Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Balancing efficiencies and tradeoffs in epidemiological field studies : evaluating EMG exposure assessment… Trask, Catherine Mary 2008

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

Item Metadata

Download

Media
24-ubc_2008_fall_trask_catherine.pdf [ 1.49MB ]
Metadata
JSON: 24-1.0066568.json
JSON-LD: 24-1.0066568-ld.json
RDF/XML (Pretty): 24-1.0066568-rdf.xml
RDF/JSON: 24-1.0066568-rdf.json
Turtle: 24-1.0066568-turtle.txt
N-Triples: 24-1.0066568-rdf-ntriples.txt
Original Record: 24-1.0066568-source.json
Full Text
24-1.0066568-fulltext.txt
Citation
24-1.0066568.ris

Full Text

BALANCING EFFICIENCIES AND TRADEOFFS IN EPIDEMIOLOGICAL FIELD STUDIES: EVALUATING EMG EXPOSURE ASSESSMENT FOR LOW BACK INJURY RISK FACTORS IN HEAVY INDUSTRY  by Catherine Mary Trask BSc Kines (hons) Simon Fraser University 2003  A THESIS SUBMITTED IN PARTIAL REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES  (Occupational and Environmental Hygiene)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) AUGUST 2008 © Catherine Mary Trask, 2008  Abstract In order to investigate the etiology of and evaluate interventions for work-related back injuries, researchers need efficient, accurate occupational exposure assessment methods suitable for large samples. The chapters in this thesis examine critical decisions using electromyography (EMG): How should exposure be measured? For what duration? Who should be measured, and how many times? Low-back EMG, or muscle activity data, was collected during 138 full-shift field measurements over 30 different job titles at 50 different worksites in 5 heavy industries: forestry, transportation, wood products, construction, and warehousing. Observations and self-reports of posture, manual materials handling (MMH), and driving exposures were collected concurrently. 1) Variability of EMG calibration measurements was investigated on right/left sides, multiple trials, 4 positions, and pre/post-shift. Position accounts for the majority of explained variability; there is little to gain by measuring multiple trials or pre- and post-shift, but measuring both sides and multiple positions is worthwhile. 2) Observation and self-report data were easier to collect and cheaper than the EMG direct measure. Costs and successful field performance need to be weighed against the added data detail when making choices about exposure assessment techniques for epidemiological studies. 3) Observed and self-reported exposures were used to predict EMG exposure metrics using mixed multiple linear regression models. Regression models using observed variables predicted 43-50% of the variability in the EMG metrics, while self-reported variables predicted 21%-36%. The observation exposure model provides a low-cost alternative to direct measurement. The selfreported exposure model should be considered with more caution. 4) Full-shift EMG data was resampled for 4, 2, and 1 hour, and for 10 and 2 minute durations to determine the optimal sampling duration. Bias was consistently low, but shorter durations had higher absolute error, percentage error, and limits of agreement. Durations of 4 and 2 hours may be acceptable but those less than 1 hour had large errors. 5) Components of EMG variance were calculated between- and within-subject, and betweenindustry, company, job, and post hoc grouping. Resolution, contrast, and exposure-response relationship attenuation were calculated for each grouping scheme. The post hoc scheme had the highest contrast and lowest resolution.  ii  Table of Contents Abstract ..................................................................................................................................... ii Table of Contents...................................................................................................................... iii List of Tables ............................................................................................................................ iv List of Figures........................................................................................................................... vi Acknowledgements .................................................................................................................. vii Dedication............................................................................................................................... viii Co-Authorship statement........................................................................................................... ix Chapter 1: Introduction .............................................................................................................. 1 Chapter 2: Context of the thesis: the UBC Back Study ............................................................. 19 Chapter 3: Evaluating EMG as an exposure metric: implications for calibration and shift-long measurement ............................................................................................................................ 24 Chapter 4: Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility ............................................................................................. 37 Chapter 5: Predicting exposure for mean, 90th percentile, and cumulative EMG activity in heavy industry.................................................................................................................................... 52 Chapter 6: How long is long enough? Selecting efficient sampling durations for low-back EMG assessment ............................................................................................................................... 68 Chapter 7: Optimizing sampling strategies: components of low-back EMG variability in five heavy industries ....................................................................................................................... 80 Chapter 8: Discussion .............................................................................................................. 95 Appendix A: Detailed recruitment methodology......................................................................113 Appendix B: Detailed exposure assessment methodology........................................................116 Appendix C: Observation tool .................................................................................................121 Appendix D: Interview tool (worker copy) ..............................................................................122 Appendix E: Interview record sheet (researcher copy).............................................................138 Appendix F: EMG data collection an processing schematic.....................................................145 Appendix G: Detailed EMG data cleaning methods................................................................146 Appendix H: Detailed EMG analysis methods........................................................................149 Appendix I: Supplementary results: worker population ...........................................................152 Appendix J: Supplementary results for Chapter 5: full bivariable results .................................154 Appendix K: Supplementary results for Chapter 5: cumulative EMG using % time variables ..162 Appendix L: Supplementary results for Chapter 5: prediction modeling findings for collapsed variables..................................................................................................................................166 Appendix L: Supplementary results for Chapter 5: prediction modeling findings for collapsed variables..................................................................................................................................167 Appendix M: Supplementary results for Chapter 7: exposure by post-hoc groups....................168 Appendix N: Copy of research ethics certificate......................................................................171  iii  List of Tables Table 3.1: Mean age, height, weight, and sex of study participants (n = 103 workers) .............. 27 Table 3.2: T-test results for EMG activities by pre- and post-shift, trial 1 and 2, left and right side................................................................................................................................... 29 Table 3.3: Summary of simple linear regressions of the relationships between each of the following independent variables and muscle activity in µV............................................... 29 Table 3.4: Mixed-effect multiple regression model examining the adjusted effect of calibration variables, load weight and trunk posture on muscle activity.............................................. 30 Table 3.5: Mixed-effect multiple regression model examining the adjusted effect of calibration variables and task condition (combined load weight and trunk posture) on muscle activity30 Table 4.1: Summary of reasons for missed or incomplete measurements for each of the five measurement techniques................................................................................................... 41 Table 4.2: The total and per-day costs (in $CDN) associated with data collection for five ergonomic measurement methods used in this study ......................................................... 44 Table 4.3: Risk factors assessed, data detail, advantages and disadvantages of the exposure assessment methods used in this study.............................................................................. 46 Table 5.1: Sex, age, height, weight of study participants in heavy industry (n=103 workers) .... 56 Table 5.2: EMG exposure metrics for five heavy industries. All metrics are based on full-shift data collection .................................................................................................................. 56 Table 5.3: Observed ergonomic variables associated with mean, 90th percentile, and cumulative EMG exposure (expressed at % of reference calibration contraction or %RC) in final multiple linear regression models, with subject as a random effect.................................... 57 Table 5.4: Self-reported ergonomic variables associated with mean, 90th percentile, and cumulative EMG exposure (expressed at percentage of reference calibration contraction or %RC) in final multiple linear regression models, with subject as a random effect. ............ 58 Table 6.5: Mean and standard deviation of differences, bias, and limits of agreement between full shift and partial shift exposure metrics ....................................................................... 74 Table 7.1: Pearson Correlation coefficients between EMG metrics........................................... 84 Table 7.2: Industry and job averages (over all person-shifts) for three EMG exposure metrics.. 85 Table 7.3: Summary of EMG exposure metrics for post hoc groupings (approximate quintiles of exposures based on stratification by both industry and job)............................................... 86 Table 7.4: The proportions of variance in four EMG exposure metrics accounted for by betweengroup, between-worker and within-worker components using 5 different grouping schemes ......................................................................................................................................... 86 Table 7.5: Attenuation factors for EMG exposure-response relationships estimated using each grouping strategy and exposure metric.............................................................................. 87 Table 7.6: Required number of workers per groups (k) to achieve attenuation factors of greater than 0.95 for the mean EMG metric.................................................................................. 88 Table G.1: Record of Changes to the EMG data ......................................................................148 Table H.1: Original Exposure channels, variable codes, and full variable name .......................149 Table H.2: Modified exposure channels, variable codes, and full variable name ......................150 Table H.3: Summary Statistics, variable codes, and full variable name....................................151 Table I.1: Participants’ personal and demographic factors by industry.....................................152 Table I.2: Participants’’ work hours and related factors ...........................................................152  iv  Table I.3: Participants’ pain and activity reporting ..................................................................153 Table J.1: Descriptive statistics s and simple linear regression results for significant relationships between observed posture variables and EMG metrics by conceptual exposure groups....154 Table J.2. Descriptive statistics and simple linear regression results for significant relationships between observed MMH variables and EMG metric s by conceptual exposure groups.....155 Table J.3. Descriptive statistics and simple linear regression results for significant relationships between additional observed MMH variables and EMG metric s by conceptual exposure groups .............................................................................................................................156 Table J.4. Descriptive statistics and simple linear regression results for significant relationships between self-reported posture variables and EMG metrics by conceptual groups .............157 Table J.5. Descriptive statistics and simple linear regression results for significant relationships between additional self-reported posture variables and EMG metrics by conceptual groups ........................................................................................................................................158 Table J.6: Descriptive statistics and simple linear regression results for significant relationships between additional self-reported MMH variables and EMG metrics by conceptual groups ........................................................................................................................................159 Table J.7. Descriptive statistics and simple linear regression results for significant relationships between self-reported pushing and pulling variables and EMG metrics by conceptual groups ........................................................................................................................................160 Table J.8. Descriptive statistics and simple linear regression results for significant relationships between additional self-reported posture variables and EMG metrics by conceptual groups ........................................................................................................................................161 Table K.1: Descriptive statistics and simple linear regression results for significant relationships between observed postural variables and EMG outcomes by conceptual exposure groups162 Table K.2: Descriptive statistics and simple linear regression results for significant relationships between additional observed postural variables and EMG outcomes by conceptual exposure groups .............................................................................................................................163 Table K.3: Descriptive statistics and simple linear regression results for significant relationships between observed MMH variables and EMG outcomes by conceptual exposure groups ..164 Table K.4: Descriptive statistics and simple linear regression results for significant relationships between additional observed MMH variables and EMG outcomes by conceptual exposure groups .............................................................................................................................165 Table K.5. Observations and self-reported ergonomic variables associated with median, 90th percentile, and cumulative EMG exposure in final linear regression equations.................166 Table L.1: Bivariable regression for derived MMH and trunk posture variables ......................167 Table M.1: Development of Post-hoc groups for mean EMG ..................................................168 Table M.2: Development of Post-hoc groups for peak EMG....................................................169 Table M.3: Development of Post-hoc groups for cumulative EMG..........................................170  v  List of Figures Figure 1.1: a) vertical interpretation of exposure assessment hierarchy (left), b) horizontal interpretation of exposure assessment as a continuum (right) .............................................. 5 Figure 2.1: The stages of the UBC Back Study research program ............................................. 19 Figure 2.2: Flow chart outlining the methods development steps of the UBC Back Study......... 20 Figure 2.3: Conceptual model of the determinants of physical back injury risk factor exposure 21 Figure 3.1 Multipoint Calibration Data Collection Schematic................................................... 26 Figure 3.2: Box plots for muscle activity separated by a) pre/post b) left/right c) task condition and d) trial number ........................................................................................................... 28 Figure 6.1: Schematic of the resampling process for shorter sampling durations....................... 71 Figure 8.1: Four examples of simulated working exposure varying over time ..........................101 Figure B.1: Anatomical landmarking for the location of the EMG electrodes ..........................116 Figure B.2: Calibration was performed without a weight while standing upright, and forward o o o flexed at 45 . The 45 and 60 positions were measured with a 11.5 kg weight. ...............118  vi  Acknowledgements I am very fortunate to have worked with so many great people and benefited from the help of some great organizations. I would like to thank the following individuals and groups for their help and support while this research was being conducted: • I would like to thank the workers and employers who participated in this study. In scheduling and completing two full shifts of measurement they showed substantial commitment, patience, and goodwill, and for this I am very grateful. •  Dr. Mieke Koehoorn, my supervisor, whose patience, grace, and trust in me throughout all the challenges helped me be not only a better researcher, but a stronger person as well.  •  Dr. Kay Teschke, a committee member whose gentle mentoring since the start of the Back Study helped me to find the answers by asking the right questions and allowed me to learn by leading the project.  •  Dr. Jim Morrison, a committee member whose experience, attention to detail, and selfdescribed ‘caginess’ and has been invaluable during analysis and write-up.  •  A special thanks to Judy Village, Dr. Pete Johnson, Jim Ploger, and Dr. Geoff Wright who provided mentoring support throughout my degree.  •  Kevin Hong, Nancy Luong, Melissa Knott, James Cooper , and Yat Chow, my fellow data collectors, who shared much travel, long shifts, and BC coastal weather in all seasons! Thank you for your enthusiasm, your dedication to the process, and your friendship.  •  The CIHR/MSFHR Bridge Fellowship Program, Imelda Wong, Linda Bonamis, the faculty mentors, and all the Bridge Fellows. The Bridge Program provided not only financial support, but also fostered my development as a researcher.  •  The Research Secretariat of the Workers’ Compensation Board of BC (now WorkSafeBC) for financial support of both the UBC Back Study and me as a research trainee. I am happy to see the importance of occupational health research continue to be recognized and supported by WorkSafeBC.  •  Michael Smith Foundation for Health Research for their financial support of me as a research trainee as well as the British Columbia Occupational and Environmental Health Research Network (BCEOHRN) and the Centre for Health and Environment Research (CHER).  •  UBC School of Occupational and Environmental Hygiene, its faculty, staff and students, particularly Andrea L., for support and friendship throughout my degree.  •  Erica McArthur, Jordan Einarson, Andrea Dillon, Kamini Jain, and everyone else on the False Creek Racing Canoe Team and Canadian National Dragonboat Team.  vii  Dedication Completing this thesis was rewarding but involved some tough times and sacrifices, not just on my part but for my family as well. Data collection meant about 18 months of late nights, early mornings, extreme conditions, and travel. Data analysis and write up also included lots of ‘screen time’ cleaning and processing data, time spent writing, editing and incorporating comments, then reading and reviewing. Every time I came home from working on my research I was met with love, support, and usually a warm meal as well. I would like to dedicate this thesis to my family: my parents Les and Annabelle Trask, my siblings Elizabeth, David, and Bridget, and most of all, my husband Graham Smith. Thank you for encouraging me and helping span the gaps. I could not have made this journey alone and I would not have wanted to.  viii  Co-Authorship statement Identification and design of research program The research topic and research design were developed by the candidate with the help of the thesis committee and paper co-authors. Although conducted within the UBC Back Study as described in Chapter 2, the current work presents a unique contribution. Four of the six research manuscript chapters investigated in the current work were not included in any way in the original UBC Back Study. Performing the research The development of data collection protocols as well as pilot testing and protocol evaluations was led by the candidate with the help of research assistants and supervised by the thesis committee. Recruitment of subjects and worksites, scheduling of worksite measurement visits, and data collection was managed and conducted by the candidate, with input from bi-weekly or weekly meetings with Dr Koehoorn and Dr Teschke. Data Analyses Data analysis plans, including data clean-up, data compilation, and statistical analysis, were developed by the candidate, circulated to the committee for review, and then carried out solely by the candidate. The software developed for analyzing the EMG data was developed with input and documentation from the candidate. Preparation of manuscripts Original manuscript outlines and drafts were prepared by the candidate and then reviewed by the thesis committee and co-authors. The manuscript chapters as they appear in this thesis are the result of comments, editing, and redirection from the committee. A list of the co-authors is included in a footnote for each of Chapters 2-7.  ix  Chapter 1: Introduction The problem of back injury Back injury is a prevalent and expensive problem in modern industrialized societies. Estimates of point prevalence of back pain from the United States (US) range from 12% to 33%, 1-year prevalence from 22% to 65%, and lifetime prevalence from 11% to 84% (Deyo and Weinstein, 2001; Papageorgiou et al. 1995; van Tulder et al. 2002). In 1988, the 22.4 million back pain cases in the US resulted in 149.1 million lost workdays, resulting in considerable direct and indirect costs (Guo et al. 1999). In the Netherlands, the estimated total societal costs of back injury is 1.7% of the country’s GDP (US$367.6 million in 1991). The direct medical costs of treating back pain, although substantial, represented only 7% of the total cost, while indirect costs of lost production due to absenteeism and persistent disability made up the remaining 93% (van Tulder et al. 1995; Hutubessy et al. 1999). In the US 65% of all back-injury cases are attributable to occupational activities, and back pain point prevalence in the working population is 17.6%, with some occupations like construction as high as 22.6% (Guo et al. 1999). Between 1987 and 1995, back injury claims decreased 34% in Washington State but still remained the most common work-related injury claim, with the last year of the study showing an estimated $8.8 billion spent on low back pain claims and a claim rate of 1.8 per 100 workers (Murphy and Volinn, 1999). In the Canadian province of British Columbia (BC), there were 167,480 accepted compensation claims for back strain between 1996 and 2005, representing 25% of all claims, 23% of workdays lost, and 20% of claims cost. These statistics have remained fairly consistent or decreased only slightly during the 10-year time period (WorkSafeBC, 2005). One-quarter of all back claims in British Columbia come from the heavy industrial sectors of construction, forestry, transportation, warehousing and wood products. Clearly work-related back injury remains a highly prevalent and expensive problem, and although controls and interventions may have shown some decreases or prevented increases, heavy industry continues to have high physical demands and correspondingly high back injury rates.  Risk factors for back injury Whatever debate there might be on the specific causes of back injury, it is generally accepted that back injury has a multifactorial etiology. Numerous reviews of back injury have been conducted, and multiple personal, psychosocial, and physical risk factors have been identified (Keyserling, 2000; Gerr and Mani, 2000; Shelerud, 2006; Burdorf and Sorock, 1997; Hartvigsen et al. 2004; Pope and Novotny, 1993; Shelerud, 1998; Manek and MacGregor, 2005). Personal risk factors such as genetics (Manek and MacGregor, 2005), age, previous back injury, anthropometry, and smoking are important (Riihimaki et al. 1989; Burdorf and Sorock, 1997) but are under limited control in a workplace context. Psychosocial factors such as social support, job status, control, and job satisfaction are thought to have a relationship with back injury (Koehoorn et al, 2006; Hoogendoorn et al. 2001), but may have more of an impact on the  1  transition from pain to disability, or on the chronicity of back pain (Shelerud, 2006) rather than on the mechanisms of injury themselves.  Physical risk factors for back injury There is generally consensus that awkward and sustained postures, heavy manual materials handling, and repetitive tasks contribute to work-related back disorders. A comprehensive review of risk factors identified in epidemilogical studies (Burdorf and Sorock, 1997) concluded that there is “a clear relationship between back disorders and physical load, that is, between back disorders and materials handling, frequent bending and twisting, physically heavy work and whole body vibration”. This conclusion was based on the finding that the majority (84%) of studies evaluating lifting and carrying showed these activities associated with increased back injury, as did 57% of studies investigating heavy physical loads. In vitro studies of tests of spinal loading and tissue tolerance (Aultman et al. 2004; Aultman et al. 2005; Parkinson and Callaghan, 2007; Yingling et al. 1997), and the neuromuscular effects of different loading conditions (Sbriccoli et al. 2004) show how physical loads culminate in mechanical and physiological effects (Courville et al. 2005). These studies have led to estimates of the mechanical limits of anatomical structures and theories of how physical exposures lead to injury (Marras, 2005).  Challenges in back injury epidemiology Exposure assessment The controversies about relationships between working exposures and back injuries are related to difficulties in epidemiological studies of this outcome: in identifying cases of the health outcome, or in quantifying exposure, or both. Case ascertainment, and particularly identifying an ‘incident’ case of back injury which tends to be chronic or recurrent, is discussed elsewhere (Hagberg et al., 1997; Von Koff 1994; Wasiak et al., 2003) and will not be covered here. However, there has been general agreement that one of the largest barriers to elucidating back injury risk factors and creating guidelines has been inadequate exposure assessment methods (Burdorf, 1992a). Parnianpour et al. state that the fundamental inability to determine “How much of a risk factor is too much?” has been one of the most critical hindrances to the development of an ergonomic guideline for design of safe manual material handling tasks (Parnianpour et al. 1997). This barrier to understanding may well spring from the inability to determine ‘How much of the risk factor is present?’  Exposure metrics There are many ways to summarize ‘How much of the risk factor is present’. In the past, a single peak or ‘worst-case’ measure of exposure was the critical exposure metric for risk assessment in studies of manual materials handling tasks, the rationale being that injury would occur if acute loading exceeded the tolerance level of a tissue. However, changes in the conceptual model of injury that includes chronic pathways or mechanisms has lead to an appreciation for the importance of cumulative physical exposures (Marras, 2003; Waters et al. 2006;Morrison, 1999; Waters et al. 1993; McAtamney and Nigel-Corlett, 1993). Studies of care  2  aids (Kumar, 1990; Village et al. 2005), auto workers (Norman et al. 1998) and mixed occupational groups (Seidler et al. 2001) have shown that cumulative loads (that is, the sum of loads accumulated over time) are related to back injury. Burdorf and van Riel (1996) have stated that exposure metrics should include an exposure factor’s level, frequency, and duration, as well as some measure of the patterns of exposure over time. Peak and mean summary metrics provide only a measure of exposure level. Cumulative loads as reported in the literature contain both level and duration, although these dimensions are indistinguishable when combined in a single metric. The frequency dimension has been investigated in electromyography (EMG) studies as the number of excursions below a certain threshold (called ‘gap time’) (Hansson et al. 2000; Veiersted et al. 1993) or above a certain threshold (called ‘jerk time’) (Mathiassen et al. 2003a; Mathiassen et al. 2005), but overall has been studied far less than the other two factors. Given that the many factors of exposure are likely to interact with the body in very different ways, a single summary of exposure is an incomplete representation of the exposure and the risk. Although causal pathways can be theorized, one way to test the suitability of an exposure metric is through predictive validity in epidemiological study. Since it is not known which exposure metrics are the best prior to conducting research, multiple ‘good candidates’ with plausible injury mechanisms might be examined in a given study. This dissertation investigates a number of EMG exposure metrics, comparing their feasibility and utility for epidemiological studies.  Challenges in exposure assessment In Burdorf’s review (Burdorf, 1992a) of postural exposure assessment in epidemiological studies of back disorders, 47 of 81 reviewed articles did not report quantifiable exposure data. Exposures in the remaining papers were assessed by questionnaire (33%), observation (9%), and direct measurement (5%), and exposure was rarely analyzed as a continuous variable but rather in discrete categories. The overall conclusion was that the quality of exposure assessment was poor in most epidemiological studies of back injury, and that quantitative assessment methods need to be improved upon for use in occupational field studies. Five years later the same author (Burdorf and Sorock, 1997) reviewed 140 articles on the risk factors related to back injury. Many studies were excluded due to poor description of physical exposure assessment or lowquality exposure assessment methods. Nearly half (49%) of the included studies used job title as a measure of exposure, which does not provide a quantitative estimate nor allows for appropriate targeting of intervention efforts in the jobs identified as ‘higher risk’. A more recent review by David (2005) examining ergonomic exposure assessment methods for any kind of musculoskeletal complaint found 8 articles using self-report methods, 27 articles using different kinds of observational techniques ranging from checklists to video-based methods, and 9 using direct measurement methods. The author pointed out that some guidance or decision tool outlining the best method given the situation would be very useful for researchers. Poorly measured exposure can lead to bias or random error. Bias in the exposure assessment can yield spurious relationships between variables. Just as problematic are unbiased measures with large random error, because the relationship tends towards null and may not reach significance even when a relationship exists. Unfortunately most studies can’t afford the time and money it would take to estimate the misclassification error, and as a result it is not clear whether bias or 3  measurement error has influenced the results (Wiker, 2003). Challenges are ubiquitous in epidemiological exposure assessment, but the nature, extent, and effect of these challenges can vary with the specific method chosen.  Physical exposure assessment paradigm Physical exposure assessment methods are often grouped into three broad categories: direct measurement, observation by trained personnel, and worker self report. Self-reports via questionnaire, interview, or diary are “stand-ins for economically and socially impractical round the clock measures” (Wiker, 2003), but their subjective nature may produce unreliable, imprecise, or even systematically biased results. Observation represents a ‘middle ground’ in that there is more objectivity in an expert’s assessment, but it still lacks precision. Objective direct measurement methods are preferred for accuracy (Burdorf and van der Beek, 1999; van der Beek and Frings-Dresen, 1998; Burdorf, 1992b; David, 2005). These three groups of methods are often presented as a hierarchy, with self-report at the bottom as ‘least desirable’ and direct measure at the top as ‘most desirable’ (Figure 1.1a). Although this hierarchy holds true in terms of the precision of measurements, thinking about the benefits and advantages of each method might best be facilitated by moving from a vertical hierarchy to a horizontal continuum, as in Figure 1.1b. A continuum was similarly used by Winkel and Mathiassen (Winkel and Mathiassen, 1994) to convey the tradeoffs in the three types of methods; some advantages and disadvantages have also been tabulated by van der Beek and Frings-Dresen (van der Beek and Frings-Dresen, 1998). Contrasted with precision is cost, running on an inverse continuum. The result is that the cheaper methods on the right-hand side can be applied to much larger numbers of people, and repeats within people, tasks, or trials of a task. The horizontal structure also gives an insight into the optimization that takes place in the selection of methods. The microscope and telescope represent the fact that methods on both ends of the spectrum are valuable, but they are tools giving different information with a very different ‘scope’ or ‘resolution’. High-resolution lab-based studies typically have 10-20 subjects, depending on the depth of information collected. Observation and self-report data used in large-scale injury epidemiological studies may have hundreds to thousands of subjects. This dissertation will quantify and compare the tradeoffs between methods from the three areas of measurement and determine their suitability for use in epidemiological studies of back injury.  4  a)  b)  Figure 1.1: a) vertical interpretation of exposure assessment hierarchy (left), b) horizontal interpretation of exposure assessment as a continuum (right)  EMG as a direct measurement method Portable electromyography (EMG) instrumentation can make practical, high-resolution assessments of physical risks required for epidemiological studies. In addition, EMG makes a good choice for assessing multiple summary metrics since measurements can be made for a full shift, allowing for estimation of cumulative exposure, peak loads, and other metrics. EMG has been used in studies of working exposures to the back (Village et al. 2005; Keir and MacDonell, 2004; de Oliveira et al. 2001; Jones and Kumar, 2007b; Jones and Kumar, 2007a) and shoulder (Akesson et al. 1997; Blangsted et al. 2003; Hansson et al. 2000; Jensen et al. 1993; Capodaglio et al. 1996; Nakata et al. 1993; Nordander et al. 2004; Sporrong et al. 1999). The nature of the EMG electrode-skin interface is unique to the measurement session, so a unique calibration equation must be developed for each session. Calibration can be accomplished in several different ways, but all involve making measurements of standardized movements to act as a reference prior to a field measurement to act as a reference. Muscle activities can be transformed into estimates of spinal compression (Potvin et al. 1990; Potvin et al. 1996; Mientjes et al. 1999; Village et al. 2005), or expressed as a percentage of a maximum voluntary contraction (%MVC) (Jones and Kumar, 2007a; Jones and Kumar, 2007b), or as a percentage of a sub-maximal reference contraction (%RC) (Burden and Bartlett, 1999; Sporrong et al. 1999). It is not clear is how EMG calibration measurements vary between trials, pre-and post-shift, or in different reference positions. In addition to comparing multiple exposure metrics from full-shift measurements, this dissertation will identify and quantify these sources of variability in EMG calibration measures.  5  Sampling strategies Once an exposure assessment method has been selected, a researcher is faced with additional choices of how to apply that method. Even the best, most precise method is limited if the sampling strategy is haphazard. Given that research budgets are limited and not everyone can be measured multiple times, Burdorf and van Riel (1996) have described strategy design options as ‘necessary choices’. Specifically, these choices involve whom to measure, when, for how long, and how many times (i.e. repeated measurement of individuals). Often workers are measured once or twice in an effort to relate their exposure to a health effect (such as back injury) that we expect is the result of long-term cumulative effect of many weeks, months, or years of exposure. In doing this, assumptions are made about the representativeness of the measurements (Gold et al. 2006). Determining the ‘whom’, ‘how many’, and ‘how long’ aspects of sampling strategy are critical to the representativeness of the exposure assessments assigned to each individual, and as such have considerable impact on the ability of a study to identify existing dose-response relationships.  Whom to measure The decision of whom to measure and how to group or stratify individuals exposed to physical loading is an important one, but one on which there has been little ergonomic research. Grouping schemes have been used in the past for convenience (Nieuwenhuijsen, 1997) and are useful if all workers within in occupational group or company department have comparable exposures. This allows for a streamlined sampling strategy. The notion of a ‘homogenous exposure group’ or HEG has been discussed in the context of chemical exposures, where the th th range between the 5 and 95 percentiles within the group is less than a two-fold difference (Rappaport, 1991). However, over a range of chemical exposures, only about 20% of job groupings were HEGs, and another 20% had 15-fold differences between the 5th and 95th percentile mean (Rappaport et al. 1993). When the exposure within a group is highly varied (that is, the within-group variability is high), comparisons between groups are less likely to show differences in exposure and subsequently, significant exposure-response relationships will be harder to identify. Despite this challenge, appropriate groupings for exposure assessment can make a study more efficient, because the larger numbers of samples that are possible for each group provide a precise estimate of the group’s mean exposure. Even when precision is lost for each individual, the increased precision of the group can lead to stronger dose-response relationships (Kromhout et al. 1995; Seixas and Sheppard, 1996; Tielemans et al. 1998).  When to measure There are a number of possible sampling strategies to provide guidance as to when to sample. This can include decisions of the day of the week or time of day, as well as which tasks to target. Some examples of previously used sampling strategies include full-shift, task-based, fixed interval, random interval, and worst-case (Mathiassen et al. 2003a). Unlike other methods that select sampling times without knowledge of the work, task-based sampling requires advance understanding about the type of work tasks and the scheduling of these tasks, so that each task may be targeted for measurement.  6  How long to measure The duration of exposure assessment can also affect the outcome of a study. Even when highprecision direct measurement equipment is used, brief sampling durations can be unrepresentative (Burdorf and van der Beek, 1999) since they neglect the temporal variations in exposure throughout the work day (Burdorf A et al. 1992; Ensink et al. 1996). Short sampling durations assume (implicitly or explicitly) that measured characteristics are representative of the larger period over which they are extrapolated (Gold et al. 2006). Despite this, “whole-day recordings of exposure…are rare in ergonomics” (Mathiassen et al. 2002), even though the purpose of sampling is to assign an exposure “that is representative for the individual during an extended period of time”. When work is cyclical (Mathiassen et al. 2003b; Veiersted et al. 1993), or highly repetitive, a short sampling duration may deliver an accurate estimate of exposure if enough task cycles are included to account for variability between cycles. Three task cycles have been suggested as an appropriate number for workplace posture assessments (Allread et al. 2000), though this number was based on very strictly defined MMH tasks (i.e. lifting box A from height B to height C) and it is possible that ‘lifting’ tasks that varied in weight, distance, or symmetry would require more cycles. Short sampling durations seem less likely to capture the range of exposure for non-cyclical or highly variable work typical of much of the workforce in heavy industries. EMG studies have a wide variety of sampling durations, as shown in a review of shoulder EMG by Mathiassen et al. (2002). Studies assessed cyclical tasks on the order of minutes (Hammarskjold and Harms-Ringdahl, 1992; Jensen et al. 1993), 1-2 hours (Lundberg et al. 1999; Rissen et al. 2000; Nakata et al. 1993) (Moller et al. 2004) and a full work day (Hansson et al. 2000). Even studies of non-cyclical work, such as patient transfer tasks in health care, measured 2 repeats of 92-145 seconds in duration (Keir 2004; Keir and MacDonell, 2004). This is typical of sampling durations in ergonomics, although a study of care aides by Village and colleagues (Village et al. 2005) measured back EMG for a full shift. Despite the importance of the tradeoffs between sampling long, more representative durations and short, inexpensive durations, there is very little guidance in the literature as to how long sampling durations should be.  