrix implied by specific factor structures. In each case the structure intended by the scales original authors was tested against a number of alternatives, derived from previous literature. Of all the instruments, the PCS demonstrated the best fit to the factor structure proposed by its authors in this sample. As was found in other populations (e.g., Osman et al., 1997, Osman et al., 2000; Sullivan et al., 1995; Van Damme et al., 2002) a three factor model of catastrophizing, with factors representing; 1) a tendency to dwell on negative aspects of one's pain 2) a tendency to exaggerate the threat value of pain stimuli and 3) a tendency to doubt one's ability to cope with pain. This is an important finding from the perspective of those wishing to use the PCS to predict or prevent the development of chronic pain. The finding that adequate fit to the three factor model of the PCS could be achieved with no post hoc model modifications is striking, given that previous confirmatory work required at  Catastrophizing, Fear  195  least some, and frequently considerable modification to achieve comparable fit levels (e.g., Osman et al, 1997, Osman et al., 2001, Sullivan et al., 2001). In the present study, a previously reported two factor model (Osman et al., 1997) did not provide adequate fit to the data, which appears to allay some concerns expressed by those authors that the three factor model might not be stable. Item-scale correlations were all high, indicating that all items should be considered important indicators of their respective subscales. As in prior studies, the three catastrophizing factors were found to be highly correlated. Interscale correlations ranged from .81 between helplessness and rumination to .90 between magnification and rumination. Correlations of this magnitude suggest that the three subscales might usefully be conceptualized as reflecting a single higher order catastrophizing factor. However, because a three factor first order model is statistically equivalent to a higher order model in S E M , it was not possible to test this hypothesis. The finding that the three subscales are differentially related to other constructs in, subacute and chronic samples suggests that the use of the subscale scores in addition or instead of the full-scale score may be useful, particularly if fine discrimination is desired, particularly for the purpose of clinical intervention. Also supporting this contention, is the finding in the current work that magnification at the subacute stage was uniquely predictive of depressive symptoms at three months. The original four factor model of the PASS did not fare well in the current sample. This contrasts with work supporting the four factor model to hold in chronic pain (McCracken, Gross, Hexum, et al., 1993) and community (Osman et al., 1994) samples. It must be noted here that the only study previously utilizing confirmatory factor analysis (Osman et al., 1994) employed different fit criteria than those employed in the current work. However, even according to the criteria Osman et al used (GFI > .90, AGFI > .80, R M R < .10), the four factor model of the full  Catastrophizing, Fear  196  data set did not approach adequate fit in the current sample (GFI = .73, A G F I = .70, and R M R = .141). These differences may be attributable to differences in the population sampled, i.e., a community (non pain) sample in Osman et al., versus a subacute pain sample in the present study. However, it may also reflect some of the factorial weaknesses in the instrument highlighted by Larsen et al. (1997). It is equally noteworthy that neither the alternative four factor model, nor the favored five factor model indicated in Larsen et al (1997) provided adequate fit in the current sample. In contrast, the abbreviated item set constituting the PASS-20 (McCracken & Dhingra (2002) did conform to the four factor structure, although one item (item 37 "I worry when I am in pain") loaded on fearful appraisals rather than cognitive anxiety as intended by the scale authors. It is noteworthy that this item is very similar to an item from the PCS (item 1. "When I am in pain .. I worry all the time about whether the pain will end". The finding that this item loaded on the fearful appraisal subscale replicates the finding by Larsen et al. (1997) that this item loaded with fearful appraisal items in their five factor model. The fact that this item appears to tap catastrophizing supports their contention that it and the fearful appraisal items measure catastrophizing. It is also noteworthy that all of the reverse scored items from the original scale were dropped in creating the PASS-20 scale. This fact may account for some of the difference of fit of the four factor model to the two item sets, since, as evidenced by previous work on the TSK (e.g., Clarke et al, 1996), the inclusion of reverse scored items can introduce complexity into a factor structure, by introducing method variance which is not accounted for by the specified factors. If this were the case, and in fact the negative scoring were introducing added method variance into the model, accounting for that variance by specifying all of the negatively scored  Catastrophizing, Fear  197  items onto a fifth (negative scored) factor, might be expected to improve model fit. Examination of residuals from the four factor model of the full item set on the current sample indicated that these variables were indeed contributing to the poor fit. The three largest residuals involved covariances between two of the reverse scored items (item 16 and 8, items 40 and 2, and items 16 and 2) and many others involved one of these items, suggesting that specification of a method variance factor for reverse scored items may have improved fit substantially. This possibility was examined post-hoc by creating a five factor model of the full PASS with the original four factors and a fifth Negative Scored factor. A l l negatively scored items were specified to load on both their original factor and the Negative Scored factor, while positively scored items loaded only on their original factors. S E M analysis revealed that such a model improved fit substantially (A% = 93.81 on Adf= 5, p <0.001), but not to a level where fit could be considered 2  adequate (e.g., CFI = .83). This indicates that the difficulties with the factorial validity of the full PASS extend beyond common method variance among negatively scored items. Also noteworthy is that 2 of the items dropped in creating the PASS-20 Fearful appraisal subscale (items 8 and 16), were found by Asmundson to load on factors other than the catastrophic thoughts factor in their five factor model. Although McCracken and LaDhingra (2002) based decisions on what items to drop in creating the PASS-20, on criteria other than factor analysis, it appears that factor analytic results, including the current work, support their choices. In summary, confirmatory factor analyses support the contention that the PASS, at least in its abbreviated form, the PASS-20, with item 37 specified to load on fearful appraisals, and one theoretically justifiable correlation between two item error terms, assesses four dimensions, fearful appraisals (catastrophic thinking), cognitive symptoms, physiological symptoms and  Catastrophizing, Fear  198  escape/avoidance behavior, which are moderately to strongly related in a subacute pain population This is the first demonstration of the validity of the four factor structure in a subacute pain population, as well as the first demonstration of the factorial validity of the abbreviated form of the PASS, the PASS-20. It does suggest that item 37 is problematic and might fruitfully be dropped or considered an indicator of fearful appraisals rather than cognitive symptoms. Superior factorial validity, combined with the increased efficiency and reduced demands on respondent and scorer, suggest that the PASS-20 version might be more appropriate for use in a subacute pain population than the full version. These findings will require replication, particularly as the adequate fit was achieved with some modification to the model. Confirmatory factor analysis of the F A B Q indicated that the decision by Waddell et al. (1993) to drop several items was warranted, at least from the perspective of factorial validity. While both single and two factor structures provided poor fit to the data for the full version of the F A B Q , the two factor model when applied to the abbreviated item set provided adequate fit with only one easily justified modification. The nature of this modification, a correlation between two measurement error terms, however, points to a more general difficulty with the scale, namely, the close similarity between many items. Several pairs of items represent nearly identical statements, for example, the two items which had correlated errors were, item 4. "I should not do physical activities which (might) make my pain worse" and item 5. "I cannot do physical activities which (might) make my pain worse". Such subtle differences in item content suggests that the items may be redundant. Indeed the significant correlation between their error terms (.45) despite their common loading on a single factor suggests that those two items may be redundant. Examination of the remaining Lagrange multiplier statistics indicates that a number of other item pairs also share correlated errors.  Catastrophizing, Fear  199  The correlation between the two subscales was moderate (0.51) and consistent with that found by others (Waddell et al., 1993, r = 0.39). While not high, this level of correlation indicates that the scales are not orthogonal, despite having been derived through orthogonal rotation (Waddell et al., 1993). In summary, despite considerable overlap in item wording, in some cases across subscales, the F A B Q appears to measure two moderately related dimensions of fear avoidance beliefs, fear avoidance beliefs about work and fear avoidance beliefs about activity. Confirmatory factor analysis of the T S K was less conclusive. The analyses failed to indicate adequate fit to any of the established factor structures. Several factors may account for this. The first is that the majority of factor analytic work to date on the T S K has involved the Dutch version of the instrument. It is possible that translation issues and/or cultural differences may underlie the failure of the current work to replicate any of the previous factor structures. Even the best translation is likely to alter the meanings of items slightly, and therefore potentially alter the way they interrelate. Similarly, language differences probably also reflect cultural differences, and individuals of different cultures may appraise particular constructs in different ways, leading to different groupings of items. In addition, the majority of work on the factor structure of the T S K has been conducted with chronic pain patients and the current sample was subacute. Given the apparent complexity of pain related fear, it is quite possible that it changes not only in magnitude, but also in quality, such that a different factor structure might be expected in samples of differing pain duration. However, two pieces of evidence argue against this interpretation. First, there is one other study examining the factor structure of the TSK in an acute/subacute sample (Swinkels- Meewise, Swinkels, et al., 2003). The factor structure indicated in that sample also did not provide an adequate fit to the data in the current sample.  Catastrophizing, Fear  200  Secondly, for all of the other pain related fear and catastrophizing measures, a factor structure previously found in chronic populations and in the case of the PCS and PASS, replicated in nonclinical samples, was replicated in the present subacute pain sample. It should be noted that the current sample was restricted to compensation claimants while the Swinkels-Meewise sample was not, and this difference may also have contributed to the failure to replicate. The success in replicating factor structures from other samples for the PCS, PASS, and F A B Q argues against this interpretation but does not rule it out, since there may be some aspect of the TSK, which is particularly sensitive to the receipt of compensation. It is possible to assess this possibility through confirmatory factor analysis by collecting data on a broader sample including individuals receiving compensation and others who are not, and conducting tests of factorial invariance. However, the current sample does not permit such analysis. As indicated in the preceding review, the literature on the T S K includes some replication of a two factor model of the TSK, but the evidence is not entirely convincing. Based on that review and the current work, the T S K while having demonstrated predictive validity, would appear to be in need of further work from the perspective of factorial validity. Without such work, it is difficult to place the instrument and the construct(s) it is measuring into a theoretical context with the other aspects of pain related fear and catastrophizing. Structural Model The construction of the structural model was informed by the findings from the CFAs. Specifically, it was decided to utilize the PCS to represent catastrophizing. The fact that the three factor structure proposed by the scale authors (Sullivan et al., 1995) provided good fit to the data for the current sample, suggested that it could be utilized as a measure of catastrophizing  Catastrophizing, Fear  201  in the structural model with no modification. Two subscales from the PASS were chosen to represent fear of pain, because they appeared to be stable in the current sample and of all the available indicators, they appeared to best capture the proposed fear of pain construct. The F A B Q activity subscale items were selected as indicators of fear of activity, because they too appeared to reflect a stable factor in the C F A of the F A B Q , and most closely captured the proposed fear of activity construct. While the C F A analysis was intended to ascertain the degree to which the items comprising each instrument fit together conceptually, the structural analyses were designed to determine how the concepts captured by the scales related to one another. The structural model provided a test of the factor structure of the instruments employed, in the context of other variables The fact that a well fitting measurement model could be constructed from subscales of existing instruments serves as evidence for the construct validity of the subscales utilized. The measurement model served as a test for the distinctiveness of the constructs specified and the good fit of that model indicates that pain related fear and catastrophizing as measured by current instruments can be construed as consisting of separate but related constructs. Testing of the structural model indicated that the relationships between those constructs are directional. Specifically structural model analyses provided partial support for predictive relationships implied in Vlaeyen and Linton's (2000) model (see figure 3). The structural model which best accounted for the data in the current analyses, indicated that the effect of pain on fear of pain is fully mediated by catastrophizing; the effect of pain on fear of activity is partially mediated by catastrophizing and the effect of catastrophizing on avoidance is completely mediated by fear of pain. The best fitting model retained a direct path from pain intensity to fear of activity, the paths indicating partial mediation of that relationship by catastrophizing. The  Catastrophizing, Fear  202  finding that the prediction of fear of activity from pain is only partially mediated by catastrophizing deviates slightly but not entirely from the prediction that the relationship would be fully mediated. Perhaps, most importantly, the model accounted for more than 19 percent of the variance in self-reported avoidance behaviour. However, two of the most important propositions inherent in the hypothesized model were not supported. First the proposition derived from the work of Taylor & Rachman (1992) that the relationship, between catastrophizing and fear of activity would be partially mediated by fear of pain was not supported. The model containing this path did not provide better fit than the simpler model not including it. While it does support the contention that fear of pain (as measured by the PASS cognitive and physiological symptoms subscales) and fear of activity (as measured by the F A B Q fear avoidance beliefs about activity) are distinct constructs, the current work does not support the contention that there is a causal relationship between them. Perhaps more surprisingly, the structural analysis did not indicate fear of activity (as measured by F A B Q fearful beliefs about activity) to predict avoidance behavior. Although the path from fear of activity to avoidance was retained in the best fitting of the apriori specified models, the parameter estimated for that path was not found to be statistically different from zero and dropping it from the model did not significantly detract from model fit. It is somewhat difficult to explain this finding, particularly given that the scale authors specified that the construct should bear on long-term disability through pain behavior, including avoidance and there is strong evidence for the contention that F A B Q scores do contribute to disability. Previous research indicated that the F A B Q activity subscale was significantly related to concurrent disability (Waddell et al., 1993) and that the total F A B Q score predicted outcome in acute pain patients (Fritz et a l , 2001). The current regression analyses indicated that the  Catastrophizing, Fear  203  F A B Q Activity subscale predicted three month perceived disability scores. It is noteworthy, however, that when entered together with the other pain related fear full scale scores in the current regression analyses, the F A B Q total score did not add significant unique variance to the prediction of time two disability. In fact, the variance contributed to the prediction by the F A B Q total score rounded to zero. This suggests that in the context of other pain related fear variables, the F A B Q , including the fearful beliefs about activity subscale may not be as important in the prediction of disability as has been suggested by studies which examined it in isolation. For example, Fritz et al. did not employ other measures of pain related fear or catastrophizing in their longitudinal study. Likewise, Crombez, Vlaeyen, et al. (1999) did not directly compare the contribution of F A B Q scores to other measures of pain related fear and catastrophizing. A n alternative explanation for the failure to support a role for fear of activity in avoidance behavior is the way in which various constructs in the model were measured. The specification of Avoidance behavior as the proportion of usual activities currently avoided may have contributed to the failure to find a significant effect for fear of activity. Questions on the PBC, the instrument specified to measure avoidance allows only a "yes" or "no" response indicating whether or not the individual ever avoids particular activities because of pain. Thus, the measure provided a count of the number of activities avoided but not the degree to which each is avoided. When avoidance is measured in this way, an activity, which has been avoided even once, is considered equal to an activity, which is consistently avoided. Fear of activity might better the extent to which an individual will avoid particular activities believed to be dangerous, but not the breadth of activities avoided. The discrimination between fear and anxiety and their likely effects provided by Asmundson, et al. (in press) suggests a potential mechanism for such an effect. The PASS items  Catastrophizing, Fear  204  used as indicators of fear of pain (despite being labeled anxiety symptoms) may reflect a fear response, while the F A B Q beliefs about activity items (used in the current work as indicating fear of activity) appear to reflect the kind of the conscious deliberation described by Asmundson et al as more typical of anxiety. They argued that the response to anxiety is likely to be more intentional and considered than the response to fear. A n individual who experiences pain related anxiety would be more likely to appraise particular activities for the degree of danger they represent, and avoid those activities extensively. They describe the more typical response to fear as being escape. However, the discrimination between escape and avoidance is a fine one, particularly with regard to activity. The only difference appears to be how close one gets to the feared stimuli. Escape is stopping once you are engaged, while avoidance implies never approaching. However, becoming involved in any activity is a gradual process and it is sometimes difficult to define at what point you actually became engaged. For example, does walking begin with the act of rising from the chair, with raising one foot, or with moving that foot forward? At what point in this progression does stopping imply escape from the act of walking as opposed to avoidance of walking. Similarly, i f one stops at the point of rising from the chair because of twinge of leg pain, has one avoided or escaped? However, i f one considers the difference between fear of pain and fear of activity, this same example speaks to the difference between broad generalized avoidance and avoidance of specific activities. Fear of pain might lead one to avoid any activity which has triggered pain in the past or which is anticipated to trigger pain, where fear of specific activities might not generalize to other activities in the same manner. If I have learned to fear twisting my head in a certain direction because I fear it will cause damage to my neck, I may not fear bending my arthritic leg, which also hurts, because the same fear is not attached to that activity. However, i f I fear pain itself,  Catastrophizing, Fear  205  then I may avoid both activities, or stop them as soon as I experience pain. Thus the range of activities avoided may actually be broader i f one is afraid of fear as opposed to being afraid of activity. Yet another possible explanation, following similar logic, can be derived from the work of Gheldof, de Jong, Vinck, & Rudd (in press). Those authors draw on the broader literature on beliefs to inform the understanding of pain related fear beliefs. They argue that beliefs about pain and its consequences can be implicit or explicit. Implicit beliefs are more long-standing habitual ways of viewing the world and tend to influence thinking and behavior in situations where time or other factors limit our ability for conscious deliberation. Explicit beliefs on the other hand are more accessible to conscious thought and are based on a more thorough appraisal including information from others and social demand characteristics. Gheldof et al. argue that when asked about beliefs we may be more likely to report our explicit beliefs due to demand characteristics and the luxury of time to deliberate. However, when faced with an imminent threat we may fall back on implicit beliefs. Therefore, when we measure beliefs we may not be gaining access to the actual beliefs, which dictate behavior. The measure of fear of pain (PASS cognitive and physiological subscales) ask about responses to stimuli while the measure for fear of activity (the F A B Q activity subscale) measures stated beliefs. Based on Gheldof et al.'s analysis the latter are more likely to be susceptible to response bias, e.g., individuals asked for their belief about the safety of activity may be reporting as their own belief what they have been told by health care practitioners. However, the same individual may hold an intrinsic belief which runs counter to that extrinsic belief and in circumstances where the threat is more imminent, e.g., when pain is anticipated or experienced, that intrinsic belief may be triggered and lead to avoidance behaviour. Asking the same individual if they experience fear symptoms in  Catastrophizing, Fear  206  the presence of pain may not elicit the same response bias, since they have not likely been informed as to the appropriateness or inappropriateness of those symptoms. Likewise, selfreport of behaviour may not be as susceptible to response bias, as beliefs because the former are more salient. That is, I can remember if I went dancing last week, but I may not remember what my reasons for doing it or not doing it were. To follow this line of reasoning one step further, the very fact that one reports experiencing fear symptoms in response to pain, particularly cognitive interference, would also indicate that one is likely to respond to pain based on implicit beliefs rather than explicit beliefs which take more cognitive effort. If this were the case, then as fear of pain increased, one would expect the connection between explicit fearful beliefs and behaviour to become weaker. That interactive hypothesis was not tested in the current work, but might represent an interesting avenue for future work. In contrast to the relative specificity of avoidance, which might result from anxiety, fear (in this case fear of pain) might result in a more diffuse avoidance. Asmundson has argued that fear should contribute to escape behaviors more than avoidance, because avoidance by its nature is a considered and more deliberate action of the kind, which would arise from longer-term diffuse arousal, such as that attributable to anxiety. Escape on the other hand is a more discreet immediate response such as might be occasioned by the specific and relatively immediate arousal response, which is fear. However, as argued earlier, the line between escape and avoidance is not always clear. A n individual who escapes a stimulus is doing so, at least partly to avoid escalation of that stimulus. A final potential explanation for the failure to find fearful beliefs about activity to predict avoidance in the structural model, lies in the connection between fear of pain and affective  Catastrophizing, Fear  207  responses to pain (e.g., McNeil & Vowles, in press). It may be that general tendencies toward avoidance are related to issues such as self-efficacy and depressed mood, which are more related to fear of pain and catastrophizing than to fear avoidance beliefs about activity. While the current findings contrast with previous work indicating F A B Q scores to be predictive of avoidance behavior, the analysis utilized here is somewhat different in that it considers that relationship within the context of several other variables, perhaps most importantly within the context of catastrophizing and fear of pain. It may be that the addition of those variables is what leads to the non-significant contribution of fear of activity. Indeed the F A B Q avoidance scale showed a significant zero order correlation with both the pain behavior questionnaire total score and with the PASS escape/avoidance subscale. The results may also reflect a change in the role of pain related fear and catastrophizing over time. Linton et al., found that initial levels of pain related fear and catastrophizing were not as important as change over time, such that individuals with high fear at three months had poor outcome at nine months. It may be that as originally hypothesized here, initial levels of catastrophizing predict fear of pain immediately following injury and that fear of activity develops more slowly. Such relationships may not have been captured by the cross sectional methodology employed here. Thus the relationship between fear of pain and fear of activity may not have appeared at the early stage examined in this work and likewise the relationship between fear of activity and avoidance may not have appeared this early. The fact that fear of activity was predictive of outcome suggests that beliefs may have an impact beyond immediate avoidance. Pain Related Fear and Catastrophizing in the Prediction of Outcome at Three Months The regression analyses were intended to extend the C F A and structural model analyses by testing the predictive validity of the various instruments in a prospective manner. While the  Catastrophizing, Fear  208  structural model analysis provided a test of the internal components of Vlaeyen's model, the regression analyses provided a test of the external predictions, namely does pain related fear and catastrophizing predict the persistence of pain, perceived disability and depressed mood which characterize chronic pain and does it predict functional outcome operationalized as return to work. In addition, the use of multiple measures of pain related fear and catastrophizing allowed for comparison between them in this regard. In the case of depressive symptoms, each of the pain related fear scales (PASS, F A B Q and TSK) contributed significantly to the prediction of time two depressive symptoms. In the analysis combining all of the scales, catastrophizing (PCS) scores contributed unique variance beyond that contributed by any of the pain related fear. Of the three components of catastrophizing, helplessness contributed uniquely over and above the shared contribution of the other two aspects of catastrophizing, magnification and rumination. This would seem consistent with the connection between self-efficacy and depression (Bandura, 1986). Individuals who are prone to feeling little power over their pain or have learned to believe they have little power over it would also seem likely to develop similar beliefs about other aspects of their life. Given the similarities between the fearful appraisal subscale of the PASS and catastrophizing, highlighted earlier, it is interesting that fearful appraisals did not provide a unique contribution to the prediction of depressive symptoms beyond the contribution it shared with the other PASS subscales (cognitive symptoms, physiological symptoms, and escape/avoidance). However, as noted earlier, the fearful appraisal items appear to bear the most similarity to the PCS rumination and magnification subscales. Therefore, the finding that fearful appraisals did not contribute significant unique variance in the context of the other PASS subscale scores, further supports the contention that the helplessness dimension of catastrophizing is particularly important in  Catastrophizing, Fear  209  predicting later depressive symptoms. Also interesting is that the F A B Q activity beliefs subscale contributed unique variance to the prediction of depressive symptoms at three months over and above that shared with the F A B Q work beliefs. This is surprising, given the previously demonstrated higher association of the work subscale with concurrent depressive symptoms (Waddell, et al., 1993) and other work showing F A B Q total scores to be more highly associated with disability and work status than with affective variables in a chronic sample (McNeil & Vowles, in press). It suggests that activity avoidance beliefs are more important in predicting later depressive symptoms than are work avoidance beliefs. Equally surprising, for the same reasons, was the finding that fear of movement/(re)injury contributed significantly to the prediction of depressive symptoms at three months. With regard to the prediction of perceived disability at three months, the combination of the four pain related fear scales did contribute significantly to the prediction of perceived disability at three months even after controlling for initial pain level and demographic variables, but none of the pain related fear total scale scores contributed significant unique variance to the prediction. As might be expected from previous reports that they are most closely connected with disability and functional outcome (e.g., Crombez et al., 1999), when each scale was analyzed in isolation from the other pain related fear scales, both fear avoidance beliefs (FABQ) and fear of movement/(re)injury (TSK) contributed significantly to the prediction of perceived disability at three months while pain anxiety symptoms and catastrophizing did not. Of the two subscales, based on the unique variance contributed to prediction, activity avoidance beliefs were clearly responsible for most of this effect. Again this is surprising given previous indications that fear avoidance beliefs about work was a stronger predictor of disability (Waddell et al.,  Catastrophizing, Fear  210  1993). Given the results of the other regression analyses, it was somewhat surprising that the pain related fear and catastrophizing variables together did not contribute significantly to the prediction of three month pain levels, after controlling for demographics, and time one pain levels. Likewise, the only scale, which contributed significantly to the prediction of three-month pain levels when each scale was entered independently was the F A B Q . O f the two subscales making up that instrument, only activity avoidance beliefs contributed uniquely. The relatively weak prediction of pain level from pain related fear and catastrophizing, relative to the prediction of depressive symptoms and perceived disability, may partly have been attributable to the fact that time one pain level was included as a control variable. Indeed, the total amount of variance accounted for in the models predicting pain level were similar to those obtained in predicting depressive symptoms and perceived disability. Therefore it seems likely that initial pain level shares more variance with pain related fear and catastrophizing in the prediction of future pain, than in the prediction of future depressive symptoms and future perceived disability. The pain related fear and catastrophizing variables together did not contribute to the prediction of three-month work status, although their contribution approached statistical significance. When each of the scales was analyzed in isolation, the fear avoidance beliefs, as assessed by the two subscale scores from the F A B Q combined, did contribute significantly to the prediction of return to work at three months, and fear avoidance beliefs about work contributed uniquely to that prediction. A one point increase in F A B Q work scale scores indicated a 25% decreased chance of returning to work. This finding is consistent with previous work also finding the F A B Q work subscale to predict failure to return to work (Fritz, et al., 2001). Perhaps more surprisingly, the PASS subscales as a group contributed significantly to the  Catastrophizing, Fear  211  prediction of failure to return to work, and fearful appraisals contributed uniquely to that prediction. As might be expected, given the degree of overlap between fearful appraisals and catastrophizing, particularly magnification, the magnification subscale of the PCS also contributed significantly to the prediction of three month return to work status. Strangely, both magnification and fearful appraisals contributed to prediction of return to work status in the opposite direction expected i.e., low scores on both predicted a lower likelihood of having returned to work at three months. Examination of the zero order correlations of these variables with failure to return to work revealed non-significant but negative values for both fearful appraisals (r = -0.03) and magnification (r_= -0.06). In both cases, other subscales from the same instrument showed higher zero-order correlation with failure to return to work, but in a positive direction. This suggests that the significant regression effects for fearful appraisals and magnification may have been the result of suppression effects (Tabachnick & Fidell, 2001) and therefore represented statistical artifacts. However, the 95% confidence intervals around the odds ratios for both variables fall entirely below one, suggesting that they are both meaningful predictors of failure to return to work. In addition, the fact that similar results occurred across two separate analyses, indicates that they may not be simply statistical artifacts and this possibility warrants further study. It is difficult to find strong theoretical reasons to expect such a relationship, and there is little in the other results from the current work which would inform speculation, but should the effect be replicable, it would be important to explore its meaning. Two potential explanations can be proposed tentatively. One possibility, is that individuals who are prone to catastrophizing also catastrophize about the impact of being off work and thus are more highly motivated to return.  