How many to measure Often the total number of measurements is constrained by budget, but efficient allocation of these measurements between different individuals and repeated measures within individuals can increase the probability of identifying existing dose-response relationships. Statisticians evaluating the effects of measurement error have developed formulae to determine whether the focus should be on increasing the number of different subjects or the number of repeated samples; the efficiency of choosing one over the other depends on where the variability lies: within workers or between workers (Brunekreef et al 1987; Seixas et al. 1996; Jansen and Burdorf, 2003). For example, if within-worker variability is high, i.e. a worker’s exposure varies a lot from day to day, more repeated measurements will need to be made on individual workers in order to get an accurate idea of their exposures (Waddell, 1996). This requires a level of knowledge about the exposures and their distributions (Burdorf, 1995a; Nicas and Spear, 1993).  7  Methods to guide sampling strategy decisions As Burdorf and van Riel (1996) point out, “there are no simple universal rules as to choose which workers and work conditions to monitor”. Few ergonomic measurement tools explicitly state when and how to observe exposures (Gold et al. 2006). Informed decisions can be made with some baseline knowledge or estimates of the components of variability in the population, such as the between-group, between-subject, and within-subject variability (Rappaport et al. 1995). Since data on the components of variance is needed to make efficient sampling choices, it is generally agreed that a pilot study should be conducted to get an estimate of variance components prior to a wholesale sampling campaign (Brunekreef et al. 1987; Burdorf, 1995b; Burdorf and van Riel, 1996; Burdorf, 2005; Loomis and Kromhout, 2004). Although there are many gaps, there are some tools available to help determine sampling strategy even for those methods where strategy is not explicit. In ergonomic epidemiology, Burdorf developed formulae to determine study power and the number of individuals versus repeats on individuals based on within- and between-worker variability in postural exposure (Burdorf, 1995a). The ratio of between-worker variability over the sum of within and between-worker variability is called the ‘contrast’ between workers (Kromhout et al. 1997). In 1987 Brunekreef et al used the ‘variance ratio’ (within worker variance divided by the between-worker variance) to estimate the bias or attenuation in the regression coefficients that would result from using those exposure estimates (Brunekreef et al. 1987). Since then, several methods have been proposed using the components of variance to measure the attenuation of exposure response relationships with the purpose of optimizing the categorization and sampling strategies for epidemiological studies (Seixas and Sheppard, 1996; Tielemans et al. 1998; Kromhout et al. 1995; Teschke et al. 2004; Burdorf, 1995b; van der Beek et al. 1995). When exposures are grouped, stratified random selection within these groups has been recommended (Burdorf and van Riel, 1996). However, deciding how to categorize or group workers is a challenge. Analogous to the ‘contrast’ between workers is the ratio of betweengroup variability over the sum of within and between-group variability, called the ‘resolution’ between groups (Kromhout et al. 1995); exposure response relationships are least attenuated when the difference between subjects and groups is large compared to within-subject variability. The question of ‘when’ and ‘how long’ are poorly addressed in the literature, except to say that days should be randomly selected. These issues will be addressed in the current work by making shift-long EMG measurements concurrent with observation and self-report of working physical exposures. With these measures, methodological tools such as resampling of smaller durations, determinants of exposure modeling, contrast, and resolution will be used to answer questions of when and how long.  8  Objectives and rationale In order to move towards a richer understanding of the relationships between working exposures and back injury and eventually to intervening and controlling them, researchers need evidence for informed decision-making on the following components of physical exposure assessment. The purpose of this PhD thesis is to address knowledge gaps in exposure assessment and sampling strategy for workplace physical exposures. 1)  HOW TO MEASURE a) Requires evidence on how best to calibrate EMG for field studies EMG needs to be calibrated before it can be interpreted, but there are many methods available. The objective of chapter 3 is to compare EMG calibration measurements of 4 different reference postures and examine the variability within and between pre and post shift measures, right and left side of the body, and repeated trials of the same position. b) Requires evidence on the tradeoffs between direct, observation and self report exposure assessment methods Epidemiology requires methods that are inexpensive and feasible for large numbers of workers. The objective of Chapter 4 is to compare EMG, vibration, trunk inclinometry, observation, and self-reports of physical exposures with respect to cost, reliability, and feasibility in heavy industrial workplaces. This will provide an objective evaluation of how exposure assessment methods perform in the field to allow for a more informed choice by researchers. c) Requires evidence on the exposure assessment methods that balance sample power with precision Direct measurements have the highest accuracy, but observation and self-report allow for large numbers of measurements in epidemiological study. The objective of Chapter 5 is to examine what proportion of the variance for a directly measured exposure (EMG in this case) can be explained by observed or self-reported variables.  2) HOW LONG TO MEASURE a) Requires evidence on the duration of measurement to deliver representative exposure assessment data Long-term, cumulative measurement allows for analysis of trends, but it may not be necessary to measure a whole shift to obtain this information. The objective of Chapter 6 is to examine the effect of decreasing sampling lengths on the representativeness of resulting exposure estimates in order to identify the optimum sampling length for different EMG exposure metrics to capture representative shift-long exposures. 3) HOW MANY TIMES TO MEASURE a) Requires evidence on the number of repeated measures within a worker or group Research budgets have limits, so measurements should be judiciously apportioned within and between subjects. The objective of Chapter 7 is to find the most efficient sampling strategies for EMG exposure assessments by identifying the total number of samples as well as the number of repeated measures for subjects.  9  4) WHOM TO MEASURE a) Requires evidence on which workers should be measured, and from which groups: industries, companies, job titles, and working groups Having an estimate of exposure variability is important when making decisions about sampling strategy, particularly when distributing the total number of samples allowed by the budget to among and within companies, work sections, and subjects. The objective of Chapter 7 is to examine the sources of variability within and between workers, industries, companies, and job titles to enable a priori decisions about sampling strategy.  10  References Akesson, I., Hansson, G.A., Balogh, I., Moritz, U. and Skerfving, S. (1997) Quantifying work load in neck, shoulders and wrists in female dentists. Int Arch Occup Environ Health 69, 46174. Allread, W.G., Marras, W.S. and Burr, D.L. (2000) Measuring trunk motions in industry: variability due to task factors, individual differences, and the amount of data collected. Ergonomics 43, 691-701. Aultman, C.D., Drake, J.D., Callaghan, J.P. and McGill, S.M. (2004) The effect of static torsion on the compressive strength of the spine: an in vitro analysis using a porcine spine model. Spine 29, E304-9. Aultman, C.D., Scannell, J. and McGill, S.M. (2005) The direction of progressive herniation in porcine spine motion segments is influenced by the orientation of the bending axis. Clin Biomech (Bristol, Avon) 20, 126-9. Blangsted, A.K., Hansen, K. and Jensen, C. (2003) Muscle activity during computer-based office work in relation to self-reported job demands and gender. Eur J Appl Physiol 89, 352-8. Brunekreef, B., Noy, D. and Clausing, P. (1987) Variability of Exposure Measurements in Environmental Epidemiology. Am J Epidemiol 122 , 892-898. Burden, A. and Bartlett, R. (1999) Normalisation of EMG amplitude: an evaluation and comparison of old and new methods. Med Eng Phys 21, 247-57. Burdorf, A. (1992a) Exposure assessment of risk factors for disorders of the back in occupational epidemiology. Scand J Work Environ Health 18, 1-9. Burdorf, A. (1992b) Sources of variance in exposure to postural load on the back in occupational groups. Scand J Work Environ Health 18, 361-7. Burdorf, A. (1995a) Reducing random measurement error in assessment of postural load on the back in epidemiological surveys. Scand J Work Environ Health 21, 15-23. Burdorf, A. (1995b) Reducing random measurement error in assessing postural load on the back in epidemiologic surveys. Scand J Work Environ Health 21, 15-23. Burdorf, A. (2005) Identification of determinants of exposure: consequences for measurement and control strategies. Occup Environ Med 62, 344-50. Burdorf A, Derksen J, Naaktgeboren B and van Riel M. (1992) Measurement of trunk bending during work by direct observation and continuous measurement. Appl Ergon 23, 263-7. Burdorf, A., Rossignol, M., Fathallah, F.A., Snook, S.H. and Herrick, R.F. (1997) Challenges in assessing risk factors in epidemiologic studies on back disorders. Am J Ind Med 32, 142-52.  11  Burdorf, A. and Sorock, G. (1997) Positive and negative evidence of risk factors for back disorders. Scand J Work Environ Health 23, 243-56. Burdorf, A. and van der Beek, A.J. (1999) In musculoskeletal epidemiology are we asking the unanswerable in questionnaires on physical load? Scand J Work Environ Health 25, 81-3. Burdorf, A. and van Riel, M. (1996) Design of strategies to assess lumbar posture during work . Int J of Ind Ergo 18, 239-249. Burton, A.K. and Erg, E. (1997) Back injury and work loss. Biomechanical and psychosocial influences. Spine 22, 2575-80. Capodaglio, P., Jensen, C. and Christensen, H. (1996) Quantification of muscular activity in the shoulder region during monotonous repetitive work. Med Lav 87, 305-13. Courville, A., Sbriccoli, P., Zhou, B.H., Solomonow, M., Lu, Y. and Burger, E.L. (2005) Short rest periods after static lumbar flexion are a risk factor for cumulative low back disorder. J Electromyogr Kinesiol 15, 37-52. David GC. (2005) Ergonomic methods for assessing exposure to risk factors for work-related musculoskeletal disorders. Occup Med (Lond); 55(3):190-9. de Oliveira, C.G., Simpson, D.M. and Nadal, J. (2001) Lumbar back muscle activity of helicopter pilots and whole-body vibration. J Biomech 34, 1309-15. Deyo, R.A. and Weinstein, J.N. (2001) Low back pain. N Engl J Med 344, 363-70. Ensink, F.B., Saur, P.M., Frese, K., Seeger, D. and Hildebrandt, J. (1996) Lumbar range of motion: influence of time of day and individual factors on measurements. Spine 21, 1339-43. Foliart, D.E., Iriye, R.N., Tarr, K.J., Silva, J.M., Kavet, R. and Ebi, K.L. (2001) Alternative magnetic field exposure metrics: relationship to TWA, appliance use, and demographic characteristics of children in a leukemia survival study. Bioelectromagnetics 22, 574-80. Gerr, F. and Mani, L. (2000) Work-related low back pain. Prim Care 27, 865-76. Gold, J.E., Park, J.S. and Punnett, L. (2006) Work routinization and implications for ergonomic exposure assessment. Ergonomics 49, 12-27. Guo, H.R., Tanaka, S., Halperin, W.E. and Cameron, L.L. (1999) Back pain prevalence in US industry and estimates of lost workdays. Am J Public Health 89, 1029-35. Hammarskjold, E. and Harms-Ringdahl, K. (1992) Effect of arm-shoulder fatigue on carpenters at work. Eur J Appl Physiol Occup Physiol 64, 402-9. Hansson, G.A., Nordander, C., Asterland, P., Ohlsson, K., Stromberg, U., Skerfving, S. and Rempel, D. (2000) Sensitivity of trapezius electromyography to differences between work tasks - influence of gap definition and normalisation methods. J Electromyogr Kinesiol 10, 103-15.  12  Hagberg M, Christiani D, Courtney TK, Halperin W, Leamon TB, Smith TJ. (1997) Conceptual and definitional issues in occupational injury epidemiology. Am J Ind Med. Aug;32(2):106-15. Hartvigsen, J., Lings, S., Leboeuf-Yde, C. and Bakketeig, L. (2004) Psychosocial factors at work in relation to low back pain and consequences of low back pain; a systematic, critical review of prospective cohort studies. Occup Environ Med 61, e2 Hoogendoorn, W.E., Bongers, P.M., de Vet, H.C., Houtman, I.L., Ariens, G.A., van Mechelen, W. and Bouter, L.M. (2001) Psychosocial work characteristics and psychological strain in relation to low-back pain. Scand J Work Environ Health 27, 258-67. Hutubessy, R.C., van Tulder, M.W., Vondeling, H. and Bouter, L.M. (1999) Indirect costs of back pain in the Netherlands: a comparison of the human capital method with the friction cost method. Pain 80, 201-7. Ibbotson, J., Yost, M., Johnson, P. (2003). Using various exposure analyses to explore difference in mouse usage patterns. IEA International Ergonomics Association 15th Triennial Congress. 2003. ISO 2631-5. Mechanical vibration and shock - Evaluation of human exposure to whole-body vibration - Part 5: Method for evaluation of vibration containing multiple shocks . Jensen, B.R., Schibye, B., Sogaard, K., Simonsen, E.B. and Sjogaard, G. (1993) Shoulder muscle load and muscle fatigue among industrial sewing-machine operators. Eur J Appl Physiol Occup Physiol 67, 467-75. Jones, T. and Kumar, S. (2007a) Assessment of physical demands and comparison of multiple exposure definitions in a repetitive sawmill job: board edger operator. Ergonomics 50, 676-93. Jones, T. and Kumar, T. (2007b) Assessment of physical exposures and comparison of exposure definitions in a repetitive sawmill occupation: trim-saw operator. Work 28, 183-96. Keir, P.J. and MacDonell, C.W. (2004) Muscle activity during patient transfers: a preliminary study on the influence of lift assists and experience. Ergonomics 47, 296-306. Keyserling, W.M. (2000) Workplace risk factors and occupational musculoskeletal disorders, Part 1: A review of biomechanical and psychophysical research on risk factors associated with low-back pain. AIHAJ: a Journal For The Science Of Occupational And Environmental Health And Safety 61, 39-50. Koehoorn M, Demers PA, Hertzman C, Village J, Kennedy SM. Work organization and musculoskeletal injuries among a cohort of health care workers . Scand J Work Environ Health 2006; 32(4):285-93. Kromhout, H., Loomis, D.P., Kleckner, R.C. and Savitz, D.A. (1997) Sensitivity of the relation between cumulative magnetic field exposure and brain cancer mortality to choice of monitoring data grouping scheme. Epidemiology 8, 442-5. Kromhout, H., Loomis, D.P., Mihlan, G.J., Peipins, L.A., Kleckner, R.C., Iriye, R. and Savitz,  13  D.A. (1995) Assessment and grouping of occupational magnetic field exposure in five electric utility companies. Scand J Work Environ Health 21, 43-50. Kumar, S. (1990) Cumulative load as a risk factor for back pain. Spine 15, 1311-6. Loomis, D. and Kromhout, H. (2004) Exposure variability: concepts and applications in occupational epidemiology. Am J Ind Med 45, 113-22. Lundberg, U., Dohns, I.E., Melin, B., Sandsjo, L., Palmerud, G., Kadefors, R., Ekstrom, M. and Parr, D. (1999) Psychophysiological stress responses, muscle tension, and neck and shoulder pain among supermarket cashiers. J Occup Health Psychol 4, 245-55. Manek, N.J. and MacGregor, A.J. (2005) Epidemiology of back disorders: prevalence, risk factors, and prognosis. Curr Opin Rheumatol 17, 134-40. Marras, W.S. (2003) The case for cumulative trauma in low back disorders. The Spine Journal 3, 177-179. Marras, W.S. (2005) The future of research in understanding and controlling work-related low back disorders. Ergonomics 48, 464-77. Mathiassen, S.E., Burdorf, A. and van der Beek, A.J. (2002) Statistical power and measurement allocation in ergonomic intervention studies assessing upper trapezius EMG amplitude. A case study of assembly work. J Electromyogr Kinesiol 12, 45-57. Mathiassen, S.E., Burdorf, A., van der Beek, A.J. and Hansson, G.A. (2003a) Efficient one-day sampling of mechanical job exposure data--a study based on upper trapezius activity in cleaners and office workers. AIHA J (Fairfax, Va) 64, 196-211. Mathiassen, S.E., Moller, T. and Forsman, M. (2003b) Variability in mechanical exposure within and between individuals performing a highly constrained industrial work task. Ergonomics 46, 800-24. Mathiassen, S.E., Nordander, C., Svendsen, S.W., Wellman, H.M. and Dempsey, P.G. (2005) Task-based estimation of mechanical job exposure in occupational groups. Scand J Work Environ Health 31, 138-51. Mientjes, M.I., Norman, R.W., Wells, R.P. and McGill, S.M. (1999) Assessment of an EMGbased method for continuous estimates of low back compression during asymmetrical occupational tasks. Ergonomics 42, 868-79. Moller, T., Mathiassen, S.E., Franzon, H. and Kihlberg, S. (2004) Job enlargement and mechanical exposure variability in cyclic assembly work. Ergonomics 47, 19-40. Morrison, J., Robinson, D.G., Nicol, J.J., Roddan, G., Martin, S.H., Springer, M.J-N., Cameron, B.J., Albano, J.P. ( 1999) A biomechanical approach to evaluating the health effects of repeated mechanical shocks. Proc. Human Factors and Med. Panel, RTO-MP-20. Neuilly-sur-Seine Cedex, France.: Published by NATO Research and Technology Organisation.  14  Murphy, P.L. and Volinn, E. (1999) Is occupational low back pain on the rise? Spine 24, 6917. Nakata, M., Hagner, I.M. and Jonsson, B. (1993) Trapezius muscle pressure pain threshold and strain in the neck and shoulder regions during repetitive light work. Scand J Rehabil Med 25, 131-7. Nicas, M. and Spear, R.C. (1993) A task-based statistical model of a worker's exposure distribution: Part I--Description of the model. Am Ind Hyg Assoc J 54, 211-20. Nieuwenhuijsen, M.J. (1997) Exposure assessment in occupational epidemiology: measuring present exposures with an example of a study of occupational asthma. Int Arch Occup Environ Health 70, 295-308. Nordander, C., Balogh, I., Mathiassen, S.E., Ohlsson, K., Unge, J., Skerfving, S. and Hansson, G.A. (2004) Precision of measurements of physical workload during standardised manual handling. Part I: surface electromyography of m. trapezius, m. infraspinatus and the forearm extensors. J Electromyogr Kinesiol 14, 443-54. Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clinical Biomechanics 13 (8):561-573. Papageorgiou, A.C., Croft, P.R., Ferry, S., Jayson, M.I. and Silman, A.J. (1995) Estimating the prevalence of low back pain in the general population. Evidence from the South Manchester Back Pain Survey. Spine 20, 1889-94. Parkinson, R.J. and Callaghan, J.P. (2007) Can periods of static loading be used to enhance the resistance of the spine to cumulative compression? J Biomech 40, 2944-52. Parnianpour, M., Sparto, P.J. and Chen, M.C. (1997) Validation of electrolytic-liquid tilt sensors for human motion measurement. Biomed Sci Instrum 33, 465-70. Pope, M.H. and Novotny, J.E. (1993) Spinal biomechanics. Journal Of Biomechanical Engineering 115 (4B):569-574. Potvin, J., Norman, R. and Wells, R. A field method for continuous estimation of dynamic compressive forces on the L4/L5 disc during the performance of repetitive industrial tasks. Proceedings of 23rd Annual Conference of the Human Factors Association of Canada. September 26-28. Ottawa, Ontario Potvin, J.R., Norman, R.W. and McGill, S.M. (1996) Mechanically corrected EMG for the continuous estimation of erector spinae muscle loading during repetitive lifting. Eur J Appl Physiol Occup Physiol 74, 119-32. Rappaport, S.M. (1991) Assessment of long-term exposures to toxic substances in air. Ann Occup Hyg 35, 61-121. Rappaport, S.M., Kromhout, H. and Symanski, E. (1993) Variation of exposure between  15  workers in homogeneous exposure groups. Am Ind Hyg Assoc J 54 , 654-62. Rappaport, S.M., Lyles, R.H. and Kupper, L.L. (1995) An exposure-assessments strategy accounting for within- and between-worker sources of variability. Ann Occup Hyg 39, 469-95. Riihimaki, H., Wickstrom, G., Hanninen, K. and Luopajarvi, T. (1989) Predictors of sciatic pain among concrete reinforcement workers and house painters--a five-year follow-up. Scand J Work Environ Health 15, 415-23. Rissen, D., Melin, B., Sandsjo, L., Dohns, I. and Lundberg, U. (2000) Surface EMG and psychophysiological stress reactions in women during repetitive work. Eur J Appl Physiol 83, 215-22. Sbriccoli, P., Solomonow, M., Zhou, B.H., Baratta, R.V., Lu, Y., Zhu, M.P. and Burger, E.L. (2004) Static load magnitude is a risk factor in the development of cumulative low back disorder. Muscle Nerve 29, 300-8. Seidler, A., Bolm-Audorff, U., Heiskel, H., Henkel, N., Roth-Kuver, B., Kaiser, U., Bickeboller, R., Willingstorfer, W.J., Beck, W. and Elsner, G. (2001) The role of cumulative physical work load in lumbar spine disease: risk factors for lumbar osteochondrosis and spondylosis associated with chronic complaints. Occup Environ Med 58, 735-46. Seixas, N.S. and Sheppard, L. (1996) Maximizing accuracy and precision using individual and grouped exposure assessments. Scand J Work Environ Health 22, 94-101. Shelerud, R. (1998) Epidemiology of occupational low back pain. Occup Med 13, 1-22. Shelerud, R.A. (2006) Epidemiology of occupational low back pain. Clin Occup Environ Med 5, 501-28, v. Sporrong, H., Sandsjo, L., Kadefors, R. and Herberts, P. (1999) Assessment of workload and arm position during different work sequences: a study with portable devices on construction workers. Appl Ergon 30, 495-503. Teschke, K., Spierings, J., Marion, S.A., Demers, P.A., Davies, H.W. and Kennedy, S.M. (2004) Reducing attenuation in exposure-response relationships by exposure modeling and grouping: the relationship between wood dust exposure and lung function. Am J Ind Med 46, 663-7. Tielemans, E., Kupper, L.L., Kromhout, H., Heederik, D. and Houba, R. (1998) Individualbased and group-based occupational exposure assessment: some equations to evaluate different strategies. Ann Occup Hyg 42, 115-9. van der Beek, A.J. and Frings-Dresen, M.H. (1998) Assessment of mechanical exposure in ergonomic epidemiology. Occup Environ Med 55, 291-9. van der Beek, A.J., Kuiper, J.I., Dawson, M., Burdorf, A., Bongers, P.M. and Frings-Dresen, M.H. (1995) Sources of variance in exposure to nonneutral trunk postures in varying work situations. Scand J Work Environ Health 21, 215-22.  16  van Tulder, M., Koes, B. and Bombardier, C. (2002) Low back pain. Best Pract Res Clin Rheumatol 16, 761-75. van Tulder, M.W., Koes, B.W. and Bouter, L.M. (1995) A cost-of-illness study of back pain in The Netherlands. Pain 62, 233-40. Veiersted, K., Westgaard, R. and Andersen, P. (1993) Electromyographic evaluation of muecular work pattern as a predictor of trapezius myalgia. Scandanavian Journal of Work, Environment, and Health 19, 284-290. Village, J., Frazer, M., Cohen, M., Leyland, A., Park, I. and Yassi, A. (2005) Electromyography as a measure of peak and cumulative workload in intermediate care and its relationship to musculoskeletal injury: an exploratory ergonomic study. Appl Ergon 36, 609-18. Village, J., Trask, C., Luong, N., Chow, Y., Johnson, P., Koehoorn, M. and Teschke, K. (2007) Development and Evaluation of an Observational Back Exposure Sampling Tool (Back-EST) for Work-Related Back Injury Risk Factors. Applied Ergonomics Von Korff M. (1994) Studying the natural history of back pain. Spine. 1994 Sep 15;19(18 Suppl):2041S-6S. Waddell, G. (1996) Low back pain: a twentieth century health care enigma. Spine 21, 2820-5. Wasiak R, Pransky GS, Webster BS. (2003) Methodological challenges in studying recurrence of low back pain. J Occup Rehabil. Mar;13(1):21-31. Waters, T., Yeung, S., Genaidy, A., Callaghan, J., Barriera-Viruet, H. and Deddens, J. (2006) Cumulative Spinal loading exposure methods for maunal material handling tasks. Part 1: is cumulative spinal loading associated with lower back disorders? Theoret Issues Ergo Sci 7, 113-130. Wells, R., Norman, R., Neumann, P., Andrews, D., Frank, J., Shannon, H. and Kerr , M. (1997) Assessment of physical work load in epidemiologic studies: common measurement metrics for exposure assessment. Ergonomics 40, 51-61. Wiker, S. (2003) Statistical challenges facing development of epidemiologically validated low back injury risk predictive and descriptive models using biomechanical, anthropometric, or consensus-based ergonomic risk prevention guidelines. Seminars in Spine Surgery 15, 3-15. Winkel, J. and Mathiassen, S.E. (1994) Assessment of physical work load in epidemiologic studies: concepts, issues and operational considerations. Ergonomics 37, 979-88. WorkSafeBC. WorkSafeBC Annual Report: Statistical Supplement. 2005. 2006. Yingling, V.R., Callaghan, J.P. and McGill, S.M. (1997) Dynamic loading affects the mechanical properties and failure site of porcine spines. Clin Biomech (Bristol, Avon) 12, 301305.  17  18  Chapter 2: Context of the thesis: the UBC Back Study1 The University of British Columbia (UBC) Back Study was a large, field-based study of the risk factors for back injury in heavy industry. Funding was granted by the Workers’ Compensation Board of BC (now WorkSafeBC) to co-principal investigators Mieke Koehoorn and Kay Teschke in September 2003. Pilot testing of the study started in May of 2004, and field assessments were conducted from September 2004 to January 2006.  III  Figure 2.1: The stages of the UBC Back Study research program  The studies contained in this thesis were included as part of the first phase in a program of research designed to ultimately examine the etiology of back injuries in the five at-risk heavy industries and to test interventions to reduce these injuries (Figure 2.1). In Phase I, we evaluated ways to recruit workers at a variety of worksites, improve exposure assessment of back injury risk factors (the focus of this dissertation), and identified and modeled the work environment factors that contributed to increased and decreased exposure levels. Based on the findings from Phase I, Phase II will investigate the relative importance of many postulated risk factors and their interactions in the etiology and progression of acute and chronic back injuries in heavy industry. The data from Phase II will be used to design control measures. Phase III will be a randomized workplace trial of the effectiveness of the control measures. Phase I addressed a persistent problem in back injury epidemiology: the difficulty of assessing exposures for large scale epidemiological studies. The goal of the first portion of the study was to identify the suite of measurements that allow the most efficient and accurate collection of 1  Part of this chapter has been published. Trask, C., Koehoorn, M., Village, J., Johnson, P. and Teschke, K. (2006) Modeling determinants of working exposures and exposure variability. IEA2006, 16th World Conference on Ergonomics. Maastricht, the Netherlands  19  exposure data for epidemiological studies (see Figure 2.2) and to develop predictive exposure models, and to develop an “exposure assessment tool kit” to facilitate future research.  Figure 2.2: Flow chart outlining the methods development steps of the UBC Back Study  A literature review was conducted to identify risk factors with consistent evidence of a link to work-related back injuries (Figure 2.3): work postures, materials handling, and whole body vibration. These factors were further evaluated for feasible measurement methods in heavy industries. Different measurement methods for these risk factors were designed, tested, and piloted on a convenience sample of 7 university utility operations and 2 lab staff employees with multiple exposures, with iterative refinement after each sampling trial. A final “battery” of methods was adopted that were feasible for full-shift (8 hours) on-site use and flexible enough to measure risk factors in a wide variety of work environments (e.g. grain elevator, helicopter, log boom, etc.). The recruitment methods were designed to identify workers and workplaces that had high risk factors for back injury. WorkSafeBC identified a random sample of 50 employees in the Greater Vancouver area who had had an accepted workers’ compensation claim for back strain in 2001 and who agreed to have their information released to researchers. Due to current legislative requirements of the Freedom of Information and Protection of Privacy Act, representatives of WorkSafeBC (as the data steward) needed to contact each of the workers to obtain their permission to release their contact information to the study team. Once this consent was obtained and the workers’ contact information was forwarded to researchers, workers were contacted first by letter, then by telephone and invited to participate. The employers of participating workers were contacted to gain permission to conduct measurements at their worksite.  20  Once at the worksite, up to 5 co-workers were also invited to participate. This two-stage recruitment process met several objectives. Firstly, it was meant to capture a wide variety of tasks, jobs, worksites, work conditions, and industries. Secondly, it expanded the sample size beyond the initially-contacted workers with claims. It also tested the feasibility of this recruitment method for potential use in a case-control study at some point in the future. See Appendix I for a summary of the characteristics of the participants. Wherever feasible, one follow-up measurement was made of each worker for a total of two full-shift measurements per individual. Recruiting workers directly entailed a considerable amount of contacting workers and employers, developing trust, explaining and answering questions about the study, scheduling and attending meetings with management, and scheduling measurement visits. Despite the challenges, recruiting workers directly and then visiting their worksites allowed for a tremendous variety of tasks, job titles, departments, and worksites, from large multi-national business to selfemployed workers. More about the recruitment methods can be found in Appendix A.  Figure 2.3: Conceptual model of the determinants of physical back injury risk factor exposure  Five exposure assessment methods for workplace physical demands were employed for each shift: observation by trained personnel, self report by workers, and direct measures of trunk inclinometry measured with an inclinometer, whole body vibration (WBV) measured with a seat pad accelerometer, and electromyography measured with a portable EMG monitor of the low back muscles. With the exception of self-report conducted at the end of the shift and WBV measured only while driving, all methods were conducted for the entire work shift, excluding lunch and non-working breaks. The exposure assessment methods used in this study are  21  described in more detail in Appendix B. The observation methods are also described in detail elsewhere in Village et al. (2007); a copy of the observation tool is included in Appendix C. A copy of the self-report instrument is included in Appendices D and E.  22  References Village, J., Trask, C., Luong, N., Chow, Y., Johnson, P., Koehoorn, M. and Teschke, K. (2007) Development and Evaluation of an Observational Back Exposure Sampling Tool (Back-EST) for Work-Related Back Injury Risk Factors. Submitted in June 2007 to: Applied Ergonomics; Submission number JERG-D-07-00096.  23  Chapter 3: Evaluating EMG as an exposure metric: implications for calibration and shift-long measurement Introduction Typically ergonomic studies use intensive exposure assessments of short duration (McAtamney and Nigel-Corlett, 1993; Buchholz B et al. 1996; Moore and Garg, 1995; Hignett and McAtamney, 2000; Waters et al. 1993; ACGIH, 2005; Snook and Ciriello, 1991), but short duration measurements may not be representative of long-term or cumulative exposures experienced by workers in their work environments . Research budgets seldom allow for the time and cost associated with long-term, in-depth exposure assessments, and even when they do, some working conditions do not facilitate the capture of body segments’ orientation because of privacy concerns (like patient care) or because workers move in confined spaces. Portable electromyography (EMG) instrumentation can make practical, high-resolution assessments of physical exposures over the full work shift, while maintaining privacy and not constraining worker movement. EMG measures muscle activity as electrical potential in microvolts at the site of electrode placement (in this case the lower back). The nature of the EMG electrode-skin interface is unique to each measurement session. This means that the raw muscle activity cannot be directly compared between individuals, between measurement days within an individual, or to ergonomic guidelines. As a result, a unique calibration equation is developed for each individual session and applied only to the data collected during that session. Calibration can be accomplished in several different ways, but all involve making measurements of standardized movements prior to a field measurement to act as a reference. The maximum voluntary contraction method involves having subjects perform a maximum static contraction with the muscle in question, then dividing all values by a maximum effort to produce a percentage of maximum voluntary contraction (%MVC) (Jones and Kumar, 2007; Village et al. 2005). Similarly, subjects may perform a standardized, non-maximal static effort, often maintaining a posture against gravity, and dividing all measured values by this value produces a percentage of the reference contraction (%RC) (Burden and Bartlett, 1999; Sporrong et al. 1999). The %RC is sub-maximal, so it is preferred for subjects with pain or history of injury in the measured muscle groups or related joints (Marras et al. 2001). EMG can feasibly be measured for a whole shift, thereby delivering a more comprehensive estimate of exposure. However, fatigue or drift in the electrode-skin interface between the start and the end of the day may change exposure estimates over the course of the day, suggesting that when long measurements are conducted, calibration should be performed pre-and post shift. The repeatability of measurements from reference positions is also of interest, since low betweentrial correlation of the reference task would lead to fluctuating exposure estimates. Although adding repeats of calibration measurements for a position and measuring pre- and post-shift could enhance the accuracy of the calibration, the additional calibration time takes time away from exposure assessment. It would be helpful to know if there are any differences between preshift and post-shift measurements, between trials of the reference posture, and how this variability compares to the measurements that one expects to collect in order to make informed decisions about the type and frequency of calibration measurements.  24  This study compares calibration reference postures collected from workers in the University of British Columbia (UBC) Back Study. Analyses investigated sources of variability in EMG calibration measurements, such as pre-shift versus post-shift collection, repeated trials of a reference contraction, left versus right side of the back, and different reference tasks/postures. The purpose of this study was to identify the sources of variability in EMG calibration measurements and determine an efficient combination of calibration measurements that will provide adequate information in minimum time.  Methods Participant recruitment and sampling strategy As part of a larger study, WorkSafeBC (the Workers’ Compensation Board of British Columbia) identified a random sample of 50 employees in the five heavy industries of transportation, warehousing, forestry, wood and paper products, and construction. These workers had accepted worker’s compensation claims for back injury in 2001, resided in the Greater Vancouver area, and agreed to have their contact information released to researchers. Workers were contacted first by letter, then by telephone and invited to participate. The employers of participating workers were contacted to gain permission to conduct measurements at the worksite, and to recruit additional co-worker volunteers from the worksite. Participation was completely voluntary and human subject procedures were approved by the University of British Columbia’s Behavioral Research Ethics Board (#B03-0644). EMG data collection Disposable 12mm Ag-AgCl electrodes (Blue Sensor N-00-S, Ambu, Denmark) with a 20 mm inter-electrode spacing were placed over the erector spinae at approximately the level of L4 with a ground electrode and preamplifier on the posterior aspect of the iliac crest. EMG data were collected in real time using a portable EMG measurement unit (Mega M3000P4 Mega Electronics, Finland) and downloaded to a computer. Signals were collected at 1000 Hz and filtered internally using an 8-500 Hz band pass filter. EMG data were collected at industrial worksites for a full work shift (excluding breaks). A schematic of EMG data collection and processing is found in Appendix F. Calibration was performed during work time but before work tasks began. When time allowed, calibration was also performed at the end of the shift. The calibration protocol involved testing o four task conditions on each participant: standing upright, standing with the trunk flexed at 45 , o o and standing at 45 with an 11.5 kg load and at 60 with an 11.5 kg load. There were two repetitions for each condition, electrode placement on both the right and left side of the lower back, and the protocol was repeated before and after the work shift (time permitting), totaling a maximum of 32 potential separate maneuvers per subject (Figure 3.1).  25  L Position 1 standing  Pre  Trial 1  R L  Post Trial 1  R L  Trial 1 Pre Trial 2  R L R  Position 2 45o  L Trial 1 Post Trial 2  R L R  L Trial 1  Pre Trial 2  R L R  Position 3 45o /weight  L Trial 1 Post Trial 2  R L R L  Trial 1 Pre Trial 2  R L R  Position 4 60o /weight  L Trial 1 Post Trial 2  R L R  Figure 3.1 Multipoint Calibration Data Collection Schematic  26  Trunk angle was measured using a 12-inch hand-held goniometer (Baseline Instruments Inc). Participants were instructed to keep the head up, the shoulders back, and the pelvis tilted in lordosis. Each task condition was held for 5 seconds and repeated twice with at least a 30second rest between each repetition. The average voltage of the central 3 seconds of each repetition was used to estimate mean muscle activity for that trial. The data collection scheme is outlined in Figure 3.1. Statistical analysis Paired t-tests were used to calculate mean differences between the following pairings of calibration maneuvers and whether the differences were significantly different than zero: preand post-shift; trials 1 and 2; and the left and right side muscles. Pairs of measurements were matched for subject, measurement session, posture and weight, etc. Box plots were also produced to visualize the distribution of the measurements in each group. Linear regression models were used to identify any relationships between muscle activity, in µV, and the grouping variables: pre- versus post-shift, left versus right side, trial 1 versus trial 2, and task conditions 1 through 4. As alternatives to categorizing trunk loads by task condition category, ‘weight in hand’ (kg) and ‘trunk angle’ (degrees) were tested as continuous variables. Simple linear regression was used to find the strength of relationship between EMG measures and each of the independent variables. Multiple linear regression mixed models were developed using ‘individual’ and ‘electrode session’ as random effects variables. An ‘electrode session’ represents the unique electrode/skin interface for each shift measurement of individual. Task condition, pre/post shift, right/left side, and trial number were fixed effects. Two mixed models were developed, one that included task condition as a categorical variable, and another that included both weight and trunk angle as continuous variables.  Results EMG calibration measures were collected for 103 individuals, with repeated measurement days on 40 (38%) individuals for a total of 143 measurement days. The total number of muscle activity calibration measurements including all subjects, measurement days, pre- and post-shift measurements, task conditions, first and second trials, and left and right sides was 3141. A demographic and anthropometric summary of participants is found in Table 3.1. Table 3.1: Mean age, height, weight, and sex of study participants (n = 103 workers) Variable Mean age in years (sd) Mean height in cm (sd) Mean weight in kg (sd) % Male  Value 42.2 (12) 178.1 (7.9) 85.2 (16.1) 95.3%  The mean muscle activity over all calibration measurements was 31.8 µV (standard deviation = 22.0 µV). The distributions of the measurements in each category can be seen in Figure 3.2.  27  160  160 3537 1529 1530  140  3538 3913 1531 2345 3097 2449 1532 2009 2010 1697 2450 2689 1698 2451 1441 1442 2347 1833 1851 2201 2801 1834 2011 2802 1473 2349 2351 1852 2012 3914 1193 529 2690 2452 2577 3098 963 2691 530 964  120 100  60  60  40  40  20 0 -20 943  pre  post  5634 4228 5636 4230 4232 4233 5027 5028 1562 1882 1883 5637 1563 1884 2401 5031 5032 1564 5123 5640 1565 5033 2820 5379 1885 5039  20 0 -20 N=  PREPST a) pre/post shift  1534  1607  left  right  left/right side b) left or right side  160  160 4545  140  140  4225 4226  4226  4546  120  4547  100  4549  80  5038 1885 5037 5041 1566 1888 1889 5044  60  5026 5025 4228 5635 5636 4230 1148 4231 5027 5028 1882 5637 1884 5029 5030 5031 5032 5638 5121 1151 1564 4234  4546 4547 4227 4548 5025 5026 4228 1146 1147 5635 5636 2817 1148 4231 4232 5027 5028 1883 5637 1563 1884 5638 5121 5122 5123 1565 5641 5034 2819  120  5633 4227 4548 5634 1146 1147 4229 2817 2818 4232 4233 1562 1883 1149 1563 2401 1150 5122 4550 5639 5123 5640 5124  100 80  4545 4225  5633 5634 4229 4230 2818 4233 1562 1882 1149 2401 5029 5030 5031 5032 1150 4549 1151 1564 4234 4550 1152 5639 5640 5033 5377 5124 5642 2820 5035 5036 5379 1885 4552 5037 5038 5039 1153 3715 4705  60  2151 771 2403  40  40  4729 775 2117 1170 2118 1172 6757 4089 6149 6150 1356 4897 6151 4730 4090 4091 4997 4998 6152 7053 3725 5389 4731 1184 1234 4999 5000 4092 5001  20  MUSCLEAC  MUSCLEAC  4225 4226  80  MUSCLEAC  MUSCLEAC  100  2433 2435 2438 3482 3825 1705 2013 3081 2439 881 3065 3521  2198  4546 4547 5633 4227 4548 5025 5026 1146 1147 4229 5635 2817 1148 2818 4231 1149 5029 5030 1150 4549 5638 5121 1151 5122 4234 4550 1152 5639 5641 5034 5377 5124 5642  120  961 2346 962 1849 1850 2348 2350 2352  80  N=  4545  140  0 -20 N=  457  919  standing upright  45  reference position c) position  888  877  45 w ith w eight 60 w ith w eight  20 0 -20 N=  1808  1333  trial one  trial tw o  d)TRIAL trial number  Figure 3.2: Box plots for muscle activity separated by a) pre/post b) left/right c) task condition and d) trial number  The results of the paired-tests are available in Table 3.2. With the large sample size there were significant differences between all variables tested; trial 2 was higher than trial 1, pre-shift values were higher than post-shift values, and the left side was higher than the right side. Although significant, the magnitude was not large, with 3.25 µV between pre and post-shift trials being the largest difference.  