Catastrophizing, Fear  212  A second possible explanation for the reverse effect of elements of catastrophizing in the prediction of return to work, lies in the possibility that individuals who catastrophize about pain have higher expectations of recovery. Higher expectations of recovery have been shown to predict better outcome with regard to occupational disability (Schultz et al., 2004). The proposition that catastrophizing might be related to higher expectations of recovery seems counterintuitive, since one would expect the tendency of catastrophizers to have negative expectations around their pain to extend to their expectations of recovery. However, Sullivan et al. (2001), found in a non-clinical sample, that individuals high in catastrophizing actually underestimated future pain levels. This may represent a general tendency of catastrophizers to cope through moderating their expectancies. This might lead to an increased likelihood of early return to work. However, it might be expected that such individuals would subsequently experience a higher rate of failed return to work. Sullivan et al., point out that this strategy of underestimation led to an increased experience of emotional distress during a painful procedure. It was not possible to evaluate any of these possibilities in the current work, but they present interesting avenues for future research, should the reverse effect of catastrophizing on return to work be found to represent a reliable finding. It is surprising that fear of movement (re)injury (TSK) scores did not predict failure to return to work at three months, given previous research suggesting that it should be more predictive of functional outcomes (McNeil & Vowles, in press) than PCS or PASS scores. The overall modest degree of variance in outcome attributable to pain related fear and catastrophizing, particularly in prediction of return to work suggests the importance of other factors. Indeed from previous research it is clear that a range of other biopsychosocial factors may be operative (Crook et al., 2002, Schultz et al., 2002). One possibility is systemic issues.  Catastrophizing, Fear  213  Krause and Ragland (1994) suggested that it is likely that injured workers will be encouraged to attempt light duty or restricted hours return to work by Compensation providers. It may be that workers in the current sample had returned to work due to such encouragement from the Workers' Compensation Board. This possibility points to the importance of considering broader biopsychosocial variables in predicting return to work, and also to the value of assessing work disability in other ways, such as duration or wage loss (e.g., Schultz et al., 2002). However, these variables are somewhat more complex to assess and for the current work, being focussed less on specific outcomes than on sampling a range of outcomes, dichotomous categorization of return to work was considered adequate. In addition, it was expected that over the course of the three month period covered by the current work, that relatively few patients would evidence the pattern of pain recurrence which would make days lost an important outcome measure. Another issue, which may have complicated prediction of return to work, is variability in job demands. It is quite likely that return to work is to some degree determined by the nature of the job, and this might reduce the impact of psychological processes. A n individual who has a lighter job is more likely to return to work at an earlier point in recovery, since there is less chance of pain exacerbation or reinjury regardless of psychological variables. Overall Summary Taken together, the results from the current work, consistent with previous findings, suggest that pain related fear early after the onset of low-back pain is important in the prediction of outcome. The results expand on previous work in this regard by including multiple measures of pain related fear and catastrophizing and examining outcome across multiple domains including affective distress, pain level, perceived disability and return to work. Each of these variables is an important part of the chronic pain experience and previous research indicates that  Catastrophizing, Fear  214  they are related but also somewhat independent. Each has weaknesses as a sole measure of outcome (e.g., Turk, 1997), but together, they provide a relatively comprehensive picture. In the current work, pain related fear and catastrophizing, in individuals experiencing subacute work related back pain, were shown as a group to add significantly to the prediction of perceived disability, depressive symptoms and failure to return to work at three months post injury, even after controlling for demographic variables, and initial pain level. While the amount of variance in the different outcomes attributable to pain related fear was typically moderate, ranging from approximately 10 percent in failure to return to work to 23 percent in depressive symptoms, those percentages are not negligible. Pain related fear and/or catastrophizing added to the prediction of all outcome variables at a statistically significant level. Despite significant overlap between the various pain related fear and catastrophizing instruments in the prediction of outcome, different pain related fear and catastrophizing variables contributed significantly, supporting the contention by McNeil and Vowles (In press) that the choice of instrument may depend on the purpose. The pattern of results was somewhat inconsistent with what might have been expected based on prior research, which demonstrates the benefit of using the four pain related fear and catastrophizing scales in one study. Pain catastrophizing (PCS) and pain fear/anxiety (PASS), which in the past have been associated more with affective aspects of pain, did predict future depressive symptoms, but they (or constituent factor(s)) also contributed to the prediction of future perceived disability and, in the opposite direction expected, to the failure to return to work. Similarly, while fear of movement and (re)injury (TSK) and fear avoidance beliefs (FABQ) contributed significantly to later disability, both also predicted future depressive symptoms. The F A B Q subscales in particular demonstrated rather specific effects. The contribution of F A B Q scores to predicting depressive  Catastrophizing, Fear  215  symptoms was nearly exclusively attributable to Fear-avoidance beliefs about activity, while the contribution of the F A B Q in predicting failure to return to work appeared to be exclusively attributable to the contribution of the work subscale. The effect of fear avoidance beliefs about work appears to be very specific to predicting work status, as that subscale did not contribute uniquely to the prediction of any other outcome variable. The most surprising finding was that higher scores on the magnification component of catastrophizing, and higher scores on the fearful appraisals component of the PASS significantly predicted an increased chance of having returned to work at three months. This runs counter to what would be expected i f catastrophizing leads to increased functional disability. While the direction of that effect is difficult to explain, it further adds to the contention that different aspects of pain related fear and catastrophizing are discriminable and that there is utility in attempting to discriminate between them. Fearful appraisals and magnification are clearly related based on their item content, and this similarly unusual finding across the two scales further supports that contention. However, it must be acknowledged that for the most part, the shared contribution of the various pain related fear and catastrophizing scales in predicting outcome was much greater than the unique contribution of any scale or subscale and therefore, the current instruments may not be sensitive or specific enough to provide significant discrimination between aspects of pain related fear and catastrophizing. The current results suggest that pain related fear should not be considered a unitary ' construct and that different aspects of pain related fear are predictive of different aspects of outcome. Most authors acknowledge that there are a number of interrelated aspects to pain related fear, and recent theoretical developments, particularly the work of Asmundson, et al. (in press) and for catastrophizing, the work of Sullivan and colleagues (e.g., review 2001) have  Catastrophizing, Fear  216  refined the constructs involved dramatically. However, the existing instruments do not appear to have kept pace. Each is based on a slightly different conceptualization of pain related fear and it is not clear from the instruments themselves how they relate. Clearly, they are tapping into different constructs or at least different aspects of one construct, but it is not clear what those constructs are or how they interrelate. The existing instruments have proven utility in their use in developing the current literature. The instruments have also been used in a variety of research which has demonstrated the importance of pain related fear and catastrophizing, and in driving the development of theory, but many are difficult to understand within the framework of current conceptualizations and they appear to be limited in terms of providing further contribution to the refinement of knowledge on pain related fear and catastrophizing. The PCS appears to be the exception in this regard. Interestingly, despite the fact that it was developed outside the theoretical framework of pain related fear (see review by Sullivan, 2001), it appears to mesh very well with that theory and the current work demonstrated that it can be used in research guided by models of pain related fear. The current work builds on previous research demonstrating the PCS to have a relatively robust factor structure across pain and community samples. In addition, the findings reported here support the construct validity of the instrument by demonstrating, that within the context of pain, pain related fear and avoidance, it performs as expected, mediating the relationship between pain and pain related fear. The predictive validity of the PCS and by extension of catastrophizing was demonstrated by the finding that PCS scores contributed significantly to the prediction of three-month outcome. The validity of considering catastrophizing as three related dimensions was further supported by the finding of differential contributions for two of the subscales in predicting outcome. Specifically, the helplessness subscale contributed independently of the other subscales in the prediction of  Catastrophizing, Fear  217  three-month depressive symptoms. The other instruments tested here do appear to hold some promise. The PASS in its abbreviated form (PASS-20) appears to reliably assess the factors intended in this subacute, work disabled sample. However, there is some debate about aspects of those factors. For instance, Larsen, et al. (1997) raised the concern that physiological and cognitive symptoms subscales may actually be assessing pain responses, rather than fear per se. This issue is at least partly addressed by the current work in that the relationship between cognitive and physiological symptoms were demonstrated to load together in the context of a model containing pain, catastrophizing, avoidance and another pain related fear construct. In addition, the fact that a mediating role for catastrophizing in the relationship between sensory pain and pain related fear suggests that the cognitive and physiological symptoms are somewhat independent from pain levels. However, this evidence is tentative given the way in which the constructs were measured. The measure of pain intensity (MPQ-Sensory items) related to current pain, while the PASS subscales specifically refer to "when I am in pain", which is more general than current pain level. A second issue with regard to the PASS, which the current work highlights is the utility and validity of measuring disparate aspects of pain related fear in one instrument and combining them in a sum score. The PASS clearly does not cover all aspects of pain related fear and/or anxiety, but the current work suggests it does cover at least three. There were several indications in the current work that fearful appraisals as measured in the PASS are actually measuring pain catastrophizing. This is apparent from careful examination of the items themselves, from the loading of one cognitive symptoms item with similarities to catastrophizing on the fearful appraisal subscale in the C F A and from similar patterns of association with outcome variables between the fearful appraisals subscale and aspects of pain catastrophizing as measured by the  Catastrophizing, Fear  218  PCS. Finally the PASS-20 conformed to the intended four factor structure, better than the full version, and the items dropped in creating the PASS-20 version have been found to reflect constructs other than catastrophizing (Larsen, Asmundson, & Norton, 1997). It would appear that retaining only the items best reflecting catastrophizing improved the degree to which the PASS conformed to its intended factor structure, including a separate fearful appraisals/catastrophizing factor. The overlap of fearful appraisals the constructs assessed by the PCS and the well demonstrated factorial validity of the PCS with only 13 items suggest that there may be little utility in having a component of the PASS, which also measures these constructs. If the intent is to have a comprehensive inventory, assessing a range of pain related fear and anxiety dimensions, the PASS would need to be expanded. Alternatively i f the intent were to measure only pain related fear, it would appear unnecessary and somewhat confusing in terms of interpretation to include fearful appraisals (catastrophizing). It must be noted in making this argument, that several of the premises were not directly tested. Specifically, the degree to which the fearful appraisals subscale measures catastrophizing is based on item similarity and disparate empirical findings suggestive of relationship, but not conclusive in that regard. Demonstrating redundancy would require more detailed factor analytic work on catastrophizing and Fearful Appraisals. Although the structural model analyses demonstrated that catastrophizing could profitably be separated from other aspects of the PASS and in fact predicted them in the context of other pain related variables, that demonstration did not use the Fearful Appraisal subscale and the conclusion as to the standing of that subscale can therefore, only be viewed as tentative. The F A B Q demonstrated reasonable adherence to the factor structure derived from a sample of chronic pain patients in the current subacute sample, and the results of those analyses suggest that the selection of items by Waddell et al. (1993) was appropriate. However, the  Catastrophizing, Fear  219  moderate correlation between subscales suggests that they are not orthogonal as originally intended. In addition, the apparent correlations between several item error terms, sometimes for items on different subscales, suggests that there is a significant degree of method variance due to similar wording in many items, which is not accounted for by the specified factors. This method variance may in part account for the lack of orthogonality. The finding that the subscales when measured at the subacute stage, differentially contribute to the prediction of later depressive symptoms (activity subscale) and failure to return to work (work subscale), does indicate that there is utility in differentiating the constructs, suggesting that reducing overlap due to method variance such as common wording might usefully increase discriminant validity. The failure to support any of the previously reported factor structures raises concerns about the construct validity of the T S K in a subacute sample. The instrument has been demonstrated to predict concurrent perceived disability and performance on physical tasks (e.g. Crombez, Vlaeyen, et al., 1999) in chronic samples and in the current subacute sample, to significantly predicted perceived disability and depressive symptoms at three months post injury. The fact that none of the instruments when entered together contributed significant unique variance to the prediction of any of the three month post injury outcome variables, suggests that there is significant overlap between them, despite the fact that the constructs they assess do seem to be separable. This points to the need for further refinement of the instruments to more precisely assess the separate aspects of pain related fear and catastrophizing. Clinical Relevance of the Findings The current work has several clinical implications. First it was demonstrated that aspects of pain related fear and catastrophizing in recent onset low-back pain, as measured by the most commonly utilized instruments, the PCS, PASS, F A B Q and T S K are predictive of depressive  Catastrophizing, Fear  220  symptoms, perceived disability, pain level and failure to return to work at three months postonset, even after controlling for initial pain levels and demographic variables. This suggests that the assessment of pain related fear at the early stages following injury is an important undertaking for prediction and potentially to guide interventions to prevent chronicity. Secondly, the current work demonstrated that for three of the instruments, the PCS, PASS and F A B Q , factor structures previously found in other populations could be replicated in an subacute lowback pain sample. This suggests that, for at least these three instruments, the structure of their component subscales, and, therefore, of the constructs they assess, is similar in individuals with subacute pain and other populations. The third broad implication of the current work is that catastrophizing predicts other aspects of pain related fear, including fear of pain and fearavoidance beliefs about activity, and that fear of pain predicts the number of activities avoided. The hypothesis that Fear Avoidance beliefs about activity predict avoidance behaviour was not supported although that relationship approached statistical significance. Together, these findings largely support the pain related fear-avoidance model proposed by Vlaeyen and Linton (2000), which posits that pain related fear is instrumental in the development and persistence of chronic pain problems. The current work also supports the contention implicit in that model, that pain related fear and catastrophizing represent important targets for early assessment and intervention to predict and prevent the development of chronic pain. If the promise of utilizing measures of pain related fear and catastrophizing to predict and prevent chronic pain problems is to be met, it will be essential to assess pain related fear and catastrophizing at the earliest stage possible in order to prevent individuals with subacute pain from falling into the self perpetuating fear-avoidance cycle. The current work indicates that existing instruments might be very useful in this regard. Despite the conceptual concerns raised  Catastrophizing, Fear  221  earlier, all of the existing instruments demonstrated predictive validity in the current subacute pain sample. However, assessment of pain related fear at the early stages of pain poses problems. The. majority of individuals with acute and subacute pain problems present for primary care, where the resources and motivation to assess psychosocial variables is limited (Linton & Boersma, in press). Assessment in such settings would of necessity be brief but relatively comprehensive, in order that it could inform the need for further intervention or referral. Steps toward developing brief measures of pain related fear have been taken (e.g., Linton & Hallden, 1988; Vlaeyen, et al., 2001; Vlaeyen, et al., 2002). The screening instrument developed by Linton & Hallden (1998), contained three items reflecting fear avoidance beliefs, two from the F A B Q ("Physical activity makes my pain worse", and "I should not do my normal work with my current pain.") and one from a related instrument ("An increase in pain is an indication that I should stop what I am doing until the pain decreases"). When administered to individuals early after the onset of pain, these items significantly contributed to the prediction of pain, self reported impairment in physical activity and accumulated sick days six months later. In a second study utilizing the same questionnaire, Linton & Boersman (in press), found that the fear-avoidance items were predictive of future sick leave over the subsequent six months, but a stepwise regression analysis resulted in these variables being dropped from the final model, indicating that other psychosocial variables are also important in the prediction of sick leave. The breadth of pain related fear variables included in this instrument was relatively narrow and the current work suggests that there might be benefit in including items reflecting other aspects of pain related fear in such an instrument, particularly, i f as Linton and Boersma (In press) argue, such an instrument could be used to inform a brief behavioural analysis which could then guide intervention.  Catastrophizing, Fear  222  In order to effectively address pain related fear, it may be important to identify the specific threat which is anticipated (Crombez et al. 2002) and such identification would be added by a broader assessment of pain related fear and catastrophizing. For example, the inclusion of items assessing catastrophizing in such an instrument may be important, since catastrophizing has been implicated in the degree to which interventions to address pain related fear might generalize across stimuli (Crombez et al. 2002). The finding that catastrophizing mediated the relationship between pain levels and pain related fear in the current work adds further to this assertion, since it is possible that catastrophizing may arise earlier in the fear/avoidance cycle and therefore represent the best target for early intervention. Interventions for panic (e.g. Barlow, 2002) and health anxiety/Hypochondriasis (e.g. Warwick et al., 1996) have demonstrated success in treating those disorders by specifically addressing catastrophic cognitions. Likewise, the simple provision of written information, which counters catastrophic beliefs and interpretations (e.g. describing the usual course of back pain and the importance of staying active), early after pain onset has been demonstrated to reduce pain related fear and improve outcome (Burton et al., 1999). Another instrument, designed for the purpose of monitoring treatment progress as opposed to prediction, did include items assessing catastrophizing and other aspects of pain related fear (Vlaeyen et al., 2001; Vlaeyen et al., 2002). Vlaeyen and colleagues developed a brief instrument assessing multiple dimension of pain related fear for daily administration in two separate treatment case studies with chronic pain patients (Vlaeyen et al., 2001; Vlaeyen et al., 2002). The instrument consisted of 11 items selected from the PASS, PCS and TSK, to assess the various dimensions of pain related fear and catastrophizing tapped by the three instruments. They analyzed the effects of the items from each instrument separately in both papers. In one paper  Catastrophizing, Fear  223  (2002) they found the pattern of changes in catastrophizing (PCS items), fear of pain (PASS items) and fear of movement/(re)injury (TSK items) measured daily, to be very similar, suggesting redundancy across the constructs. However, in one of the studies, (Vlaeyen et al., 2001) a differential effect across items from the different scales was demonstrated.  In that  study, the T S K items demonstrated a larger effect in response to the treatment manipulation than did the PCS and PASS items. The intervention involved exposure exercises targeted at feared movements, so it is perhaps not surprising that the T S K was most sensitive in this context. It would be very interesting to determine i f interventions specific to other aspects of pain related fear and catastrophizing produced similar specific effects. The most relevant point which can be derived from these findings, is that the three aspects of pain related fear and catastrophizing could be discriminated at all, given the limited number of items, which apparently were not selected for maximal discriminant validity. Examination of the items indicates that one of the catastrophizing items is nearly identical to the item selected to represent fearful appraisals on the PASS. As indicated by this overlap, the intent seemed more to represent each of the subscales of the pain related fear and catastrophizing instruments than to represent theoretically distinct constructs. Together, the work of Vlaeyen and colleagues (Vlaeyen et al., 2001; Vlaeyen et al., 2002) and Linton and colleagues (Linton & Buer, 1995; Linton & Hallden, 1988), indicates that brief screening instruments to assess pain related fear and catastrophizing is possible and useful, and the work of Linton and colleagues in particular points to the value of utilizing such an instrument early after the onset of pain. However the two instruments developed also point to issues, which are highlighted by the current work. Namely, that it is important to consider different dimensions of pain related fear in assessment, and that a theoretical framework used be used to  Catastrophizing, Fear  224  guide the selection of items. Selecting items based on a theoretical model of pain related fear and catastrophizing and the ability of particular items to assess the relevant theoretical constructs would increase the efficiency and interpretability of the results. For example, if one considers fearful appraisals to measure catastrophizing, then it would be uninformative to include both catastrophizing and fearful appraisal items. The current work holds similar implications for the more in depth assessment of pain related fear and catastrophizing which might occur in clinical settings beyond primary care or in research settings, namely that careful attention needs to be paid to issues of construct definition and measurement. The existing pain related fear instruments were developed from different theoretical foundations and were not brought together under a common theoretical framework until the synthesizing reviews by Asmundson and colleagues (1997) and Vlaeyen and colleagues (2001). That framework in turn, has been refined (e.g., Asmundson et al., in press) and it appears time for the refinement of measures, based on these theoretical advances, which would allow finer discrimination between aspects of pain related fear and provide more conceptual clarity. That may involve revision of existing measure to achieve more theoretical clarity as appears has begun to occur with the PASS (e.g., M c Williams & Asmundson, 1999). Ironically, the current work suggests that the PCS, which arose entirely outside of the pain related fear framework appears to best fit within that framework as evidenced by the performance of that instrument in the structural model tested in the current work. It is worth noting however, that the PCS was developed on a theoretical basis and subsequently has undergone a great deal of psychometric and theoretical examination of its conceptual properties, for example, extensive debate about the degree to which it is differentiable from depression and other aspects of negative affect (Sullivan et al., 2000). Such conceptual issues are less well examined for the  Catastrophizing, Fear  225  current measures of pain related fear (McNeil & Vowles, in press), although the current work contributes important information. The current findings indicate that although there is considerable overlap and perhaps redundancy among current measures there is utility in conceptualizing pain related fear and catastrophizing as multidimensional. The structural model demonstrated that the constructs can be differentiated, even based on the existing measures, and more generally the findings suggest that through consideration of recent theoretical developments, the existing instruments might be refined or new ones developed, which would allow finer discrimination. Development of new instruments or refinement of existing ones toward increased conceptual clarity would aid in testing elaborations to the fear-avoidance model of pain related fear, such as that proposed by Asmundson et al. (in press), and provide data to inform further theoretical developments. In clinical work, more specific and clearly defined instruments will assist in more efficiently planning treatment. For example, knowing details about the specific object of fear (Crombez et al., 2002), i.e., pain versus activity may be important in directing treatment. The finding in the current study that pain related fear contributes only modestly predicted return to work at three months suggests, consistent with Linton and Boersma (in press) that other psychosocial variables also need to be considered. Indeed, recent work suggests that a broad range of psychosocial factors predict outcome among injured workers receiving W C B benefits (Schultz et al., 2002). Krause and Ragland (1994) point out that after three months of work absence, most Workers' compensation systems establish a "light duty" or limited hours return to work. Systemic issues such as this may have caused some workers in the current sample to have returned to work despite ongoing pain and disability. As stated above, one goal driving improved assessment is to direct treatment more  Catastrophizing, Fear  226  effectively. The current findings also hold implications for treatment. The findings clearly indicate that pain related fear and catastrophizing early after the onset of pain are important in predicting outcome at three months. Recent studies (e.g., Burton et al., 1999; Vlaeyen et al., 2001; Vlaeyen et al., 2002) indicate that intervention to directly address pain related fear can be valuable at different stages of pain and disability. Interventions as simple as providing information to patients with relatively recent pain onset, attending primary care regarding the safety and benefits of resuming normal activity have been demonstrated to reduce the likelihood of long-term disability (Burton et al. 1999). Interventions targeting pain related fear through graduated exposure have shown promise in reducing pain related fear and catastrophizing in chronic pain patients and perhaps more importantly in reducing perceived disability in those patients who were initially fearful (Vlaeyen et al., 2001; Vlaeyen et al., 2002). Graded increases in activity are a component in many multidisciplinary chronic pain treatment programs and have also been conceptualized of as providing exposure to feared activities (Mullen, 1999). This hypothesis was tested by Vlaeyen, de Jong, Onghena, Kerkhoffs-Hanssen, & Kole-Snijders, (2002) who found that a graded activity did result in reductions in T S K scores of chronic pain patients, but these reductions were modest. Likewise, Mullen (1999), found that a work hardening program, based on similar principles reduced PASS scores, although changes in PASS scores did not predict response to the treatment. In a series of case study reports, Vlaeyen and colleagues (Linton, Overmeer, Janson, Vlaeyen, & de Jong, 2002; Vlaeyen et al., 2001; Vlaeyen et al., 2002) demonstrated that a cognitive behavioural exposure program consisting of education around the fear avoidance model, development of an exposure hierarchy and behavioral experiments in the form of graded exposure exercises to test and disconfirm catastrophic appraisals and beliefs was successful in reducing pain related fear and catastrophizing and  Catastrophizing, Fear  227  ultimately perceived disability in chronic pain patients. Further, these studies indicated that this targeted exposure is more effective than traditional graded activity programs based on the operant learning paradigm. Interestingly from the perspective of the current work, a study by Crombez et al. (2002) examined the degree to which the effects of exposure to one feared activity generalized to new activities. Arguing that the reduction of avoidance due to exposure results from a decrease negative expectations, they measured expected and actual pain, and expected and actual perceived harm during a series of exposure exercises. As hypothesized, they found that fearful patients tended to over predict pain, but readily corrected their predictions with experience. However, this correction in expectancies did not carry over to other, even very similar activities. If the activity was changed slightly the over prediction of pain recurred with the new activity. The current results, specifically the discrimination between fear of pain and fear of activity, although based on a subacute sample may be relevant to interpreting the lack of generalization found in chronic patients. Specifically, it may be that exposure to feared activities/movements does not generalize, because it does not directly address fear of pain. In line with this hypothesis, the authors found that level of catastrophizing was related to the failure to generalize, Only individuals high in catastrophizing and not those high in fear of (re)injury showed the failure to generalize the effects of exposure. Catastrophizing appears to be more closely related to fear of pain than to fear of activity/movement and therefore the individuals who did not show a generalization of exposure effects may have experienced greater fear of pain. Patients in the exposure based treatments in Vlaeyen, and colleagues (Linton et al., 2002; Vlaeyen et al., 2001; Vlaeyen et al., 2002) studies, were selected based on high T S K scores. If T S K scores reflect a broader fear