28  Table 3.2: T-test results for EMG activities by pre- and post-shift, trial 1 and 2, left and right side. Comparison Pre-shift trials vs post-shift trials Trial 1 vs trial 2 Left side vs right side  # pairs used in analysis 916  Mean Difference ( µV) 3.25  p-value  1333 1512  1.00 1.59  <.0001 <.0001  <.0001  Table 3.3 shows the results of the bivariable models. All variables were significantly related to muscle activity, but task condition 3 (45o with an 11.5 kg weight) was not significant in the model when compared to reference task condition 4 (60o with an 11.5 kg weight). Mixed modeling results for weight and trunk angle or task condition are found in Tables 3.4 and 3.5, respectively. The fixed effects of weight and posture explains 21.3% of the variability in measured muscle activity, while the model that includes task condition (weight and posture combined) explains 23%. The model with only the random effects terms of ‘subject’ and ‘measurement session’ explained 44.6% of the variance in measured EMG.  Table 3.3: Summary of simple linear regressions of the relationships between each of the following independent variables and muscle activity in µV Variable Post-shift (reference) Pre-shift  Intercept 29.8  Β  p 0.0005  R2 0.0038  <.0001  0.020  0.0431  0.0013  2.97  Trial 2 (reference) Trial 1  35.5  Right (reference) Left  31.1  Weight (continuous)  23.1  1.37  <.0001  0.13  Posture (continuous)  9.86  0.517  <.0001  0.19  Task condition 1  -30.0  <.0001  0.23  Task condition 2  -7.39  <.0001  -  Task condition 3  1.70 -  0.0648  -  -  -  Task condition 4 (reference)  -6.26 1.59  37.9  29  Table 3.4: Mixed-effect multiple regression model examining the adjusted effect of calibration variables, load weight and trunk posture on muscle activity Intercept  Variable  Β 7.81  p <.0001  Pre-shift (post shift is the reference)  3.09  <.0001  Trial 1 (Trial 2 is the reference) Left (right is the reference) Weight  0.132 1.53  0.7898 0.0011  0.566  <.0001  Posture  0.416  <.0001  Table 3.5: Mixed-effect multiple regression model examining the adjusted effect of calibration variables and task condition (combined load weight and trunk posture) on muscle activity Variable Intercept  Β 34.9  p <.0001  Pre-shift (post shift is the reference)  3.06  <.0001  Trial 1 (Trial 2 is the reference) Left (right is the reference) Task condition 1  1.00 1.53  0.0375 0.0007  -30.4  <.0001  Task condition 2  -7.48  <.0001  Task condition 3  1.68 -  0.0047 -  Task condition 4 (reference)  Discussion Evaluation of the calibration method Although maximal voluntary contractions (MVC) and reference contractions (RC) have been used extensively to calibrate EMG measurements, this is the first study known to the author that examines the stability of static reference contractions over the course of a shift, from trial to trial, between task conditions, and between the left and right side of the body. As a result, this study is useful not only to EMG assessment methods that use MVC and %RC as outcome measures, but also to studies that translate EMG into measurements of spinal compression or ‘compressionnormalized EMG’ (CNEMG), since these studies use static reference contractions to create linear calibration equations between muscle activity and spinal compression (Mientjes et al. 1999; Village et al. 2005; Potvin et al. 1990 ; Potvin et al. 1996). The sources of variability examined in this study are far from exhaustive, but they can nonetheless provide some guidance for future studies of shift-long EMG exposure. For example, the largest proportion of explained variance is explained by the random effects of ‘subject’ and measurement session, highlighting how ‘normalizing’ EMG using a reference contraction or other calibration procedure is necessary. The fact that there is a difference between trials suggests that measuring repeated trials is a good idea. However, the differences  30  between trials represented a small amount of variability between measurements, so could be considered the least important factor, and could be the first to be removed from the protocol if time were an issue. The fact that task condition accounts for the bulk of the variation in EMG calibration measures suggests that in cases where EMG is being translated into spinal compression via a linear regression, calibration time and effort would be better spent on measuring several task conditions rather than several repeats of the same task condition, as in previous studies (Mientjes et al. 1999; (Village et al. 2005). Only slightly more important than repeated trials was the difference between right and left side. In studies where both right and left sides of the back muscle are studied, each EMG channel should be calibrated using measurements from the corresponding side. Generally, both channels would be averaged to give an overall exposure during the shift (Trask et al. 2007a). However, it is not uncommon for one channel of EMG to be lost due to failed adhesion of electrodes, excessive motion artifacts from uncoiled cables, or excessive contact pressure on one side of the body (Trask et al. 2007b). In this case, only the channel without those problems is used. Differences between the left and right side averaged 1.5 µV (3-4% of the average EMG measurement), so the difference imparted by using either the right or left channel in lieu of the average appears to be small, at least for the static calibration contractions examined in this study. At 3.5 µV or 11% of the average recorded EMG, the difference between pre- and postmeasurement is larger than between-trial or left-right differences. This suggests that the reference contraction value used as the denominator for all shift-long measurements expressed as a %RC should include the average of pre-shift and post-shift measurements. The assumption involved with averaging the pre-shift and post-shift values is that the difference between preshift and post-shift is due to a monotonically increasing function, or that there is a step function occurring at the midpoint of the day. However, it is unclear if the difference between pre- and post-shift calibration measures is representative of a similar shift in working exposures, or at what point during the day the shift is occurring. Nonetheless, this assumption seems more appropriate and matches the data presented far better than assuming that there is no difference pre- and post-shift and using only pre-shift values for calibration. Summary of results With the exception of task condition, the differences in the means and inter-quartile ranges between levels of the factors examined were not very large, especially considering the range of measurement values. However, due to the large sample size, significant differences were found using paired t-tests for all of these variables. Although statistically significant, the question of how important these differences are to calibration remains. The mean difference between trials and between left and right side was less than 2 µV, while the mean difference between pre and post was less than 3.5 µV. Given that the mean measurement from calibration task conditions is 31.8 µV, these differences range from 3-11% of the mean. The variation between left versus right or pre versus post can also be compared to the variation in repeated trials on the same side at the same time of the shift; this difference is about 1.0 µV, or about 3% of the mean. The variation due to right versus left side is comparable to the variation between repeated trials on the same side, whereas the difference in pre-shift and post shift is about 3 times the trial-to-trial difference.  31  Sources of variability The explained variability in measured EMG came mostly from the task condition or the combination of trunk posture and weight lifted, as seen in Tables 3.3 to 3.5. This is consistent with the goal of using EMG as an exposure assessment method; the goal is to quantify the aspects of exposure that we know to be risk factors for back injury: posture and manual materials handling. However, the fact that only 23% of the variance in EMG calibration measurements indicates that the bulk of the variance is coming from other sources. Muscle activity, as measured by EMG, is a multi-factorial phenomenon, incorporating not only the postural and loading demands of muscle force, but also the length-tension relationship of the muscle, the velocity-tension relationship of the muscle, and fatigue (NIOSH, 1992). However, for static calibration efforts, the length is constant and the velocity is zero, so one would not expect these components to contribute to variability in the signal. EMG picks up the initial electrical signal at the neuromuscular junction, so there is a lag between the detection of a signal and the development of muscle tension. This electromechanical delay is reported to be between 30 and 100 ms (Cavanagh and Komi, 1979). Still, the mean muscle activity was only selected from the central 3 seconds of a 5-second effort, so the initiation lag would not be expected to be a factor either. It would appear that a larger portion of EMG measurement variance comes from subject and measurement session sources, which are accounted for in shift-long measurements by dividing all within-session values by a reference measurement value. This variances is accounted for here in the current study by entering subject and measurement session in the model as random effect terms. For a sagittally symmetrical task, one would not expect differential activation on the left and right side. The fact that the left side is marginally but significantly higher may relate to lateral dominance in recruitment of these muscles, or perhaps the postures assumed by the workers are imperceptibly asymmetrical. This postural difference may also relate to left or right side dominance in task performance. The differences in pre-post measurements could arise from several sources. Although changes can occur in the electrode-skin interface over the course of the day, this tends to be a slow drift that would be removed by the 4-500 Hz band pass filter applied to the data. Heavy lifting over the course of the day may produce muscle fatigue, although generally fatigue-related differences are seen more in EMG frequency rather than in voltage level (NIOSH, 1992). In addition, fatigue-related effects due to slowed re-uptake of Calcium ions into the sarcoplasmic reticulum (Roy et al. 1998) are likely to have dissipated in the 10-15 minutes between the worker completing work, removing any safety equipment, walking to the calibration testing area, and setting up the testing equipment. It may be then that there are changes in muscle recruitment or biomechanical strategies for task performance that change the profile of EMG measurements from the beginning to the end of the shift. Exposure may also vary within repeats of the same task. Substantial variation between repeated trials of the same task has been observed even with very comprehensive exposure assessment methods, including an EMG-assisted biomechanical model of spinal loading showing trial-to trial differences accounting for 14% of spinal compression and 32% of lateral shear (Granata et al. 1999). Bonato et al (2002) found that when lifting 15% body mass at 12 times per minute for 5 minutes, instantaneous median frequency went down in most subjects (indicating muscle fatigue), and this was accompanied by biomechanical adaptations, including greater accelerations that increased estimates of spinal compression. Although these tests were performed on dynamic tasks rather than static tasks (as for the calibration), it shows that there is likely to be variation within tasks over a shift, and the 32  physiological and mechanical factors mediating the changes in dynamic exposure profiles may well apply to repeated static exposure during calibration measurement. Strengths and limitations This study measured a sample of workers in heavy industry in occupational settings. These settings undoubtedly offered conditions that were less consistent and controlled than a laboratory setting, providing instead the setting required for occupational epidemiological study. In addition, the workers were older (mean age = 42.2 years) than the subjects that participate in many lab-based studies of lifting tasks (McGill, 1992; Mirka et al. 1997; Marras et al. 2001; Kee and Chung, 1996), and were less likely to be as fit and injury free as university students. In fact, the recruitment method targeted workers who had back injury claims in the past and their co-workers. The lack of controlled measurement conditions and the varied personal characteristics of the subject sample are likely to have added to the variability in measured EMG. However, consideration of the limitations involved with field conditions and working subjects must be part of any field study. The purpose of this study was to measure the effect of different calibration sampling protocols on the EMG calibration values. The circumstances of the current study represent the variability seen under field conditions and this enhances the generalizability of the results. Frequency analysis of pre and post shift measurements, as well as trial 1 and trial 2 measurements would help understand what type of fatigue effects are occurring throughout the shift or from repeated trials of the same task condition. In addition, future studies could collect information on the right-left dominance of workers to determine if this factor is contributing to the left-right differences. Conclusion Task condition accounts for the bulk of explained EMG variability, with trial 1 versus trial 2, left versus right, and pre- versus post-shift measurement accounting for only 0.1-3.3% of EMG variability. It appears that there is little to be gained by measuring multiple trials or pre- and post-shift; measurement time might be better spent collecting more working exposures. There is little variability between left and right side during symmetrical calibration tasks, but measuring both sides takes very little additional effort and could prove very useful if the signal on one side is lost. However, since all fixed effect variables together account for only 23% of variability in measured EMG, there remains unknown individual-level and other factors that contribute substantially to the variation in EMG.  33  References ACGIH (American Conference of Governmental Industrial Hygienists) (2001) Hand activity level. TLVs and BEIs - Threshold Limit Values for Chemical Substances and Physical Agents pp. 110-112. Cincinnati, OH: ACGIH . ACGIH (American Conference of Governmental Industrial Hygienists) (2005) TLVs and BEIs: Threshold Limit Values for Chemical Substances and Physical Agents and Biological Exposure Indices. Cincinnati, OH.: ACGIH . Bonato, P., Boissy, P., Della Croce, U. and Roy, S.H. (2002) Changes in the surface EMG signal and the biomechanics of motion during a repetitive lifting task. IEEE Trans Neural Syst Rehabil Eng 10, 38-47. Buchholz B, Paquet V, Punnett L , Lee D and Moir S. (1996) PATH: a work sampling-based approach to ergonomic job analysis for construction and other non-repetitive work. Appl Ergon 27, 177-87. Burden, A. and Bartlett, R. (1999) Normalisation of EMG amplitude: an evaluation and comparison of old and new methods. Med Eng Phys 21, 247-57. Cavanagh, P.R. and Komi, P.V. (1979) Electromechanical delay in human skeletal muscle under concentric and eccentric contractions. Eur J Appl Physiol Occup Physiol 42, 159-63. Granata, K.P., Marras, W.S. and Davis, K.G. (1999) Variation in spinal load and trunk dynamics during repeated lifting exertions. Clin Biomech (Bristol, Avon) 14, 367-75. Hignett, S. and McAtamney, L. (2000) Rapid entire body assessment (REBA). Appl Ergon 31, 201-5. Jones, T. and Kumar, S. (2007) Assessment of physical demands and comparison of multiple exposure definitions in a repetitive sawmill job: board edger operator. Ergonomics 50, 676-93. Kee, D. and Chung, M.K. (1996) Comparison of prediction models for the compression force on the lumbosacral disc. Ergonomics 39, 1419-29. Marras, W.S., Jorgensen, M.J., Granata, K.P. and Wiand, B. (2001) Female and male trunk geometry: size and prediction of the spine loading trunk muscles derived from MRI. Clin Biomech (Bristol, Avon) 16, 38-46. Marras, W. S. and Davis, K. G., A non-MVC EMG normalization technique for the trunk musculature: Part 1. Method development J Electromyogr Kinesiol, vol. 11, pp. 1-9, Feb, 2001. McAtamney L and Nigel-Corlett E. (1993) RULA: a survey method for the investigation of work-related upper limb disorders. Appl Ergon 24, 91-9. McGill, S.M. (1992) A myoelectrically based dynamic three-dimensional model to predict loads on lumbar spine tissues during lateral bending. J Biomech 25, 395-414.  34  Mientjes, M.I., Norman, R.W., Wells, R.P. and McGill, S.M. (1999) Assessment of an EMGbased method for continuous estimates of low back compression during asymmetrical occupational tasks. Ergonomics 42, 868-79. Mirka, G., Kelaher, D., Baker, A., Harrison, A. and Davis, J. (1997) Selective activation of the external oblique musculature during axial torque production. Clin Biomech (Bristol, Avon) 12, 172-180. Moore, J.S. and Garg, A. (1995) The Strain Index: a proposed method to analyze jobs for risk of distal upper extremity disorders. Am Ind Hyg Assoc J 56, 443-58. NIOSH (1992) Selected Topics in Surface Electromyography for Use in the Occupational Setting: Expert Perspective. Publication No. 91-100, DHHS (NIOSH) . Nordander, C., Balogh, I., Mathiassen, S.E., Ohlsson, K., Unge, J., Skerfving, S. and Hansson, G.A. (2004) Precision of measurements of physical workload during standardised manual handling. Part I: surface electromyography of m. trapezius, m. infraspinatus and the forearm extensors. J Electromyogr Kinesiol 14, 443-54. Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clinical Biomechanics 13 (8):561-573. Potvin, J., Norman, R. and Wells, R. A field method for continuous estimation of dynamic compressive forces on the L4/L5 disc during the performance of repetitive industrial tasks. Proceedings of 23rd Annual Conference of the Human Factors Association of Canada. September 26-28. Ottawa, Ontario Potvin, J.R., Norman, R.W. and McGill, S.M. (1996) Mechanically corrected EMG for the continuous estimation of erector spinae muscle loading during repetitive lifting. Eur J Appl Physiol Occup Physiol 74, 119-32. Roy, S.H., Bonato, P. and Knaflitz, M. (1998) EMG assessment of back muscle function during cyclical lifting. J Electromyogr Kinesiol 8, 233-45. Snook, S.H. and Ciriello, V.M. (1991) The design of manual handling tasks: revised tables of maximum acceptable weights and forces. Ergonomics 34, 1197-213. Solomonow, M., Baratta, R.V., Banks, A., Freudenberger, C. and Zhou, B.H. (2003) Flexionrelaxation response to static lumbar flexion in males and females. Clin Biomech (Bristol, Avon) 18, 273-9. Sporrong, H., Sandsjo, L., Kadefors, R. and Herberts, P. (1999) Assessment of workload and arm position during different work sequences: a study with portable devices on construction workers. Appl Ergon 30, 495-503. Trask, C.M., Teschke, K., Village, J. and Koehoorn, M. (2007a) Determinants of exposure for mean, 90th percentile, and cumulative muscle activity in heavy industry. Submitted to: Appl Ergon  35  Trask C, Teschke K, Village J, Chow Y, Johnson P, Luong N and Koehoorn M. (2007b) Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility. Am J Inl Med 50, 687-696. Village, J., Frazer, M., Cohen, M., Leyland, A., Park, I. and Yassi, A. (2005) Electromyography as a measure of peak and cumulative workload in intermediate care and its relationship to musculoskeletal injury: an exploratory ergonomic study. Appl Ergon 36, 609-18. Waters, T.R., Putz-Anderson, V., Garg, A. and Fine, L.J. (1993) Revised NIOSH equation for the design and evaluation of manual lifting tasks. Ergonomics 36, 749-76.  36  Chapter 4: Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility 2 Introduction Measurement of physical risk factors for low back injuries presents many challenges, particularly in field research. Nonetheless, detailed workplace exposure measurements of physical demands are valuable and necessary for studies examining the determinants of exposure and exposure-response relationships. Exposure assessment for the most commonly cited back injury risk factors (manual materials handling, frequent bending and twisting, and whole body vibration) is not simple. Many exposures might be present in a single job or task, and this might require multiple exposure assessment approaches. Several authors have discussed the array of techniques for measuring exposures in musculoskeletal studies (Burdorf and van der Beek, 1999, Genaidy et al. 1994, Spielholz et al. 2001, van der Beek & Frings-Dresen 1998, Li & Buckle 1999, Danneels, Coorevits, et al. 2002, and Wells, Norman, et al. 1997). They can be broadly classified into three categories: self-reports by workers; observations by trained experts; and measurements using monitoring equipment. The choice of exposure assessment methods will largely be based on the exposure of interest as well as the resolution, validity, and reliability of the method. Laboratory-based investigations of the validity and reliability of data collection methods is often part of the research process. However, information on the practical implications or feasibility of transferring measurement techniques developed in the controlled milieu of a laboratory to diverse and complex work environments for real-world data collection is rarely available. If potential difficulties are not known by researchers embarking on studies, they may underestimate the time, personnel and monetary resources required to overcome them. There are reports in the literature about overcoming challenges in recruitment (Kidd et al. 2004), participation (LaMontagne and Needleman, 1996), sampling strategy (Burdorf and van der Beek, 1999), and study design (Cole et al. 2003), but no thorough descriptions of challenges related to the use of ergonomic equipment or ergonomic measurements in occupational field studies appear to have been published to date. The purpose of this paper is to describe successes in and obstacles to assessment of exposures to three back injury risk factors (manual materials handling, posture, and whole body vibration) using five measurement methods. We describe the potential effects of conditions encountered in worksites in heavy industry on data collection and suggest strategies to mitigate these effects.  2  A version of this chapter has been published. Trask, C., Teschke, K., Village, J., Chow, Y., Johnson, P., Luong, N., and Koehoorn, M. (2007). Evaluating methods to measure low back injury risk factors in challenging work environments. American Journal of Industrial Medicine 50(9):687-96.  37  Methods The UBC Back Study This analysis was performed as part of a study of risk factors for back injury in heavy industry in the Greater Vancouver region of Canada: the UBC Back Study. The main goal of the study was to develop a set of exposure assessment tools to quantify risk factors for back injury, and to compare and evaluate these tools with respect to their suitability for future epidemiological studies. Exposure assessment tools were selected to be general enough to be applicable to a wide range of industrial tasks, from sedentary work like truck driving to very active work like construction labour, and to allow comparisons between worker-days, occupations, and industries for each of the three risk factors: posture, manual materials handling, and vibration. The study provided a unique opportunity to study the costs, successes and challenges of different ergonomic exposure assessment methods in a variety of worksites. Exposure assessment methods Study sample This study measured 125 individual workers, each monitored on one or two days, for a total of 223 worker-days. Data were collected from 50 different worksites within the heavy industrial sectors of construction, forestry, transportation, warehousing, and wood and paper products. Setup, measurements, and interviews were conducted during regular work time. Five measurement methods were used: posture inclinometry; EMG of lumbar muscles, whole body vibration monitoring, worker observations, and worker self-reports. Vibration monitoring was attempted only on workdays when the worker was in a vehicle for at least 5 minutes (n=128), whereas the other four measures were attempted on all 223 worker-days. Measurement techniques Continuous posture measurements were made in two dimensions relative to gravity using a Virtual Corset (VC-323, Microstrain, Inc., Willeston, VT, USA). The Virtual Corset consists of two low-pass filtered accelerometers (inclinometers) packaged in a battery powered, pager-sized logger that was harnessed tightly to the trunk at the level of the T6 spinous process. Posture in two axes (flexion-extension, and lateral bending) was measured throughout the shift, data-logged at 7 Hz, and subsequently downloaded to a laptop computer. Muscle activity was measured via electromyography (EMG), using a portable data collection system (ME3000 P4 and ME3000 P8, Mega Electronics, Kupio, Finland) that gathered data from electrodes placed bilaterally over the erector spinae at the level of L4/L5. The EMG equipment was mounted on the worker, then calibrated relative to a 45o trunk flexion reference posture holding an 11.5 kg weight at the start and end of each shift. Data were stored at 10 Hz, and downloaded from the EMG monitor onto laptop computers during work breaks. Whole body vibration (WBV) intensity, direction and duration were measured according to ISO 2631-1 (1997) guidelines using a Larson Davis triaxial seatpan accelerometer and Larson Davis HVM100 Monitor (Larson Davis Laboratories, Provo, UT, USA) during periods of vehicle use.  38  These data were averaged over each 1-second interval, stored and downloaded to a laptop computer during work breaks and at the end of the work shift. Observations of each subject were made by trained observers once every minute throughout the work shift excluding breaks, starting just after mounting the monitors on the worker before the shift began and ending just prior to removing the monitors at the end of the shift. Data were recorded on paper using a coding system to indicate task or activity, item or power tool in hands, items worn (such as a tool belt), general body posture (e.g., standing, walking, kneeling), trunk angle, presence of trunk support, lateral bend or twist, and manual materials handling (horizontal distance, weight and force estimate). Where possible, photos were taken of the typical materials handled. When workers were in vehicles, additional observations included vehicle type, terrain and slope, speed, driving style, vehicle load, operating duration, gross vehicle weight, wheel characteristics, type of transmission, seat type and suspension type, back support, armrests, and location of cab in relation to the load. Where possible, photos were taken of the vehicle and where applicable, the seat area and tires. A post-shift interview was conducted with each worker to obtain their self-report of physical exposures during the day’s work shift. Workers were asked to give estimates of the duration of their shift associated with their main tasks, various body postures, trunk angles, manual materials handling, and vehicle use using representative drawings of these exposures. Measurement Success For each of the 5 exposure assessment methods, measurement success rates were calculated as the number of worker days with useable exposure data divided by the number of attempted exposure measurement days. An ‘attempted day’ was defined as a work shift where the exposure measurement method was appropriate for that subject’s job. This definition provided a different denominator for whole body vibration measurements, since not all worker-days involved vehicle use. Cost estimates We calculated the estimated costs of each method, including the capital, maintenance, comestibles, and personnel expenses associated with data collection and data entry. Cost estimates did not include expenses associated with training personnel, development of operating procedures, travel to research sites, or data cleaning or analysis. Costs also did not account for the remaining value of the sampling equipment at the conclusion of the study. Personnel costs were calculated for the total measuring time spent on site for each method individually (i.e., without taking into account the economies of scale available to our study because we used multiple measurement methods simultaneously). Personnel costs were based on a research assistant wage ($20 Canadian/hour) and did not include holiday time or other benefits.  39  Consent and approval The study protocol was approved by the Behavioural Research Ethics Board at the University of British Columbia (Approval #B03-0644). Each worker signed an informed consent form prior to participation in the study.  Results Measurement success The observation and self-report methods provided the most comprehensive dataset, with 222 (99.6%) observation forms and 218 (97.8%) self-reports completed over 223 worker-days. All three types of monitoring equipment methods had lower success rates. Of the three monitoring methods, the virtual corset inclinometer was the most successful, with complete posture data for 199 worker-days (89.2%). EMG was measured for 139 worker days (62.3%); of these, 20 included data for only the left side of the erector spinae muscles and 22 for only the right side. Of the 128 days when subjects spent at least 5 minutes in a vehicle, whole body vibration was measured successfully on 54 (42.2%). In addition to missed measurement days, some measurements were not complete for the full exposure period, thus providing incomplete, but usable, data. For the inclinometer, 29 days had measurements lasting less than 50% of the shift, as did 41 days for EMG; these data were lost due to monitoring equipment or data logging problems. For vibration, there were 6 shifts with data for less than half the time the subject was on the vehicle, mainly due to the difficulty in turning the vibration monitor on in sync with subjects who were on and off vehicles. One selfreport of exposures resulted in fewer than 50% completed questions. Observations were always made for at least half the shift. Subjects never refused to be monitored or observed, but one subject refused to provide selfreported data at the end of the shift (Table 4.1). Other factors affecting measurement success and quality are outlined in Table 4.1 and characterized qualitatively below in further detail. Challenges of workplace environments Workplace conditions in the construction, transportation, warehousing, forestry, and wood and paper products industries were diverse and some hindered exposure assessment in surprising ways. Adverse conditions included wet, cold, heat, dust, and noise. Wet outdoor environments included log boom boats in which seats were occasionally fully immersed, making it impossible to use electronic equipment for vibration monitoring. During rainfall at forestry and construction worksites, the EMG equipment stopped recording, likely due to an electrical short circuit. Cold storage warehouses with sustained extreme temperatures below -25oC caused condensation in electrical circuits and halted their function, as well as stopping ink flow in pens used to record observations. Although dust could clog monitoring equipment, it did not appear to alter the function of the monitoring equipment when used in sawmills, paper mills, or concrete work. However, none of the monitors were designed with grounded, arc-free circuits, a requirement for the explosive atmosphere of grain elevators. Hot and humid conditions in paper manufacturing and at outdoor worksites in summer increased worker sweating, which limited EMG electrode  40  adhesion and made all subject-mounted equipment uncomfortable. Noise, ubiquitous in most industrial workplaces, posed a challenge to thoughtful communication during the self-report data collection period when a quiet room was not available. Workplaces with noisy conditions obscured the audio alarm set to time the sampling intervals for the one-minute observations. Conversely, an alarm with a high sampling frequency was annoying to workers in quiet situations. A silent vibrating timer was used to time observations in such circumstances. Table 4.1: Summary of reasons for missed or incomplete measurements for each of the five measurement techniques Reasons for not measuring Subject Refusal: Worker declined method Working Conditions: Explosive environments Removed monitor for fear of damage Right or left electrode came off Difficulty measuring vibration when standing on vehicle Not enough time to conduct method Equipment Malfunction: Waiting for equipment repairs Monitoring equipment not operating properly Data too noisy to use Data with Noise or Errors Data for < 50% of shift Wrong settings used by researchers Insufficient Exposure: Not exposed Exposed for < 5 minutes  Number of Measurement Days Affected Inclinometer EMG Vibration SelfObservations report 0  0  0  1  0  9 0  9 10  0 12  0 –  0 –  –  †  42  –  –  –  –  –  5  –  –  0  0  15  4  0  0  36  2  –  –  8  5  4  –  –  7  24  7  –  –  29 0  †  †  †  0 0  41 0  6 29  †  1 –  0 –  0 0  *79 *16  0 0  0 0  † – = not applicable = usable data still available * = vibration measurements were not attempted on these days, therefore these days were not included in the denominator for calculated success rates or costs  Challenges of work tasks and postures Even under the most pleasant working conditions, successful data collection was complicated by the nature of work tasks. Monitoring equipment had to be comfortable in all working postures so that workers would wear it, and unobtrusive enough that it did not alter task performance. Workers in this study used lifting belts, tool belts, and fall protection harnesses that had the potential to interfere with EMG cables and electrodes, and the inclinometer mounted on the 41  trunk. This required careful consideration of the position of monitoring equipment and occasional repositioning throughout the day (e.g., the inclinometer could be positioned on the back or chest). During pilot testing, the carrying pouches for monitoring equipment were altered to have breathable, elasticized straps that allowed for movement and sweat evaporation when mounted on workers, and not to interfere with protective work clothing or equipment. Measurement protocols were designed to allow for bathroom breaks and clothing changes. Despite this preparatory work, unanticipated tasks and positions were still encountered, and required some creative problem-solving at the worksite. The postures adopted in workplace tasks can also present a challenge. Although lying down, kneeling, and crawling are infrequent in most jobs, maintenance and construction workers (and some others) performed these activities for substantial periods of time. Monitors worn on the body did not lend themselves to these postures without modification. The opportunity for cable movement artifacts and contact interference with electrodes was quite high in these industrial environments. Even sitting, which is a fairly common work position, could create unwanted contacts with electrodes. At times, worker-mounted measurement equipment was struck or compressed, and cables or harnesses snagged on scaffolding or machinery. This could have placed the worker at risk of injury and damaged the measurement equipment. The inclinometer was removed on 12 worker-days and the EMG was removed on 10 days for fear of injury to the worker or damage to the equipment. Observing very dynamic work, such as a maintenance worker walking to different tasks, challenged the observers to keep up and stay conscious of workplace hazards such as forklift traffic or cranes and wrecking balls. Despite the efforts of the observers, occasionally a worker ‘escaped’ while the observer recorded notes. The quality of observations decreased when the worker was in a single-occupant vehicle. For example, when a worker drove a forklift around a loading dock, the forklift was not always in the line of sight of the observer, sometimes for substantial time periods. When the observer watched from a site allowing a broader field of view (e.g., a supervisor’s tower), the distance was often too great to allow details of posture to be recorded, and this introduced a trade-off in data quality. Damage and repairs to monitoring equipment Both EMG and vibration equipment sustained damage in the field as a result of working conditions. The vibration monitor was used in moving vehicles and was padded and fixed to the frame of the vehicle, but despite efforts to protect it, bumpy terrain coupled with stop-and-start vehicle operation often resulted in movement of the monitor. On one occasion, it was caught under the vehicle seat and the LCD screen was broken (though data collection was not affected). Since cables from the EMG monitor to the electrodes were vulnerable to snagging and tugging, the cables were taped to the skin leading to a waist pouch holding the equipment. There were still many days when at least one electrode came off, and compression caused cable fatigue, requiring repairs to the cables on multiple occasions. Repairs to the EMG, including delivery time, resulted in 36 lost measurement days. The EMG and vibration monitors also had problems in pilot testing: both had faulty cables. The EMG equipment came from the factory with a cracked optic fibre cable; this took months to  42  repair and return. The whole body vibration accelerometer adapter cable was not properly shielded to protect against vibration artifact; the manufacturer provided an alternate accelerometer that did not require an adapter. In addition to cable issues, there were hardware incompatibilities between laptop computers and the two monitors that took time to remedy with new hardware. The inclinometer was ready for operation when received and was not damaged during sampling. It had no external cables and was more compact than the other two monitoring systems. Identifying noise or errors in monitoring data During data collection, data were downloaded and inspected 2 to 3 times per day. This allowed for noise or other problems to be identified and corrected without losing a whole shift’s worth of data. However, systematic data inspection and cleaning was not performed until data collection in the field was complete. There were several instances where collected data passed the brief initial inspection, but during later analysis was found to be uninterpretable. In one instance, this involved human error on the part of the data collectors: the pre-set menu for the whole body vibration monitor was changed, resulting in inadvertent selection of the wrong settings and loss of 29 measurement days. The post-measurement discovery of uninterpretable data presents an argument for analyzing data in parallel with data collection. However, this option would require an increase in data analysis personnel time during the data collection period. The effect of the presence of researchers It was possible for the presence of researchers and other people to influence exposure assessment. This was particularly true for self-reports and observations. At some workplaces it was difficult to identify a private space in which to collect the self-reported data from workers. Having the supervisor present while a subject estimated the amount of time he or she spent sitting or lifting may have affected the validity of these self-reported exposures. Many subjects in this study expressed reluctance to estimate exposure and a desire to match the observer’s record, although this was likely a challenge unique to this study, since the two methods were used on the same day. During observations, workers occasionally tried to interact with the observer by explaining what they were doing or making conversation. Although this helped to build rapport conducive to the post-shift self-report data collection, conversation was likely to alter the work tasks, and so was kept to a minimum by indicating to the worker a desire to “stay out of the way” and “not interrupt work,” and by increasing the distance between the worker and the observer. It was important to avoid changing the nature of work activities when assessing exposure. If equipment restricted movement or workers performed tasks differently because they were being observed, the measurements would be unlikely to accurately reflect typical exposures. Making observations over the whole shift so that workers became accustomed to being observed and developing protocols to decrease the impact of the monitoring equipment helped to minimize alterations to the work routine.  