of stimuli expected to cause pain, then exposure to the specific feared activities  Catastrophizing, Fear.  228  should be adequate to reduce avoidance, as was demonstrated in the studies. However, if an individual is high in fear of pain, they may be more predisposed to fear any new activity, which could potentially cause pain. A corollary, based on the work of Taylor and Rachman (1991) would be observed in an individual who fears a particular movie because they have heard it is sad. If they are exposed to that movie and find out it is not sad, they may not fear that movie any longer but may continue to fear sadness and any other activity they believe might cause sadness. Based on the foregoing it is interesting to speculate on how the discrimination between fear of pain and fear of activity/movement/work might be utilized in treatment. As reported earlier, a successful cognitive behavioural treatment for health anxiety/Hypochondriasis (Warwick et al., 1996) involved not only exposure to illness related stimuli and situations but also to the feared physiological symptoms. Individuals who are high in fear of pain in addition to or rather than fear of activity/work or movement, might benefit from a similar approach where they are encouraged to focus on and control pain stimuli while engaging in exposure to feared movements/activities. Supporting the potential value for such intervention, Sullivan et al. (2001), found that individuals from an undergraduate sample, who were high in catastrophizing tended to underestimate the level of pain they would experience in a laboratory task and that this underprediction led to greater emotional distress during the procedure. The inaccuracy appeared to be attributable to high pain experience rather than low expectations of pain. Sullivan et al. tentatively proposed that individuals who are high in catastrophizing might actually underpredict pain as a strategy for coping with a tendency to experience greater pain as it might help to reduce their anticipatory anxiety. However, that this strategy appears to be counter productive when the actual pain level is greater than expected and leads to greater emotional distress. When faced with a painful injury, such individuals might be more likely to develop a fearful response to pain,  Catastrophizing, Fear  229  because of the increased negative valence it assumes. As fear of pain develops, the same cognitive avoidance those individuals demonstrate in underpredicting their pain may shift to attempts to avoid attending to pain. In this case, exposure to movements or activities expected to cause pain might not generalize. It is possible that such exposure would reduce fear of the particular activity or movement, but without reducing the underlying fear of pain. The cognitive behavioural treatment of Hypochondriasis and health anxiety, where in addition to addressing fears of external stimuli such as illness information, fears of bodily sensations are addressed through exposure with increased attention to the stimuli, may be relevant to this issue. It suggests that fear of pain as opposed to fear of activity and/or movement, might best be treated by encouraging attention to pain, during exposure exercises. Focusing attention toward pain is likely to lead to increased arousal and pain perception in the short term, but it may eventually habituate, ultimately reducing fear of pain. From a cognitive perspective such exposure, would serve as a behavioural experiment in which and individual could challenge their catastrophic misinterpretations of pain. For example, through such exposure, an individual may learn that they can manage their pain and that it does not need to overwhelm them. This would be likely to generalize more broadly than would simply learning that a particular activity could be tolerated. Limitations of the current work. Several limitations to the current work bear mention. First, the generalizeability of the current results may be limited to some extent by characteristics of the sample. Specifically, the sample was composed entirely of worker's compensation claimants with work related injuries. The process of self-selection inherent in the mailing procedure meant that the participants likely represented a select group even within worker's compensation claimants. The nearly equal balance between male and female participants in the current sample contrasts with other work on  Catastrophizing, Fear  230  a similar population where nearly 75% of the participants were male. It is likely that worker's compensation claimants differ from individuals with non-work related injuries and those not involved in the compensation system. Compensation may interact with pain related catastrophizing, fear and anxiety in any number of ways. For example, the availability of financial compensation may reduce overall anxiety about financial issues. Alternatively, the need to demonstrate one's disability which is inherent in any compensation system may encourage a focus on negative aspects of one's condition. These systemic issues in turn may be influenced by other demographic issues such as financial need, and social support. In addition, systemic and psychosocial variables change over time (Krause & Rangland, 1994), and their interaction with fear and catastrophizing is also likely to change. Exploration of these possibilities was beyond the scope of the current project, but they warrant attention in future work. A further comment with regard to generalizeability is that the relationships demonstrated between aspects of pain related fear and catastrophizing in the current subacute sample may not hold at other stages of pain and disability, as the relationships between variables are likely to change over time (Krause, & Ragland, 1994). Fear and catastrophizing represent only two of a subset (emotional distress) of psychosocial variables implicated in the development of persistent pain and disability. Psychosocial variables are, in turn, only one subset of a much broader set of biopsychosocial factors, which appear to be predictive of outcome (e.g., Schultz et al., 2002). While the current work demonstrates that pain related fear and catastrophizing provide an important contribution to prediction of persistent, pain, disability, depressive symptoms and to a lesser degree to return to work, the contribution is moderate at best. It will be important for future work to integrate these variables into the larger context provided by biopsychosocial models. Such work will be  Catastrophizing, Fear  231  facilitated by the improved understanding of pain related fear and catastrophizing provided by the current findings. Similarly, it is acknowledged that the outcome variables selected for the longitudinal analyses in the current study, represented only a subset of potential outcomes which have been identified and there is some controversy over the most appropriate measure of outcome (e.g. Krause, Dasinger, & Neuhauser, 1998; Turk et a l , 2003). For example, the operationalization of return to work as a dichotomous outcome, measured at one point in time, is limited in that many injuries continue to have an impact for many years after the first return to work and first return to work may overestimate the rate of successful return to work (Krause et al., 1998). Other work disability outcomes, such as costs incurred or number of work days lost provide valuable information (Crook et al., 2002), but it was beyond the scope of the current work to include more than the four selected measures as the predictive aspect of the current work was one aspect of the overall goal of demonstrating the validity of pain related fear and catastrophizing constructs and theory. The current work did utilize a range of variables covering many aspects of biopsychosocial outcome but there are many other potential indices and it will be important for future work to demonstrate whether pain related fear and catastrophizing are implicated in other aspects of outcome. The focus of the structural model in the work was a relatively narrow aspect of the pain related fear model (Vlaeyen et al., 2000). The variables examined; pain, fear of pain, fear of activity and avoidance represent an important central part of the model, but not the entire model. Avoidance, while a step in the fear avoidance model is not the ultimate outcome of interest. Unless one takes a very strict and perhaps outdated (e.g., see discussion by McCracken, in press, for updated and comprehensive behavioral theory) perspective that behaviours such as avoidance  Catastrophizing, Fear  232  are the defining and only relevant characteristics of pain, broader outcomes such as perceived disability, affective distress and subjective pain levels are of more interest. In the fear avoidance model (Vlaeyen & Linton, 2000), avoidance is simply one mechanism contributing to Depression, Disability and Disuse. Although the current work addressed the role if pain related fear and catastrophizing prospectively on those outcomes in the regression analyses, this was not done within the context of the structural model. Although the exclusive reliance on self-report measures is potentially a limitation of the current work, the constructs of interest largely dictate the use of such instruments. The potential for response bias in self report measures exists and where possible collateral measurement such as observation is valuable (Hadjistavropoulos & Craig, 2002), however the difficulty in measuring internal states in other ways means that self report is often the best available option (Kazdin, 1992). The variables of interest in this study, fear and catastrophizing are largely subjective. Although they are likely to have physiological and behavioural correlates, the current work did not permit those types of measurement. Likewise, more objective measures of outcome were not available. In addition, the current study was designed to examine the performance of pain related fear and catastrophizing as they are currently measured and current measures are almost exclusively self-report. The likely result of reliance on one measurement modality is that correlations between instruments may have been inflated due to shared method variance. However, such shared method variance would also be expected to obscure the discriminant validity of the constructs involved, and the test of the structural model demonstrated that the data could reliably differentiate constructs. Future work might profitably include multimethod assessment, such as physiological and/or behavioural measurement of pain related fear, and pain and objective verification of outcome data, particularly work related disability.  Catastrophizing, Fear  233  A n additional point in this regard is use of single time data for the structural model to test relationships proposed to develop over time. While structural equation modeling, in testing several mediational relationships simultaneously, allows for somewhat stronger statements as to causality than do other correctional approaches such as regression, some of the same cautions apply, namely that correlation does not imply causality. Therefore, the results of the current work should not be viewed as demonstrating causality. A related issue is that of generalizeability. While S E M modifications were made very judiciously in the current work, there exists that possibility that the degree of fit achieved capitalized on sample specific characteristics of the data. While ideally a model, which has been modified in a post hoc manner, should be retested on a second sample, this type of reanalysis was not practical in the present study, given the practical challenges of collecting a second sample of adequate size. A final note with regard to the structural model is that the choice of measurement indicators imposed certain constraints on the extent to which the results can be interpreted. The present work attempted to address two goals in one analysis. The first goal was to examine the relationships between hypothetical constructs and the second was to use existing instruments in doing so in order that the results would speak to pain related fear as it is currently assessed. These goals imposed somewhat conflicting demands that to a certain degree compromised achievement of either. Selecting subsets of items from current questionnaires increased the relevance of the findings to the current measurement of pain related fear and catastrophizing, but it compromised the purity of the derived latent constructs. Particularly in the differentiation of fear of activity from fear of pain, it is acknowledged that the choice of items to measure these constructs meant that they differed on several dimensions in addition to the focus of fear. For example, the PASS items utilized as measures of fear of pain assess more immediate symptoms  Catastrophizing, Fear  234  in response to a feared stimulus, where the F A B Q activity items assess beliefs as to the danger inherent in a stimulus (e.g., activity). In addition, the F A B Q items assess cognitive content (e.g., beliefs), while the PASS items measure the consequences of fear (e.g., disturbed cognition and physiological responses). The inclusion of physiological symptoms in the fear of pain variable also differentiates it from the fear of activity variable. Finally, Asmundson, et al. (in press) provided a salient argument for the differentiation between fear and anxiety, noting that they may interrelated but have different effects. The degree to which anxiety and fear are confounded in the PASS and F A B Q may have obscured the degree to which that differentiation is relevant to the current work. Those factors may also underlie the failure to confirm an effect of fear of activity on avoidance behavior. It is acknowledged that many of the relationships demonstrated in the current structural model may confound differences in fear stimuli (i.e. pain vs. activity) with these other factors. However, the advantage of being able to discuss the results of the model within the context of variables as they are actually measured in practice (e.g., PASS, F A B Q & PCS scores) outweigh that concern. The choice of structural equation modeling, while allowing stronger inferences about the relationships between constructs, constrained the degree to which full instruments could be used. The need to achieve relatively well-defined constructs in the form of a well-fitting measurement model, and the limitations on the number of indicators relative to sample size, meant that the author had to select sections of available instruments rather than use them in their entirety. This meant that generalization beyond the actual instruments used (PASS cognitive interference and physiological symptoms, and the F A B Q activity subscale) can only be tentative. However, since  Catastrophizing, Fear  235  each latent variable consisted of entire subscale(s) of existing instruments, it was possible to interpret the results with regard to those subscales. The degree to which the PASS items from the physiological symptoms and cognitive interference subscales even measure a fear response has been debated (e.g., Larsen et al., 1997). However, the demonstration that catastrophizing mediated between these symptoms and pain in the current study provide evidence that they are not measuring a direct pain response. This example also indicates the utility of having used full subscales of the existing instruments in the analysis. More general limitations to the current work include the sample characteristics. The choice of individuals with worker's compensation claims may have had some impact on the results. There is some evidence that compensation status can affect response to injury (Turk, 1999), and it is quite possible that it has some impact on pain related fear and avoidance. In particular, the fact that all participants had a claim for work related disability likely had an impact of fear avoidance beliefs about work in particular. The need to delete F A B Q item 6 ("My injury was a result of my work") because the majority of respondents highly endorsed it, speaks to this likelihood. However, the very fact that compensation status has an impact of pain presentation also speaks to the need to focus on that population as was done in the current work. The choice of this population does to some degree limit the generalizeability to other populations of pain patients, as does the restriction of participants to individuals with low-back pain.  (  In order to achieve an adequate sample size, it was necessary to collect data by mail, and that data collection procedure also created some limitations to the current work. Specifically, the experimenter had limited control over application of exclusion criteria. For example, while care was taken to send questionnaires only to appropriate claimants, some individuals reported worst  Catastrophizing, Fear  236  pain somewhere other than their low-back. Likewise, while checks were made to ensure participants met diagnostic inclusion criteria, it is possible that unsuitable individuals were included because they did not report and their file did not indicate an existing, more serious medical condition The collection of data by mail may also have compromised the reliability to some extent, since the conditions under which participants completed the questionnaires could not be controlled. However, the steps taken, including provision of explicit written instructions, phone follow-up and the inclusion of reliability questions, suggest that the data were not significantly less reliable than what might have been collected in a laboratory, and likely at least as reliable as what might have been obtained in a busy clinical setting. The degree to which participants were self selected is another potential issue, since approximately 1500 questionnaires were sent out to achieve a final sample of 210. The relatively high time one scores on the pain related fear and catastrophizing variables, with means approaching what has previously been reported in chronic samples, suggest that the participants may have been more distressed initially than would be the average subacute pain sample. However, the recovery rate as measured by return to work, was quite high (approximately 85 percent) suggesting that this sample was not biased in the direction of poor outcome. A limitation with regard to the regression analyses is the high correlation between the Pain related fear and catastrophizing variables. While there were no issues with multicolinearity as the VIFF values for all analyses were examined and found to be satisfactory, the high intercorrelation of predictors in the regression analyses, resulted in significant shared variance in the prediction of most of the models, to the extent that the finding of unique effects was relatively rare. However, this finding in itself is informative, since it suggests that either the  Catastrophizing, Fear  237  current instruments are tapping aspects of one larger construct, or more likely according to the current results, the instruments lack specificity in the way they measure their intended constructs. In addition the finding of some significant unique effects in this work is even more impressive in the context of high shared variance, and speaks to the likelihood that separate pain related fear constructs can be differentiated. Since a primary goal of the current work was to test a structural model using concepts, which are already measured in the literature, the concepts employed in the model do not correspond exactly to those, which were intended in the structural model. Specifically, differences between fear of pain and fear of activity as measured in the current study are confounded by other differences in the measures employed. For example, the F A B Q activity subscale, which was used to measure fear of activity, taps into fearful beliefs, which may be somewhat distinct from fear itself. Indeed, as argued by Asmundson et al. (in press), such beliefs might better be considered anxiety than fear. The fear of pain measure (cognitive and physiological symptoms subscales of the PASS), again contrary to the use of the word "anxiety" in the instrument title, do appear to measure fear, although, it has been argued, that they perhaps deal more with the aversiveness of pain as opposed to a fear response per se. This assertion is supported by the high correlation between the PASS physiological anxiety subscale and the McGill Pain Questionnaire affective scale. However, the reverse argument can as easily be made, i.e. the M P Q affective scale may be measuring fear. This argument is supported by the fact that the M P Q affective scale specifically uses the word "fearful" in item 14. Again, it appears that at least some of the overlap between instruments results from item wording and therefore imprecision in measurement, rather than actual overlap between contracts.  Catastrophizing, Fear  238  This highlights the importance of studies along the line of the current work and more careful attention to the subtleties of construct validity. The PCS appears exemplary in this regard. There is a strong theoretical framework underlying the instrument and the authors have considered in depth the issue of construct validity (e.g., Sullivan and D'Eon, 1990). A number of studies have examined how the constructs measured by the instrument are differentiable from other affective (e.g., depression, negative affectivity) responses (see review by Sullivan et al., 2001). The work of Larsen, et al. (1997) and Mc Williams & Asmundson (1999) also represents a step in this direction. Specifically, their careful analysis of the PASS and what it measures and attempts to respectively the questionnaire in the interest of achieving better construct validity. S E M as utilized in the current work, can also provide valuable insights. Turk (1999) pointed to the issue of the proliferating measurement tools in the absence of a corresponding theoretical framework for understanding the interrelationships of the instruments and the constructs they assess. S E M allows the integration of measurement error into a model and that error can provide important clues as to the theoretical structure of instruments. For example the indication of correlated errors between items suggests some common, unmodeled cause for those items. This information can be on obscured in standard exploratory factor analysis or regression analysis where error of measurement is attributed to other parameters. A final limitation to the current work is relatively short follow-up period. While three months post onset is suggested by the IASP (Merskey & Bogduk, 1994) as representing the beginning of chronicity, outcome at that time may not accurately reflect longer-term output. Some research has demonstrated that psychosocial variables measured at three months post injury is more predictive of one year outcome, than is the same set of variables at acute presentation (Philips & Grant, 1991b). Still, intervention in the first few months following onset  Catastrophizing, Fear  239  might prevent the fear/avoidance cycle from becoming self-perpetuating and thereby prevent chronicity. Three-month status may be predictive, because by that point the cycle has already become entrenched. Implications for Future Research From the current work it is clear that while some of the current measures of pain related fear and catastrophizing, assess constructs, which are highly interrelated but can be separated. It further suggests that there is merit in measuring the various constructs comprising pain related fear and catastrophizing. However, the existing instruments overlap substantially, and are of a length, which precludes the use of more than one or two in, most clinical or research settings. This is the likely reason for the paucity of studies comparing instruments, which has been noted by others (McNeil & Vowles, in press). As discussed already, initial steps have been taken toward developing screening instruments, which detects pain related fear and catastrophizing with a small number of items. (Linton & Hallden, 1998; Linton et al., 2002; Vlaeyen et al., 2001; Vlaeyen et al., 2002). The three pain related fear items instrument devised by Linton and Hallden, in particular have been demonstrated to possess predictive validity, which is particularly impressive for such a small item set. However, in order for such an instrument to inform clinical practice, it will be useful for it to assess multiple dimensions of pain related fear. A n instrument devised by Balderson, Lin, & Von Korff (in press), for use in primary care settings, does assess a number of specific fears, such as the fear that "back pain may worsen or become chronic" but focuses exclusively on feared outcomes of pain. This is also a useful approach, but again does not assess multiple dimensions of pain related fear. Vlaeyen et al (Vlaeyen et al., 2001) devised an instrument which involves items from the PASS, PCS and TSK, which has proven useful as a measure of pain related fear in chronic pain treatment studies Linton et al., 2002; Vlaeyen et al.,  Catastrophizing, Fear  240  2001; Vlaeyen et al., 2002). This suggests that a broader range of pain related fear can be assessed in a brief instrument. It would be useful to extend this research in an attempt to develop a psychometrically sound measure of multiple dimensions of pain related fear derived from a theoretical framework such as that provided by Asmundson, Norton, & Vlaeyen, (in press) for use in acute and subacute pain. Along similar lines it appears that the current pain related fear instruments each derived from slightly different theoretical perspectives have triggered a burgeoning area of research that has culminated in the development of comprehensive well-specified theoretical models. However, it appears that the specificity of the model has outpaced the development of the instruments and it will be important to develop instruments capable of finer discernment between the constructs identified in the Asmundson et al (in press) model. The difficulty developing a clearly defined conceptual model based on existing measures points to the importance of such an undertaking. The development of new instruments within a nomological network informed by theory promised to increase the efficiency and effectiveness of assessment of pain related fear and the further refinement of theory. Another area of research, which will inform the development of new measures and the refinement of theory, is further examination of the hypothesized structure of pain related fear, catastrophizing and other related constructs. For example the proposition by Keogh and Asmundson (in press) that pain catastrophizing and fear of pain both represent first order factors in a four level hierarchical model of negative affect lends itself well to confirmatory factor analysis methods. Similarly, research more directly exploring causal aspects of the pain related fear model would be useful. Brown & Barlow (2002) provide an excellent example of this approach in their use of S E M to examine a model of interrelationships of emotional disorders  Catastrophizing, Fear  241  and purported higher order constructs of negative affect, positive affect and autonomic arousal. If such work is freed from the constraint imposed on the current work of utilizing existing instruments and instead involves the creation of items designed specifically to measure the constructs of interest, it will serve as both a test of theoretical assumptions and a first step in developing new instruments. One of the benefits of structural equation modeling is that it allows testing of measurement and structural hypotheses simultaneously and therefore psychometrically sound instruments can be designed in the context of a predefined nomological net. Items can be selected in such a way to maximize the measurement of constructs as defined by their hypothesized relationships to other constructs. For example a researcher might test the model presented in the current work by devising items specifically designed to measure fear of pain and fear of activity. Testing such a model would provide evidence as to the psychometric properties of the selected items for measuring the specified constructs fear of pain and fear of activity, within the context of specified relationships between fear of pain and fear of activity. Testing the predictive validity and the interrelationship of various pain related fear and catastrophizing variables over longer time durations will also be important. It may be that the variables, which are most predictive at one time point, will be different than those most predictive at another point. A n example of such a relationship was provided by Philips and Grant, (1991b), who measured a number of psychosocial variables in a group of pain patients presenting with acute pain over the subsequent year. Philips and Grant found that the factors predicting three-month outcome from initial presentation were different than the variables predicting one-year outcome from three-month outcome.  Catastrophizing, Fear  242  REFERENCES American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, D C : Author. Anderson, J. C , & Gerbing, D. W. (1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103, 411-423. Andersson, G. B . J. (1981). Epidemic aspects on low-back pain in industry. Spine, 6, 53-60. Arntz, A., Dreessen, L., & Merckelbach, H . (1991). Attention, not anxiety, influences pain. Behaviour Research and Therapy, 29, 41-50. Arntz, A., van Eck, M . , & Ffeijmans, M . (1990). Predictions of dental pain: The fear of any expected evil, is worse than the evil itself. Behaviour Research and Therapy, 28, 29-41. Asmundson, G. J. G., Frombach, I. K., & Hadjistavropoulos, H . D . (1998). Anxiety sensitivity: Assessing factor structure and relationship to multidimensional aspects of pain in injured workers. Journal of Occupational Rehabilitation, 8, 223-234. Asmundson, G. J. G., Norton, G. R., & Allerdings, M . D. (1997). Fear and avoidance in dysfunctional chronic back pain patients. Pain, 69, 231-236. Asmundson, G. J. G., Norton, P. J., & Norton, G. R. (1999). Beyond pain: The role of fear and avoidance in chronicity. Clinical Psychology Review, 19, 97-119. Asmundson, G. J. G., Norton, P. J., & Vlaeyen, J. W. S. (in press). Fear-avoidance models of chronic pain: A n overview. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.), Understanding and treating fear ofpain. Oxford, U K : Oxford University Press. Asmundson, G. J. G., & Taylor, S. (1996). Role of anxiety sensitivity in pain related fear and avoidance. Journal of Behavioral Medicine, 19, 577-586.  Catastrophizing, Fear  243  Atkinson, J. H., Slater, M . A., Patterson, T. L., Grant, I., & Garfin, S. R. (1991). Prevalence, onset and risk of psychiatric disorder in men with chronic low-back pain: A controlled study. Pain, 45, 111-122. Bandura, A. (1969). Principles of behavior modification. New York: Holt, Rinehart, & Winston. Barlow, D. H . (2002). The phenomenon of Panic. In D. H . Barlow (Ed.). Anxiety and Its Disorders: The Nature and Treatment of Anxiety and Panic. New York: The Guilford Press (pp. 105-138). Beecher, H . K . (1956). Relationship of significance of wound to the pain experienced. Journal of the American Medical Association, 161, 1609-1613. Bentler, P. M . (1995). EQS Structural Equations Program Manual. Encino, C A : Multivariate Software Inc. Bentler, P. M . (1988). Comparative fit indices in structural models. Psychological Bulletin, 107, 238-246. Biering-Sorenson, F. (1983). A prospective study of low-back pain in a general'population: Occurrence, recurrence and etiology. Scandinavian Journal of Rehabilitation Medicine, 75,71-79. Bradley, L. A., & McKendree-Smith, N . L . (2001). Assessment of psychological status using interviews and self-report instruments. In D.C. Turk, & R. Melzack (Eds.). Handbook of Pain Assessment. New York: The Guilford Press. Breckler, S. J. (1990). Applications of covariance structure modeling in psychology: Cause for concern? Psychological Bulletin, 707,260-273. Browne (1984). Asymptotically distribution-free methods for the analysis of covariance structures. British Journal of Mathematics and Statistical Psychology, 37, 62-83.  Catastrophizing, Fear  244  Browne, M.W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K . A . Bollen & J.S. Long (Eds.), Testing structural equation models (p. 10-39). Brown, T. A . & Barlow, D. H . (2002). Classification of Anxiety and Mood Disorders. In D. H . Barlow (Ed.). Anxiety and Its Disorders: The nature and Treatment of Anxiety and Panic. New York: The Guilford Press. Buer, N . , Linton, S. J., Samuelsson, L., & Harms-Ringdahl, K . (submitted). The role of fearavoidance beliefs and catastrophizing in patients with fractures. Burns, J., Mullen, J. T., Higdon, L. J., Wei, J. M . , & Lansky, D . (2000). Validity of the pain anxiety symptoms scale (PASS): Prediction of physical capacity variables. Pain, 84, 247252. Burton, A. K., Tillotson, K . M . , Main, C. J., & Hollis, S. (1995). Psychosocial predictors of outcome in acute and subchronic low-back trouble. Spine, 20, 722-728. Burton, A. K., Waddell, G., Tillotson, K . M . , & Summerton, N . (1999). Information and advice to patients with back pain can have a positive effect: A randomized controlled trial of a novel educational booklet in primary care. Spine, 24, 2484-2491. Bussman, J. B . J., van de Laar, Y . M . , Neleman, M . P., & Stam, H . J. (1998). Ambulatory, accelerometry to quantify motor behaviour in patients after failed back surgery: A validation study. Pain, 74, 153-161. Byrne, B. M . , (1994). Structural Equation Modeling with EQS & EQS/WINDOWS: Basic Concepts, Applications, & Programming. Published by Sage Publications (February 1994). Byrne, B. M . , Baron, P., Larsson, B., & Melin, L. (1995). The Beck Depression Inventory: Testing and cross-validating a second-order factorial structure for Swedish nonclinical adolescents. Behaviour Research and Therapy, 33, 345-356.  Catastrophizing, Fear  245  Cats-Baril, W. L., & Frymoyer, J. W. (1991). Identifying patients at risk of becoming disabled because of low-back pain: The Vermont rehabilitation engineering center predictive model. Spine, 16, 605-607. Cheung, G. W., & Rensvold, R. B . (2002). Evaluating goodness-of-fit indices for testing measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal, 9, • 233 -255. Chou, C. P., Bentler, P. M . , & Satorra, A . (1991). Scaled test statistics and robust standard errors for non-normal data in covariance structural analysis: A Monte Carlo study, British Journal of Mathematical and Statistical Psychology, 44, 347-357. Clark, D. M . (1986). A cognitive approach to panic. Behaviour Research and Therapy, 24, 461470. Clark, L. A . & Watson, D. (1991) Tripartite model of anxiety and depression: psychometric evidence and taxonomic implications. Journal of Abnormal Psychology, 100, 316-336. Clark, M . F., Kori, S. H., & Brockel, J. (1996). Kinesiophobia and chronic pain: psychometric characteristics and factor analysis of the Tampa Scale. American Pain Society Abstracts. 