43  Cost of measurement methods Table 4.2 lists the costs of equipment, equipment repairs, personnel, data entry, and consumable supplies for each measurement method in this study. Table 4.2: The total and per-day costs (in $CDN) associated with data collection for five ergonomic measurement methods used in this study Cost per work shift when attempted measurements n = 223  Cost per work shift with successful measurements n = 199  $31 $160 $1 $192 n = 223 $149 $8 $160 $12 $329 n = 128  $35 $179 $1 $215 n = 139 $239 $13 $257 $19 $528 n = 54  $34,439 $20,480  $269 $160  $638 $379  $155  $1  $3  $55,074  $430 n = 223  $1,020 n = 222  $197  $1  $1  $35,680 $11,128 $611  $160 $50 $3  $161 $50 $3  $47,616  $214 n = 223  $214 n = 218  Total for this study Inclinometer Equipment Personnel Consumable supplies Total  $7,015 $35,680 $132 $42,827  EMG Equipment Equipment repairs Personnel Consumable supplies Total Vibration* Equipment Personnel (only for days when vehicles were used) Consumable supplies Total Observation Equipment (vibrating timers, clipboards) Personnel Data entry Consumable supplies Total  $33,238 $1,852 $35,680 $2,644 $73,414  Self-report $50 $0 $0 Equipment (pens, clipboards) $4,460 $20 $20 Personnel $389 $2 $2 Data entry $1 $1 Consumable supplies $165 Total $5,064 $23 $23 *the sample size for vibration is lower because not all measured workers used vehicles and had associated whole body vibration exposure  Self-report was the least expensive method because it demanded the least personnel time. It was 10-fold cheaper than observations and inclinometry, the next lowest cost methods. Whole body vibration monitoring and EMG were the most expensive methods, both because of the high capital costs of the monitoring equipment, and because fewer shifts were successfully measured. Not all workers used vehicles and the cost of measuring whole body vibration per attempted shift 44  was inflated compared to the other measures used in this study. There were no equipment repair charges for the vibration monitor as all repairs needed were covered by warranty. There were no direct data entry costs for the methods using monitoring equipment, since data was stored electronically during the measurement period. Risk factors assessed and data detail Table 4.3 summarizes the scope of each method in terms of the risk factors assessed, as well as the advantages and disadvantages of each within the context of field work. The self-reports provided a broad overview of exposures to all three risk factors (posture, materials handling, and whole body vibration); the observations did the same, but with more detailed information, since the data were recorded on a minute-by-minute basis. Each of the methods using monitoring equipment provided data focused on one risk factor (though EMG provides data on muscle activity due to posture and materials handling), but in tremendous detail (data logging at 1second or smaller intervals) that could be summarized in many metrics (e.g., mean, peaks, percentiles, cumulative exposure).  45  46  Self-report  Observation  Vibration  EMG  Inclinometer  Measurement Method  Does not require worker to be in view of research staff.  Provides 7 measurements per second that can be used to multiple exposure metrics, including average, cumulative, or peak postures, angular velocity or acceleration/deceleration.  Data in categorical format, rather than continuous measures.  Data summarized over full work shift.  Assesses multiple risk factors, including posture, manual materials handling, vehicle use, job title.  Provides data once per minute, that can be summarized as duration exposed or percent of shift exposed for each risk factor.  Assesses multiple risk factors, including posture, manual materials handling, vehicle use, tasks and items held or worn.  Provides 1 measurement per second that can be used to calculate multiple exposure metrics, average (RMS or VDV) and peaks.  Whole body vibration magnitude and direction.  Can question worker about shift, ‘typical’ or historical exposures.  Takes little time.  Low equipment requirements, few technical problems, can observe a large number of risk factors.  Can collect for long periods before downloading.  Does not encumber worker since it is not attached to worker.  Does not require worker to be in view of research staff.  Portable, can be set up on vehicle before shift.  Does not require worker to be in view of research staff.  Can be calibrated to estimate disc compression.  Provides 10 measurements per second that can be used to calculate multiple exposure metrics, including average, cumulative, or peak exposures.  Portable; maintains privacy and allows for movement throughout work space.  Muscle activity due to lifting, handling loads, and bending postures.  No cables and small in size; not easily damaged.  Can collect for long periods before downloading.  Portable; maintains privacy and allows for movement throughout work space.  Advantages  Flexion/extension and lateral bending angles.  Risk Factors and Data Detail  Privacy required for candid report not always readily available.  Subjective report, questionable validity due to difficulties with recall.  Requires extensive training and inter-observer comparison to ensure measurement consistency  Human resource intensive.  Requires visual contact with worker and sustained vigilance by observer to avoid danger and loss of data.  Not suitable for explosive environments.  Vulnerable to movements of vehicle and subject.  Difficult to stop and start monitor during intermittent vehicle use.  Must be downloaded at every break.  Requires time at beginning of shift for set-up and calibration.  Optic cables fragile and may break if bumped or compressed.  Not suitable for explosive environments.  Needs care not to interfere with clothing and personal protective equipment.  Vulnerable to inclement or dusty conditions.  Needs time at beginning of shift for set-up and calibration.  Not suitable for explosive environments.  Needs care not to interfere with clothing and personal protective equipment.  Disadvantages  Table 4.3: Risk factors assessed, data detail, advantages and disadvantages of the exposure assessment methods used in this study  Discussion In this study of back injury risk factors in heavy industries, exposure measurement was very demanding on the monitoring equipment. Problems arose from many sources: human error setting up the equipment; equipment failure; extreme and dynamic worksite conditions; and unusual or unanticipated work tasks, positions, or machinery. This resulted in less complete data for the monitoring equipment than for self-reports and observations. Of the equipment used, the inclinometer was best able to handle the stresses of the industrial work settings, likely due to its small size, its lack of cables, and its potential to be positioned on the chest or the back. The cost of this device was similar to the observation method and it was used successfully on almost 90% of the measurement days. It provided data only on trunk flexion/extension and lateral bending, but did so in great detail (7 measurements per second), and the logged data can be summarized into many possible metrics, providing depth of data rather than breadth. The EMG and vibration monitoring provided similar data depth, but at considerably more cost, though the relative cost of vibration monitoring would be lower in studies focused on driving occupations alone. Others have commented that direct measurements using monitoring equipment are preferred since they provide abundant detail on exposure level, duration and frequency using precise and unbiased measures about very specific components of physical exposures. However, these advantages need to be balanced against the high costs for equipment and personnel (Burdorf & van der Beek 1999). Exposure-response relationships will be stronger when more precise exposure assessment measures are used (Armstrong BG, 1990). Herein lies the classic tradeoff between measurement accuracy and complexity versus expense. The self-reports and observations were less prone to environmental interference than the monitoring equipment, but were not without drawbacks. The one-minute observation sampling rate was difficult for research staff to maintain throughout the work shift, especially when subjects were moving throughout the workplace or when workplace hazards required vigilance on the part of the observers. Despite the difficulty of this recording schedule, all measured days included observations for more than 50% of the work shift. Observations were more expensive than self-reports, but had the most complete measurement success rate, included considerably more detail than the self-reported data, and were conducted during the work cycle, so were not subject to problems with recall. Various researchers have developed observation tools for recording ergonomic risk factors and tested their validity and reliability (Karhu, Harkonen, et al. 1981, Burdorf, Derksen, et al. 1992, McAtamney & Nigel Corlett 1993, Buchholz, Paquet, et al. 1996, Paquet, Punnett, et al. 2001). For example, Burdorf et al. (1992) compared measurements of trunk bending by observation to direct measurements and found reasonable agreement for mobile workers doing maintenance jobs, but poorer agreement for those who were more sedentary, including drivers and office workers. The most commonly reported disadvantage of observational methods is that unless the job is cyclical and has a short cycle time, the method is very labour intensive. Interestingly, we found that using monitoring equipment was also labour intensive. Although it did not require the research staff to follow each worker through every minute of the shift, it did require routine checks to make sure it was functioning, to download data, and to ensure it was not being damaged. Self-reports were both successfully completed and the least expensive method, but the tradeoff was potential biases associated with retrospective recall of the work shift at the end of the day.  47  Workers struggled to recall risk factors encountered during the day and were reluctant to provide answers that might be perceived as ‘inaccurate’, especially given that observers had been present throughout the shift. The self-reported exposure data were less detailed than the data collected by any of the other methods. Other authors have commented about self-reports for assessing ergonomic risk factors, and have suggested that information collected by self-report is subject to systematic bias and lack of precision (Burdorf & van der Beek 1999, Spielholz, Silverstein, et al. 2001). Wells et al. (1997) stated that in self-reports, respondents can identify whether exposure to a stress such as vibration or lifting occurred, but they do not tend to give reliable information on either the nature or magnitude of the exposure. Palmer et al. (2000) found that both the use and duration of use of large equipment are readily identified by users and well reported as sources of whole body vibration. Limitations The evaluation provided here is specific to the conditions we encountered in the study worksites and the measurement methods we used, though it provides information that should be useful to other investigators. The cost estimates deliberately excluded methods development, training, travel, and analysis, because these costs were expected to be less generalizable to other research conditions. For example, travel time and associated transport and/or accommodation costs would vary considerably by geographical location, and personnel training may or may not be required, depending on experience and education. These costs do need to be considered, but need to be estimated on a more study-specific basis. Vibration measures were attempted only on workers who used vehicles for longer than 5 consecutive minutes during their shift. This decreased the total number of attempted worker days for vibration measurements to 128. Two other conditions made it difficult to measure vibration exposures: work tasks in which subjects moved on and off their vehicles so frequently that it was impossible to set up the vibration meter on time; and exposure to vibration via the feet because the subject was standing, e.g., on boats. The resulting lower numbers of attempted and successful measurements inflated the costs per attempted day and per successful day for the vibration equipment. Conclusions In our experience in heavy industry worksites, observation and self-report methods were less expensive and successfully completed more often than were methods using monitoring equipment. Unsuitability of monitoring equipment for some workplace conditions, interference with postures and work gear, malfunction, and human error contributed to their lower success rate, while substantial capital investment was the main factor in their higher cost. However, one monitoring method, inclinometry, was similar in utility to observations. Both methods had only moderate costs and nearly complete shift measurements. These two methods were complementary in data detail: observations were broad in scope with measurements of all the  48  injury risk factors of interest; and inclinometry provided data depth and precision on postural exposures. The choice of exposure measurement methods depends on the tasks being assessed and the objectives of the evaluation. In many cases, higher costs may be warranted to achieve the more detailed, precise and valid measures offered by monitoring equipment. Depending on the environment and the research question, a combination of methods may also be appropriate. The information collected in this study underscores the importance of another criterion when selecting exposure assessment methods: the practicability of the methods within a work setting. Methods should be robust enough to withstand the demands of work environments, as well as the range of worker tasks and positions. Finally, assessment tools should combine valid assessments and reliable functioning with reasonable cost to provide good value in terms of the price per successful sample.  49  References Armstrong BG. 1990. The effects of measurement errors on relative risk regressions. Am J Epi 132, 1176-1184. Buchholz B, Paquet V, Punnett L , Lee D and Moir S. 1996. PATH: a work sampling-based approach to ergonomic job analysis for construction and other non-repetitive work. Appl Ergon 27, 177-87. Burdorf A, Derksen J, Naaktgeboren B and van Riel M. 1992. Measurement of trunk bending during work by direct observation and continuous measurement. Appl Ergon 23, 263-7. Cole D C, Wells RP, Frazer MB , Kerr MS, Neumann WP and Laing AC. 2003. Methodological issues in evaluating workplace interventions to reduce work-related musculoskeletal disorders through mechanical exposure reduction. Scand J Work Environ Health 29, 396-405. Danneels LA, Coorevits PL, Cools AM, Vanderstraeten GG , Cambier DC, Witvrouw EE and De CH. 2002. Differences in electromyographic activity in the multifidus muscle and the iliocostalis lumborum between healthy subjects and patients with sub-acute and chronic low back pain. Eur Spine J 11, 13-9. Genaidy AM, Al-Shedi AA and Karwowski W. 1994. Postural stress analysis in industry. Appl Ergon 25, 77-87. Karhu O., Harkonen R, Sorvali P and Vepsalainen P. 1981. Observing working postures in industry: Examples of OWAS application . Appl Ergon 12, 13-7. Kidd P, Parshall M, Wojcik S and Struttmann T. 2004 Overcoming recruitment challenges in construction safety intervention research. Am J Ind Med 45, 297-304. LaMontagne AD and Needleman C. 1996. Overcoming practical challenges in intervention research in occupational health and safety. Am J Ind Med 29, 367-72. Li G and Buckle P. 1999. Current techniques for assessing physical exposure to work-related musculoskeletal risks, with emphasis on posture-based methods. Ergonomics 42, 674-95. McAtamney L and Nigel-Corlett E. 1993. RULA: a survey method for the investigation of work-related upper limb disorders. Appl Ergon 24, 91-9. Palmer KT, Haward B, Griffin MJ, Bendall H and Coggon D. 2000. Validity of self reported occupational exposures to hand transmitted and whole body vibration. Occup Environ Med 57, 237-41. Paquet VL, Punnett L and Buchholz B. 2001. Validity of fixed-interval observations for postural assessment in construction work. Appl Ergon 32, 215-24. Spielholz P, Silverstein B, Morgan M, Checkoway H and Kaufman J. 2001. Comparison of self-  50  report, video observation and direct measurement methods for upper extremity musculoskeletal disorder physical risk factors. Ergonomics 44, 588-613. van der Beek AJ and Frings-Dresen MH. 1998. Assessment of mechanical exposure in ergonomic epidemiology. Occup Environ Med 55, 291-9. Wells R, Norman R, Neumann P, Andrews D, Frank J, Shannon H and Kerr M. 1997. Assessment of physical work load in epidemiologic studies: common measurement metrics for exposure assessment. Ergonomics 40, 51-61.  51  Chapter 5: Predicting exposure for mean, 90th percentile, and cumulative EMG activity in heavy industry3 Introduction Despite increasing technological advances, heavy industry continues to have high physical demands and correspondingly high back injury rates. In the Canadian province of British Columbia (BC) there were 167,480 accepted compensation claims for back strain between 1996 and 2005, representing 25% of all claims, 23% workdays lost, and 20% of claims costs (WorkSafeBC, 2005). These results are mirrored in other regions, including the United States (Guo et al. 1999). In BC, more than a quarter of all back strain claims were from employees in the following heavy industries: forestry, wood and paper products, construction, transportation, and warehousing (Workers' Compensation Board of British Columbia., 2002). The awkward and sustained postures, forceful exertions, and repetition of manual materials handling and other tasks typically seen in heavy industry jobs place loads on the spine that can contribute to back injuries and back disorders (Keyserling, 2000). Peak spinal loading is one measurement of the physical demands of a job or workday that has been shown to be associated with back pain. However, cumulative spinal load is gaining attention as a workplace exposure related to back pain independently of peak loads (Norman et al. 1998). Typical ergonomic exposure assessments have used expensive and time-intensive direct measurment techniques and biomechanical modeling to measure exposures for short durations or small numbers of workers, (Cooper and Ghassemieh, 2007; Village et al. 2005; Jones and Kumar, 2007) or in simulated work tasks (Lavender et al. 2007; Keir and MacDonell, 2004; Moore and Garg, 1995; van der Beek and Frings-Dresen, 1998). Although this type of highresolution measurement is often considered a preferred method for occupational exposure assessment (Wells et al. 1997; Houba et al. 1997; Wells et al. 1997), large-scale epidemiological studies require efficient methods applied over large samples. Observations and worker self-reporting allow for inexpensive ergonomic assessments for large numbers of workers, without the high cost and limited feasibility associated with direct measurement (Trask et al. 2007). However, the relationship between direct measurement methods and observation or self-report methods needs to be determined. “Determinants of exposure” modeling involves predicting a measured exposure using characteristics that directly or indirectly increase or decrease that exposure. The intent is to identify the key measured characteristics that best predict exposure as an efficient and effective sampling strategy. Although this methodology is in the process of being applied to physical exposures related to low-back disorders, it has been used more commonly in industrial hygiene to estimate chemical exposures (Burstyn et al. 1997; Burstyn and Teschke, 1999; You et al. 2007; Nieuwenhuijsen et al. 1995). The purpose of this study was to use a determinants of exposure modeling approach to develop exposure prediction models for mean, 90th percentile, and cumulative low back EMG activity. Two models were developed, one to identify observed physical predictors of EMG exposure, and another to identify self-reported physical predictors of EMG exposures. The larger goal is to 3  A version of this chapter has been submitted for publication. Trask, C., Teschke, K., Morrison, J., Village, J., Johnson, P., Koehoorn, M. (2008) Predicting Exposure for Mean, 90th Percentile, and Cumulative EMG Activity in Heavy Industry.  52  help researchers efficiently and effectively measure ergonomic exposures using a combination of exposure assessment methods for a specified number of work tasks for studies of work-related injury in heavy industries.  Methods This study measured full-shift EMG of the lumbar muscles on workers from 50 different worksites in British Columbia within the heavy industrial sectors of construction, forestry, transportation, warehousing, and wood and paper products from September 2004 to January 2006. Human subject procedures were approved by the University of British Columbia’s Behavioural Research Ethics Board. Workers were selected at random from those with accepted back strain claims for the year 2001. Eligible workers were working in the five target industries, did ‘shop floor’ rather than administrative jobs, and resided in the Vancouver area. After subjects agreed to participate, researchers contacted their employers to obtain permission to conduct worksite measurements and to recruit an additional 1 to 4 co-workers. The recruitment methods are described in detail in Appendix A. Set-up, measurements, and interviews were conducted during regular work time. Concurrent measurements were made over a full work shift using observations by trained personnel and worker self-reports. EMG measurement Field sampling Full-shift EMG measurements were made using a portable data collection system with on-board memory (ME3000P4/ME3000P8, Mega Electronics, Finland) and disposable Ag-AgCl electrodes (Blue Sensor N-00-S, Ambu, Denmark). Electrodes were placed bilaterally over the erector spinae at approximately the level of L4/L5, with a 20 mm inter-electrode spacing and a ground electrode and preamplifier placed on the posterior aspect of the iliac crest. Signals were collected at 1000 Hz and filtered internally using an 8-500 Hz band pass filter. Root-meansquare values were data-logged at 10 Hz. During work breaks, data from the portable system were downloaded to a laptop computer. EMG data were collected for the full shift excluding breaks (mean 6.2 hours, range of 5.5 to 10.3 hours). EMG data cleaning and analysis methods are described in more detail in Appendices G and H, respectively. Calibration and calculation of exposure metrics A sub-maximal reference contraction was employed to calibrate EMG data collected during the shift. The reference contraction involved a static 45o forward trunk flexion while holding an 11.5 kg weight. The reference contraction was performed twice for 5-seconds at the beginning of each shift. All EMG data collected during the shift were expressed as a percentage of this reference contraction (%RC) and all EMG exposure metrics or outcomes are in units of %RC. EMG exposure metrics were calculated for each individual’s work shift data. Mean represented central tendency, while 90th percentile represented a measure of peak exposure level. Both metrics were measured in %RC. Cumulative EMG activity represented the sum of instantaneous %RC values throughout the shift, expressed as %RC-seconds of cumulative EMG activity. Cumulative exposure was calculated over the whole shift regardless of shift length. For peak exposure, the 100th percentile (maximum value) was not used, as it may represent a single value  53  contaminated by noise such as movement artifacts, and may not be representative of repeated peak loadings. The 95th percentile was highly and significantly correlated with 90th percentile (Pearson r = 0.981) and the 90th percentile has been used as a measure of ‘peak’ load in previous studies (Jonsson, 1988; Mathiassen et al. 2002; Mientjes et al. 1999; Nordander et al. 2004; Moller et al. 2004). Investigation of EMG calibration An additional set of calibration tests were conducted at the beginning and end of the shift to alleviate concerns that differences such as drift over the course of a full shift might change the value of the reference contraction. In addition to the 45o flexion reference contraction with the 11.5 kg weight, muscle activity was recorded for standing upright, 45o flexion without a weight, and 60o flexion with an 11.5 kg weight to represent some typical positions and weights seen in industrial tasks. Two repeats of each maneuver were collected both before and after the shift. Altogether, 16 calibrations were collected per subject per day totaling 3247 discrete calibration measurements. Pre-shift measures were, on average, 3 µV higher than post-shift measures for the same position, trial number, and side of the body. In the context of the average EMG calibration recording (37 µV), this difference was not considered sufficient to affect the results of this study. In multivariate modeling of the factors affecting the EMG signal during the various calibration maneuvers, controlling for subject and measurement session, position explained the largest proportion of EMG variance at 23%, compared to less than 0.4% for prevs. post-shift timing. Observation data collection Observations of physical exposures were made by trained observers throughout the work shift (excluding breaks) starting after EMG instrumentation and calibration and ending with deinstrumentation at the shift’s completion. Observations were recorded once every minute using the BackEST (Back Exposure Sampling Tool) observation tool to code variables into categories (Village et al. 2007). In brief, these variables were: general task or activity, item or power tool in hands, items worn (such as a tool belt), general body posture (such as standing, walking, kneeling), trunk posture (twisting, lateral flexion, and categories of trunk flexion), presence of trunk support, and manual materials handling (type of load, horizontal distance, weight and force estimate). The total summed time for observed exposure to a given risk factor was used as the outcome of interest for cumulative EMG models, whereas proportions of time for observed exposure were used for mean and 90th percentile EMG models. For example, the sum of oneminute observations where a worker was observed to be standing was divided by the total number of one-minute observations over the work shift to yield a proportion of time spent standing. Pilot testing methods, validity and reliability data, and a sample of the BackEST observational tool are reported in detail in Village et al. (2007). Interview data collection A post-shift interview was conducted with each worker to collect self-reported exposures during that shift. Using diagrams of activities and postures as a visual cue, workers were asked to identify the presence or absence of general activities such as standing, walking, kneeling, trunk  54  postures (including twisting, lateral flexion, and categories of trunk flexion) and manual materials handling activities (including type of load, horizontal distance, weight and force estimate). For each exposure, workers were asked to estimate the duration of time spent in a posture or performing an activity during the work day by selecting a time category: (< 5 min; 515 min; 15-30 min; 30-45 min; 45-60 min; 1-2 hrs; 2-4 hrs; 4-6 hrs; 6-8 hrs; >8 hrs.). These times were converted to a percentage of work time by dividing by the shift length. Industry, job title, age, height, weight, hours worked per shift, shifts per week, and the number of consecutive shifts worked were also assessed by the questionnaire. A copy of the final survey instrument can be found in Appendix D. Statistical analysis EMG exposure prediction models were developed for each of the three exposure metrics: mean, 90th percentile (peak), and cumulative EMG activity. Separate models were constructed for observed variables and self-reported variables as predictors of exposure for the EMG metrics. Modeling was performed in three stages: simple linear regression to examine bivariable relationships, mixed multiple regression modeling with independent variables in conceptual groups, and combining conceptual groups into a final mixed model. Initially, simple linear regression (SAS version 9.1, SAS Institute Inc., Cary, NC USA) was used to determine the bivariable relationship between EMG exposure metrics and each of the observed or self-reported exposure variables. Independent variables were retained for input into multiple regression models if they were significantly associated with the outcome; observation variables were offered to subsequent models if p < 0.1, while self-reported variables were offered if p< 0.05 (more restrictive because there were many more individual self-reported variables available). After determining that variables were significant in bivariable analysis, a correlation matrix of all independent variables was consulted to ensure that highly correlated variables were not offered to the same model. If independent variables were correlated at Pearson r ≥ 0.70, then the variable with the lowest bivariable p-value was selected for input to the multivariable model. Mixed modeling methods incorporate both fixed and random effects in order to account for repeated measurements on the same individuals (Burdorf, 2005; Burstyn and Teschke, 1999). For this study, ‘subject’ and ‘company’ were initially offered as random effects to account for the within-subject and within-company correlation not accounted for in the fixed effects; only ‘subject’ was retained in the final models as significantly related to EMG exposure. Mixed effects modeling was conducted using backwards stepwise multiple linear regression in SAS. As the second step of modeling, variables significant in bivariable modeling were offered in ‘conceptual groups’. In order of offering, all postural variables were offered as a group, followed by manual materials handling variables, demographic variables, and then job factors. A list of observed and self-reported variables within their categories can be seen in Tables 5.3 and 5.4. Variables that remained in the conceptual group model were forced into the final model. Lifting and bending often coincided in occupational tasks, so the possibility of an interaction between these exposures was explored via an additional variable created by calculating the percentage of time spent handling a load greater than 4.5 kg while simultaneously flexing the trunk forward 20o or more. This variable was offered only when both a lifting variable and forward flexion variable were significant in the initial model.  55  The proportion of variance explained by each model was estimated by comparing the predicted exposure levels to the measured exposure levels. The R2 between estimated and measured EMG exposures was used as an estimate of the proportion of variance explained by the mixed model.  Results Worker sample Full-shift EMG was successfully measured on 103 individual workers (34% measured on two days) for a total of 138 worker-days; the lag between repeat measurement days on the same worker ranged from 1 to 439 days (mean = 93 SD=64). Observation and self-report data was available for all these worker-days. The 138 out of 223 attempted worker-days represented a 62% success rate. Demographic data on the sample are summarized in Table 5.1. Additional worker characteristics can be found in Appendix I. Table 5.1: Sex, age, height, weight of study participants in heavy industry (n=103 workers) Variable % Male Mean Height in cm (sd) Mean Weight in kg (sd) Mean Age, in years, on sampling day (sd)  Value 95.3% 178.1 (7.9) 85.2 (16.1) 42.2 (12)  EMG exposure metrics by industry The measured mean, 90th percentile, and cumulative EMG activities in the five heavy industries are presented in Table 5.2. The construction industry had significantly higher measures of mean and 90th percentile EMG activity than the transportation industry, while the forestry industry was significantly higher than transportation for all three exposure metrics. Longer measurement periods in forestry (7.4 hours) versus in construction (5.8 hours) contributed to the difference for the cumulative metric. Warehousing and wood and paper products had similar exposures. Overall, the ordinal exposure ranking of the industries was fairly consistent across the metrics (Table 5.2). Table 5.2: EMG exposure metrics for five heavy industries. All metrics are based on full-shift data collection Exposure metric  Mean %RC (sd) 90th %ile %RC Cumulative (%RC.sec)  Construction (n=25 worker days)  Forestry (n=30 worker days)  Transportation (n=33 worker days)  Warehousing (n=23 worker days)  Wood products (n=27 worker days)  All industries combined (n=138 worker days)  49.1 (14.1)  41.9 (23.8)  28.7 (12.2)  39.8 (21.5)  37.8 (24.1)  38.9 (20.5)  103.4  84.2  67.5  84.9  80.0  83.1  993,943  1,084,283  681,270  982,756  895,931  918,320  56  Prediction models The final mixed models using self-reported or observed variables to predict mean, 90th percentile, and cumulative EMG activities are found in Tables 5.3 and 5.4. Bivariable results were extensive and are included in Appendix J. Cumulative EMG prediction models were developed using total time spent engaged in an activity; results for prediction models based on the percentage time engaged in activities are found in Appendix K. Table 5.3: Observed ergonomic variables associated with mean, 90th percentile, and cumulative EMG exposure (expressed at % of reference calibration contraction or %RC) in final multiple linear regression models, with subject as a random effect. Variable  Mean %RC % of total observed time  Intercept (average for all subjects) Standing (% time)  β (slope)  p  19.8 46.6%  0.115  90th percentile %RC β p (slope) 43.8  <0.001*  0.166  0.236  .1026  0.970  <.0001*  -3118  0.041*  7295  0.0004*  17182  0.0004*  22581  <.0001*  7535  0.114  0.010  0.612  .0018*  1.25  0.0014  0.134  0.633  0.659  0.229  0.910  <0.001*  0.987  0.009  0.325  .0641  0.236  0.144 0.298  595895 0.316  30.1%  Sitting (% time) Trunk position 10-20o 30.1% (% time) Trunk position 20-45o 12.0% (% time) Trunk position 45-60o 3.9% (% time) Trunk position >60o (% 5.3% time) Handling load at 5.7% extended horizontal distance (% time) 4.5-10kg load in hands 4.7% (% time) 10-20 kg load in hands 3.6% (% time) ‘Light’ Push/pull force 9.5% (% time) Handling loads with 19.1% two hands Estimated proportion of variance explained by model Within-subject correlation *variables significant at p<0.05  Cumulative %RC** β p (slope)  0.289  47.2%  42.9%  30.7%  48.6%  50.0%  42.7%  **independent variables for cumulative EMG activity are in total time  57  Table 5.4: Self-reported ergonomic variables associated with mean, 90th percentile, and cumulative EMG exposure (expressed at percentage of reference calibration contraction or %RC) in final multiple linear regression models, with subject as a random effect. Variable  % of total Selfreported time  Mean %RC β (slope)  Intercept (average of subjects) Sitting (% time) Walking with trunk twisted (% time) Crouching, kneeling, or squatting (% time) Sitting and twisting (% time) Manual materials handling (% time) 4.5-10 kg load in hands (% time) Handling load at extended horizontal distance (% time) Construction industry  18.8%  14.8  Forestry industry  18.8%  Wood product industry Warehousing industry  p  33.4 31.3%  -0.181  0.0023*  90th percentile %RC β p (slope) 78.7  <.0001  -0.376  0.0009  1.7% 9.5%  0.202  0.0543  4.4%  -0.226  0.0664  -0.498  0.0403  41.9% 9.3%  0.165  0.3367  5.0%  0.710  0.0040*  0.0054*  24.3  0.0131*  13.3  0.0109*  20.02  0.0323 *  18.8%  4.44  0.369  4.15  0.6498  19.3%  8.75  0.1024  13.8  0.1581  Cumulative %RC** β p (slope) 627029  <.0001*  22566  0.0012*  -7157.  0.0268*  1332  0.159  5214  0.1243  Reference Reference Transportation industry 24.2% 0 0 Estimated proportion 36.0% 21.0% of variance explained 36.0% by model Within-subject 65.9% 58% 49.1% correlation *variables significant at p<0.05 **independent variables for cumulative EMG activity are in total time  Observed variables At the bivariable level, a large number of observed postural variables were significant predictors of EMG: percent time observed sitting, standing, and crouching/kneeling and percent time spent with trunk flexed 0-10o, 20-45o, 45-60o, and >60o were significant for all three EMG metrics. Several other variables were significant for one or two of the EMG metrics. However, only a portion of the postural variables that were significant in bivariable analysis were significant in the final models, and the variable categories that were included in the final models were not consistent across EMG metrics.  58  The final model for mean EMG included observed percent time spent standing, trunk flexed 2045o, and trunk flexed >60o; the 90th percentile EMG model also included percent time standing. The postural variables included for cumulative EMG activity were total time spent sitting, and time with trunk flexed 10-20o or 45-60o. Many manual materials handling variables were also significant in bivariable analysis. All EMG metrics were significantly related to: total time spent manual materials handling, as well as subcategories of time spent lifting, lowering, and holding, as well as percent time handling a power hand tool, with the tool ‘idling’ and with the ‘tool on’. As with postural variables, only a subset was significant in the final models. There was some consistency between models: the time spent handling 4.5-10kg loads was significant for all models. The mean EMG model also included the observed percent time handling 10-20 kg loads, handling loads at an extended distance from the body, and light pushing/pulling. The model for 90th percentile EMG also included handling extended loads, as well as percent time spent handling loads with two hands. Self-report variables For some of the variables, the maximum reported percentage of day exposed exceeded 100%. This occurred when the worker’s estimate exceeded the length of the shift. Since such misclassifications by the worker could occur anywhere in the range of reported exposure, cumulative exposure times were not corrected. Fewer self-reported variables were significant in bivariable analysis than were observed variables, though a number of self-reported postural and manual material handling variables were offered to the mixed models. In terms of postural variables, time spent sitting, sitting with lateral bending, and sitting while twisting were significant predictors for all EMG metrics. Time sitting with the trunk twisted was included in the final version of all the models. Mean EMG was predicted by percent time reported sitting, and crouching/kneeling/squatting; the 90th percentile model also included self-reported percent time sitting. The cumulative EMG model included the total amount of time walking with trunk twisted. At a bivariable level, all EMG metrics were significantly related to self-reported time spent performing manual materials handing (MMH), handling 10-20 kg loads, and handling loads at extended distance from the body. However, the final model for mean EMG did not include any manual materials handling variables (Table 5.4). Handling an extended load was a significant predictor of both 90th percentile and cumulative EMG; handling 4.5-10 kg loads was significant only for 90th percentile EMG, while total amount of manual materials handling was significant only for cumulative EMG. Other variables Industry was significant in two self-report models and one observation models. None of the demographic variables were significant, nor were the job charecterics. Whether or not the worker was identified as a back strain claimant in 2001 was also not significant in any of the models.  59  Predictive power of the final models The observation models explained higher proportions of the EMG variability than the models based on self-report. The observation-based models for mean, 90th percentile, and cumulative EMG explained 47.2%, 42.9% and 30.7% of the variability in measured EMG respectively. The models based on self report explained 36.0% of mean, 36.0% of 90th percentile, and 21.0% of cumulative EMG variability.  Discussion Performance of EMG prediction models It is fair to question whether explaining 20-50% of the variability in EMG activity using observed and self-reported ergonomic exposures is adequate for epidemiological study. Studies of risk factors for back injury have found a relationship with ergonomic work factors using selfreported and observed exposures (Kumar, 1990; Burdorf and Sorock, 1997; Myers et al. 1999; Morrison, 1999; Knibbe and Friele, 1996; Macfarlane et al. 1997). This shows predictive validity with these methods, even in cases where their relationship to directly-measured exposure was not explicitly tested or described. In occupational hygiene, ‘determinants of exposure’ studies of airborne chemical exposures typically explain 30-60% of the variance in a directlymeasured exposure, and have been subsequently used successfully for exposure assessment in epidemiological studies (Burstyn and Teschke, 1999). EMG differs from airborne chemical exposures in that it is dependent on multiple external factors (MMH loading, posture) and internal factors (muscle recruitment patterns, fitness, muscle fatigue). Studies modeling determinants of exposure for ergonomic exposures as in the current study were not found by the authors, so comparing the predictive power of the models to those examining similar exposures was not possible. For a measure of physical exposure as complex and multifactorial as EMG activity, it is not surprising that the proportion of variability explained was on the lower end of what is typical for external chemical exposures. The estimates of EMG activity given by the models presented do not (and are not expected to) deliver a perfect estimate of an individual’s true EMG exposure. Rather, the value in the ‘determinants of exposure’ methodology is for the identification of the tasks and methods that best predict working exposure for epidemiological study of large numbers of workers. This has been acknowledged in prior comparisons of direct measurements versus observation and selfreport methods. A study comparing spinal compression estimates from checklists, video digitization, work sampling, and self-report in a study of low-back pain found that all methods gave significant odds ratios, but were not interchangeably appropriate for assessing risk at an individual level (Neumann et al. 1999). Mean EMG was the best-predicted exposure metric for both observation and self-report models, followed closely by 90th percentile EMG. Observation variables offered to the cumulative model were summarized as total time rather than percentage time in an effort to account for the duration component of the cumulative exposure metric. Nonetheless, the performance of exposure models for cumulative EMG activity was lower than for the other metrics. This is especially interesting given that the total time observed and the total time reported by workers are essentially cumulative measures, rather than peak or mean measures of exposure.  60  As one might expect given the higher subjectivity, self-reported variables were poorer predictors of exposure than observed variables. However, the increased time and cost associated with observation measurement compared to self-report measurements makes it a less attractive option for epidemiological studies of back injury (Trask et al. 2007). Although there was a lower percentage of variability explained for self-report models, self-reported data can be collected in a fraction of the time and allow for larger sample sizes and/or more repeated samples. Depending on the aims, hypotheses, target population and budget of an epidemiological study, self-report of significant variables may be the more effective option. For all models, the majority of significant relationships with EMG metrics were with posture or manual materials handling variables. Both lifting and trunk bending require low-back muscle activity, so this result is not surprising. These exposures are important to back injury as direct and indirect risk factors; sustained and non-neutral postures are related to low back disorders and pain reporting (Burdorf et al. 1991; Keyserling, 2000; Keyserling, 2000), as are manual materials handling (Kerr et al. 2001; Barriera-Viruet et al. 2006; Keyserling, 2000). Posture In the final observation-based models, the total time observed sitting decreased the predicted cumulative EMG. Sitting requires very little back muscle activity and workers are unlikely to be performing manual materials handling or extreme trunk flexion while seated. Four out of five observed trunk flexion variables were significant in bivariable analysis; the 20-45o flexion category had the largest effect estimates. Interestingly, the trunk flexion categories of 20-45o and >60o were significant predictors of mean and 90th percentile activity, but cumulative activity was predicted by intermediate categories of 10-20o and 45-60o. Although increased trunk flexion clearly increases predicted EMG activity over all metrics, the ranges of posture which are included differs between metrics, likely because of correlation between these categories. Observed time spent walking with the trunk twisted was a surprising addition to the final model since it comprised less than 2% of total observed time and most (70%) of worker-days reported no ‘walking while twisting’. Self-reported ‘sitting’ decreased mean and 90th percentile EMG, while self-reported ‘sitting while twisting’ increased predicted mean, cumulative, and 90th percentile EMG. ‘Walking while twisting’ may be a surrogate for carrying or loading tasks, but given that the variables included as predictors in the final self-report model are counter-intuitive, it seems more likely that their inclusion is a consequence of poor precision of the self-reports, or chance since it is such a small proportion of working time. Self-reported trunk postures did not predict EMG metrics in the final models; only low-risk categories like ‘10-20o’ and ‘extended trunk’ were significant in bivariable analyses and thus offered to the model. Previous studies have shown self-reports are less precise than observation and direct measure (Neumann et al. 1999) and that workers tend to over-report physical exposures (Spielholz et al. 2001). Manual materials handling Manual materials handling has a clear and documented relationship to muscle force and spinal loads (Waters et al. 1993). As was expected, observed and self-reported time spent handling  61  loads were included in several models. Overall, 3.7% of all observed time was spent handling loads in the 4.5-10 kg range and 2.1% for the next most common range, 10-20 kg. Although this is not a large proportion of observed time, these variables were fairly evenly distributed among the workers, so that nearly all individuals handled such weights at least once. All three EMG metrics were also associated with observed and/or self-reported handling of loads at extended distance from the body. The horizontal distance of loads in the hands from the body increases the moment arm at the lumbar joint and requires more torque to be generated by the back extensor muscles (Waters et al. 1993). Observation versus self-report Observation and self report are cheaper than direct measurement, making large numbers of measurements feasible. Observation requires more investment per measure than self-report, but it predicts more of the variability in measured EMG and the variables for posture and manual materials handing included in observation models are intuitive. The self-report methodology used in this study had one-tenth the cost of the observation method (Trask et al. 2007) and would allow for a larger sample size with a given budget relative to the observation method. However, the large proportions of unexplained variance in the self-report models of cumulative exposure indicates that a large number of measurements will be required to detect exposureresponse relationships (Siemiatycki et al. 1989). Self-report is not as refined a measure of exposure if gross variables such as ‘industry’ are included in the model to the exclusion of specific manual materials handling and posture variables; the inclusion of several unusual and potentially spurious variables also does not enhance confidence in the self-report models. The precision tradeoffs involved can limit the efficiencies of the self report model; lower cost means that more measurements can be made, but due to low precision many more will need to be made to reach the same level of significance as with direct measure (Siemiatycki et al. 1989). For these reasons, observation seems like a more prudent choice than self-report. Industry ‘Industry’, a categorical variable for the five sectors included in this study, was a significant predictor in two self-report models but one of the observation models. Industry added a substantial amount to these models; without the ‘industry’ variable, the self-report-based 90th percentile model explained 33% and the mean model explained 29% of the variance (compared to 36% for both models with industry included). Multiple self-reported posture and manual materials handling variables were significant predictors in bivariable models but were not retained in the final multi-variable models. The fact that they were not included suggests that the ‘industry’ variable is better at accounting for the differences between individuals than manual materials handling or trunk posture. EMG activity was significantly higher in construction and forestry, which included jobs with lots of bending and MMH, than in transportation, which included many sedentary vehicle operating jobs; clearly there is a systematic difference in manual materials handling and postural exposures between industries. In terms of data collection, ‘industry’ is one of the fastest, easiest, and most reliable pieces of data to acquire since it can be obtained either during an interview and/or from employer or workers’ compensation classification registries. Regardless, the inclusion of ‘industry’ in the model limits the ability to apply the models to other heavy industries, such as mining or oil and gas.  62  However, the small proportion of variance explained by the self-report model and the inclusion of unexpected or under-represented activities (such as walking while twisting) may limit the utility of the self-report model in any industry. Measurement issues EMG measurement has some limitations that should be considered. Worksite conditions can include extreme heat, cold, wet, dust, and vibration, as well as pressure or contact from tight spaces, seat backs or safety equipment, sweating caused by extended exertion, or tugging on the electrode cables (Trask et al. 2007). Although these factors can introduce noise and can result in misclassification of exposure, signals were filtered to remove electrical noise and care was taken to remove data with identifiable artifacts. The amount removed ranged from 1% to 5% of the shift for roughly 30% of measured work shifts. It is harder to predict how missing data affected the subsequent models. If EMG data collection was interrupted only when work became very strenuous due to sweating or increased tugging on cables during dynamic movements, then the net effect would be EMG measures underestimating true working exposures. It is possible that the flexion-relaxation response may also have an impact on the validity of EMG measurements as a ‘gold standard’ upon which the models are based. Deep and sustained forward flexion has been shown to inhibit back extensor muscle activity (Solomonow et al. 2003; Schultz et al. 1985), even though spinal loads (and ostensibly back injury risk) continue to increase. In cases where forward flexion is frequent or sustained, EMG could be showing ‘low exposure’ even when observed, self-reported, and biomechanically-modeled measures of exposure would estimate ‘high exposure’. If this were the case, the mismatch would decrease the ability of observed and self-report-based models to estimate the true exposures. This does not appear to be the case in the current study; flexion of more than 60o accounted for about 5% of the total observed time and was positively related to EMG activity. The quality of exposures assessed via observation and self-report may also influence the observed relationships. Observation-based estimates, although recorded by trained, impartial observers, are still subject to some level of subjective interpretation of each observer and cannot be considered completely objective. Difficulties were encountered when trying to observe workers that were moving throughout the worksite (i.e. maintenance personnel or forklift drivers), resulting in some missing or incomplete observations. In a previous study, the observation technique compared well to direct measures of trunk posture using an inclinometer. A full description of the observation methods, pilot testing and validation testing is reported elsewhere (Village et al. 2007), as is a description of some of the challenges encountered with observation and self-report data collection (Trask et al. 2007). Care was taken during the interview to provide diagrams of the tasks/postures/loads in question, to establish a level of trust that would facilitate candid responses, and not to influence worker responses. Nonetheless, workers expressed difficulty in reporting cumulative exposures such as the ‘total amount of time spent walking during the day’ or the ‘total amount of time spent lifting, lowering, pushing or pulling’. This difficulty is reflected in self-reported durations summing to  63  over 100% of working time. This identifiable misclassification at long durations suggests that there is also unidentifiable misclassification at shorter durations. Conclusion The ability to identify exposure-response relationships is dependent on the quality of the exposure assessment. There are undoubtedly trade-offs between the precision of exposure measurements, their cost, and the number of measurements that can be made. Differences in predictive power between the models should therefore be evaluated within the context of a proposed study’s goals and hypotheses. The results of this study indicate that observation-based estimates of 90th percentile or mean EMG could be expected to perform well, but self-reportbased models, particularly for cumulative EMG, should be treated with caution.  64  References Barriera-Viruet, H., Sobeih, T., Daraiseh, N. and Salem , S. (2006) Questionnaires vs observational and direct measurements: systematic review. Theoret Issues Ergon Sci 7, 261-284. Burdorf, A. (2005) Identification of determinants of exposure: consequences for measurement and control strategies. Occup Environ Med 62, 344-50. Burdorf, A., Govaert, G. and Elders, L. (1991) Postural load and back pain of workers in the manufacturing of prefabricated concrete elements. Ergonomics. 34, 909-918. Burdorf, A. and Sorock, G. (1997) Positive and negative evidence of risk factors for back disorders. Scand J Work Environ Health 23, 243-56. Burstyn, I. and Teschke, K. (1999) Studying the determinants of exposure: a review of methods. Am Ind Hyg Assoc J 60, 57-72. Burstyn, I., Teschke, K. and Kennedy, S.M. (1997) Exposure levels and determinants of inhalable dust exposure in bakeries. Ann Occup Hyg 41, 609-24. Cooper, G. and Ghassemieh, E. (2007) Risk assessment of patient handling with ambulance stretcher systems (ramp/(winch), easi-loader, tail-lift) using biomechanical failure criteria. Med Eng Phys 29, 775-87. Guo, H.R., Tanaka, S., Halperin, W.E. and Cameron, L.L. (1999) Back pain prevalence in US industry and estimates of lost workdays. Am J Public Health 89, 1029-35. Houba, R., Heederik, D. and Kromhout, H. (1997) Grouping strategies for exposure to inhalable dust, wheat allergens and alpha-amylase allergens in bakeries. Ann Occup Hyg 41, 287-96. Jones, T. and Kumar, S. (2007) Assessment of physical demands and comparison of multiple exposure definitions in a repetitive sawmill job: board edger operator. Ergonomics 50, 676-93. Jonsson B (1988) The static load component in muscle work. Eur J Appl Physiol Occup Physiol 57, 305-310. Keir, P.J. and MacDonell, C.W. (2004) Muscle activity during patient transfers: a preliminary study on the influence of lift assists and experience. Ergonomics 47, 296-306. Kerr, M.S., Frank, J.W., Shannon, H.S., Norman, R.W., Wells, R.P., Neumann, W.P. and Bombardier, C. (2001) Biomechanical and psychosocial risk factors for low back pain at work. Am J Public Health 91, 1069-75. Keyserling, W.M. (2000) Workplace risk factors and occupational musculoskeletal disorders, Part 1: A review of biomechanical and psychophysical research on risk factors associated with low-back pain. AIHAJ 61, 39-50. Knibbe, J.J. and Friele, R.D. (1996) Prevalence of back pain and characteristics of the physical  65  workload of community nurses. Ergonomics 39, 186-98. Kumar, S. (1990) Cumulative load as a risk factor for back pain. Spine 15, 1311-6. Lavender, S.A., Conrad, K.M., Reichelt, P.A., Gacki-Smith, J. and Kohok, A.K. (2007) Designing ergonomic interventions for EMS workers, Part I: transporting patients down the stairs. Appl Ergon 38, 71-81. Macfarlane, G.J., Thomas, E., Papageorgiou, A.C., Croft, P.R., Jayson, M.I. and Silman, A.J. (1997) Employment and physical work activities as predictors of future low back pain. Spine 22, 1143-9. Mathiassen, S.E., Burdorf, A. and van der Beek, A.J. (2002) Statistical power and measurement allocation in ergonomic intervention studies assessing upper trapezius EMG amplitude. A case study of assembly work. J Electromyogr Kinesiol 12, 45-57. Mientjes, M.I., Norman, R.W., Wells, R.P. and McGill, S.M. (1999) Assessment of an EMGbased method for continuous estimates of low back compression during asymmetrical occupational tasks. Ergonomics 42, 868-79. Moller, T., Mathiassen, S.E., Franzon, H. and Kihlberg, S. (2004) Job enlargement and mechanical exposure variability in cyclic assembly work. Ergonomics 47, 19-40. Moore, J.S. and Garg, A. (1995) The Strain Index: a proposed method to analyze jobs for risk of distal upper extremity disorders. Am Ind Hyg Assoc J 56, 443-58. Morrison, J., Robinson, D.G., Nicol, J.J., Roddan, G., et al. ( 1999) A biomechanical approach to evaluating the health effects of repeated mechanical shocks. Proc. Human Factors and Med. Panel, RTO-MP-20. Published by NATO Research and Technology Organisation. Myers, A.H., Baker, S.P., Li, G., Smith, G.S., Wiker, S., Liang, K.Y. and Johnson, J.V. (1999) Back injury in municipal workers: a case-control study. Am J Public Health 89, 1036-41. Neumann, W.P., Wells, R.P., Norman, R.W., Andrews, D.M., Frank, J., Shannon, H.S. and Kerr, M.S. (1999) Comparison of four peak spinal loading exposure measurement methods and their association with low-back pain. Scand J Work Environ Health 25, 404-9. Nieuwenhuijsen, M.J., Gordon, S., Harris, J.M., Tee, R.D., Venables, K.M. and Newman Taylor, A.J. (1995) Variation in rat urinary aeroallergen levels explained by differences in site, task and exposure group. Ann Occup Hyg 39 , 819-25. Nordander, C., Balogh, I., Mathiassen, S.E., Ohlsson, K., Unge, J., Skerfving, S. and Hansson, G.A. (2004) Precision of measurements of physical workload during standardised manual handling. Part I: surface electromyography of m. trapezius, m. infraspinatus and the forearm extensors. J Electromyogr Kinesiol 14, 443-54. Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clinical Biomechanics 13 (8):561-573.  66  Schultz, A.B., Haderspeck-Grib, K., Sinkora, G. and Warwick, D.N. (1985) Quantitative studies of the flexion-relaxation phenomenon in the back muscles. J Orthop Res 3, 189-97. Siemiatycki, J., Dewar, R. and Richardson, L. (1989) Costs and statistical power associated with five methods of collecting occupation exposure information for population-based case-control studies. Am J Epidemiol 130, 1236-46. Solomonow, M., Baratta, R.V., Banks, A., Freudenberger, C. and Zhou, B.H. (2003) Flexionrelaxation response to static lumbar flexion in males and females. Clin Biomech (Bristol, Avon) 18, 273-9. Spielholz, P., Silverstein, B., Morgan, M., Checkoway, H. and Kaufman, J. (2001) Comparison of self-report, video observation and direct measurement methods for upper extremity musculoskeletal disorder physical risk factors. Ergonomics 44, 588-613. Trask, C., Teschke, K., Village, J., Chow, Y., Johnson, P., Luong, N. and Koehoorn, M. (2007) Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility. Am J Ind Med 50, 687-696. van der Beek, A.J. and Frings-Dresen, M.H. (1998) Assessment of mechanical exposure in ergonomic epidemiology. Occup Environ Med 55, 291-9. Village, J., Frazer, M., Cohen, M., Leyland, A., Park, I. and Yassi, A. (2005) Electromyography as a measure of peak and cumulative workload in intermediate care and its relationship to musculoskeletal injury: an exploratory ergonomic study. Appl Ergon 36, 609-18. Village, J., Trask, C., Luong, N., Chow, Y., Johnson, P., Koehoorn, M. and Teschke, K. (2007) Development and Evaluation of an Observational Back Exposure Sampling Tool (Back-EST) for Work-Related Back Injury Risk Factors. Submitted to: Applied Ergonomics; Submission number JERG-D-07-00096. Waters, T.R., Putz-Anderson, V., Garg, A. and Fine, L.J. (1993) Revised NIOSH equation for the design and evaluation of manual lifting tasks. Ergonomics 36, 749-76. Wells, R., Norman, R., Neumann, P., Andrews, D., Frank, J., Shannon, H. and Kerr , M. (1997) Assessment of physical work load in epidemiologic studies: common measurement metrics for exposure assessment. Ergonomics 40, 51-61. Workers' Compensation Board of British Columbia. Annual Report: Statistical Supplement. Richmond, BC. 2002. WorkSafeBC. WorkSafeBC Annual Report: Statistical Supplement. 2005. 2006. You, X.I., Senthilselvan, A., Cherry, N.M., Kim, H.G. and Burstyn, I. (2007) Determinants of airborne concentrations of volatile organic compounds in rural areas of Western Canada. J Expo Sci Environ Epidemiol 18(2):117-28.  67  Chapter 6: How long is long enough? Selecting efficient sampling durations for low-back EMG assessment4  Introduction Mathiassen et al (2003) observed that long-term or ‘whole-day recordings of exposure…are rare in ergonomics’ yet the implied goal of an ergonomic exposure measurement is to find an exposure ‘that is representative for the individual during an extended period of time’. The majority of ergonomic exposure assessments rely on a sampling strategy of selecting the task(s) or element(s) deemed to be most demanding, and assessing risk based on this data (ACGIH, 2001; Moore and Garg, 1995; McAtamney and Nigel Corlett, 1993; Hignett and McAtamney, 2000; Waters et al. 1993). Data using this approach have been extrapolated to summarize a full day (Norman et al. 1998) and a working lifetime of exposure (Kumar, 1990) despite the assumptions about the representativeness of the data for these longer time periods (Gold et al. 2006). This illustrates how measurement goals are often at odds with the constraints of real-life data collection. Short, task-based or worst-case exposure assessments minimize data collection time and equipment resources, but may misrepresent exposures since they neglect the cumulative load as well as temporal variations throughout the work day that influence average and peak exposures. Failing to gather representative exposure measurements can result in over- or underestimation risk (McAtamney and Nigel Corlett, 1993) and the development of interventions that result in little or no reduction in risk (Rappaport et al. 1993). However, few ergonomic exposure assessment tools explicitly state when and how to sample exposures to provide a representative sample (Gold et al. 2006). While an appropriate sampling duration might be estimated for cyclic work tasks such as on an assembly line, non-cyclic work is much harder to estimate. It is not clear whether assessments ranging from minutes to hours are representative of longer term exposures such as those over a full-shift (Winkel and Mathiassen, 1994). In order to begin to think about employing a sampling strategy beyond assessing selected peak or ‘worst case’ exposures, industrial ergonomists and occupational researchers also need exposure assessment tools that are efficient, delivering maximum information for a minimum investment of resources (Burdorf and van Riel, 1996). Electromyography (EMG) is increasingly used to assess the exposures during simulated tasks and actual work (Village et al. 2005; Lavender et al. 2007; Marshall and Burnett, 2004). Portable EMG is an appropriate measurement method when privacy concerns preclude video or direct observation of work (as in healthcare (Village et al. 2005)), as well as during very active work when movement about a worksite would limit the ability to observe or video record postures (Trask et al. 2007). As such, EMG is feasible and practical for collecting multiple data samples over a long term work window. Both shoulder and low-back EMG studies (Hui et al. 2001) show a variety of sampling durations, (Keir and MacDonell, 2004; Mathiassen et al. 2002) typically on the order of minutes but some EMG studies have sampled for 1-2 hours, (Rissen et al. 2000; Nakata et al. 1993; Lundberg et al. 1999) or even a whole shift (Village et al. 2005). 4  A version of this chapter has been published. Trask, C., Koehoorn, M., Village, J., Johnson, P., Teschke, K. (2008) How long is long enough? Evaluating sampling durations for low-back EMG assessment. Journal of Occupational and Environmental Hygiene. 5(10) 664 -670  68  The purpose of this study is to compare a variety of low-back EMG exposure metrics measured over a full-shift with the same metrics sampled over shorter durations to evaluate these durations as representative measures of exposure. Sampling durations were tested on a variety of jobs with both non-cyclical and some cyclical tasks in a range of heavy industries. It is hoped that the results will allow for identification of an optimal sampling duration combining both representative accuracy and measurement efficiency.  Methods Worker recruitment and data collection This analysis was undertaken as part of a larger study examining the risk factors for low back injury in five heavy industries: transportation, warehousing, forestry, wood and paper products, and construction (Trask et al. 2006). As seen in Table 6.1, a wide range of jobs were represented in the study sample. The tasks and task patterns observed were equally diverse. The sample included static, unvaried tasks, as seen in delivery truck drivers who drove continuously with little or no manual materials handling, as well as dynamic, monotonous tasks as seen in a sawmill work repeatedly stacking wood off a conveyor belt. Work that was both dynamic and varied was also included via maintenance workers who rarely repeated tasks within a day. There was substantial variation within- and between-industries with respect to the tasks performed and the degree of repetitiveness or serialization involved (Trask et al. 2008). Table 6.1: Industries and job titles included in the UBC Back Study Industry  Job titles included in the UBC Back Study  Construction  Asphalt worker(3), Bricklayers (2), Cabinet maker (1), Construction carpenter (8), Construction labourer (8), Construction supervisor (4), Construction trades (unspecified) (3), Floor Layer (1), Forklift operator (1)  Forestry  Construction labourer (2), Boomman (14), Faller (6), Heavy-duty equipment mechanic (4), Heavy equipment operator (5), Helicopter pilot (1), Logging Machinery Operators(4), Saw filer (2), Truck driver (1)  Wood Products  Cabinet maker(8), Forklift operator(10), Log chipper/grinder (3), Lumber grader, puller (6), Papermaking and Coating Control Operator(9)  Warehousing  Construction trades (unspecified) (5), Forklift operator(28), Grain elevator worker(3), Warehouse person(3)  Transportation  Construction supervisor (1), Forklift operator (2), Heavy equipment operator (3) Truck driver(7), Warehouse person(2), Air Transport Ramp Attendants(6), Automotive mechanic(8), Bus cleaner (2), Bus driver(2), Ferry worker(9), Storekeepers and Parts Clerks(4)  WorkSafeBC (The Workers’ Compensation Board of British Columbia) identified a random sample of 50 employees in the five study industries who had an accepted workers’ compensation claim for back injury in 2001, who resided in the study area, and who agreed to have their  69  information released to researchers. These workers were contacted first by letter, then by telephone and invited to participate. The employers of participating workers were contacted to gain permission to conduct measurements at the worksite, and to recruit an additional one to four co-workers. Participation was entirely voluntary and signed consent was received from each participant. Human subject procedures were approved by the University of British Columbia’s Behavioural Research Ethics Board. Full-shift EMG measurements were made for 103 workers with a second day of measurement for 35 workers (34%) for a total of 138 worker days. EMG was measured using a portable data collection system with on-board memory (ME3000P4/ME3000P8, Mega Electronics, Finland) and disposable Ag-AgCl electrodes (Blue Sensor N-00-S, Ambu, Denmark). Electrodes were placed over the erector spinae at approximately the level of L4/L5, with a 20 mm inter-electrode spacing and a ground electrode and preamplifier placed on the posterior aspect of the iliac crest. Signals were collected at 1000 Hz and filtered internally using an 8-500 Hz band pass filter. Averaged values were data logged at 10 Hz. During work breaks, data from the portable system were downloaded onto a laptop computer. EMG data were collected for the full shift excluding breaks (5.5 to 10.3 hours of working time, mean = 6.32, sd = 1.32). A sub-maximal reference contraction was employed to calibrate EMG data collected during the shift. The reference contraction involved a static 45o forward trunk flexion while holding an 11.5 kg weight and was performed twice for 5-seconds at the beginning of each shift. Set-up time and calibration took approximately 20 minutes. All EMG data collected during the shift was expressed as a percentage of this reference contraction (%RC). A set of summary statistics or ‘exposure metrics’ was calculated for each individual’s work shift data. Mean and 50th percentiles were included as a measure of central tendency or exposure level, while standard deviation and percentiles from the tails of the amplitude probability distribution function (1st, 5th, 10th, 90th, 95th, and 99th percentiles) were included to represent the range of exposure. Most of these exposure metrics had a relatively normal distribution. However, the distribution of the standard deviations was log-normal, therefore standard deviations were log-transformed (base e) before performing any parametric tests. Due to a non-negligible number of 0 values, the 1st, 5th, and 10th percentiles distributions were left-censored. To account for this, the 0 values of these exposure metrics were replaced by the limit of detection divided by √2 ; the limit of detection was the lowest resolution of the EMG recorder (2.95 µV) divided by the individuals’ reference contraction EMG measurement. Re-sampling EMG data The measured full-shift data were used to derive data for shorter measurement durations via a posteriori re-sampling within single full-work shifts, similar to the post-collection within-shift re-sampling described by Mathiassen et al. (2003) Within each shift, the data were re-sampled for shorter durations: a 4-hour duration, 2-hour duration, 1-hour duration, 10-minute duration, and 2-minute duration. Such short duration EMG measurements have been made in the past, for example in assessments of patient transfer tasks in health care (Keir and MacDonell, 2004). However, the instrumentation and calibration time required for EMG make 10-minute and 2minute samples undesirable in most cases. These durations were included in the present study not as a suggestion that only 2 minutes of EMG should be collected in the workplace, but rather  70  to further a theoretical goal of finding the shortest required sampling time and to identify any potential ‘point of diminishing returns’. For each duration, a randomly-selected start time was used to resample the full-shift data. This protocol was conducted for each worker-day, where one sample of each of the shorter durations was made within the full-shift, then replaced. Random start times with replacement meant that the data may or may not overlap in the selected durations. This process resulted in six sets of data for each worker as illustrated in Figure 6.1. Summary statistics or ‘exposure metrics,’ were calculated for each of these durations for all 138 worker-days. For the 35 workers with two measured days, the full-shift exposure metrics were also compared to the summary statistics from the combined data of the two days. This was done to investigate the representativeness of multi-day sampling durations on a subset of workers. 10 minutes 2 minutes  2 hours  4 hours  1 hours  14  EMG value  12 10 8 6 4 2 0 1  4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 time  Figure 6.1: Schematic of the resampling process for shorter sampling durations  Comparing sampling lengths Agreement between the full-shift exposure metrics and the metrics for the re-sampled, shorter durations was investigated in several ways. Pearson’s correlation coefficients were used to determine the strength of relationship between the metrics for full-shift and re-sampled durations. Repeated measures ANOVA was used to test for significant differences in results between full shift and shorter sampling lengths. Findings were considered statistically significant at the p<0.05 level. The error or ‘absolute difference’ as well as ‘percentage difference’ (the absolute error divided by the full-shift value) were calculated to determine the level of deviation from the full-shift value for each sampling duration. Agreement between the full-shift exposure metrics and those from each of the shorter durations was also assessed by calculating the bias and limits of  71  agreement as described by Bland and Altman (Bland and Altman, 1986). The bias is calculated as the mean difference between results from full-shift measurements and shorter measurements, divided by the standard deviation of those differences. With perfect agreement between samples of different durations, the bias would be zero. The limits of agreement represent the mean difference plus or minus two standard deviations of the difference. Analyses were performed in SPSS v11.5 (SPSS Inc., Chicago, IL) and SAS v9.1 (SAS Institute Inc., Cary, NC).  Results Exposure estimates All EMG data collected during the shift were expressed as a percentage of the reference calibration contraction (%RC) and all EMG exposure outcomes are in units of %RC. Full-shift exposure metrics were positively and significantly correlated to those estimated in shorter durations (Pearson r ranging from 0.344-0.969) for all but the standard deviation at the 2-minute duration (Pearson r = 0.293). As might be expected, the strength of the correlations between full-shift and re-sampled durations were highest for the 2-day and 4-hour duration (Pearson r 0.969 to 0.854), and tended to decrease with shorter sampling durations. As seen in Table 6.2, the group mean and 50th percentile do not vary substantially by sampling duration. However, shorter sampling durations tended to overestimate the full-shift exposure for percentiles below the median and underestimate the full-shift exposure at percentiles above the median. That is, longer measurement periods are better able to capture the extremes of exposure. st th ANOVA revealed significant differences between full shift and 2-minute values for the 1 , 5 , 10th, 95th and 99th percentiles; and between full shift and 10-minute values for the 5thpercentile. This corresponds with the slight (though not monotonic) trend for decreasing exposure range with decreasing duration of sampling, as observed in the standard deviation. Table 6.2: Muscle activity expressed as a percentage of the reference contraction (%RC): means for each exposure metric at different sampling durations Two shifts full shift 4hrs 2 hrs 1hr 10 min (n = 35) 41.8 38.9 38.5 37 35.5 37 Mean † 45.4 40.4 41.3 38.1 34.5 33.8 sd 1.07 1.31 1.56 1.70 2.61 1st %ile 1.50 3.56 4.14 3.93 4.26 6.13* 5th %ile 4.40 th 6.46 6.54 6.53 7.41 8.77 10 %ile 7.27 th 27 25 26 25 28 50 %ile 28.9 90th %ile 82.1 78.7 78.0 74.2 77.0 87.0 95th %ile 107 104 100 96.2 99.0 113 99th %ile 187. 187 167 164 155 209 * indicates a significant difference between shorter duration metric and full-shift metric (p<0.05) † The geometric mean (antilog of mean of ln-transformed values) Exposure metric  2 min 35.8 25.8 4.48* 7.85* 9.75* 28 70.4 89.1* 141*  72  30.0  160.0  25.0  140.0 20.9  20.0 14.8  15.0 10.3  10.0  9.3  9.1 7.1  6.2  5.3  5.0  4.2 2.9 1.6  1.0 1.4  1.8  2.4  5.5  4.7 3.0 1.4  percentage error in %RC  absolute error in %RC  Absolute and percentage difference The absolute and percentage differences between the full-shift and other sampling durations for th th th the mean and the 10 , 50 , and 90 percentiles are presented in Figure 6.2.  120.0 100.0 80.0  75.5  60.0  46.9 39.9 39.7 39.5  40.0  30.1  20.0  2.4  4.4  0.0  41.1 28.3  25.0 17.6 12.1 6.1  18.8 13.2 7.8  2.9  18.3 11.5 7.6  0.0 mean  10 %ile  full shift - 2 shifts full shift - 1 hours  50 %ile  full shift - 4 hours full shift - 10min  90%ile full shift - 2 hours full shift - 2min  mean  10 %ile  full-shift - 2 shifts full shift - 1 hours  50 %ile  full shift - 4 hours full shift - 10min  90%ile  full shift - 2 hours full shift - 2min  Figure 6.2: Absolute and percentage difference from full-shift value at each sampling duration  An increase in error with shorter sampling durations is consistent across all exposure metrics for both absolute and percentage difference. Note that there is a tradeoff between absolute and percentage error; i.e. as the magnitude of one increases the magnitude of the other decreases. In the case of 10th percentile EMG, errors for the 2-shift, 4-hour and 2-hour durations are very similar but there are notable increases for the 1-hour, 10-minute and 2-minute durations. Bias and limits of agreement Bias did not exceed 1 %RC for mean, 10th percentile, and 90th percentile metrics at any of the sampling durations (Table 6.3). Half the biases were negative, indicating an over-prediction by the shorter sampling duration. All 10th percentile sampling durations and all 4-hour metrics overpredicted the full shift exposures. The limits of agreement, in %RC, were always smallest for th th 10 percentile metrics, followed by mean and then 90 percentile. The width of the limits of agreement increased consistently with decreasing sampling duration. For the subset of workers with 2 measured shifts, the bias was large and positive, while the width of the limits of agreement were small.  Discussion The central question raised in this study was: How long should one sample EMG in a single day to represent full-shift exposure? Regarding EMG measures, the extensive set-up and calibration time make the marginal cost of additional measured time small, and so full-shift sampling is ‘feasible’ (Macfarlane et al. 1997). However, shorter measurement durations are attractive because more assessments could be made during the same day. This allows for cost-and timeefficient measurements of more individuals, vital for epidemiological studies that compare the health effects between individuals.  73  Table 6.5: Mean and standard deviation of differences, bias, and limits of agreement between full shift and partial shift exposure metrics Exposure metric  Assessment of agreement  mean  Mean difference Bias Upper limit of agreement Lower limit of agreement Width of limit of agreement Mean difference Bias Upper limit of agreement Lower limit of agreement Width of limit of agreement Mean difference Bias Upper limit of agreement Lower limit of agreement Width of limit of agreement  10th %ile  90th %ile  full shift vs 4 hours  full shift vs 2 hours  full shift vs 1 hours  full shift vs 10min  full shift vs 2 min  full shift vs 2 shifts (n=35)  -0.181  0.681  1.88  0.2692  1.45  1.65  -0.04  0.09  0.20  0.02  0.08  0.90  7.98  16.26  21.02  28.27  36.65  5.31  -8.34  -14.90  -17.26  -27.73  -33.75  -2.02  16.32  31.16  38.28  56.00  70.40  7.33  -0.4266  -0.2325  -1.08  -2.4  -3.64  1.01  -0.14  -0.07  -0.25  -0.32  -0.45  0.83  5.75  6.67  7.54  12.44  12.66  3.45  -6.61  -7.13  -9.70  -17.24  -19.94  -1.42  12.36  13.80  17.24  29.68  32.60  4.87  -0.204  2.12  5.27  3.97  10  2.44  -0.02  0.15  0.25  0.14  0.28  0.88  17.30  30.92  46.63  60.91  80.86  7.99  -17.70  -26.68  -36.09  -52.97  -60.86  -3.11  35.00  57.60  82.72  113.88  141.72  11.10  As might be expected, absolute error, percentage error, and the limits of agreement increase with shorter sampling durations. Of the within-shift samples, the 4-hour sampling duration most closely reflected full-day measurements of both exposure level and exposure range in this study. A 4-hour sampling duration has been suggested by other authors. For example, a study of sampling strategies for noise measurements, Brunn et al state that if exposure variability is uniform throughout the 8-hr shift, a 4-hr sample can provide a “good estimate” with “negligible loss of precision” (Brunn et al., 1986). Given the scale of the measurements, the limits of agreement are non-negligible even at the 4-hour duration. At all sampling durations, absolute error, bias and limits of agreement were lowest for the 10th percentile EMG metric, while percentage errors were lowest for the 90th percentile. The higher bias and limits of agreement with 90th percentile EMG would suggest that longer durations need to be used for this measure. However, if the limits of agreement are considered as a percentage of the measurement range (analogous to percentage error) the agreement of mean, 10th and 90th percentile is closer. The relative error may be more relevant than the absolute error for many applications. In general, selecting the optimal sampling length may depend on the exposure metrics of choice for the  74  study. For example, the means and 50th percentiles (i.e. measures of central tendency or average snapshots of exposure) of shorter durations did not vary substantially from the full-shift measures. However, percentage errors for the mean ranged from 4.4% for 2 shifts to 30% at 2 minutes. Clearly this represents a substantial mismatch in exposure estimates at shorter durations. Differences for shorter durations were especially evident in the metrics representing exposure variability: percentiles below the median tended to overestimate the full-shift estimate, and percentiles above the median tended to underestimate it. Correlations for the more extreme percentiles (5th, 95th) and standard deviation also tended to be lower than metrics representing th central tendency such as mean and 50 percentile. This makes intuitive sense, as the distribution from which a shorter duration is estimating has less opportunity to capture extreme values. This is particularly important when trying to capture peak values, as short durations may not th encounter a sufficient number of peaks to adequately represent peak exposures. For 10 percentile EMG, absolute errors are much lower than for 90th percentile, while percentage errors th are higher. This occurs because the 10 percentile values tend to be much lower, so a small absolute error can nonetheless be a large proportion of the measured value. To our knowledge, this is the first study to compare exposure metrics estimated using different sampling durations using EMG. The analyses presented here were made possible because we had full-shift measurements from a field study of a large number of employees in 5 heavy industries. In contrast, many EMG studies are conducted on small samples of 10-20 workers for short durations likely due to time and cost limitations (Mathiassen et al. 2002). Examples include a study of grocery cashiers that monitored 8 employees for 74 - 166 seconds (mean 104 seconds) (Lannersten and Harms-Ringdahl, 1990), a study of 12 dentists with shoulder EMG measurements of 20 minutes (Akesson et al. 1997), and a study of 16 board edging operators which measured 18 seconds for each of 2-5 tasks (Jones and Kumar, 2007). This study sought an answer to the ‘how long’ question both within the context of a single shift, and also, for a sub-set of the data, between shifts. Comparing the full-shift and two-shift measurements can be considered an example of how well a single day’s measurement represents ‘typical’ exposure. Given that the representativeness decreases with durations less than a full shift, it follows that the exposures of a single day may not represent a worker’s overall exposure. This is of particular interest in chronic health outcomes such as back injury, which is thought to develop according to a cumulative tissue damage model (Marras, 2003). Typical hygiene studies of airborne exposures find little autocorrelation between measurements on consecutive days, but variability does increase with the interval between measurements (Burstyn and Teschke, 1999), and there are systematic changes when measures are taken for longer than a year (Peretz et al. 1997). In the current study, the second day’s measurement was taken on average 93 days after the first day (range 1-439 days), so some degree of production and seasonal variation is anticipated to be included in the repeated samples. This may explain why the bias was highest even though errors were lowest between full-shift and 2-shift measures of exposure. Although measuring a full-shift does yield lower error and bias than sampling fractions of a shift, it is important to note that even a full shift is not a perfect measure of ‘typical’ exposure. Many work characteristics are likely to play a role in the representativeness of shorter samples: whether the work is cyclical or non-cyclical, the cycle length, and even qualities as broad as work tasks, job title, and industry. For example, cyclical work with short cycle times may allow  75  shorter sampling durations than would long-cycle or non-cyclical work. Related research has employed statistical modeling to identify workplace and worker characteristics which could be identified a priori to identify sources of work variability and help guide the choice of measurement duration.(Trask et al. 2008) Conclusion When comparing a full-shift measurement to two full-shifts, the width of limits of agreement was far smaller, although the bias was higher. As sampling durations decreased from a full-shift to a few minutes, the absolute error, percentage error, and limits of agreement for exposure estimates deviate more from full day estimates even though bias is very low. Estimates of mean and 90th percentile exposure average 8% error for 4-hour and 14% error for 2-hour durations. This loss of accuracy may be acceptable when weighed against the tradeoff of acquiring more subjects. Sampling for 2 or 4 hours provides reasonable estimates of the full-shift measures of central tendency, while sampling durations of 1 hour or less seem likely to produce very large errors over all exposure metrics, and particularly for the range and peak exposures. Depending on the purpose of measurement and the level of detail required, 4 or even 2 hours should be long enough to reasonably estimate exposure, whereas shorter durations might be used with caution.  