15: 77. Cornwall, A., & Donderi, D. C. (1988). The effect of experimentally induced anxiety on the experience of pressure pain. Pain, 35, 105-113. Craig, K . D. (1986). Social modeling influences: Pain in context. In R. A . Stembach (Ed.), The psychology of pain 2nd ed., pp. 67-96. New York: Raven Press. Craig, K . D. (1994). Emotional aspects of pain. In P. D. Wall & R. Melzack (Eds.), Textbook of Pain (3rd ed., pp.261-274). Edinburgh: Churchill Livingstone. Craig, K . D. (1999). Emotions and psychobiology. In P. D. Wall & R. Melzack (Eds.), Textbook of Pain (4th ed., pp.261-274). Edinburgh: Churchill Livingstone.  Catastrophizing, Fear  246  Craig, K . D., Best, FL, & Reith, G. (1974). Social determinants of reports of pain in the absence of painful stimulation. Canadian Journal of Behavioural Sciences, 6, 169-177. Craig, K. D., Hill; M . L., & McMurtry, B. (1999). Detecting deception and malingering. In A. R. Block, E. F. Kremer, & E. Fernandez (Eds.), Handbook of pain syndromes: Biopsychosocialperspectives. Mahwah, NJ: Lawrence Erlbaum. Craig, K . D., & Patrick, C. J. (1985). Facial expression during induced pain. Journal of Personality and Social Psychology, 44, 1080-1091. Craig, K . D., & Weiss, S. M . (1971). Vicarious influences on pain threshold determinations. Journal of Personality and Social Psychology, 19, 53-59. Crombez, G., Eccleston, C , Baeyens, F., & Eelen, P. (1998). When somatic information threatens, catastrophic thinking enhances attentional interference. Pain, 75, 187-198. Crombez, G., Eccleston, C , Baeyens, F., Van Houdenhove, B . & Van Den Broeck, J. (1999). Attention to chronic pain is dependent on pain related fear. Journal of Psychosomatic Research, 47, 403-410. Crombez, G., Eccleston, Van den Broeck, Van Houdenhove, B., & Goubert, L. (2002). The effects of catastrophic thinking about pain upon attentional interference by pain: No mediation of negative affectivity in healthy volunteers and in low-back pain patients. Pain Research and Management, 7, 31-39. Crombez, G., Eccleston, C , Vlaeyen, J. W. S., Eccleston, C , Vansteenweegen, D., Lysens, R., & Eelen, P. (2002). Exposure to physical movements in low-back pain patients: Restricted effects of generalization. Health Psychology, 21, 573-578. Crombez, G., Eccleston, C , Vlaeyen, J.W.S., Vansteenwegen, D., Lysens, R. & Eclan, P. (2002). Exposure to physical movements in low-back paint patients: Restricted effects of generalization. Health Psychology, 21. (pp. 573-578.  Catastrophizing, Fear  247  Crombez, G., Vervaet, L., Baeyens, F., Lysens, R. & Eelen, P. (1996). Do pain expectancies cause pain in chronic low-back patients? A clinical investigation. Behaviour Research and Therapy, 34, 919-925. Crombez, G., Vlaeyen, J. W. S., Heuts, P. H . T. G., & Lysens, R. (1999). Pain related fear is more disabling than pain itself: Evidence on the role of pain related fear in chronic back pain disability. Pain, 80, 329-339. Crook, J., Milner, R., Schultz, I. Z., & Stringer, B . (2002). Determinants of occupational disability following a low back injury: A critical review of the literature. Journal of Occupational Rehabilitation, 12, 277-295. Deacon, B . J. & Abramowitz, J. S. (2004). Cognitive and behavioral treatments for anxiety disorders: A review of meta-analytic findings. Journal of Clinical Psychology, 60, 429441. Deyo, R. A., & Diehl, A . K . (1988). Psychosocial predictors of disability in patients with low back pain. Journal of Rheumatology, 15, 1557-1564. Dworkin, S. F., & Whitney, C. W. (1992). Relying on objective and subjective measures of chronic pain: Guidelines for use and interpretation. In D. C. Turk and R. Melzack (Eds.), Handbook of pain assessment (pp. 429-446). New York: Guildford Press. Eccleston, C , Bayens, F., & Eelen, P. (1998). When somatic information threatens, pain catastrophizing enhances attentional interference. Pain, 75, 197-198. Els, L. M . , Gheldof, P. J., de Jong, J. V., & Ruud, M . A . H . (in press). Attitudes toward physical activity: The role of implicit versus explicit associations. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.), Understanding and treating fear of pain (p. 117-137). Oxford, U K : Oxford University Press.  Catastrophizing, Fear  248  Fernandez, E., & Turk, D . C. (1989). The utility of cognitive coping strategies for altering pain perception: A meta-analysis. Pain, 38, 123-135. Feuerstein, M . , & Thebarge, R. W. (1991). Perceptions of disability and occupational stress as discriminators of work disability in patients with chronic pain. Journal of Occupational Rehabilitation, 1, 185-195. Flor, FL, & Turk, D. C. (1999). A biobehavioral perspective on chronic pain: A comprehensive assessment and treatment guide. New York: Plenum. Fordyce, W. E. (1976). Behavioral methods for chronic pain and illness. St. Louis, M O : Mosby. Fordyce, W. E. (1995). Back pain in the workplace: Management of disability in nonspecific conditions. Seattle, W A : IASP Press. Fritz, J. M . , George, S.Z., & Delitto, A . (2001). The role of fear-avoidance beliefs in acute lowback pain: relationships with current and future disability and work status. Pain, 94, 7-15. Frymoyer, J. W., & Cats-Baril, W. (1987). Predictors of low-back pain disability. Clinical Orthopedics, 221, 89-98. Geisser, M . E., Ffaig, A . J., & Theisen, M . E. (2000). Activity avoidance and function in persons with chronic back pain. Journal of Occupational Rehabilitation, 10, 215-227. Geisser, M . , E., Roth, R., S., & Robinson, M . , E., (1997). Assessing depression among persons with chronic pain using the Center for Epidemiological Studies-Depression Scale and the Beck Depression Inventory; A comparative analysis. Clinical Journal of Pain, 13, 163170. Gheldof, E . L . M . , de Jong, P.J., Vinck, J. & Houben, R. M . A. (In Press).Attitudes, forward physical activity: The role of implicit versus explicit associations. In G. J.G. Asmundson, J.W.S. Vlaeyen & G. Crombez (Eds.) Understanding and treating fear ofpain. Oxford, U K : Oxford University Press.  >  Catastrophizing, Fear  249  Goubert, L., Crombez, G., Can Damme, S., Vlaeyen, J. S., Bijttebier, P., & Roeflofs, J. (2004). Confirmatory factor analysis of the Tampa Scale for Kinesiophobia: Invariant two factor model across low-back pain patients and fibromyalgia patients. Clinical Journal of Pain, 20,103-110. Gramling, S. E., & Elliot, T. R. (1992). Efficient pain assessment in clinical settings. Behaviour Research and Therapy, 30, 71-73. Greenwood, J. G., Wolf, H . J., Pearson, J. C , Woon, C. L., Posey, P., & Main, C. (1990). Early intervention in low-back disability among coal miners in West Virginia: Negative findings. Journal of Occupational Medicine, 32, 1047-1052. Gronblad, M . , Lukinmaa, A., & Konttinnen, Y . T.(1990). Chronic low-back pain: Intercorrelation of repeated measures for pain and disability. Scandinavian Journal of Rehabilitation Medicine, 22, 11-11. Gross, R. T., & Collins, F. L . (1981). On the relationship between anxiety and pain: A methodological confounding. Clinical Psychology Review, 1, 375-386. Guieu, R., Tardy-Gervet, M . , F., & Roll, J. P. (1991). Analgesic effects of vibration and transcutaneous electrical nerve stimulation applied separately and simultaneously to patients with chronic pain. Canadian Journal of Neurological sciences, 18, 113-119. Hadjistavropoulos, H . D., & Craig, K . D. (1994). Acute and chronic low-back pain: Cognitive, affective, and behavioral dimensions. Journal of Consulting and Clinical Psychology, 62, 341-349. Hadjistavropoulos, H . D., Craig, K . D., & Hadjistavropoulos, T. (1998). Cognitive and behavioural responses to illness information: The role of health anxiety. Behaviour Research and Therapy, 36, 149-164.  Catastrophizing, Fear  250  Haldeman, S. (1990). Failure of the pathology model to predict back pain. Presidential address, North American Spine Society. Spine, 15, 718-723. Hazard, R. G., Haugh, L . D., Reid, S., Preble, J. B., & MacDonald, L . (1996). Early prediction of chronic disability after occupational low-back injury. Spine, 21, 945-951. Helander, N . (1984). Depression caused by chronic pain. Journal of Clinical Psychiatry, 45, 3036. Hoyle, R. H . (1995). The structural equation modeling approach: Basic concepts and fundamental issues. In R. H . Hoyle (Ed.), Structural equation modeling; Concepts, issues, and applications (pp. 1-16). London, U K : Sage Publications. Hu, L. & Bentler, P . M . (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6 (1), 1-55. Jensen, M . P., & Karoly, P. (1991). Control beliefs, coping efforts, and adjustment to chronic pain. Journal of Consulting and Clinical Psychology, 59, 431-438. Jensen, M . P., Karoly, P., & Huger, R. (1987). The development and preliminary validation of an instrument to assess patient's attitudes towards pain. Journal of Psychosomatic Research, J i , 393-400. Jensen, M . P., Turner, J. A., & Romano, J. M . (1991). Self-efficacy and outcome-expectancies: Relationship to chronic pain coping strategies and adjustment. Pain, 44, 263-269. Jensen, M . P., Turner, J. A., Romano, J. M . , & Lawler, B . K . (1994). Relationship of painspecific beliefs to chronic pain adjustment. Pain, 57, 301-309. Kazdin, A . E . (2002). Research design in clinical psychology (4th ed.). Needham Heights, M A : Allyn & Bacon  C atastrophizing, Fear  251  Keefe, F. J., & Williams, D. A. (1989). Assessment of pain behaviors. In D. C. Turk & R. Melzack (Eds.), Handbook of pain assessment (pp. 275-294). New York: Guildford Press. Keogh, E., & Asmundson, G. J. G. (in press). Negative affectivity, catastrophizing, and anxiety sensitivity. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.), Understanding and treating fear ofpain. Oxford, U K : Oxford University Press. Klein, B. P., Jensen, R. C , & Sanderson, L. M . (1984). Assessment of worker's compensation claims for back strains/sprains. Journal of Occupational Medicine, 26, 443-448. Klenerman, L., Slade, P. D., Stanley, I. M . , Pennie, B., Reilly, J. P., Atchison, L. E., et al., (1995). The prediction of chronicity in patients with an acute attack of low-back pain in a general practice setting. Spine, 20, 478-484. Kori, S. H., Miller, R. P., & Todd, D. D. (1990). Kinisophobia: A new view of chronic pain behavior. Pain Management, 35-43. Krause, N . , Ragland, D. (1994). Occupational disability due to low back pain: A new interdisciplinary classification based on a phase model of disability. Spine, 19, 10111020. LaCroix, R., & Barbaree, H . E. (1992). Pain-elicited responses and their role in predicting future pain duration and severity. Behaviour Research and Therapy, 30, 471-478. LaCroix, J. M . , Powell, J., Lloyd, G. J., Doxey, M . C. S., Mitson, G. L., & Aldam, C. F. (1990). Low-back pain: Factors of value in predicting outcome. Spine, 15, 495-499. Larsen, D. K., Taylor, S., & Asmundson, G. J. G. (1997). Exploratory factor analysis of the Pain Anxiety Symptoms Scale in patients with chronic pain complaints. Pain, 69, 27-34. Lehmann, T. R., & Brand, R. A . (1982). Disability of the patient with low-back pain. Orthopedic Clinics of North America, 73,559-568.  Catastrophizing, Fear  252  Lethem, J., Slade, P. D., Troup, J. D. G., & Bentley, G. (1983). Outline of a fear-avoidance model of exaggerated pain perception-I, Behaviour Research and Therapy, 21, 401-408. Linton, S. J. (1999). Prevention with special reference to chronic musculoskeletal disorders. In R. J. Gatchel & D. C. Turk (Eds.), Psychosocial factors in pain: Critical perspectives (pp. 374-389). New York: Guilford Press. Linton, S. J., Althoff, B., Melin, L., et al., (1994). Psychological factors related to health, back pain, and dysfunction. Journal of Occupational Rehabilitation, 5, 1-10. Linton, S. J., Bradley, L. A., Jensen, I., Spangfort, E., & Sundell, L. (1989). The secondary prevention of low-back pain: A controlled study with follow-up. Pain, 36, 197-207. Linton, S. J., & Boersma, K . (in press). The role of fear-avoidance in the early identification of patients risking the development of disability. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.). Understanding and treating fear ofpain. Oxford, U K : Oxford University Press, pp. 213-234. Linton, S. J., & Buer, N . (1995). Working despite pain: Factors associated with work attendance versus dysfunction. International Journal of Behavioral Medicine, 2, 252-262. Linton, S. J., Buer, N . , Vlaeyen, J., & Hellsing, A. (2000). Are fear-avoidance beliefs related to the inception of an episode of back pain? A prospective study. Psychology and Health, 14, 1051-1059. Linton, S. J., & Hallden, K . (1998). Can we screen for problematic back pain? A screening questionnaire for predicting outcome in acute and subacute back pain. Clinical Journal of Pain, 14, 209-215. Linton, S. J., Overmeer, T., Janson, M . , Vlaeyen, J. W. S. & de Jong, J. R. (2002). Graded in vivo exposure treatment for fear-avoidance pain patients with functional disability: A case study. Cognitive Behaviour Therapy, 31, 49-58.  Catastrophizing, Fear  253  Main, C. J. (1983). The modified somatic perception questionnaire. Journal of Psychosomatic Research, 27, 503-514. Marks, I. M . , & Mathews, A . M . (1979). Brief standard self rating for phobic patients. Behaviour Research and Therapy, 17, 263 267. Marsh H. W., Balla, J. R., & McDonald, R. P. (1988). Goodness-of-fit indexes in confirmatory factor analysis: The effect of sample size. Psychological Bulletin, 103, 391-411. Martin, M . Y . , Bradley, L . A., Alexander, R. W., Alarcon, G. S., Tri ana-Alexander, M . , Aaron, L. A., & Alberts, K . R. (1996). Coping strategies predict disability in patients with primary fibromyalgia. Pain, 68, 45-53. McCracken, L . M . (1998). Learning to live with the pain: acceptance of pain predicts adjustment in persons with chronic pain. Pain, 74, 21-27. McCracken, L. M . (in press). A behavioral analysis of pain related fear responses. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.), Understanding and treatingfear of pain. Oxford, U K : Oxford University Press. McCracken, L . M . , & Dhingra, L . (2002). A short version of the Pain Anxiety Symptoms Scale (PASS-20): Preliminary development and validity. McCracken, L . M . , Faber, S. D. & Janek, A. S. (1998). Pain related anxiety predicts non-specific physical complaints in persons with chronic pain. Behaviour Research and Therapy, 36, 621-630. McCracken, L. M . , & Gross, R. T. (1993). Does anxiety affect coping with pain? Clinical Journal of Pain, 9, 253-259. McCracken, L . M . , Gross, R. T., Aikens, J., & Carnrike, C. L . M . (1996). The assessment of anxiety and fear in persons with chronic pain: A comparison of instruments. Behaviour Research and Therapy, 34, 927-933.  Catastrophizing, Fear  254  McCracken, L . M . , Gross, R. T., Hexum, C. L., & Semenchuk, E. M . (1993, March). Factor structure and temporal stability of the Pain Anxiety Symptoms Scale. Paper presented at the annual convention of the Society of Behavioural Medicine, San Francisco, CA. McCracken, L . M . , Gross, R. T., Sorg, P. J., & Edmands, T. A . (1993). Prediction of pain in patients with chronic low-back pain: Effects of inaccurate prediction and pain related anxiety. Behavior Research and Therapy, 31, 647-652. McCracken, L . M . , Gross, R. T., Zayfert, C , Sorg, P. J., & Edmands, T. A . (1992, November). Validation of the Pain Anxiety Symptoms Scale (PASS) in chronic low-back pain: A study of directly observed behavior during a physical examination. Paper presented at the annual convention of the Association for Advancement of Behavior Therapy, Boston, MA. McCracken, L . M . , Zayfert, C , & Gross, R. T. (1992). The Pain Anxiety Symptoms Scale: Development and validation of a scale to measure fear of pain, Pain, 50, 67-73. McCracken, L . M . , Zayfert, C , & Gross, R. T. (1993). The Pain Anxiety Symptoms Scale (PASS): A multi-modal measure of pain-specific anxiety symptoms. Behavior Therapist, 7t5, 183-184. McLean, P. D., & Woody, S. R. (2001). Anxiety disorders in adults: An evidence-based approach to psychological treatment. New York: Oxford University Press. McNeil, & Vowles, (in press). Assessment of fear and anxiety associated with pain: Conceptualization, methods and measures. In G. J. G. Asmundson, J. W. S. Vlaeyen, & G. Crombez (Eds.), Understanding and treating fear ofpain. Oxford, U K : Oxford University Press. Melzack, R. (1975). The M c G i l l Pain Questionnaire: Major properties and scoring methods. Pain, 7,277-199.  Catastrophizing, Fear  255  Melzack, R., & Casey, K . L . (1968). Sensory, motivational and central control determinants of pain. In D. R. Kennshalo (Ed.), The skin senses (pp. 423-443). Springfield, IL: Thomas. Melzack, R.(1987). The short-form McGill Pain Questionnaire. Pain, 30, 191-197. Melzack, R., & Katz, J. (1992). The McGill Pain Questionnaire: Appraisal and current status. In D.C. Turk & R. Melzack (Eds.). Handbook of Pain Assessment (ppl52-168). New York: Guilford Press. Melzack, R., & Torgerson,, W. S., (1971). On the language of pain. Anesthesiology, 34, 50-59. Melzack, R., & Wall, P. D. (1965). Pain mechanisms: A new theory. Science, 150, 971-979. Merskey, FL, & Bogduk, N . (1994). Classification of chronic pain: Descriptions of chronic pain syndromes and definitions of pain terms (2nd ed.). Seattle: IASP Press. Morle, S. Eccleston, C. & Williams, A . (1999). Systematic review and meta-analysis of randomized controlled trials of cognitive behaviour therapy and behaviour therapy for chronic pain in adults, excluding headache. Pain, 80, 1-13. Mullen, J. T. (1998). Effects of the reduction of fear of pain on treatment outcome in a work hardening program. Unpublished doctoral dissertation, Chicago Medical School, Chicago. Nachemson, A . L . (1992). Newest knowledge of low-back pain. Clinical Orthopaedics, 279, 820. Novy, D. M . , Nelson, D. V . , Francis, D. J., & Turk, D. C. (1995). Perspectives of chronic pain: An evaluative comparison of restrictive and comprehensive models. Psychological Bulletin, 118, 238-247. Osman, A., Barrios, F. X . , Osman, J. R., & Schneekloth, R. (1994). The pain anxiety symptoms scale: Psychometric properties in a community sample. Journal of Behavioral Medicine, 77, 511-522.  