76  References ACGIH (American Conference of Governmental Industrial Hygienists) (2001) Hand activity level. TLVs and BEIs - Threshold Limit Values for Chemical Substances and Physical Agents pp. 110-112. Cincinnati, OH: ACGIH . Akesson, I., Hansson, G.A., Balogh, I., Moritz, U. and Skerfving, S. (1997) Quantifying work load in neck, shoulders and wrists in female dentists. Int Arch Occup Environ Health 69, 46174. Bland, J.M. and Altman, D.G. (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 8, 307-310. Brunn IO, Campbell JS, Hutzel RL. (1986) Evaluation of occupational exposures: a proposed sampling method. Am Ind Hyg Assoc J; 47(4):229-35. Burdorf, A. and van Riel, M. (1996) Design of strategies to assess lumbar posture during work . Int J Ind Ergon 18, 239-249. Burstyn, I. and Teschke, K. (1999) Studying the determinants of exposure: a review of methods. Am Ind Hyg Assoc J 60, 57-72. Gold, J.E., Park, J.S. and Punnett, L. (2006) Work routinization and implications for ergonomic exposure assessment. Ergonomics 49, 12-27. Hignett, S. and McAtamney, L. (2000) Rapid entire body assessment (REBA). Appl Ergon 31, 201-5. Hui, L., Ng, G.Y., Yeung, S.S. and Hui-Chan, C.W. (2001) Evaluation of physiological work demands and low back neuromuscular fatigue on nurses working in geriatric wards. Appl Ergon 32, 479-83. Jeebhay, MF., Robins, TG., Seixas, N., Baatjies, R., George, DA., Rusford, E., Lehrer, SB. and Lopata, AL. (2005) Environmental exposure characterization of fish processing workers. Ann Occup Hyg 49, 423-37. Jones, T. and Kumar, S. (2007) Assessment of physical demands and comparison of multiple exposure definitions in a repetitive sawmill job: board edger operator. Ergonomics 50, 676-93. Keir, P.J. and MacDonell, C.W. (2004) Muscle activity during patient transfers: a preliminary study on the influence of lift assists and experience. Ergonomics 47, 296-306. Kumar, S. (1990) Cumulative load as a risk factor for back pain. Spine 15, 1311-6. Lannersten, L. and Harms-Ringdahl, K. (1990) Neck and shoulder muscle activity during work with different cash register systems. Ergonomics 33, 49-65. Lavender, S.A., Conrad, K.M., Reichelt, P.A., Kohok, A.K. and Gacki-Smith, J. (2007) Designing ergonomic interventions for emergency medical services workers--part III: Bed to 77  stairchair transfers. Appl Ergon 38, 581-9. Lundberg, U., Dohns, I.E., Melin, B., Sandsjo, L., Palmerud, G., Kadefors, R., Ekstrom, M. and Parr, D. (1999) Psychophysiological stress responses, muscle tension, and neck and shoulder pain among supermarket cashiers. J Occup Health Psychol 4, 245-55. Macfarlane, G.J., Thomas, E., Papageorgiou, A.C., Croft, P.R., Jayson, M.I. and Silman, A.J. (1997) Employment and physical work activities as predictors of future low back pain. Spine 22, 1143-9. Marras, W.S. (2003) The case for cumulative trauma in low back disorders. The Spine Journal 3, 177-179. Marshall, R.N. and Burnett, A.F. (2004) A kinematic, kinetic and electromyographic comparison of stooped sheep shearing techniques and shearing with a sheep manipulator. Appl Ergon 35, 137-45. Mathiassen, S.E., Burdorf, A. and van der Beek, A.J. (2002) Statistical power and measurement allocation in ergonomic intervention studies assessing upper trapezius EMG amplitude. A case study of assembly work. J Electromyogr Kinesiol 12, 45-57. Mathiassen SE, Burdorf A, van der Beek AJ, Hansson GA. (2003) Efficient one-day sampling of mechanical job exposure data--a study based on upper trapezius activity in cleaners and office workers. AIHA J; 64(2):196-211. McAtamney, L. and Nigel Corlett, E. (1993) RULA: a survey method for the investigation of work-related upper limb disorders. Appl Ergon 24, 91-9. Moore, J.S. and Garg, A. (1995) The Strain Index: a proposed method to analyze jobs for risk of distal upper extremity disorders. Am Ind Hyg Assoc J 56, 443-58. Nakata, M., Hagner, I.M. and Jonsson, B. (1993) Trapezius muscle pressure pain threshold and strain in the neck and shoulder regions during repetitive light work. Scand J Rehabil Med 25, 131-7. Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clin Biomech 13 (8):561-573. Peretz, C., Goldberg, P., Kahan, E., Grady, S. and Goren, A. (1997) The variability of exposure over time: a prospective longitudinal study. Ann Occup Hyg 41, 485-500. Rappaport, S.M., Kromhout, H. and Symanski, E. (1993) Variation of exposure between workers in homogeneous exposure groups. Am Ind Hyg Assoc J 54 , 654-62. Rissen, D., Melin, B., Sandsjo, L., Dohns, I. and Lundberg, U. (2000) Surface EMG and psychophysiological stress reactions in women during repetitive work. Eur J Appl Physiol 83, 215-22.  78  Trask, C., Koehoorn, M., Morrison, J. and Teschke, K. (2008) Optimizing Sampling Strategies: Components of Low-Back EMG Variability in Five Heavy Industries. Submitted to: Occup Env Med Trask C, Koehoorn M, Village J, Johnson P and Teschke K. (2006) Modeling determinants of working exposures and exposure variability. IEA2006, 16th World Conference on Ergonomics. Maastricht, the Netherlands Trask, C., Teschke, K., Village, J., Chow, Y., Johnson, P., Luong, N. and Koehoorn, M. (2007) Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility. Am J Ind Med 50, 687-696. Village, J., Frazer, M., Cohen, M., Leyland, A., Park, I. and Yassi, A. (2005) Electromyography as a measure of peak and cumulative workload in intermediate care and its relationship to musculoskeletal injury: an exploratory ergonomic study. Appl Ergon 36, 609-18. Waters, T.R., Putz-Anderson, V., Garg, A. and Fine, L.J. (1993) Revised NIOSH equation for the design and evaluation of manual lifting tasks. Ergonomics 36, 749-76. Winkel, J. and Mathiassen, S.E. (1994) Assessment of physical work load in epidemiologic studies: concepts, issues and operational considerations. Ergonomics 37, 979-88.  79  Chapter 7: Optimizing sampling strategies: components of low-back EMG variability in five heavy industries5 Introduction Occupational back injury is an expensive and prevalent problem (Guo, 2002). In order to reduce the occurrence of back injuries, researchers need a better understanding of exposure response relationships in the workplace. Previous studies of exposure response relationships have been limited by the quality of the exposure assessment (Neumann et al. 1999), with few using direct exposure measurement despite the fact that it is generally preferred to more subjective measures (Houba et al. 1997; Burdorf, 1992). Unfortunately, direct measurement at worksites can be expensive and difficult (Trask et al. 2007), so researchers need to use their measurement resources efficiently in order to elucidate the relationships between exposures and the back injury response. If there is imprecision in exposure measurement, then there is attenuation of the exposure response relationship (i.e., decrements in both the slope and the correlation between exposure and response). The potential attenuation due to imprecision in exposure measurement can be estimated based on the exposure data, even when the response is not measured and there is no explicit exposure-response relationship (Brunkreef et al. 1987; Tielemans et al. 1998). Awkward postures and manual materials handling (MMH) (e.g. lifting, pushing, pulling) are leading risk factors for the onset of low back disorders (Keyserling, 2000; Burdorf and Sorock, 1997). Electromyography (EMG) provides a way to estimate overall physical exposure arising from MMH and posture since these factors contribute to low-back muscle activity. Peak exposures are not the only metric for physical exposures; cumulative exposure (Norman et al. 1998; Seidler et al. 2001) as well as frequency and duration of exposure (Winkel and Mathiassen, 1994) are thought to be important in the development of back injury. Since EMG is suitable for full-shift measurement, it can assess all these dimensions of exposure simultaneously th using various metrics such mean, 90 percentile, and cumulative exposure. To determine an efficient sampling strategy, researchers need to consider how many individuals to measure, whom to measure (where ‘whom’ can mean from which job or company or industry), and how many measurements to take per person. Very little information exists on how to select these parameters. The magnitude and sources of variability can help inform these decisions. For example, if within-worker variability is high, i.e., a worker’s exposure varies a great deal from day to day, more repeated measurements of each worker will be needed (Waddell, 1996). Grouping schemes based on job, company, or industry, or other factors have also been used to help make sampling strategies more efficient. A grouping scheme that applies the mean exposure of all measurements within a working group to all workers in that group offers the benefit of a Berkson error structure, in which the attenuation of an exposure-response relationship is less than when each individual is assigned the mean of their own exposure measurements (Armstrong et al., 1990; Seixas and Sheppard, 1996). Residual classical error and exposure-response attenuation can be further minimized when between-group variability is large compared to between-worker (within-group) variability (Kromhout et al. 1995; Seixas and Sheppard, 1996; Tielemans et al. 1998). 5  A version of this chapter has been submitted for publication. Trask, C., Teschke, K., Morrison, J., Koehoorn, M. (2007) Optimizing Sampling Strategies: Components of Low-Back EMG Variability in Five Heavy Industries.  80  It is generally agreed that a pilot study should be conducted to get an estimate of variance components prior to a wholesale sampling campaign (Brunekreef et al. 1987; Burdorf, 1995b; Burdorf and van Riel, 1996; Burdorf, 2005; Loomis and Kromhout, 2004). Multiple methods have been proposed as ways to evaluate pilot data. In 1987 Brunekreef et al used the ‘variance ratio’ (within-worker variance divided by the between-worker variance) to estimate the bias or attenuation in the regression coefficients that would result from using the exposure measurements from the study (Brunekreef et al. 1987). For individual-based strategies, Kromhout et al. (1997, 1995) proposed maximizing the ratio of between-worker variability to the sum of within and between-worker variability (the contrast between workers) and analogously for group-based strategies, between-group variability to the sum of within and between-group variability (the contrast between groups) (Burdorf, 1995a; Burdorf and van Riel, 1996; Loomis and Kromhout, 2004; Burdorf, 2005). In ergonomic epidemiology, Burdorf developed formulae to determine study power and the number of individuals versus the number of repeats on individuals to measure, based on withinand between-worker variability in postural exposure (Burdorf, 1995a). Since then, statistical methods and sample size estimation equations using the components of variance have been refined for assessment of the attenuation of exposure-response relationships and optimization of exposure grouping for epidemiological studies (Seixas and Sheppard, 1996; Tielemans et al. 1998). Reports of the components of variance, the effects of grouping strategy, and estimates of exposure-response attenuation have been published for many exposures (albeit not for ergonomic exposures). For example, magnetic field exposure (Kromhout et al. 1995), carbon black (van Tongeren et al. 1997; van Tongeren et al. 1999), wood dust (Teschke et al. 2004), flour dust (Houba et al. 1997), fish proteins and antigens (Jeebhay et al. 2005), cement dust (Mwaiselage et al. 2005), rubber plant exposures (Vermeulen and Kromhout, 2005), and simulated data (Seixas and Sheppard, 1996). Using the components of variability to optimize sampling and grouping strategies has not been reported for low-back EMG studies in heavy industry. Given that research budgets are limited, a more informed exposure assessment design would enhance understanding of back injuries and their prevention. Decisions about where to focus measurement resources necessarily involve some tradeoffs, particularly between measuring more individuals or more repeats of individuals. The current study endeavours to address the following research questions related to planning a sampling strategy for low back EMG exposures in five heavy industries: 1) What are the components of variance (within-worker, between-worker, and between group) when using different grouping schemes and different EMG metrics? 2) Given the components of variance observed in this study, what level of attenuation can be expected in the exposure-response relationships? and 3) How many measures per worker and workers per group are recommended for low-back EMG studies in heavy industry?  Methods Study sample and worker recruitment As part of a larger study, WorkSafeBC (formerly the British Columbia Workers’ Compensation Board) identified a random sample of 50 employees in construction, forestry, transportation, wood products, and warehousing for whom a workers’ compensation claim had been accepted 81  for back injury in 2001 and who agreed to be contacted by researchers. After agreeing to participate, their employers were contacted to gain permission to conduct measurements at the worksite, and to recruit additional co-workers. Human subject procedures were approved by the University of British Columbia’s Behavioural Research Ethics Board and participation was entirely voluntary. Set-up, measurements, and interviews were conducted during regular work time. Measurements were made over a full work shift using electromyography (EMG) of lumbar muscles. EMG data collection Field sampling Full-shift electromyography was performed using a portable data collection system with onboard memory (ME3000P4/ME3000P8, Mega Electronics, Finland) and disposable Ag-AgCl electrodes (Blue Sensor N-00-S, Ambu, Denmark). Electrodes were placed over the erector spinae at approximately the level of L4/L5, with a 20 mm inter-electrode spacing and a ground electrode with preamplifier placed on the posterior aspect of the iliac crest. Signals were collected at 1000 Hz and filtered internally using an 8-500 Hz band-pass filter. Root-meansquare EMG values were data logged at 10 Hz. Data were collected for the full shift excluding breaks (5.5 to 10.3 hours of working time, mean = 6.32 hours) and downloaded from the portable system onto a laptop computer during breaks. Full-shift EMG measurements were made for 103 individual workers. A second measurement was made on 24% of workers (26 subjects) for a total of 133 worker-days. The average time period between measurement days on the same worker was 93 days, range 1 to 439 days. Calibration As the electrode-skin interface is unique to each worker and measurement session, a submaximal reference contraction was employed to calibrate EMG data collected during the shift. The objective was to standardize EMG voltage across all measurement sessions with a common reference contraction. The reference contraction involved a static 45o forward trunk flexion while holding an 11.5 kg weight. The reference contraction was held for five seconds and performed twice at the beginning and end of each shift. All EMG data collected during the shift were expressed as a percentage of this reference contraction (%RC) and all EMG exposure outcomes are in units of %RC. EMG exposure metrics EMG measurements were summarized per worker shift in three exposure metrics: mean (central th tendency); 90 percentile (an estimate of peak value); and cumulative exposure (temporal accumulation). Statistical analysis Workers’ EMG exposure metrics were combined using different grouping schemes to identify the components of variance and determine the optimal grouping strategy for the EMG exposure metrics. Exposures metrics were grouped by job title (24 groups), company (31 groups), industry (five groups), and a post hoc grouping scheme (five groups). The post hoc groupings for  82  each metric were developed by sorting the EMG values of jobs within industries then grouping them into quintiles, with attention paid to natural breakpoints in the distributions. Correlation among exposure metrics was investigated using Pearson product-moment correlation coefficients in SPSS 11.5 (SPSS, Chicago, IL, USA). Summary data were calculated for the four EMG metrics for all data, by industry, by job, and by the post hoc grouping. These exposure summaries and all other analyses were performed using SAS 9.1 (SAS Institute, Cary, North Carolina, USA). Identifying components of variability The relative contribution to total variance of each of the potential components of variance was calculated by developing a series of random effects or ‘null’ models using PROC MIXED in SAS. The first model included only worker as a random effect while subsequent models included worker and job, followed by worker and company, worker and industry, and finally worker and post hoc grouping. Exposure response attenuation The success of each grouping scheme in signaling exposure differences was determined by estimating the attenuation that the grouping scheme would produce in an exposure-response relationship (Brunekreef et al. 1987; Tielemans et al. 1998). Equation 1 was used to calculate attenuation without grouping and includes only the between and within-worker variance components. Equation 2 was applied when there was a grouping scheme and includes the between-group variance component. 2   σ BS  β  β* =  2 2 ( ) σ σ + n WS   BS  Equation 1   Equation 2  β * =    σ BG 2  2 2  ) k σ BG + (σ BS β 2 2 + (σ BS ) k + (σ WS ) kn   Where is the coefficient of the true exposure response relationship and * is the attenuated 2 is the between-group coefficient. The ratio in parentheses is the ‘attenuation factor’, where σ BG 2 2 variance, σ BS is the between-worker (within-group) variance, σ WS is the within-worker variance, k is the number of workers per group, and n is the number of measurements per worker. In cases where n and k were not constant between groups and workers, the average number per group or worker was used. The attenuation factors take values from 0 to 1, and were calculated for all four EMG metrics using the components of variance calculated in the PROC MIXED models. Multiplying by an attenuation value of zero means that the exposure-response relationship is fully attenuated to the point that there is no observable relationship between the variables. An attenuation factor of 1 means that the true exposure-response relationship is preserved without any attenuation.  The number of workers per group (k) required to achieve an attenuation factor no lower than 0.95 was calculated for mean exposure. For these calculations, the number of samples per  83  worker (n) was set to 1.25, 1.5, or 2, representing repeats on 25% of the workers (n = 1.25) to repeats on every worker (n = 2).  Results EMG exposure metrics The mean and 90th percentile EMG metrics were highly correlated (see Table 7.1). The mean EMG metric had a moderate correlation with the cumulative EMG. Table 7.1: Pearson Correlation coefficients between EMG metrics EMG Metric 90 percentile Cumulative th  Mean .936 .488  90th percentile .407  The average exposures over all measurement days for all EMG metrics and for each job and industry are found in Table 7.2. Exposure averages for the post hoc groups are shown in Table 7.3; the post hoc groups had 19 to 36 measurements each. A table of exposures for each post hoc sub-group can be found in Appendix M.  84  Table 7.2: Industry and job averages (over all person-shifts) for three EMG exposure metrics Category All measurements All Construction Asphalt worker Construction carpenter Construction labourer Construction supervisor Other construction trades Other construction* All Forestry Boomman Faller Heavy equipment operator Heavy-duty equipment mechanic Logging machinery operators Saw filer Other forestry* All Wood Products  133  36.6  90th Percentile EMG (%RC) 78.6  24 2 7 6 3 3 3 29  46.9 39.3 50.0 56.6 42.0 32.4 44.8 39.8  99.0 91.0 106.4 115.6 86.6 73.0 92.0 81.0  972.2 913.0 925.3 1102.8 743.7 984.0 1076.3 1158.2  10 3  31.9 67.9  71.4 141.8  1024.2 1401.1  3 4  40.4 41.8  92.0 80.5  1315.2 1435.3  4 2  25.3 38.6  49.9 79.9  1113.0 941.1  3 25  55.0 33.4  84.0 71.5  1040.3 1007.3  5 8 2 5  36.8 25.8 39.7 38.5  77.9 54.2 87.6 80.2  924.5 1080.1 1284.0 875.2  kn  1  Mean EMG 2 (%RC )  Cumulative EMG (%RC*hrs) 984.3  Cabinet maker Forklift operator Log chipper/grinder Lumber grader, puller Papermaking and coating Control operator All Warehousing Forklift operator  5  34.3  77.7  994.9  22 20  36.6 36.0  77.8 77.1  890.9 906.6  Warehouse person All Transportation  2 33  42.3 28.7  84.5 67.5  733.7 885.3  Air transport ramp attendants  6 28.3 70.5 640.8 Automotive mechanic 6 32.9 80.2 550.3 Bus driver 2 14.3 35.6 1062.8 Ferry worker 4 38.1 72.0 1335.9 Heavy equipment operator 2 20.4 52.8 736.3 Storekeepers and parts clerks 3 31.0 68.8 1706.6 Truck driver 6 19.7 50.2 866.6 Warehouse person 2 39.1 89.5 338.3 2 23.1 57.1 710.4 Other transportation* * Jobs with only 1 measurement were combined into ‘other’ categories for the purpose of this table to avoid identifying individuals. All other analyses were preformed with these job titles ungrouped. 1 number of workers/group by the number of measurements/worker; 2 %Reference Contraction;  85  Table 7.3: Summary of EMG exposure metrics for post hoc groupings (approximate quintiles of exposures based on stratification by both industry and job) Mean EMG Quintiles  kn1  (%RC)  90th percentile EMG kn1  (%RC)  Cumulative EMG kn1  (%RC*hrs)  Group 1 29 23.5 24 47.9 24 6,366 Group 2 24 32.2 27 70.2 46 8,866 Group 3 36 36.0 48 79.2 41 10,462 Group 4 25 40.2 15 88.9 13 13,485 Group 5 19 58.4 19 122.3 9 15,806 1 number of workers per group multiplied by the number of measurements per worker (total number of measurements per group)  Components of variability Within- and between-worker variances and between-group variances for each of the four grouping schemes are listed in Table 7.4. There were differences in the components of variability both by grouping strategy and by exposure metric. Between-group variance was small for the a priori grouping schemes, especially for company and industry. However, as might be expected given the method used to form groups, between-group variance was higher for the post hoc grouping schemes, especially for the mean and 90th percentile metrics, at 46.8 and 47.8%, respectively. The within-worker variance was fairly consistent with and without grouping, indicating that little of the within-worker variance was accounted for by the group characteristics. Table 7.4: The proportions of variance in four EMG exposure metrics accounted for by between-group, between-worker and within-worker components using 5 different grouping schemes  No grouping - Worker random effect only Between-worker variance Within-worker variance Grouping by Job Between-job variance Between-worker variance Within-worker variance Grouping by Company Between-company variance Between-worker variance Within-worker variance Grouping by Industry Between-industry variance Between-worker variance Within-worker variance Grouping by Post hoc grouping Between-post hoc group variance Between-worker variance Within-worker variance  mean  90th %ile  cumulative  72.4 27.6  60.3 39.7  80.8 19.2  26.7 46 27.3  0 60.3 39.7  0 80.8 19.2  7.3 65 27.7  3.6 56.7 39.7  4.7 76.1 19.2  11.6 60.7 27.7  9.7 50.5 39.7  0 80.8 19.2  46.8 30 23.2  47.8 17.1 35.1  34.6 49.7 15.7  86  For the mean EMG metric, within-worker variance was below 30% for all grouping schemes and between-group variance was consistently among the highest of all the metrics. For 90th percentile EMG, the between-job variance component was zero, but for other grouping schemes, this metric performed similarly though slightly less well than the mean EMG. Cumulative EMG had low to modest between-group variance for all groupings (0 – 34.6%) but the between-worker variance was large (50-81%), allowing for substantial contrast between workers. As a result, very little variance remained to be explained by the grouping factors. Attenuation of exposure-response relationships Table 7.5 shows the attenuation factors for each EMG exposure metric and grouping strategy in this study. Of the grouping schemes, company performed least well for mean EMG, while job performed least well for all other metrics. The attenuation was least for all exposure metrics with the post hoc grouping strategy. Table 7.5: Attenuation factors for EMG exposure-response relationships estimated using each grouping strategy and exposure metric Number of Mean Grouping strategy groups EMG No Grouping1 0.78 2 Grouping by Job 24 0.89 Grouping by Company2 31 0.81 Grouping by Industry2 5 0.93 2 5 0.98 Grouping by Post hoc 1 Calculated using equation 1 2 Calculated using equation 2 a Same as no grouping because between-group variance was zero  90th %ile EMG 0.68 0.68a 0.70 0.89 0.97  Cumulative EMG 0.85 0.85a 0.87 a 0.85 0.98  The number of workers required per group (k) for different numbers of repeated measurements per worker (n) for mean EMG can be found in Table 7.6. The value of k decreased as n increased, reflecting the inverse relationship between n and k when the attenuation coefficient is set. As the contrast between groups (the between-group variance) decreased, more measurements would be required to achieve an attenuation factor of 0.95. For mean EMG, company was fairly unsuccessful as a grouping scheme, requiring 28 to 49 workers per group. Industry grouping reduced the number needed to 18 to 32 workers per group. Job was better, requiring only 8 to 14 workers per group. Post hoc grouping required the fewest workers per group to maintain attenuation at 0.95; post hoc groups required only 7 workers per group when repeated measures are made on 25% of workers and 5 per group when repeated measures are made on everyone.  87  Table 7.6: Required number of workers per groups (k) to achieve attenuation factors of greater than 0.95 for the mean EMG metric Exposure metric  Number of groups  with n=1.25  with n=1.5  with n=2  14 49 32 7  12 40 25 6  8 28 18 5  Mean Grouping by Job Grouping by Company Grouping by Industry Post hoc Grouping  Discussion Within- and between-worker components of variance Within-worker variability accounts for those aspects of exposure that vary within a worker over time, in this case, between measurement days. Although the workers’ personal factors and company factors stayed the same between days, many things about the tasks performed can change. These might include day-to-day changes in the volume or rate of work, changes in maintenance schedules, shift lengths, or alteration of duties. Between-worker differences are those that depend on the exposure characteristics that differ between workers. These can include job and task characteristics as well as personal characteristics such as sex, body dimensions, age, habits, or techniques of performing tasks. A number of investigators have observed within-worker variability to be higher than betweenworker variability for both chemical exposures (Symanski et al. 2000; Kromhout et al. 1993) and trunk postures (van der Beek et al. 1995; Burdorf, 1995b). However, in the current study, mean, 90th percentile, and cumulative EMG had larger between-worker variance components, ranging from 60.3-80.8%. This may result from the highly diverse sample included in our study, including workers from five different industries and 24 job titles. The observed variation in EMG exposure had good contrast between workers, leaving the within-worker variation small in comparison. If all 133 EMG measurements were within one industry or one job title, one might expect within-worker variance to be higher relative to the between worker variance. The withinworker proportion of variability was above 15% for all metrics. This consistent contribution to variability by within-worker (temporal) sources means that repeated measurements of workers on different days adds to the variability captured in a study. Effect of grouping strategies In addition to individual-level exposures, four different grouping schemes were compared in this study, with the number of observations within groups ranging from 5 to 31. Typically betweenworker (within-group) variability tends to increase with broader classification groups and more workers in each group (Symanski et al. 2000). However, that was not observed in this study; between-worker variance was higher for job (24 groups with 1-29 people per group) and company (31 groups with 1-8 people per group) than it was for industry and post hoc grouping (both 5 groups with 19-36 and 22-33 people per group, respectively). Since the within-worker variability stayed roughly the same across grouping schemes, the between-worker and between-  88  group variances tended to trade off between one another. The slight change in within-worker variability can be attributed to the movement of a few individuals between companies and job titles from their first to their second measurement. This resulted in the within-worker variability being explained in part by different group membership over time. Between-company differences can include factors such as the design of tools, layout of equipment, safety culture, policies surrounding breaks, incentives, and work rate. However, in the current study between-company differences comprised less than 9% of total variability, so these differences were either very small or had little impact on exposure. For mean EMG, grouping by company had the second highest attenuation of the grouping schemes and required the largest number of workers per group to maintain an attenuation factor of 0.95. As a result, this grouping scheme would not be very efficient for an epidemiological study of EMG exposures and back injury in heavy industry. For mean EMG, ‘job’ grouping had the second highest between-group variance after the post hoc grouping scheme. However, the ‘job’ component of variance was zero for 90th percentile and cumulative EMG. This is not to say that the 90th percentile and cumulative EMG values are equal for all jobs. Rather, job title explains some of the variability in mean EMG measurements, whereas the variability between 90th percentile measurements are accounted for entirely by within-worker and between-worker differences. There were more workers per group for the industry grouping, and as a result there was less attenuation of the exposure-response relationship for industry than for job grouping. For mean EMG, 18-32 workers were required per industry group compared to 8-14 workers per job group. However, there were only 5 industry groups compared to 24 job groups, so the total number of workers would be much lower with industry: 192 to 336 for job grouping or 90 to 160 in industry. The total number of measurements decreases marginally when repeated measurements are made on all workers rather than only 50% or 25% of them. The post hoc grouping scheme that ordered exposures grouped by job titles within industries was the most successful. Similarly, a study of magnetic field exposures found more contrast between groups when post hoc grouping schemes were used (Kromhout et al. 1995). The attenuation would be least if the post hoc grouping strategy were used. This strategy also had the minimum within-worker variance in exposure so that post hoc grouping was also the most efficient strategy, requiring 7 workers per group to achieve an attenuation factor of 0.95 when repeated measures are made on 25% of workers and 5 when repeated measures are made on everyone. Measurements of 5 people per group and two measurements per person would require about 50 measurements in total, a potential for substantial savings for future epidemiological studies. However, it is important to consider the generalizability of this grouping strategy. Dividing workers into these same job-industry categories a priori in a future study is unlikely to be as effective as it was here when grouping was performed post hoc. In addition, although a post hoc grouping method could be used in future studies, it is unclear whether it would be as successful as in this study. Although an attenuation coefficient of 0.95 might be considered ambitious or optimistic, it is nonetheless a good goal. If the true exposure-response relationship is very strong, attenuation may not affect the statistical significance of the findings. However, a more subtle relationship could become non-significant when attenuated.  89  The value of each grouping scheme varied with the exposure metric. Mean and 90th percentile EMG had most of their variability concentrated between-workers, allowing for fewer repeated measurements within workers and more economical identification of exposure-response relationships. This may be part of the reason why exposure-response relationships have been successfully found for mean and peak exposures in the past (Norman et al. 1998; Marras et al. 1999). Since cumulative EMG did not benefit from grouping, a study’s grouping strategy could be developed based on the mean and 90th percentile metrics, without affecting cumulative EMG estimates. Limitations When choosing target numbers of workers per group and numbers of measurements per worker, it is prudent to plan to recruit more than needed as estimated in the current study as there can be considerable challenges in on-site exposure assessment; time and budgets should account for contingencies (Trask et al. 2007). Recruitment also has many challenges. Direct recruitment of workers as in the current study yielded response rates of 50-75%, depending on the phase of the study (Trask et al. 2006). The total variance in this study was quite high due to the inclusion of multiple industries, companies, job titles, and individuals, mimicking the variability in exposure that might be observed in a population-based study, but not in an industry-based study. If only a subset of this population were selected, the total variability could be expected to decrease and the relative contribution of each of the components might change (e.g., if only 3 jobs within construction are selected). Relationships between EMG and muscle force are affected by muscle length, contraction velocity, and fatigue, all of which change under working conditions and may not reflect the range of movements in the calibration (Houba et al. 1997). Worksite conditions can include extreme heat, cold, wet, dust, and vibration, and create artifacts caused by contact from tight spaces, seat backs or safety equipment; sweating; or tension on the electrode cables (Trask et al. 2007). These limitations of EMG measurement methods mean that a portion of the variability observed may not be due to true working exposures, but due to errors inherent to the measurement method. As discussed above, a post hoc grouping scheme based on one sample is unlikely to perform as well when applied to a different sample. Rather than relying on a new sample to conform to the findings of the last one, it may be prudent to choose the measurements per individual (n) and individuals per group (k) estimated for the most efficient a priori grouping scheme, in this case industry, and then develop a new post hoc grouping scheme with the new data. The post hoc grouping scheme would be expected to perform better than an a priori industry grouping, and to yield even lower attenuation because the sample size was based on the less efficient industry grouping. In the event that the new sample varies substantially from the sample presented in this study, the industry grouping scheme could still be used, providing options to the researcher.  90  Conclusions Overall, EMG exposure variance was concentrated between-workers for mean, 90th percentile, and cumulative EMG metrics. The post hoc grouping scheme that was based on exposures ordered within industry and job strata had the lowest estimated attenuation of exposure-response relationships, followed by industry alone, then job title alone. Using a combined industry/job grouping scheme appears to deliver the most efficient assessments of exposure-response relationships by maximizing the contrast between groups. Due to the challenges inherent in recruitment and on-site exposure measurement, researchers are encouraged to aim for a substantially larger number of measurements than the minimum estimated for successful grouping.  91  References Brunekreef, B., Noy, D. and Clausing, P. (1987) Variability of Exposure Measurements in Environmental Epidemiology. Am J Epidemiol 122 , 892-898. Burdorf, A. (1992) Exposure assessment of risk factors for disorders of the back in occupational epidemiology. Scand J Work Environ Health 18, 1-9. Burdorf, A. (1995a) Reducing random measurement error in assessment of postural load on the back in epidemiological surveys. Scand J Work, Environ Health 21, 15-23. Burdorf, A. (1995b) Reducing random measurement error in assessing postural load on the back in epidemiologic surveys. Scand J Work Environ Health 21, 15-23. Burdorf, A. (2005) Identification of determinants of exposure: consequences for measurement and control strategies. Occup Environ Med 62, 344-50. Burdorf, A. and Sorock, G. (1997) Positive and negative evidence of risk factors for back disorders. Scand J Work Environ Health 23, 243-56. Burdorf, A. and van Riel, M. (1996) Design of strategies to assess lumbar posture during work . Int J Ind Ergon 18, 239-249. Foliart, D.E., Iriye, R.N., Tarr, K.J., Silva, J.M., Kavet, R. and Ebi, K.L. (2001) Alternative magnetic field exposure metrics: relationship to TWA, appliance use, and demographic characteristics of children in a leukemia survival study. Bioelectromagnetics 22, 574-80. Guo, H.R. (2002) Working hours spent on repeated activities and prevalence of back pain. Occup Environ Med 59, 680-8. Houba, R., Heederik, D. and Kromhout, H. (1997) Grouping strategies for exposure to inhalable dust, wheat allergens and alpha-amylase allergens in bakeries. Ann Occup Hyg 41, 287-96. Jeebhay, MF., Robins, TG., Seixas, N., Baatjies, R., George, DA., Rusford, E., Lehrer, SB. and Lopata, AL. (2005) Environmental exposure characterization of fish processing workers. Ann Occup Hyg 49, 423-37. Keyserling, W.M. (2000) Workplace risk factors and occupational musculoskeletal disorders, Part 1: A review of biomechanical and psychophysical research on risk factors associated with low-back pain. AIHAJ: a Journal For The Science Of Occupational And Environmental Health And Safety 61, 39-50. Koehoorn, M., Trask, C. and Teschke, K. (2007) Recruitment of Workers for Epidemiological Research. Am J Ind Med Kromhout, H., Loomis, D.P., Mihlan, G.J., Peipins, L.A., Kleckner, R.C., Iriye, R. and Savitz, D.A. (1995) Assessment and grouping of occupational magnetic field exposure in five electric utility companies. Scand J Work Environ Health 21, 43-50.  92  Kromhout, H., Symanski, E. and Rappaport, S.M. (1993) A comprehensive evaluation of within- and between-worker components of occupational exposure to chemical agents. Ann Occup Hyg 37, 253-70. Loomis, D. and Kromhout, H. (2004) Exposure variability: concepts and applications in occupational epidemiology. Am J Ind Med 45, 113-22. Marras, W.S., Ferguson, S.A., Gupta, P., Bose, S., Parnianpour, M., Kim, J.Y. and Crowell, R.R. (1999) The quantification of low back disorder using motion measures. Methodology and validation. Spine 24, 2091-100. Mwaiselage, J., Bratveit, M., Moen, B. and Yost, M. (2005) Variability in dust exposure in a cement factory in Tanzania. Ann Occup Hyg 49, 511-9. Neumann, W.P., Wells, R.P., Norman, R.W., Andrews, D.M., Frank, J., Shannon, H.S. and Kerr, M.S. (1999) Comparison of four peak spinal loading exposure measurement methods and their association with low-back pain. Scand J Work Environ Health 25, 404-9. Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clinical Biomechanics 13 (8):561-573. Seidler, A., Bolm-Audorff, U., Heiskel, H., Henkel, N., Roth-Kuver, B., Kaiser, U., Bickeboller, R., Willingstorfer, W.J., Beck, W. and Elsner, G. (2001) The role of cumulative physical work load in lumbar spine disease: risk factors for lumbar osteochondrosis and spondylosis associated with chronic complaints. Occup Environ Med 58, 735-46. Seixas, N.S. and Sheppard, L. (1996) Maximizing accuracy and precision using individual and grouped exposure assessments. Scand J Work Environ Health 22, 94-101. Symanski, E., Sallsten, G. and Barregard, L. (2000) Variability in airborne and biological measures of exposure to mercury in the chloralkali industry: implications for epidemiologic studies. Environ Health Perspect 108, 569-73. Teschke, K., Spierings, J., Marion, S.A., Demers, P.A., Davies, H.W. and Kennedy, S.M. (2004) Reducing attenuation in exposure-response relationships by exposure modeling and grouping: the relationship between wood dust exposure and lung function. Am J Ind Med 46, 663-7. Tielemans, E., Kupper, L.L., Kromhout, H., Heederik, D. and Houba, R. (1998) Individualbased and group-based occupational exposure assessment: some equations to evaluate different strategies. Ann Occup Hyg 42, 115-9. Trask, C., Cooper, J., Teschke, K., Luong, N., Koehoorn, M. (2006) Direct recruitment of workers and worksites in heavy industry for occupational field studies. Canadian Association for Research in Work and Health. St John’s, Newfoundland. Trask, C., Teschke, K., Village, J., Chow, Y., Johnson, P., Luong, N. and Koehoorn, M. (2007) Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility. Am J Ind Med 50, 687-696.  93  van der Beek, A.J., Kuiper, J.I., Dawson, M., Burdorf, A., Bongers, P.M. and Frings-Dresen, M.H. (1995) Sources of variance in exposure to nonneutral trunk postures in varying work situations. Scand J Work Environ Health 21, 215-22. van Tongeren, M., Gardiner , K., Calvert, I., Kromhout, H. and Harrington, JM. (1997) Efficiency of different grouping schemes for dust exposure in the European carbon black respiratory morbidity study. Occup Environ Med 54, 714-9. van Tongeren, MJ., Kromhout, H., Gardiner, K., Calvert, IA. and Harrington, JM. (1999) Assessment of the sensitivity of the relation between current exposure to carbon black and lung function parameters when using different grouping schemes. Am J Ind Med 36, 548-56. Vermeulen, R. and Kromhout, H. (2005) Historical limitations of determinant based exposure groupings in the rubber manufacturing industry. Occup Environ Med 62, 793-9. Waddell, G. (1996) Low back pain: a twentieth century health care enigma. Spine 21, 2820-5. Winkel, J. and Mathiassen, S.E. (1994) Assessment of physical work load in epidemiologic studies: concepts, issues and operational considerations. Ergonomics 37, 979-88. Yost, M. (1999) Alternative magnetic field exposure metrics: occupational measurements in trolley workers. Radiation Protection Dosimetry 83, 99-106.  94  Chapter 8: Discussion The work presented in this thesis addresses methodological problems faced by occupational epidemiologists when measuring or assessing ergonomic risk factors for work-related back injuries. Individual articles address how ergonomic exposures should be measured, the determinants of exposure, the optimum measurement duration, whom should be measured, and how many times.  How to measure? The first article, “Measuring low back injury risk factors in challenging work environments: an evaluation of cost and feasibility”, examined the efficacy of several different measurement tools in the field in terms of both cost and success rate (or failure rate). The challenges presented by workplace environments, tasks and postures are diverse and difficult to anticipate based on pilot testing in a lab. The resilience or robustness of instrumentation is tested when confronted with high demands and less than careful treatment. When the workers’ exposure is closely tied to their actions and characteristics, the presence of researchers adds an additional level of complexity to data collection. Will the workers work as they normally would when they know they are being monitored? Do the measurement devices and observation methods allow for using working behaviours? Are workers comfortable enough with the researchers to report exposures to the best of their recollection, without fear of judgment or adverse consequence? Even when these concerns have been satisfied, noise in directly measured signals and typographical or transcription errors in observation or interview notes can lead to exposure misclassification. The challenges encountered by data collection in the field go beyond what can be easily measured or quantified, but they exist and contribute to exposure assessment error nonetheless. Although not common, evaluations of the process of exposure assessment rather than just the validity of the outputs have been reported before. In a review of physical workload assessment, Winkel and Mathiassen (1994) discussed the tradeoffs between self-report, observation, and direct measurement in terms of cost, capacity (sample size), versatility (applicability to different working situations), generality (to observe different exposures) and exactness (precision or accuracy) The benefit of Chapter 4 over these previous hypothetical or theoretical studies is that it used objective field data to compare five exposure methods in real worksites, on real workers, while they performed real work; it would be hard to imagine a more demanding set of conditions than those encountered in the UBC Back Study. The results of Chapter 4 support the cost/precision/sample size tradeoffs outlined by Winkel and Mathiassen, as well as quantifying them. Nothstein et al. (2000) compared several passive and active monitors’ cost-effectiveness in hypothetical scenarios. As shown in Chapter 4 and the Nothstein study, cost-effectiveness is related to the number of samples one is able to take with the device. The outputs or measurements of an exposure assessment are generally the primary concern, but the pragmatic aspects of data collection cannot be overlooked. The article in Chapter 4 does not promote the use of one method over the other, but rather provides objective, quantitative data for researchers to make decisions for a study they may be planning. Methods need to be selected in  95  the context of a proposed study’s goals, scope, and industrial milieu; having quantitative information on the performance of measurement methods helps inform these decisions. Some methods, such as self-report, are cheaper and therefore allow larger sample sizes. Since the notion of a tradeoff between sample size and precision is well-documented (Winkel and Mathiassen, 1994); (Neumann et al. 1999) the question becomes how much precision is lost. Chapter 5, “Predicting Exposure for Mean, 90th Percentile, and Cumulative EMG Activity in Heavy Industry” seeks to answer this question by predicting directly-measured EMG metrics using observation and self-report. A description of some of the advantages and limitations of EMG as an exposure assessment method are also discussed in the ‘strengths and limitations’ section of this chapter, among them the velocity-tension relationship that is important to EMG activity but likely poorly captured by observation and self-report. Regression models using observed variables performed better than those using self-report for three common EMG exposure metrics. Models based on observed variables predicted between 30.7% (cumulative EMG) and 47.2% (mean EMG) of the outcome metrics. Additionally, the self-report models included some very surprising and somewhat suspect variables. This combines with the fact that ‘industry’ was included in both the mean and 90th percentile self-report models. It may be that industry was included in the models because it was recorded with better fidelity than the selfreported exposures. Given that researchers have previously reported that many self-reported variables are reported with very low fidelity (Neumann et al. 1999; Burdorf, 1992a; Wells et al. 1997) it seems like the most likely explanation for the spurious associations with suspect variables and the strength of the industry variable in self-reported models. Laboratory studies compare measurement methods in terms of the precision, reliability, and errors of the measurements in the hopes of one being able to replace one method with another. Generally one method is considered the standard (sometimes the ‘gold standard’ or absolute metric for that type of measurement) and the other is a ‘trial method’ being investigated as a replacement, usually because it is cheaper or easier to apply in some circumstances. Correlations are expected to be very high, and the range of errors, bias, and limits of agreement very low. In this context, a method that explained 20-50% of the variability in the standard method would be seen as very poor. However, in the context of an epidemiological field study where the units and nature of collection for the trial method are different than the standard method, such high agreement cannot be expected. Such is the case with the prediction models presented in Chapter 5. A comparison is being made between model-predicted and measured exposure estimates that employ different methods, assess different aspects of exposure, and do not use the same units for output. Observation and self-report are generally less precise than direct measurements like EMG. In addition, these methods are not assessing the same parameters. Muscle activity is a composite of all external exposures that require muscle activity as well as internal factors. As measured by EMG, muscle activity is a multifactorial phenomenon, incorporating not only the postural and loading demands of muscle force, but also the length-tension relationship and velocity-tension relationship of the muscle, and fatigue (NIOSH, 1992), as well as muscle recruitment and technique variations even within the same load/posture task conditions (van Dieen et al. 2001; Granata et al. 1999). Explaining 50% of the variation in exposure could be considered a success considering the exposure variation for EMG and the possible measurement error from observation and self-report,.  96  Aside from the challenges, the ‘Measuring low back injury risk factors…’ article suggests that explaining 50% of the variability in exposure may be sufficient to identify an exposure response relationship. Many studies that used observation or self report methods have found relationships between exposures and back injuries (Seidler et al. 2001; Kumar, 1990; Krause et al. 2004),. Unfortunately, validation studies to assess the precision of observation and self-report methods are uncommon (Wiker, 2003). In addition to the cost/sample size tradeoff, there is also a precision/ sample size tradeoff. In a comparison of cost-effectiveness of different case-control exposure assessment methods for cancer studies, interview methods of determining working history exposures had higher precision but lower sample sizes than using job title data from medical records. Despite the lower sample size, the quality of the interview data led to less misclassification and stronger relationships to the health effect than did the administrative data (Siemiatycki et al. 1989). The final determinants of exposure prediction models are not intended to replace EMG as a method to estimate muscle activity, but rather to help identify the combination of work tasks and measurement methods that best capture working exposures for large epidemiological studies. Fifty percent explained variability might be increased if there were a broader spectrum of industries and job titles included in the sample, particularly low-exposure sedentary jobs such as administrative work. The total variability measured would be larger, and it seems likely that both self-report and observation would do well in discerning between the highly active heavy industrial jobs and seated desk work, thereby explaining a greater proportion of the measured range. The corollary of this argument is that if the measured sample contained only workers from one job title with very similar exposures, the measured variability would be much lower than the resolution of the methods, making it impossible to discern between workers’ exposures using self-reports or observations. This topic is explored further in the ‘Whom to measure’ section below.  How long to measure? Even when the method of exposure assessment is selected, the choice of sampling strategy presents many decisions, including how long to sample. The prevalence of short sampling durations in ergonomics and lack of explicit sampling instructions among ergonomics exposure assessment tools led to the question ‘How long is long enough?’ and the related article in Chapter 6 ‘Selecting efficient sampling durations for low-back EMG assessment’. McGill wrote that the importance of loading over time to injury means that researchers need “rigorous examination of injury and loading for substantial periods of time” (McGill, 1997). In a parallel discussion of workplace magnetic exposure variation, Loomis and Kromhout point out that the decision about time scale should be biologically based, given what is understood about the etiology of the health outcome (Loomis and Kromhout, 2004). Although long-term (weeks, months, years) exposure may accumulate to produce the physiological and structural changes that lead to symptoms, injury, and disability, continuous and long-term sampling is very expensive and isn’t feasible in most field-based studies. As a result, researchers rely on shorterterm measurement periods and the use of this data to estimate longer-term exposures.  97  Often an assumption is made that shorter-term measurements are representative of a person’s exposure over the day, week, season or year (Gold et al. 2006), and previous studies have tended to measure cumulative load via sums or extrapolation of task-based exposure assessments (Callaghan, Salewytsch et al. 2001), particularly on assembly line work (Norman et al. 1998; Kumar, 1990). The extrapolation of cycle level and frequency to a whole shift has face validity in a manufacturing context (i.e. cyclical work tasks), and such extrapolated measures in this industrial sector has increased the understanding of the relationship between cumulative loads and injury. However, non-cyclical highly varied work is more difficult to assess in this way and may not be capturing true long-term exposure. Even cyclical work may be more variable than previously thought; Fisher and Callaghan (2007) investigated the effect of adding ‘unplanned rest’ into extrapolation-based estimations of cumulative loads for assembling line work. Unplanned rest resulted from finishing work faster than expected, interrupted supply of materials and products, or tool or machine malfunction and repair. There were substantial, significant differences when unplanned rest was included in the estimates of cumulative load, suggesting that extrapolating a few measured cycles of a task to multiple cycles may not be representative of true exposure, particularly when the expected or planned ‘typical’ work day differs from what actually occurs. ‘Typical’ days are actually fairly rare; unscheduled breakdowns and maintenance, unexpected adoption of tasks for an absent coworker, and seasonal variation in production or materials, that can all produce an overall picture of working exposures quite different from a task-extrapolation approach. The study in Chapter 6 estimated the error between full-shift measures and shorter measurement durations, as well as between full-shift and two-shift measures for a variety of work patterns (cyclical and non cyclical) across occupations and industries. Clearly two days is a far better approximation of long-term exposure than a single day and would be an appropriate ‘gold standard’ in this study than one day. However, since two measurement days were available for only 35 individuals, one day was used as the comparison to allow for more power in the analysis. Although measured bias was very low for all sampling durations, there was considerable random error that increased with shorter durations. The average absolute and percentage errors between one-day and two-day means were consistently smaller than any other comparison. However, this comparison had the highest bias that was always in a positive direction, meaning that the oneday mean was, on average, larger than the two-day mean. Although the two-shift bias was two to four times higher than that of the next highest duration, it was still less than 1 %RC among EMG measures that averaged 39 %RC for a full shift, so the bias would not likely make an impact in the context of a typical measurement. One-day versus two-day comparisons also had the lowest limit of agreement. This seemingly contradictory condition means that there is more systematic error with the two day measure than with the one day measure, but that the range of error was small; it was one-half to one-third the width of one-day to 4-hour comparisons. As sampling durations decreased from a whole shift to a few minutes, the absolute error, percentage error, and limits of agreement for exposure estimates deviated more from full-day estimates. The article suggests 4- or perhaps 2-hour sampling durations as alternative to full-shift measurement, but this will depend on the purpose of the study, the precision required, and the concomitant amount of error that is acceptable. The ‘how long’ question therefore presents intrinsic tradeoffs. For the same amount of measurement time (and cost) as measuring two  98  workers for an 8-hour shift, one could measure four workers for fours hours each, or two workers for four hours and then return for a second four-hour measurement on those workers. The added information on between-worker variability (by adding an additional worker) or within-worker variability (by adding an additional day) is the value of changing from a full-shift sample to a shorter duration.  Whom to measure? The allocation of measurements within and between individuals and groups has implications for the detection of exposure-response relationships where they exist and these decisions involve optimizing the tradeoffs between resources and data. Analysis of components of variance to inform sampling strategies and grouping schemes have been published for a variety of working exposures, including magnetic field exposure (Kromhout et al. 1995), carbon black (van Tongeren et al. 1999; van Tongeren et al. 1997), wood dust (Teschke et al. 2004), flour dust (Houba et al. 1997), fish proteins and antigens (Jeebhay et al. 2005), cement dust (Mwaiselage et al. 2005), rubber plant exposures (Vermeulen and Kromhout, 2005) and simulated data (Seixas and Sheppard, 1996). This methodology has also been applied to postural exposures (Burdorf, 1992b; Burdorf and van Riel, 1996)Burdorf, 1995a), to EMG activity in the shoulder muscles (Mathiassen et al. 2003; Moller et al. 2004), and to combined task, posture, and MMH observations (Paquet et al. 2005). The study presented in Chapter 7, “Optimizing Sampling Strategies: Components of Low-Back EMG Variability in Five Heavy Industries” is, to the author’s knowledge, the first to examine the components of variance, degree of exposureresponse attenuation, and study design implications for grouping strategies for low-back EMG. In previous studies, within-worker variability was typically higher than between-worker variability. This was seen in chemical exposures (Symanski et al. 2000), trunk posture (van der Beek et al. 1995; Burdorf, 1995), and MMH tasks (Paquet et al. 2005). Without grouping, the current study had between-worker variance two to three times larger than within-worker variance for all EMG metrics. This difference may be due to the variability in task, jobs, and industries in the sampled population. The breadth of jobs and industries in this study would be expected to result in more variation between workers than a study that samples workers from the same industry, company, job title. If all workers have similar tasks and working conditions, the differences between them are small and all that is left is the variation in their own work day-today, seen as a high proportion of within-worker variability in most studies. For mean EMG, between-group variability ranged from 7.3% for grouping by job to 48% for th grouping by the post-hoc scheme, with similar results for 90 percentile EMG. This is substantially lower than the 47-72% between-group variability seen in a study of trunk posture in five occupational groups (Burdorf, 1992b). It may be that posture and EMG metrics vary differently across subjects and working conditions. However, it may also be that having fewer groups with more clearly defined membership and less overlap between groups accounts for this difference. In the current study, there were five post hoc groups created by ordering job titles within industries. The industries, jobs, and tasks contained within each post-hoc group were therefore fairly diverse, leaving more of the variability between subjects than between groups. According to Burdorf and van Riel (1999) the variability within groups must be minimized when comparing groups. In the current study, the post hoc grouping scheme had the highest between-  99  group variability, and required the fewest workers per group, making it the most efficient grouping scheme for all EMG metrics. However, these results may be an optimistic estimate of what could be achieved in future studies doing the same kind of post hoc grouping by ordering industry-job groups by their EMG exposures, then forming groups. Nevertheless, the post-hoc groups required one-half to one-fortieth of the workers per group of the other grouping schemes, making it the most efficient grouping strategy so even if more measurements are required in a future study.  Capturing Exposure: What is important? The main advantage of direct measurement is in precision, accuracy, and ‘informational content’(Jansen et al. 2001). However, as Burdorf points out, “problems arise as to how to reduce the data to manageable amounts and still preserve the important aspects of job variation” (Burdorf et al. 1991), particularly with highly detailed exposure signals such as EMG. Choosing a single ‘best’ method or metric for summarizing exposure into a comprehensive, informative, and representative metric is difficult. Several authors agree that multiple dimensions of exposure must be assessed for the exposure to truly be characterized by magnitude (called level by some authors), frequency, and duration (Winkel and Mathiassen, 1994; van der Beek and Frings-Dresen, 1998; David, 2005). However, three summary statistics might not be enough to distinguish all the important aspects of exposure. This is illustrated in Figures 8.1a-d that represent four different exposures varying over time. These examples are for illustrative purposes and are therefore very simplified; most jobs (including those in this study) have multiple levels of exposure, making the characterization of exposure more complex than in Figure 8.1. ‘Continuous’ exposure measurement obtained by data logging of discrete points at a very high frequency to create an ‘exposure signal’ (such as EMG), can be used to examine the patterns of exposure and to evaluate different exposure measures. Although these exposures are clearly different on visual inspection, ‘mean’ exposure would not be able to differentiate between them, as they all have the same mean (10 arbitrary units). Similarly, peak exposure is the same (18 arbitrary units) for Figures 8.1 a, c, and d, and doesn’t adequately discern between the different patterns of exposure either. Figures 8.1 a and c are distinguishable from Figure 8.1 d by both frequency and duration. However, Figures 8.1 a and c are distinguishable from each other only by the time sequence of the excursions. Clearly there are some aspects of exposure that are not captured by the exposure metrics commonly in use, in particular the aspect of variability in data points over time. The concept of ‘variability’ in exposure can be defined in a few different ways. For example, ‘global variability’ as measured by the standard deviation or the range of values can be thought of as the spread in the measurements, irrespective of the order in which these measured points (sometimes called ‘quanta’) occur. Mathiassen called the degree to which exposures differ over time as ‘diversity’ and changes in exposure over time as ‘variability’ (Mathiassen, 2006). Variability in exposure seems likely to be key to the mechanism of injury, particularly the interplay of loading with fatigue, rest, and recovery. For example, full disc hydration after sleeping may increase intradiscal pressure on flexion (Gunning et al. 2001), while fatigue or a ‘warming up’ effect towards the end of the shift may change the susceptibility of tissues to injury. These patterns could also be the key to many potential administrative controls, such as job rotation or work/rest ratios. Unfortunately, there is little epidemiological evidence on the effect of exposure variability, mostly because it has proved difficult to measure. 100  20  15  15  Exposure  Exposure  20  10  10  5  5  0  0  1  3  5  7  9  11  13  15  17  19  1  3  5  7  9  Time  20  20  15  15  10  15  17  19  10  5  5  0  0  1  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20  3  5  7  9  11  13  15  17  19  Time  Tim e  c  13  b  Exposure  Exposure  a  11 T ime  d Figure 8.1: Four examples of simulated working exposure varying over time  In the current research, the collection of cumulative, shift-long data made for a unique opportunity, as presented in Chapter 7, to compare mean, peak, and cumulative measures of exposure from the same worker and time period. Similar to the description of the exposure graphs in Figure 8.1a-d, Jansen and Burdorf (Jansen et al. 2001) describe the difficulties of comparing multiple aspects of exposure via Amplitude Probability Distribution Function (APDF) and Cycle Frequency Analysis (CFA), and developed Exposure Variability Analysis (EVA) to solve this problem. EVA is a three-dimensional expression of exposure patterns over time that portions exposure into discrete duration, level, and frequency categories to produce a series of frequency bars over a surface (Mathiassen and Winkel, 1991). The category cut-off points are selected with reference to relevant physiological and biomechanical phenomena. Although originally used for EMG exposure data collected over time, it has also been applied to continuous trunk posture measurements in nurses, housekeepers,  101  and office workers (Jansen et al. 2001). EVA might be the most comprehensive exposure summary method currently in use for capturing more commonly used exposure elements related to mean and peak, as well as variability, although it still does not account for the order of exposures. Winkel and Mathiassen (1994) state that no single data reduction method is satisfactory for capturing all elements of exposure that may be related to injury mechanisms. The results of this study suggest that each of the EMG exposure metrics provide different information, that they require different grouping schemes, and that different observed and self-reported working variables can be used to predict them.  Strengths and limitations: methodological issues Recruitment This study was designed to capture a wide range of job titles and working exposures, and seemed successful in doing so. Contacting workers directly, as described in Appendix A, provided access to a far broader population than typical studies where workers are contacted through a single company or union. However, there were some biases intrinsic to the recruitment method that should be acknowledged. Workers with back injury claims in the year 2001 were selected with the intention of targeting workplaces and job titles that had exposure to back injury risk factors. When contacting workers directly, the sample was biased towards workers able to communicate in English. Workers were contacted in the fall of 2004 through to the end of 2005. This was 3-4 years after their claims were accepted, and some workers had changed jobs, companies, or industries. Workers were measured if they were still working within their target industry, but those who changed industries entirely, returned to school, or were still on disability were not included, resulting in a ‘healthy worker’ bias. It could be that the excluded workers were exposed to far greater levels of the risk factors, and this resulted in more severe or longer lasting injury. In terms of scheduling measurement visits, it may be that employers were more apt to accommodate researchers in times of lower production or in areas that would be less affected by the presence or interruption introduced by measurement. Although multiple afternoon shifts and night shifts were included, day shifts were favored by both employers and the measurement team. This may introduce bias if certain activities only took place at night. For example, food warehouses are far busier at night when they can deliver food for restocking supermarkets while the supermarkets are closed. The workers measured were fairly diverse (see worker characteristics in Appendix J). The measured sample consisted of real workers drawn from the heavy industry population, mostly male (95%), with a range of ages (mean age = 42.2 years), and often with a history of back pain (75.8% reported pain in the last 6 months). This differs substantially from the young, fit university student subjects who often participate in laboratory studies of simulated work tasks and ergonomic exposure measurements (McGill, 1992; Mirka et al. 1997; Marras et al. 2001; Kee and Chung, 1996).  102  EMG A good exposure metric has a plausible causal mechanism with the health effect under study, and this is true for EMG. Muscle activity increases the force or tension in the muscles that in turn increases the compression and shear forces in the joints of the spine as well as the intradiscal pressure. Muscle tension or spasm can also lead to ischemia (lack of blood flow) that on its own can lead to pain, and muscle strain or overuse. EMG can be considered a measure of dose, since it results from a combination of external loading exposures and characteristics of the individual (Hansson et al. 2000). Even in terms of external exposures, there are several MMH and postural components that contribute to muscle activity. This makes for a convenient way to assess several aspects of exposure, but it can make it more difficult when it comes to identifying opportunities for intervention and controls. In order to develop controls, task data should be collected alongside the EMG in order to ascertain what aspects of the external loads can or should be changed. Sources of error There are many potential sources of error when measuring EMG. Muscle tension is related to the electrical activity of the motor neurons, but also to the length and speed of movement of the muscle. Under very dynamic conditions or in extreme postures, EMG activity has a less clear relationship to force and the EMG exposure being measured may not represent the tension in the muscles. The length-tension relationship of the back muscles is unique in that it is also subject to a flexion-relaxation response. This occurs during extreme flexion (as in a stoop lift or toughing the toes) when the back muscles show decreased activation and the loads are borne passively by ligamentous structures (Solomonow et al. 2003), and a lordotic static hold procedure has been shown to be more accurate for squat lifts than stoop lifts, when calibrating for compression normalized EMG (Potvin 1990). Lordotic postures were used during calibration maneuvers, but there were substantial opportunities for extreme flexion during the work tasks observed during the study. The field setting also introduced several sources of noise, (described in Chapter 4), as did the variability of EMG measurements within a worker, (described in Chapter 3). Chapter 3 showed substantial unexplained variability in the EMG measurements that was not accounted for by the calibration tasks (comprised of trunk posture and weight in the hands). All the EMG calibration prediction models were based on linear regression when the true relationship may have a multiplicative or polynomial structure. This may account for the small amount of EMG variability explained. Nonetheless, the difference in calibration measures between the beginning and end of the shift for the same individual were small, as were repeated measures of the same position within a measurement session. This suggests that though there is substantial variation in EMG calibration measurements across positions and individuals, the same exposure is being reliably recorded. Future studies could be done with frequency analysis of pre- and post-shift measurements, as well as trial 1 and trial 2 measurements, to see if amplitude differences are related to frequency differences that indicate fatigue effects throughout the shift or between repeated trials. In addition, future studies could collect information on the right-left dominance of workers to determine if this factor contributes to left-right differences. Calibration The calibration method employed in this study tested only static, sagittal movements with a constant force in the hands. However, working exposures often incorporate complex  103  asymmetrical movements, and many lifting tasks are dynamic. Nonetheless, the static reference contraction is a constant denominator for all measurements throughout the shift, and as such does not change the relative value of measurements throughout the shift within a person. Using the same reference contraction for all subjects compares the exposure to a known task. Although using percentage of reference contraction (%RC) might be intuitive in terms of comparisons of working exposures across individuals, it may not capture differences related to injury mechanisms. The %RC metric does not incorporate a worker’s skill level or maximum capacity. For example, a small person for whom 25 lbs is 25% of their weight will find it difficult to exceed the reference contraction during the work day, but a large (and stronger) person might do work that results in EMG at several multiples of the reference contraction. The most common alternative to %RC, the maximum voluntary contraction (MVC) requires a subject to be very motivated to give a maximum effort, and the results can be influenced by muscle fatigue, the time for which the MVC is held, as well as pain or spasm in the muscles (Maras 2001). There is also some risk of pain and injury when maximally contracting muscles, and this presents an ethical problem, particularly when asking older, previously-injured workers to perform activities that may prove harmful. Suitability of EMG as a ‘gold standard’ It might be argued that due to sources of error, EMG is not an appropriate ‘gold standard method’ with which to build prediction equations as in Chapter 5. By modeling self-reported and observed variables to predict EMG, there is at least the implication that EMG is a good (perhaps the best) measure. EMG is a direct measurement method, with the objectivity and precision that comes with that. There is no doubt that there are more accurate methods, in particular the biomechanical models that incorporate joint marker, force transducer and EMG inputs to calculate moments and forces in the joints and tissues (Marras and Sommerich, 1991; McGill, 1992). However, the instrumentation and controlled environment required for these methods precludes their use in field sites. Here a tradeoff was made between a) accuracy and b) portability, cost, and feasibility at worksites, with the focus being on the latter in order to measure ergonomic exposures in large epidemiological studies of workers. To meet the goals of this study, there simply wasn’t another direct measurement method that allowed for continuous measurement of both MMH and postural exposures in real working conditions without hampering work tasks.  Observation There are a few aspects of the observation method that should be considered when interpreting the results, including the accuracy and reliability of the method, biases in recording, and the analytical methods used to summarize exposures. In a separate study of the Back-EST observation method (Village et al. 2007), comparisons of observed trunk posture data were made to measurements by a trunk inclinometer over the same period. Observations underestimated the percent of the shift spent in extension, 20 to 45° flexion, and in lateral bends greater than 20° (12%, 9.6%, and -2.3%, respectively). However, observations overestimated time spent in 0 to 10° flexion and in 10 to 20o flexion (16.4 and 6.03%, respectively). These biases of over- and under-estimation will affect the ability of observations to predict EMG as in Chapter 5.  104  There is also a source of bias in the timing of observations that has not been quantified. A vibrating watch was used to signal the time for recording. At the time of a signal, a worker might be performing a very dynamic movement (for example, quickly lifting a box off the floor) that would make the posture and force of the movement at that moment very hard to assess. In this box lifting example, the force in the hand would start out very high and then decrease as the upward movement decelerated. However, the weight in the hand was invariably recorded as the static weight of the box; this decision was made for consistency but could systematically underestimate peak loads. Recording the posture in this situation causes more of a problem since it is very difficult to take a ‘mental snapshot’ of a dynamic event. The observers may have been biased toward the end point or mid-point of the movement rather than the exact position at the time of the watch signal. Even after observer training on still photographs, video, and in the field, such biases may have persisted. The observation records used in this study were summarized into cumulative percentages of time spent with specific exposures, such as trunk posture >60o or handling loads >20 kg. This method made it possible to assess the level of exposure as well as estimate the total duration. However, the frequency of exposure was not assessed, even though it is a distinguishing feature of exposure and certainly a risk factor. This aspect of the observation method would limit its ability to distinguish relationships between frequency-related exposure and back injury.  Self report The nature of self-report makes it vulnerable to recall biases. Although most workers can report that an exposure has occurred, it is often difficult for to report the duration or level accurately. Workers tend to over-report the duration or level of an exposure (Wells et al. 1997; Palmer KT et al. 2000; Burdorf, 1992a). Errors can also arise with language, literacy, and comprehending or interpreting the meaning of the question (Spielholz et al. 2001). However, self-report is very commonly used because of its low cost for large-scale epidemiologic studies of back injury, and many studies have found relationships using this exposure assessment method (Kumar, 1990; Burdorf and Sorock, 1997; Myers et al. 1999; Knibbe and Friele, 1996; Macfarlane et al. 1997). The fact that it is commonly used means that its relationship to other exposure assessment methods should be quantitatively evaluated as in Chapter 5, and the challenges should be qualitatively assessed as in Chapter 3. In this study, we found that industrial worksites were rarely conducive to quiet consideration and workers expressed confusion or discomfort with selfreport assessment methods. Such information helps with the interpretation of prior studies and with the planning of future studies.  Generalizability of Results This study was designed as part of a pilot-study phase of a large program of research that will investigate the risk factors for back injury in five heavy industries in BC. However, the tasks, job titles, and exposures measured in the current work were so varied that the results may be applicable to other industries and job titles that have similar work and exposures. For example, much of the work in mining, oil/gas industries, agriculture and fisheries could be expected to have exposures within the range of those studied, as well as some similar job titles (e.g., millwright, heavy equipment operator). However, the challenges encountered, exposures  105  observed, prediction models developed, sampling times tested, and components of variability examined were within the context of a modern, western, industrialized society with labour codes, health and safety regulations, frequent unionization, and relatively strong safety cultures. The applicability of the results would be expected to be lower in developing nations where the exposures experienced by the workforce may be influenced by different labour, cultural and regulatory environments. Similarly, the sample in the current study was 95% male and measured within a heavy industry environment. While, nursing has a very high level of back injury (Yassi et al. 1995) and may have similar levels of exposure, the results may not be completely generalizable to the service or health sectors, or to female dominated workforces.  Implications and Future Research The current work makes contributions to the interpretation of prior research and the planning of future research. Insight into the challenges of measuring real working exposures and the relationships between exposure assessment methods may help to frame previous epidemiological research. Information on the components of variance, the utility of different grouping schemes, and the optimal sampling times for non-cyclical work helps researchers to plan and budget for future studies, allowing them to maximize their return on investment, in information per dollar spent. The direct implications of the findings for individual-level assessment by field ergonomists may be more limited. Epidemiology measures the relationships between an exposure and a health effect in people at the individual-level with the goal of identifying associations at the populationlevel to inform interventions. In contrast, ergonomic assessments within a company are often done on a person, with the goal of changing the scheduling or equipment of an individual or small group of individuals in an effort to minimize their exposure and decrease their risk. The levels of precision and error that are acceptable in a large study may leave something to be desired when assessing a single person. Nonetheless, the concept of tradeoffs in assessment method, sampling strategy, measurement time, and grouping methods are applicable even to the individual level assessments of the ‘company ergonomist’ or ‘company hygienist’, and it is hoped that the conceptual structure of the work will inform their practice as well. Among the results of the individual papers emerges a theme of optimizing the choices around study design and sampling strategy, a common decision making process in occupational epidemiology. Given the ubiquitous constraints of budget and access to industrial populations, tradeoffs are made between using a highly accurate exposure assessment method on a few people or a less precise method on a larger number of people, between measurement duration and sample size, and between measuring many individuals or repeats within individuals. Finding objective ways to address tradeoffs is important to conducting research that accurately capture exposure-response relationships. The results can then appropriately inform risk guidelines, policy, and efforts to control exposure. Although these kinds of tradeoffs are ubiquitous and longstanding, it is fair to ask whether we have to decide between ‘good quality’ data and the large numbers of subjects required by epidemiology. The current state of the research demands hard decisions, but it would be unreasonable to accept this as a universal truth lest it lead to complacency and a halt in the  106  search for better exposure assessment methods. Researchers working on both ends of the ‘exposure assessment spectrum’ can help move toward optimal methods by staying informed of the challenges and advancements at either end of the spectrum. In a recent review describing the possible contributions of cumulative loading to low back disorders, Marras (2003) suggested that: “Instead of continuing to explore low back pain causality within the confines of specific disciplines… we must more fully explore the interactions between these disciplines…” Research efforts, including the current work, are moving towards a time when low-cost methods with high accuracy will be feasible for large scale studies of working exposure. Until then, studies that inform the extent of the tradeoffs between sample size, cost, feasibility, precision, and accuracy are vital.  107  References Barriera-Viruet, H., Sobeih, T., Daraiseh, N. and Salem , S. (2006) Questionnaires vs observational and direct measurements: a systematic review . Theoretical Issues in Ergonomics Science 7, 261-284. Blangsted, A.K., Hansen, K. and Jensen, C. (2003) Muscle activity during computer-based office work in relation to self-reported job demands and gender. Eur J Appl Physiol 89, 352-8. Burdorf, A. (1992a) Exposure assessment of risk factors for disorders of the back in occupational epidemiology. Scand J Work Environ Health 18, 1-9. Burdorf, A. (1992b) Sources of variance in exposure to postural load on the back in occupational groups. Scand J Work Environ Health 18, 361-7. Burdorf, A. (1995) Reducing random measurement error in assessing postural load on the back in epidemiologic surveys. Scand J Work Environ Health 21, 15-23. Burdorf, A., Govaert, G. and Elders, L. (1991) Postural load and back pain of workers in the manufacturing of prefabricated concrete elements. Ergonomics. 