Catastrophizing, Fear  256  Osman, A., Barrios, F. X . , Copper, Hauptman, Jones, and O'Neil (1997). Factor structure, reliability, and validity of the Pain Catastrophizing Scale. Journal of behavioural Medicine, 20, 589-605. Osman, A., Barrios, F. X . , Gutierrez, P. M . , Kopper, B. A., Merrifield, T., & Grittmann, L. (2000). The Pain Catastrophizing Scale: Further psychometric evaluation with adult samples. Journal of Behavioural Medicine, 23,351-365. Philips, H. C. (1987). Avoidance behaviour and its role in sustaining chronic pain. Behaviour Research and Therapy, 25, 273 279. Philips, H. C , & Grant L. (1991a). Acute back pain: A psychological analysis. Behaviour Research and Therapy, 29, 429-434. Philips, H. C , & Grant L. (1991b). The evolution of chronic back pain problems: A longitudinal study. Behaviour Research and Therapy, 29, 435-441. Philips, H. C , Grant, L., & Berkowitz, J. (1991). The prevention of pain and disability: A preliminary investigation. Behaviour Research and Therapy, 29, 443-450. Philips, H. C , & Jahanshahi, (1985). Chronic pain: A n experimental analysis of the effects of exposure. Behaviour Research and Therapy, 23, 281-290. Philips, H. C , & Jahanshahi, (1986). The components of pain behaviour report. Behaviour Research and Therapy, 24, 117-125. Pilowsky, I., & Spence, N . D. (1975). Patterns of illness behaviour in patients with intractable patterns of illness behaviour in patients with intractable pain. Journal of Psychosomatic Research, 19, 279-287. Pilowsky, I., & Spence, N . D. (1983). Manual for illness behaviour questionnaire (IBQ) (2nd ed.). Adelaide, A U : University of Adelaide.  Catastrophizing, Fear  257  Pollard, C. A . (1984). Preliminary validity study of the Pain Disability Index. Perceptual and Motor Skills, 59, 974. Polatin, P., Kinney, R. K . , Gatchel, R. J., Lillo, E., & Mayer, T. G. (1993). Psychiatric illness and chronic low-back pain: The mind and the spine. Which goes first? Spine, 18, 66-71. Paice, J. A., & Cohen, F. L . (1997). Validity of a verbally administered numeric rating scale to measure cancer pain intensity. Cancer Nursing, 20, 88-93. Price, D. D. & Harkins, S. W. (1987). Combined use of experimental pain and visual analogue scales in providing standardized measurement of clinical pain. Clinical Journal of Pain, 3, 1-8. Prkachin, K . & Craig, K . D. (1986). Social transmission of natural variations in pain behaviour. Behaviour Research and Therapy, 24, 581-585. Rachman, S. J. (1998). Anxiety. London, U K : Psychology Press. Rachman, S. J. (1988). Panics and their consequences: A review and prospect. In S. J. Rachman & J. Maser (Eds.), Panic: Psychological perspectives (pp.). Hilsdale, NJ: Lawrence Erlbaum. Rachman S., & Arntz A . (1991). The overprediction and underprediction of pain. Clinical Psychology Review, 11, 339-355. Rachman, S. J., & Maser, J. D. (1988). Panic: Psychological contributions. In S. J. Rachman & J. Maser (Eds.), Panic: Psychological perspectives (pp.). Hilsdale, NJ: Lawrence Erlbaum. Radloff, L . (1977). The CES-D scale: A self-report depression scale for research in the general population. Journal of Applied Psychological Measurement, 1,385-401. Reesor, K. A., & Craig, K . D. (1988). Medically incongruent chronic back pain: Physical limitations, suffering, and ineffective coping. Pain, 32, 35-45.  Catastrophizing, Fear  258  Reiss, S., & McNally, R. J. (1985). Expectancy model of fear. In S. Reiss & R. R. Bootzin (Eds.), Theoretical issues in behavior therapy (pp. 107-121). New York: Academic Press. Riley, J. F., Ahern, D. K., & Follick, M . J. (1988). Chronic pain and functional impairment: Assessing beliefs about their relationship. Archives of Physical Medicine and Rehabilitation, 69, 579-582. Robinson, M . E., & Riley, J. L . (1999a). Models of pain. In A . R. Block, E. F. Kremer, & E. Fernandez (Eds.), Handbook of pain syndromes (pp. 23-40). Mahwah, NJ: Erlbaum. Robinson, M . E., & Riley, J. L. (1999b). The role of emotion in pain. In R. J. Gatchel & D. C. Turk (Eds.), Psychosocial factors in pain (pp. 74-88). New York: Guilford Press. Rosensteil, A . K., & Keefe, F. J. (1983). The use of coping strategies in chronic low-back pain, Pain, 17, 33-44. Sandstrom, J., Andersson, G. B., & Wallerstedt, S, (1984). The role of alcohol abuse in working disability in patients with low-back pain. Scandinavian Journal of Rehabilitation Medicine, 16, 147-149. Santor, D. A., Zuroff, D. C , Ramsey, J. O., Cervantes, P., & Palacios, J. (1995). Examining scale discriminability in the BDI and CES-D as a function of depressive severity. Psychological Assessment, 7, 131-139. Sattora, A. (1990). Robustness issues in structural equation modeling: A review of recent developments. Quality and Quantity, 24, 367-386. Sattora, A., & Saris, W. E., (1985). Power of the likelihood ratio test in covariance structure analysis. Psychometrika, 50, 83-90. Schultz, I. Z., Crook, J. M . , Berkowitz, J., Meloche, G. R., Milner, R., Zuberbier, O. A., & Meloche, W. (2002). Biopsychosocial multivariate predictive model of occupational low back disability. Spine, 27, 2720-2725.  Catastrophizing, Fear  259  Schultz, I. Z., Crook, J., Meloche, G. R., Berkowitz, J., Milner, R., Zuberbier, O. A., Meloche, W. (2004). Psychosocial factors predictive of occupational low-back disability: Towards development of a return to work model. Pain, 107, 77-85. Schwartz, G. E., Davidson, R. J., & Goleman, D. J. (1978). Patterning of cognitive and somatic processes in the self regulation of anxiety; effects of mediation versus exercise. Journal of Psychosomatic Medicine, 40,321-328. Sieben, J. M . , Vlaeyen, J. W. S., Tuerlinkx, S., & Portegijs, P. J. M . (2002). Pain related fear in acute low-back pain: The first two weeks of a new episode. European Journal of Pain, 6, 229-237. Slade, P. D., Troup, J. D. G., Lethem, J., & Bentley, G. (1983). The fear-avoidance model of exaggerated pain perception-II. Behavior Research and Therapy, 21, 409-416. Snow-Turek, A . L., Norris, M . P., & Tan, G. (1996). Active and passive coping strategies in chronic pain patients. Pain, 64, 455-462. Spielberger, C. D. (1983). Manual for the State-Trait Anxiety Inventory. Palo Alto, C A : Consulting Psychologists Press. Spitzer, W. O. (1987). Scientific approach to the assessment and management of activity-related spinal disorders: A monograph for clinicians. Report of the Quebec Task Force on Spinal Disorders. Spine, 12(7S), 1-59. Stratford, P. W., & Spadoni, G. (2001). The reliability, consistency, and clinical application of a numeric pain rating scale. Physiotherapy Canada, 53, 88-100. Sullivan, M . J. L., Bishop, S., & Pivak, J. (1995). The Pain Catastrophizing Scale: Development and validation. Psychological Assessment, 7, 524-532. Sullivan, M . J. L., & D'Eon, J. L . (1990). Relation between catastrophizing and depression in chronic pain patients. Journal of Abnormal Psychology, 99, 260-263.  Catastrophizing, Fear  260  Sullivan, M . J. L., Rodgers, W. M . , & Kirsch, W. M . (2001). Catastrophizing, depression and expectancies for pain and emotional distress. Pain, 91, 147-154. Sullivan, M . J. L., Stanish, W., Waite, H., Sullivan, M . , & Tripp, D. A . (1998). Catastrophizing, pain, and disability in patients with soft-tissue injuries. Pain, 77, 253-260. Sullivan, M . J. L., Thorn, B., Haythornwaite, J. A. Keefe, F., Martin, M . , Bradley, L. A. & Lefebvre, J. (2001). Theoretical Perspectives on the relation between catastrophizing and pain. The Clinical Journal of Pain, 17, 52-64. Sullivan, M . J. L., Tripp, D. A., Rodgers, W. M . , & Stanish, W. (2000). Catastrophizing and pain in perception in sport participants. Journal of Sport Psychology, 12, 151-167. Swinkels-Meewise, E. J., Roelofs, J., Verbeek, A. L. M . , Ooostendorp, R. A . B., & Vlaeyen, J. W. S. (2003). Fear of movement/(re)injury, disability and participation in acute low-back pain. Pain, 105, 2>1\-2,19. Swinkels-Meewise, E. J., Swinkels, R. H. A. M . , Verbeek, A . L . M . Vlaeyen, J. W. S. & Ooostendorp, R. A . B . (2003). Psychometric properties of the Tampa Scale for Kinesiophobia and the fear-avoidance beliefs questionnaire in acute low-back pain. Manual Therapy, 8, 29-36. Tait, R. C , Chibnall, J. T., & Krause, S. (1990). The Pain Disability Index: Psychometric properties and validity data. Archives of Physical Medicine and Rehabilitation, 68, 438441. Tait, R. C , Chibnall, J. T., & Richardson, W. D. (1990). Litigation and employment status: Effects on patients with chronic pain. Pain, 43, 37 - 46. Tanaka, J. S. (1993). Multifaceted conception of fit in structural equation models. In K . A . Bollen & J.S. Long (Eds.), Testing structural equation models (pplO-39). Taylor, S., & Rachman, S. J. (1991). Fear of sadness. Journal of Anxiety Disorders, 5, 375-381.  Catastrophizing, Fear  261  Taylor, S., & Rachman, S. J. (1992). Fear and avoidance of aversive affective states: Dimensions and causal relations. Journal of Anxiety Disorders, 6, 15-25. Tbachnik, B., G., & Fiddel, L . S., (2001). Using Multivariate Statistics. Boston; Allyn and Bacon. Teasdale, J. (1988). Cognitive models and treatments for panic: A critical evaluation. In S. Rachman & J. D. Maser (Eds.), Panic: Psychological perspectives (pp. 189-203). Hillsdale, NJ: Lawrence Erlbaum Turk, D. C. (1996). Cognitive factors in chronic pain and disability. In K . Dobson, & K. Craig (Eds.) Advances in cognitive-behavioral therapy, (pp.) Newbury Park, California: Sage Publications. Turk, D. C. (1997). The role of demographic and psychosocial factors in transition from acute to chronic pain. In T. S. Jensen, J. A . Turner, & Z. Wiesenfeld-Hallin (Eds.), Proceedings of the 8th World Congress on Pain (pp. 185-213). Seattle, W A : IASP Press. Turk, D. C , & Flor, H . (1999). Chronic pain: A biobehavioral perspective. In R. J. Gatchel & D. C. Turk (Eds.), Psychosocial Factors in Pain (pp. 18-34). New York: Guilford Press. Turk, D. C , & Rudy, T. E. (1989). A cognitive-behavioral perspective on chronic pain: Beyond the scalpel and syringe. In C. D. Tollison (Ed.), Handbook of chronic pain management (pp. 222-235). Baltimore, M D : Williams & Wilkins. Turk, D. C , Rudy, T. E., & Salovey, P. (1985). The M c G i l l Pain Questionnaire reconsidered: Confirming the factor structures and examining appropriate uses. Pain, 21, 385-397. Turner, J. A., & Jensen, M . P. (1993). Efficacy of cognitive therapy for chronic low-back pain.; Pain, 52, 169-177.  Catastrophizing, Fear  262  Turner, J. A . (1982). Comparison of group progressive relaxation training and cognitive behavioural group therapy for chronic low-back pain. Journal of Consulting and Clinical Psychology, 50, 757-765. Ullman, J. M . (2001). Structural equation modeling. In B. G. Tbachnick & L. S. Fidell (Eds.), Using multivariate statistics (4 ed., pp. 653-771). Needham Heights, M A : Pearson th  Education Company. Vallfors, B. (1985). Acute, subacute and chronic low-back pain: Clinical symptoms, absenteeism and working environment. Scandinavian Journal of Rehabilitation Medicine, Supplement 11, 1-98. Van Damme, S., Crombez, G., Bijttebier, P., Goubert, L., & Van Houdenhove, B. (2002). A confirmatory factor analysis of the Pain Catastrophizing Scale: invariant factor structure across clinical and non-clinical populations. Pain, 96, 319-324. Vlaeyen, J. W. S., de Jong, J., Geilen, M . , Heuts P. H . T. G. & van Breukelen, G. (2001). Graded exposure in vivo in the treatment of pain related fear: A replicated single-case experimental design in four patients with chronic low-back pain. Behaviour Research and Therapy, 3P, 151-166. Vlaeyen, J. W. S., de Jong, J., Onghena, P., Kerkhoffs-Hanssen, M . , & Kole-Snijders, A . M . J. (2002). Can pain related fear be reduced? The application of cognitive-behavioural exposure in vivo. Pain Research and Management, 7, 144-153. Vlaeyen, J. W. S., Kole-Snijders, A . M . J., Boeren, R. G. B., & van Eek, H . (1995). Fear of movement/(re)injury in chronic low-back pain and its relation to behavioral performance. Pain, 62, 363-372.  Catastrophizing, Fear  263  Vlaeyen, J. W. S., Kole-Snijders, A. M . J., Rotteveel, A . M . , Ruesink, R., & Heuts P. H . G. T. (1995). The role of fear of movement/(re)injury in pain disability. Journal of Occupational Rehabilitation, 5,235-252 Vlaeyen, J. W. S., & Linton, S. J. (2000). Fear-avoidance and its consequences in chronic musculoskeletal pain: A state of the art. Pain, 85, 317-332. von Korff, M . , Dworkin, S. F., LeResche, L., & Kruger, A . (1988). A n epidemiological comparison of pain complaints. Pain, 32, 173-183. Waddell, G. (1987). A new clinical model for the treatment of low-back pain. Spine, 12, 157168. Waddell, G., Newton, M . , Henderson, I., Somerville, D., & Main, C. J. (1993). A fear-avoidance beliefs questionnaire (FABQ) and the role of fear avoidance beliefs in chronic low-back pain and disability. Pain, 52, 157-168. Wall, P. (1996). Comments after 30 years of the gate control theory of pain. Pain Forum, 5, 1222. Warwick, H. M . C. (1989). A cognitive-behavioural approach to hypochondriasis and health anxiety. Special Issue: Somatization: The somatic presentation of psychiatric illness. Journal of Psychosomatic Research, 33, 705-711. Warwick, H . M . C , & Salkovskis, P. M . (1990). Hypochondriasis. Behaviour Research and Therapy, 28, 105-117. Watson, D. & Pennebaker, J. W. (1989). Health complaints, stress, and distress: Exploring the central role of negative affectivity. Psychological Review, 96, 234-254. West, S. G., Finch, J. F., & Curan, P. J. (1995). Structural equation models with nonnormal variables: Problems and remedies.  In R. H. Hoyle (Ed), Structural Equation Modeling:  Concepts, Issues, and Applications (pp. 56-75). London: Sage Publications.  Catastrophizing, Fear  264  White, K . S., & Barlow, D. H . (2002). Panic disorder and Agoraphobia. In . In D.H. Barlow (Ed.). Anxiety and Its Disorders: The Nature and Treatment of Anxiety and Panic. New York: The Guilford Press pp. 105-138. Williams, D. A., & Thorn, B . E. (1989). A n empirical assessment of pain beliefs. Pain, 36, 351358.  Catastrophizing, Fear  265  APPENDIX A Complete questionnaire package  T H E U N I V E R S I T Y OF B R I T I S H C O L U M B I A Department of Psychology 2136 West Mall Vancouver, B.C. V6T 1Z4 Tel: (604)822-5280 Fax:(604)822-6823  Dear <Participant> Would you like to help researchers from UBC and the WCB studying how people recover from low-back injuries? If you participate, you will fill out 2 sets of questionnaires, one now and the other in 3 months. Each takes about 45 minutes. You will receive $20 after returning the second set. We are partly funded by the WCB but are completely separate from their regular operations. If you take part, the questionnaires will only be used for this study. Your personal information will not be released to anyone, including anybody associated with the claims process or your employer. Your involvement will have no effect on your WCB claim or any treatment you receive. This project was approved by the U.B.C. Behavioural Ethics Research Board and the Freedom of Information and Privacy Protection Office of the WCB. Please read the following description and consent form. If you agree to participate, sign the form, then fill-out the questionnaires as soon as possible. Please return them and the signed consent form in the enclosed selfaddressed stamped envelope. You then will receive another set by mail in three months. When you return them, we will mail your $20 honorarium. Your participation in this study is voluntary and you are not obligated in any way to fill in these questionnaires. The questionnaire is identified only by a code number to protect your privacy. Please do not write your name on the questionnaires. We will telephone you in about one week to be sure you have received this package and to answer questions. You will not be asked any question which does not appear on the questionnaires. If you choose not to participate, please tell the caller and you will not be contacted again. In addition, you are welcome to contact members of the research team at  (604) 822-5280 with any questions regarding the study or your  participation. If the call is long distance for you, please call collect and ask for X X X or X X X .  Catastrophizing, Fear  266  Instructions if you choose to participate: ***************************  If you: a) are  pregnant,  b) have  had back  c) have  been  told you  have  d) have  been  told you  have  e) have  been  told you  have  f) have  been  told you  have  disk  g) have  been  told you  have  spinal  h) have  been  told you  have  sciatica  i) have  been  told you  have  degenerative  j)  been  told you  have  a vertebral  to read  English  have  k) are unable  surgery cancer an arthritic  condition  of the  spine  spondlolysthesis herniation stenosis  disk  disease  fracture  well enough  to understand  the  newspaper  Please Do Not Complete the Questionnaires (See consent form for explanation) ******************************************************************  If none of the above apply to you: 1) Read and sign consent form. 2) Fill out questionnaire package as soon as possible. Preferably all questionnaires on the same day; It is very important that you fill in the questionnaires  as close  together  as  possible.  3) P