34, 909-918. Burdorf, A. and Sorock, G. (1997) Positive and negative evidence of risk factors for back disorders. Scand J Work Environ Health 23, 243-56. Burdorf, A. and van Riel, M. (1996) Design of strategies to assess lumbar posture during work . International Journal of Industrial Ergonomics 18, 239-249. Callaghan, J.P., Salewytsch, A.J. and Andrews, D.M. (2001) An evaluation of predictive methods for estimating cumulative spinal loading. Ergonomics 44, 825-37. David GC. (2005) Ergonomic methods for assessing exposure to risk factors for work-related musculoskeletal disorders. Occup Med (Lond); 55(3):190-9. Gold, J.E., Park, J.S. and Punnett, L. (2006) Work routinization and implications for ergonomic exposure assessment. Ergonomics 49, 12-27. Granata, K.P., Marras, W.S. and Davis, K.G. (1999) Variation in spinal load and trunk dynamics during repeated lifting exertions. Clin Biomech (Bristol, Avon) 14, 367-75. Gunning, J.L., Callaghan, J.P. and McGill, S.M. (2001) Spinal posture and prior loading history modulate compressive strength and type of failure in the spine: a biomechanical study using a porcine cervical spine model. Clin Biomech (Bristol, Avon) 16, 471-80. Hansson, G.A., Nordander, C., Asterland, P., Ohlsson, K., Stromberg, U., Skerfving, S. and Rempel, D. (2000) Sensitivity of trapezius electromyography to differences between work tasks - influence of gap definition and normalisation methods. J Electromyogr Kinesiol 10, 103-15. Houba, R., Heederik, D. and Kromhout, H. (1997) Grouping strategies for exposure to inhalable dust, wheat allergens and alpha-amylase allergens in bakeries. Ann Occup Hyg 41,  108  287-96. Jansen, J.P., Burdorf, A. and Steyerberg, E. (2001) A novel approach for evaluating level, frequency and duration of lumbar posture simultaneously during work. Scand J Work Environ Health 27, 373-80. Jeebhay, MF., Robins, TG., Seixas, N., Baatjies, R., George, DA., Rusford, E., Lehrer, SB. and Lopata, AL. (2005) Environmental exposure characterization of fish processing workers. Ann Occup Hyg 49, 423-37. Knibbe, J.J. and Friele, R.D. (1996) Prevalence of back pain and characteristics of the physical workload of community nurses. Ergonomics 39, 186-98. Krause, N., Rugulies, R., Ragland, D.R. and Syme, S.L. (2004) Physical workload, ergonomic problems, and incidence of low back injury: a 7.5-year prospective study of San Francisco transit operators. Am J Ind Med 46, 570-85. Kromhout, H., Loomis, D.P., Mihlan, G.J., Peipins, L.A., Kleckner, R.C., Iriye, R. and Savitz, D.A. (1995) Assessment and grouping of occupational magnetic field exposure in five electric utility companies. Scand J Work Environ Health 21, 43-50. Kumar, S. (1990) Cumulative load as a risk factor for back pain. Spine 15, 1311-6. Loomis, D. and Kromhout, H. (2004) Exposure variability: concepts and applications in occupational epidemiology. Am J Ind Med 45, 113-22. Macfarlane, G.J., Thomas, E., Papageorgiou, A.C., Croft, P.R., Jayson, M.I. and Silman, A.J. (1997) Employment and physical work activities as predictors of future low back pain. Spine 22, 1143-9. Marras, W.S. and Sommerich, C.M. (1991) A three-dimensional motion model of loads on the lumbar spine: I. Model structure. Hum Factors 33, 123-37. Mathiassen, S.E. (2006) Diversity and variation in biomechanical exposure: what is it, and why would we like to know? Appl Ergon 37, 419-27. Mathiassen, S.E., Moller, T. and Forsman, M. (2003) Variability in mechanical exposure within and between individuals performing a highly constrained industrial work task. Ergonomics 46, 800-24. Mathiassen, S.E. and Winkel, J. (1991) Quantifying variation in physical load using exposurevs-time data. Ergonomics 34, 1455-68. McGill, S.M. (1992) A myoelectrically based dynamic three-dimensional model to predict loads on lumbar spine tissues during lateral bending. J Biomech 25, 395-414. McGill, S.M. (1997) The biomechanics of low back injury: implications on current practice in industry and the clinic. J Biomech 30, 465-75.  109  Moller, T., Mathiassen, S.E., Franzon, H. and Kihlberg, S. (2004) Job enlargement and mechanical exposure variability in cyclic assembly work. Ergonomics 47, 19-40. Mwaiselage, J., Bratveit, M., Moen, B. and Yost, M. (2005) Variability in dust exposure in a cement factory in Tanzania. Ann Occup Hyg 49, 511-9. Myers, A.H., Baker, S.P., Li, G., Smith, G.S., Wiker, S., Liang, K.Y. and Johnson, J.V. (1999) Back injury in municipal workers: a case-control study. Am J Public Health 89, 1036-41. Neumann, W.P., Wells, R.P., Norman, R.W., Andrews, D.M., Frank, J., Shannon, H.S. and Kerr, M.S. (1999) Comparison of four peak spinal loading exposure measurement methods and their association with low-back pain. Scand J Work Environ Health 25, 404-9. NIOSH (1992) Selected Topics in Surface Electromyography for Use in the Occupational Setting: Expert Perspective. Publication No. 91-100, DHHS (NIOSH) . Norman, R., Wells, R., Neumann, P., Frank, J., Shannon, H. and Kerr, M. (1998) A comparison of peak vs cumulative physical work exposure risk factors for the reporting of low back pain in the automotive industry. Clinical Biomechanics 13 (8):561-573. Nothstein, G.L., Hahne, R.M. and Spence, M.W. (2000) Evaluation of the cost-effectiveness of various passive monitors and active monitors for industrial hygiene sampling. AIHAJ 61, 64-8. Palmer KT, Haward B, Griffin MJ, Bendall H and Coggon D. (2000) Validity of self reported occupational exposures to hand transmitted and whole body vibration. Occup Environ Med 57, 237-41. Paquet, V., Punnett, L., Woskie, S. and Buchholz, B. (2005) Reliable exposure assessment strategies for physical ergonomics stressors in construction and other non-routinized work. Ergonomics 48, 1200-19. Seidler, A., Bolm-Audorff, U., Heiskel, H., Henkel, N., Roth-Kuver, B., Kaiser, U., Bickeboller, R., Willingstorfer, W.J., Beck, W. and Elsner, G. (2001) The role of cumulative physical work load in lumbar spine disease: risk factors for lumbar osteochondrosis and spondylosis associated with chronic complaints. Occup Environ Med 58, 735-46. Seixas, N.S. and Sheppard, L. (1996) Maximizing accuracy and precision using individual and grouped exposure assessments. Scand J Work Environ Health 22, 94-101. Siemiatycki, J., Dewar, R. and Richardson, L. (1989) Costs and statistical power associated with five methods of collecting occupation exposure information for population-based case-control studies. Am J Epidemiol 130, 1236-46. Solomonow, M., Baratta, R.V., Banks, A., Freudenberger, C. and Zhou, B.H. (2003) Flexionrelaxation response to static lumbar flexion in males and females. Clin Biomech (Bristol, Avon) 18, 273-9. Spielholz, P., Silverstein, B., Morgan, M., Checkoway, H., Kaufman, J. (2001) Comparison of self-report, video observation and direct measurement methods for upper extremity  110  musculoskeletal disorder physical risk factors. Ergonomics; 44(6):588-613. Symanski, E., Sallsten, G. and Barregard, L. (2000) Variability in airborne and biological measures of exposure to mercury in the chloralkali industry: implications for epidemiologic studies. Environ Health Perspect 108, 569-73. Teschke, K., Spierings, J., Marion, S.A., Demers, P.A., Davies, H.W. and Kennedy, S.M. (2004) Reducing attenuation in exposure-response relationships by exposure modeling and grouping: the relationship between wood dust exposure and lung function. Am J Ind Med 46, 663-7. van der Beek, A.J. and Frings-Dresen, M.H. (1998) Assessment of mechanical exposure in ergonomic epidemiology. Occup Environ Med 55, 291-9. van der Beek, A.J., Kuiper, J.I., Dawson, M., Burdorf, A., Bongers, P.M. and Frings-Dresen, M.H. (1995) Sources of variance in exposure to nonneutral trunk postures in varying work situations. Scand J Work Environ Health 21, 215-22. van Dieen, J.H., Dekkers, J.J., Groen, V., Toussaint, H.M. and Meijer, O.G. (2001) Withinsubject variability in low back load in a repetitively performed, mildly constrained lifting task. Spine 26, 1799-804. van Tongeren, M., Gardiner , K., Calvert, I., Kromhout, H. and Harrington, JM. (1997) Efficiency of different grouping schemes for dust exposure in the European carbon black respiratory morbidity study. Occup Environ Med 54, 714-9. van Tongeren, MJ., Kromhout, H., Gardiner, K., Calvert, IA. and Harrington, JM. (1999) Assessment of the sensitivity of the relation between current exposure to carbon black and lung function parameters when using different grouping schemes. Am J Ind Med 36, 548-56. Vermeulen, R. and Kromhout, H. (2005) Historical limitations of determinant based exposure groupings in the rubber manufacturing industry. Occup Environ Med 62, 793-9. Village, J., Trask, C., Luong, N., Chow, Y., Johnson, P., Koehoorn, M. and Teschke, K. (2007) Development and Evaluation of an Observational Back Exposure Sampling Tool (Back-EST) for Work-Related Back Injury Risk Factors. Submitted in June 2007 to: Applied Ergonomics; Submission number JERG-D-07-00096. Wells, R., Norman, R., Neumann, P., Andrews, D., Frank, J., Shannon, H. and Kerr , M. (1997) Assessment of physical work load in epidemiologic studies: common measurement metrics for exposure assessment. Ergonomics 40, 51-61. Wiker, S. (2003) Statistical challenges facing development of epidemiologically validated low back injury risk predictive and descriptive models using biomechanical, anthropometric, or consensus-based ergonomic risk prevention guidelines. Seminars in Spine Surgery 15, 3-15. Winkel, J. and Mathiassen, S.E. (1994) Assessment of physical work load in epidemiologic studies: concepts, issues and operational considerations. Ergonomics 37, 979-88. Yassi, A., Khokhar, J., Tate, R., Cooper, J., Snow, C. and Vallentyne, S. (1995) The  111  epidemiology of back injuries in nurses at a large Canadian tertiary care hospital: implications for prevention. Occup Med (Lond) 45, 215-20. Yost, M. (1999) Alternative magnetic field exposure metrics: occupational measurements in trolley workers. Radiation Protection Dosimetry 83, 99-106.  112  Appendix A: Detailed recruitment methodology Sampling Strategy A total of 126 workers in heavy industry were measured in this study. Rather than recruiting workers through their employers, unions, or membership in other organization, workers were contacted directly. This worker recruitment method used was novel in North American exposure studies, but has been successfully used by Swedish investigators studying magnetic field exposures6.Workers with a lost-time workers’ compensation claim for a back injury in 2001 were eligible for participation; this sub-group contained approximately 1900 workers. Participants were all working in one of the identified heavy industries (forestry, wood products, transportation, warehousing, and construction) within the lower mainland or sunshine coast area at the time of the study. The population was restricted to those with a lost time claims to maximize the probability that the sampled individuals will have a range of exposures of interest. The population was restricted by location to simplify logistics and minimize costs. Because this region includes about half the population of British Columbia, it was expected to include a wide range of target industry employees, different worksites, and job titles. Contacting Participants In order to comply with provincial privacy regulations, the Workers’ Compensation Board research secretariat generated a randomized list of workers with claims (referred to as gateway workers) from each industry. Workers were then contacted by the WCB and asked if their contact information could be forwarded to the research team. Those subjects who agreed to be contacted were sent a letter by the research team outlining the purpose and methods of the study, as well as an invitation to participate. During a subsequent telephone call, the research team answered any questions the worker had about the study and invited the worker to participate. Those who declined to participate were not contacted again. Those who agreed to participate, but were not currently working or who were working in non-target industries or geographical areas received a follow-up call at a later date to see if conditions had changed. Those who declined participation or who did not meet the employment criteria were replaced by another worker randomly selected from the WCB records. In addition to determining the worker’s current employment status, the worker was asked about the best way to get permission from the employer to measure at the worksite. Some gateway workers chose to ask their employer themselves and agreed to a follow-up call from the researchers to determine how to proceed; other workers preferred that the researchers contact the employer. The contact information for a corporate health and safety director was requested and used as the employer contact when available. It was emphasized when discussing with workers that management does not see their individual results, but that all data is pooled together to protect privacy. At this stage, letters were sent to all unions, worker and employer organizations thought to have some connection to the target industries. This was done to notify pertinent  6  Floderus, B., Persson, T. and Stenlund, C. (1996) Magnetic-field Exposures in the Workplace: Reference Distribution and Exposures in Occupational Groups. Int J Occup Environ Health 2, 226-238.  113  stakeholders of the study objectives and measures and to foster communications between stakeholders and the researchers. Contacting Employers The process of contacting employers varied in duration and complexity from immediate approval from the department manager to repeated calls to company headquarters in attempts to find the correct decision maker in that company. On occasion, the subject was eager to participate but the employer refused to allow access to the worksite. In these cases, the workers received a follow-up call after a few months to see if their employment situation had changed. The main goal of telephone contact with the employer was to schedule a preliminary site visit, usually with either a direct supervisor, or a human resources or occupational health and safety manager. Occasionally the site visit involved a formal power point presentation of the study objectives, methods, and the potential benefit to the company, but more often the consisted of an informal discussion with questions and answers about the study and a tour of the facility. The site visit allowed researchers to get a safety orientation and learn about the worksite, meet and get signed consent from the claimant worker and volunteer co-workers, briefly observe the participants’ tasks and work environment, and to schedule a measurement day. If vehicles were used, the vehicle was inspected briefly to ascertain how best to attach the vibration equipment and decide if additional equipment was required. The site visit also provided an opportunity to identify a staging area in which to set-up the equipment and conduct the interviews. This location was agreed upon by the worker and a meeting time was set. Ideally the room had electrical plugs, locking doors, and provided privacy so that the worker would feel comfortable removing his shirt. First aid rooms, unused offices, and break rooms were the most common locations; the set-up area was also used for the interview at the end of the day. Efforts were made to ensure that all workplace stakeholders were contacted prior to the measurement day: senior management and human resources as well as the direct supervisor, the health and safety manager or committee members, and at unionized worksites, the shop steward. Researchers asked the employer for access to the worksite to measure exposures on the gateway worker as well as 1 or 3 non-management, non-administrative employees at the same worksite. When the employer was willing, the additional workers were selected from a randomized list of employees. This was complicated by a desire to measure both workers on the same shift, and typically there were few production workers on the same shift as the gateway worker who agreed to participate. In these cases, the researchers selected for task variety. There was a particular emphasis on getting workers who had whole body vibration exposure, as these workers represented a minority of the selected workers. Timing There were two sets of measurement equipment operated by two field researchers; workers were measured in pairs on the same measurement day. Direct measurements and observations were made throughout an entire 8-hour shift. For those who worked longer than 8 hours, (10 or 12 hour shifts), often only the first 8 hours were measured. EMG measurements ranged from 5.5 to 10.3 hours of working time, excluding breaks and set-up time (mean = 6.32 hours, sd = 1.32  114  hours). Whenever possible, measurements were made on each pair of workers on two separate days at each workplace. Second EMG measurements were successfully made on 34% of workers; the second measurement day occurred from 1 to 439 days after the initial measurement (mean = 93 days SD=64 days). The initial intention was to measure 10 workers from each industry for a total of 50 participants. However, the randomization process yielded several workers from the same worksites. To avoid over-sampling at a particular workplace, fewer co-workers were recruited from the companies with multiple gateway workers. Where there were 3 or more workers at one worksite, only the gateway workers were measured. In these cases, another randomly selected gateway worker was invited to participate to achieve the target sample size.  115  Appendix B: Detailed exposure assessment methodology Exposure measurements included three components, direct ‘measurements’ of exposures using instrumental methods, ‘observations’ of factors potentially related to measured exposures, and in-person interviews of each ‘subject’ about factors potentially related to measured exposures. Measured risk factors included materials handling, trunk posture, and whole body vibration. Methods of measurement were designed, tested, and piloted on a convenience sample of up to 8 University Plant Operations employees with multiple exposures. A final “battery” of methods that can be feasibly used within a day on-site and that is flexible enough to measure risk factors in a wide variety of work environments was then adopted. On the measurement day, the research team signed in to the workplace or notified the shift manager that they had arrived. If the consent forms were not already signed, a researcher verbally reviewed the content with the participants and answered any questions. After preparing the equipment in the set up area and assigning one researcher to each of the two participants, the measurement instrumentation was set-up on the workers. The exposure assessment included shift long direct measurement using inclinometer, vibration monitor and electromyographic (EMG) data logger, as well as shift-long observations and a post-shift interview. EMG Instrumentation The electromyograph (EMG) was the first instrument to be set-up at the start of the shift as it required substantial calibration. Electromyography directly measures the activation of the muscles beneath the skin. A portable data collection system was used for this study; Mega ME3000P8 and ME3000P8 muscle tester units (Mega Electronics, Finland) provided the EMG measurements.  Figure B.1: Anatomical landmarking for the location of the EMG electrodes  The researchers explained the procedures to the worker as they performed them and answered questions as they went along. The participant was asked to lift his shirt (held in pace by a clothespin) and place the tips of the thumbs on top of the hipbones (the lateral border of the iliac crest). A small dot was made at the level of the fingers in the centre of the back over the spinous processes. The skin at this level and lateral to this mark was prepared by abrading the skin in  116  one direction for 15-30 seconds, then wiped clean with an alcohol pad. The skin was shaved only if hair prevented complete adhesion of the electrode. Using palpation and muscle contractions as a guide, the electrodes were placed over the belly of the Iliocostalis Lumborum at the level of the fingers – approximately the level of L4/L5. The ground electrode was placed on the posterolateral aspect of the iliac crest. Disposable, ovoid, Ag-AgCl wet-gel electrodes sized 30mm x 22mm were placed parallel to the muscle fibers at this location with a 25 mm centre-to-centre distance. The inter-electrode impedance was measured using a standard ohmmeter (Superex Digital Multimeter M-830B). Electrodes and preamplifiers were subsequently taped down using 3M Transpore surgical tape and all wires and connections were covered using Surgifix postsurgical sleeve to prevent motion in the cables. A portable data collection system was used for this study; Mega ME3000P4 and ME3000P8 muscle tester units (Mega Electronics, Finland) provided the EMG measurements. The Mega units have a range of -/+ 6500 µV, a total gain of 412, a common mode rejection ration of 110 dB and an input impedance greater than 10 GOhms. Raw signals are filtered internally using an 8-500 Hz band pass filter. EMG Calibration The participant performed some back flexion/extension movements for a brief signal check; data was inspected in real time for discontinuity, excess signal while at rest, or usually low signal while active, or spikes indicating motion artifacts. Following a successful signal check, calibration data was collected on the two bilateral channels using raw EMG at 1000 Hz. It was decided in the methods development stage of the study to use multiple positions for calibration described in chapter 3. This was done to allow for parallel studies of muscle activity variability between these positions. Four calibration tasks were used for the EMG: 1) 2) 3) 4)  Standing unloaded with a trunk angle of 0 o (relaxed and upright.) Standing unloaded with a trunk angle of 45 o Standing at a trunk angle of 45 o with a 25 pound load in the hands. Standing at a trunk angle of 60 o with a 25 pound load in the hands.  Trunk angle was measured by the investigator using a 12-inch hand-held goniometer (Baseline Instruments Inc) with a bubble level attached. The subject was instructed to keep the head up, the shoulders back, and the pelvis tilted back to maintain lordosis and optimize activation of the erector spinae muscles. The arms were to hang straight directly below the shoulders with and without the weight. Once the worker was in the appropriate position, the ‘mark’ button on the Mega software was clicked to identify the start and end of the measurement period. The participant assumed the upright calibration position a single time for 10 seconds. Each forward flexed calibration posture was repeated twice, each time maintained in a constant static position for 5 seconds. The same EMG calibration procedure was repeated at the end of the day.  117  o  Figure B.2: Calibration was performed without a weight while standing upright, and forward flexed at 45 . o o The 45 and 60 positions were measured with a 11.5 kg weight.  Shift-long EMG measurement Shift-long EMG data collection was initiated once the inclinometer was launched. The shiftlong EMG data was collected on averaged mode so that the data logger performed the full wave rectification and applied the 100 ms moving average window. To maximize data quality, the highest averaged sampling rate was used: 10 Hz, or one data logged point every 0.1 second. However, this meant that the EMG had to be downloaded at least once per shift, usually at lunch break and at the end of the shift. At the worker’s break, information was downloaded from EMG and the vibration monitor onto the laptop computers. Electrode leads were left attached to the worker, with the cable ends tucked into a fanny pack. In the event that electrodes became tangled or removed during the course of the shift, the worker was asked to notify the research staff. Researchers then made a visual inspection of the instrumentation as soon as it fit with the worker’s task schedule. Once a problem was identified, the electrodes were removed, the skin re-prepared, and new electrodes were applied. The worker then performed the EMG calibration procedures again. Only two such mid-shift calibrations were performed. At the end of the shift, the calibration procedure was repeated to identify any shift or drift in the signal over the course of the day. Electrodes and electrode paste/glue were removed using spray bottle, towels, and alcohol pads, and the worker was offered an unscented lotion to prevent skin irritation. Shift- long observation of workers Researchers used a VibraLITE3 Model VL300 watch with a vibrating alarm auto reload counter (Global Assistive Devices, Inc., FL) to alert them silently to take an observation of the worker. Observations were made once every minute throughout the shift, starting after the pre-shift instrumentation of the worker and ending just prior to de-instrumentation at the end of the shift. Observations were not made during work breaks.  118  Observation data recorded included: general task or activity, item or power tool in hands, and items worn (such as a tool belt); trunk posture and mobility; presence of trunk support, lateral bend or twist; type of manual materials handling, horizontal distance, weight and force estimate, and any additional pertinent comments. Photographs were taken of unique worksites, tools, postures or handling methods if the participant gave permission. Researchers trained using this observation form by evaluating photos taken of various worksite activities, as well as several pilot testing days and video sessions. There were several standards developed within an observation guide used to standardize between researchers. These included defaulting to the lower category if undecided between the two categories, such as holding an object near or mid distance, default to the near distance. If undecided whether or not there is a twist, default to no twist. If undecided on action of a manual materials handling task, default to a hold. Assumptions were allowed only when: things hadn’t changed and the task was routine and predictable; the subject must have always been visible in order to record an observation, otherwise it was marked as missed. Researchers audited one another’s forms regularly to ensure that proper recording was completed. Prior to starting the shift and the observations, the researchers explained the requirements of observations, which included that they didn’t want to interrupt the participant’s work or get in their way, and to go about work as usual, also that it is hard for the researchers to talk or answer questions when making notes. Participants were also told to alert their researcher if the instruments were uncomfortable or if they heard any snapping or if they felt something had come loose. In these cases, the researcher would stop taking observations to replace electrodes, tie up loose cables, otherwise deal with instrumentation, and resume observations once the instrumentation was repaired. Observation and Measurement of Loads In order to quantify manual materials handling (lifting, lowering, pushing, pulling, carrying), measurements of the loads and forces were made using a weigh scale (Pelouze, P114S scale, 115kg capacity) for loads lifted and carried, and a Chatillon DMG 250/CSD-100 Dynamometer (250 lb capacity) for push or pull forces. A list of objects generated throughout the day was measured towards the end of the shift by the researcher and would include such things as common tools, equipment worn, and common objects handled. Questionnaire The questionnaire was completed at the end of the shift in a quiet place, and a snack was provided as a thank you and incentive to relax and take time with responses. This usually happened in same area where electrodes are removed, after the post-shift calibration. The questionnaire was presented in paper form for the subject to refer to, and was completed orally with the researcher recording the responses. Subjects were asked to identify their work postures from simplified drawings of representative postures with descriptive phrases, as well as give categorized time estimates of exposure to including questions about type of vehicle or other vibrating equipment used, duration of exposure in the shift, weight of vehicle, type of tire, type  119  of transmission, typical speeds of operation, ground surfaces, changes in slope of surface, seat type and suspension, back support, and armrests; typical dimensions and weights of materials handled, horizontal reach distances, extent of bending and twisting, shift duration, mobility, trunk posture and tasks performed. If participants asked for clarification on a question, researchers helped to clarify in an unbiased way, they did not share their observations or opinion about exposure, only re-phrased the question There was also a short health history section regarding back injuries and their impact. These were answered using 11-point scales for average and most intense back pain over the last six months, changes in daily, social and family activities, and work activities, and average exercise habits over the last six months. In addition, information was requested using terms familiar to the work force, based on information from our pre-testing batteries and consultations with industry personnel in the initial months.  120  121  Downhill; Flat  COMMENTS  STYLE  Smooth; Jerky VEHICLE Loaded; Unloaded  SPEED Idle/station; <20km/hr; 20-40km/hr; 40-70km/hr; 70km/hr  TERRAIN Smooth pave/Cement; Broken pave/cement; Gravel; Packed earth; Soft earth; O ff road, Water; Rail; Air  VEHICLE Slope Uphill;  Force Estimate, Exertion Light; Moderate; Heavy  <l lbs (0); 1-10lbs; 10- 22lbs; 22-44lbs; >44lbs (5)  Weight Estimate  Push; pUll  Near; Mid; Extended  1 Hand; 2 Hand on the item  Horizontal location:  MMH Lift; lOwer; Hold;  |  | |  |  | |  |  | |  |  Subject ID |____|____|____|____| |____|____|____|  POSTURE Stand; Walk; sIt; Crouch/Kneel/Squat; Lay; Climb; Other Trunk 0-100; 10-20; 20-450; 45-600; >600; Extension Trunk is supported (1) Lateral Bend >200 (1) Twisting/Rotating >200 (1)  (IDLING or ON)  Powered Hand Tool  TASK / ACTIVITY ITEM in Hands ITEM Worn  Observation Time  Date |_2_|_0__|_0__|_5__| |___|___| |___|___|  Appendix C: Observation tool | |  |  | |  |  | |  |  |  CATHERINE / YAT / JAMES |  |  | |  |  |  |  Sheet # |____|____|____| |  |  |  |  |  |  Appendix D: Interview tool (worker copy)  PART A  PARTICIPANT INFORMATION  1.  O MALE  O FEMALE  2.  HEIGHT (feet, inches)  3.  WEIGHT (pounds)  4.  DATE OF BIRTH (Year/Month/Day)  5.  COMPANY NAME  6.  INDUSTRY  7.  CURRENT JOB TITLE  8.  CURRENT DEPARTMENT  9.  WORKING HOURS THIS WEEK (Hours/Day; Days/Week)  10.  NUMBER OF CONSECUTIVE DAYS WORKED INCLUDING TODAY  11.  TOTAL COMMUTING TIME TO AND FROM WORK TODAY (Minutes)  O Construction O Forestry  12. MAIN TASKS TODAY OF DAY (%) Task A. _____________________________ Task B. _____________________________ Task C. _____________________________ Task D. _____________________________ Task E. _____________________________  O Warehousing O Wood Products O Transportation  PROPORTION A1. _____________ B1. _____________ C1. _____________ D1. _____________ E1. _____________  122  PART B  MOBILITY  13. Today while working, did you do any of the FOLLOWING? If yes, how LONG? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  Sit  Stand  Crouch/Kneel  Walk  Lay down  Other Activities Not on this list Climb (Example: stairs, ladders, scaffolds)  123  STANDING 14. Today of the time you were standing while working, did you stand with your back in the following POSTURES? If yes, how LONG? o o o o o o  A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  o o o o o  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  (0-10o) Upright  (20-45o)  (45-60o)  Slightly bent  Moderately bent  Bending backwards  Bending sideways  (10-20o)  Barely bent  (More than 60o) Severely bent  Twisting  124  WALKING 15. Today of the time you were walking while working, did you walk with your back in the following POSTURES? If yes, how LONG? A. B. C. D. E. F.  None < 5 min 5 to < 15 min 15 to < 30 min 30 to < 45 min 45 to < 1 hr  G. H. I. J. K.  1 to < 2 hrs 2 to < 4 hrs 4 to < 6 hrs 6 to < 8 hrs 8 hrs  (20-45o)  Slightly bent  Bending backwards  (0-10o)  (10-20o)  Upright  (45-60o)  Moderately bent  Barely bent  (More than 60o)  Bending sideways  Severely bent  Twisting  125  SITTING 16. Today of the time you were sitting while working, did you sit with your back in the following POSTURES? If yes, how LONG? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  Upright  Leaning forward  Leaning back (with no back support)  Bending sideways  Leaning back (with back support)  Twisting  126  PART C  MANUAL MATERIALS HANDLING  LIFTING/LOWERING/CARRYING 17. Today while working, did you LIFT/LOWER/CARRY any items with your hands that were …… If yes, how LONG? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  10-22 LBS  Less than 1 LBS  22-44 LBS  1-10 LBS  More than 44 LBS  127  18. Today, of the LIFTS & LOWERS you did while working, did you … A. Spend more time lifting B. Spend more time lowering C. Spend equal time lifting & lowering 19. Today of the time you were lifting/lowering/carrying while working, how long were the loads in your hands NEAR, MID or FAR from you? Please consider only loads that are heavier than 10 lbs. A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  Near (0-10”) than 20”)  Mid (10-20”)  Far (More  128  PUSHING 20. Today while working, did you PUSH any items with your hands? A. None B. < 5 min If yes, how LONG?  Examples: Push Cart, Trolley, Wheelbarrow  C. D. E. F.  > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  21. Today of the time you were pushing while working, how long did you push items with your hands LIGHTLY, MODERATELY, or HEAVILY? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  Moderate Exertion  Light Exertion:  Bicycle, Wheeled Desk Chair, Door  Shopping cart filled with 5 40-lbs of dog food Motorcycle, Couch  Heavy Exertion  Piano, Car (uphill) 2 or 3 drawer, full file cabinet across carpet  129  PULLING 22. Today while working, did you PULL any items with your hands? A. None B. < 5 min If yes, how LONG?  Examples: Push Cart, Trolley, Wheelbarrow  C. D. E. F.  > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  21. Today of the time you were pulling while working, how long did you pull items with your hands LIGHTLY, MODERATELY, or HEAVILY? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  Moderate Exertion  Light Exertion:  Bicycle, Wheeled Desk Chair, Door  Shopping cart filled with 5 40-lbs of dog food Motorcycle, Couch  Heavy Exertion  Piano, Car (uphill) 2 or 3 drawer, full file cabinet across carpet  130  PART D  VIBRATION  WHOLE BODY VIBRATION 24. Today while working, did you OPERATE or RIDE any wholebody vibrating vehicle(s)/equipment? (Refer to Whole-Body Vibrating Equipment List)  a. Please NAME each vehicle/equipment. b. Today, how LONG did you operate or ride each vehicle/equipment? A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  c. For each vehicle/equipment, is the ARM REST adjusted for you? YES NO NOT APPLICABLE because no arm rest  d. For each vehicle/ equipment, is the SEAT adjusted for you? YES NO NOT APPLICABLE because no seat  e. For each vehicle/equipment, is the BACK REST adjusted for you? YES NO NOT APPLICABLE because no back rest  f. For each vehicle/equipment, does the BACK REST give you good back support? YES NO NOT APPLICABLE  131  g. How long did you operate or ride each vehicle/ equipment over ..… A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  SMOOTH pavement/cement  BROKEN pavement/cement  GRAVEL  PACKED EARTH  SOFT EARTH  OFF-ROAD  -GRASS, SOIL  -LOGS, ROCKS  -HARD PACKED DIRT ROAD  WATER -SHIPS, BOATS  AIR  -PLANE, HELICOPTER  RAIL 132  h. Of the time you were operating or riding each vehicle/equipment, how long did you drive it ….. SMOOTHLY (ACCELERATION/BRAKING) A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  JERKY  i. Of the time you were operating or riding each vehicle/equipment, how long was the vehicle ….. STATIONARY / IDLING LESS THAN 20KM/HR 20-40KM/HR A. B. C. D. E. F.  None < 5 min > 5 to < 15 min > 15 to < 30 min > 30 to < 45 min > 45 to < 1 hr  G. H. I. J. K.  > 1 to < 2 hrs > 2 to < 4 hrs > 4 to < 6 hrs > 6 to < 8 hrs > 8 hrs  40-70KM/HR MORE THAN 70KM/HR  133  PART E  HEALTH HISTORY  25. Today, did you experience any LOW BACK PAIN? Low back pain means aches or discomfort in the low back (shaded area) whether or not it extends from there to one or both legs (sciatica). YES  NO (Go to question 28)  26. TODAY, how would you rate your low back pain on a 0-10 scale, where 0 is “NO PAIN” and 10 is “PAIN AS BAD AS COULD BE”? 0  1  2  3  4  5  NO PAIN  6  7  8  9 10 PAIN AS BAD AS COULD BE  27. Today, did you change your usual work activities NO because of low back pain? YES  If yes, please explain how?  28. In the last 6 months, did you experience any LOW BACK PAIN? Low back pain means aches or discomfort in the low back (shaded area) whether or not it extends from there to one or both legs (sciatica). YES  NO (Go to question 35)  134  29. In the past 6 months, how intense was your WORST low back pain rated on a 0-10 scale, where 0 is “NO PAIN” and 10 is “PAIN AS BAD AS COULD BE”? 0  1  2  3  4  5  6  NO PAIN  7  8  9  10  PAIN AS BAD AS COULD BE  30. In the past 6 months, ON AVERAGE, how intense was your low back pain rated on a 0-10 scale, where 0 is “NO PAIN” and 10 is “PAIN AS BAD AS COULD BE”? (That is, your usual pain at times you were experiencing pain). 0  1  2  3  4  5  6  NO PAIN  7  8  9  10  PAIN AS BAD AS COULD BE  31. About how many days in the last 6 months have you been kept from your usual activities (work, school or housework) because of low back pain? Disability days 32. In the past 6 months, how much has low back pain interfered with your daily activities rated on a 0-10 scale where 0 is ‘no interference’ and 10 is ‘unable to carry on any activities’? 0  1  2  NO INTERFERENCE  3  4  5  6  7  8  9  10  UNABLE TO CARRY ON ANY ACTIVITIES  135  33. In the past 6 months, how much has low back pain changed your ability to take part in recreational, social and family activities where 0 is ‘no change’ and 10 is ‘extreme change’? 0  1  2  3  4  5  6  7  NO CHANGE  8  9  10  EXTREME CHANGE  34. In the past 6 months, how much has low back pain changed your ability to work where 0 is ‘no interference’ and 10 is ‘extreme change’? 0  1  2  3  4  5  6  7  NO CHANGE  8  9  10  EXTREME CHANGE  35. During the last 6 months, on average, how many days a week have you engaged in 30 minutes or more of exercise? Examples:  Walking for exercise Golfing Bicycling Rollerblading Hockey  0  1  2  3  4  5  6  7  days/week  136  CONCLUSION Thank you so much for answering our questions. You have been very helpful. 1. May we contact you in the future if we wish to clarify any answers you gave in this interview? YES NO 2. Is there anything else that you think we should know about that has not been asked? __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________  3. If you have questions about the interview or the study in the future, please feel free to contact us. The names and phone numbers of the investigators are included in the consent form I have left with you. Feel free to call collect if you are outside the lower mainland. COMMENTS: __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________ __________________________________________________________________  137  Appendix E: Interview record sheet (researcher copy) -The last thing we ask from you today is a questionnaire in an interview style. It will take approximately 30minutes. -We will be asking you the questions but you can follow along with us using this interview package. This interview will ask about your activities while working today and there will be some questions about your health history. -Some of the questions we ask may not apply to you, but it is important that we ask all our participants the same questions. We ask that you attempt to answers all the questions honestly. If you feel uncomfortable with a question, please do not hesitate to tell us so we can skip to the next question. -Your answers will be used for research purposes only and will be kept confidential. -Your employer will not see your answers.  Date (year, month, day)  |___|___|___|___| |___|___| |___|___|  Subject ID |___|___|___|  |___|___|___|___| CATHERINE | JAMES | YAT  PART A – PARTICIPANT INFORMATION 1.  SEX (Male, Female)  2.  HEIGHT  |___|___|___| cm  |___| feet |___|___| inches  3.  WEIGHT  |___|___|___| kg  |___|___|___| lbs  4.  DOB (year, month, day)  5.  COMPANY NAME  6.  INDUSTRY (wood Products, Construction, Transportation, Forestry, Warehousing)  7.  CURRENT JOB TITLE  |___|  |___|___|___|___| |___|___| |___|___| _______________________________________ |___|  _______________________________________ 8.  CURRENT DEPARTMENT  _______________________________________ 9.  WORKING HOURS THIS WEEK A. (Hours/Day)  |___|___|.|___|___|  B. (Days/Week)  |___|  10.  NUMBER OF CONSECUTIVE DAYS WORKED (INCLUDING TODAY)  11.  TOTAL COMMUTING TIME TO WORK TODAY (Minutes)  12.  MAIN TASKS TODAY (gardening example: trimming, weeding, raking) Task A Duration A (% of day)  |___|___|  |___|___|___|  _______________________________________ |___|___|___| . |___|  138  Task B  _______________________________________  Duration B (% of day) Task C  |___|___|___| . |___| _______________________________________  Duration C (% of day) Task D  |___|___|___| . |___| _______________________________________  Duration D (% of day) Task E  |___|___|___| . |___| _______________________________________  Duration E (% of day)  |___|___|___| . |___|  PART B – MOBILITY 13.  14.  MOBILITY (Did you do any of the following & how long?) A. STAND (A-K)  |___|  B. WALK (A-K)  |___|  C. SIT (A-K)  |___|  D. CROUCH (A-K)  |___|  E. LAY DOWN (A-K)  |___|  F. CLIMB (A-K)  |___|  G. OTHER ACTIVITIES – NOT ON THIS LIST (A-K)  |___|  STANDING (Did you STAND with your BACK in the following POSTURES?) A. UPRIGHT, 0-10degrees; (A-K)  |___|  B. BARELY BENT, 10-20degrees; (A-K)  |___|  C. SLIGHTLY BENT, 20-45degrees; (A-K)  |___|  D. MODERATELY BENT, 45-60degree; (A-K)  |___|  E. SEVERELY BENT, >60degree; (A-K)  |___|  F. BENDING BACKWARDS; (A-K)  |___|  G. BENDING SIDEWAYS; (A-K)  |___|  H. TWISTING; (A-K)  |___|  139  15.  16.  WALKING (Did you WALK with your BACK in the following POSTURES?) A. UPRIGHT, 0-10degrees; (A-K) |___| B. BARELY BENT, 10-20degrees; (A-K)  |___|  C. SLIGHTLY BENT, 20-45degrees; (A-K)  |___|  D. MODERATELY BENT, 45-60degree; (A-K)  |___|  E. SEVERELY BENT, >60degree; (A-K)  |___|  F. BENDING BACKWARDS; (A-K)  |___|  G. BENDING SIDEWAYS; (A-K)  |___|  H. TWISTING; (A-K)  |___|  SITTING (Did you SIT with your BACK in the following POSTURES?) A. UPRIGHT (A-K) |___| B. LEANING FORWARD; (A-K)  |___|  C. LEANING BACK with NO support; (A-K)  |___|  D. LEANING BACK with support; (A-K)  |___|  E. BENDING SIDEWAYS; (A-K)  |___|  F. TWISTING; (A-K)  |___|  PART C – MANUAL MATERIALS HANDLING (Did you LIFT, LOWER or CARRY any items & for how long?) 17.  18.  19.  A. <1LBS (A-K)  |___|  B. 1-10LBS (A-K)  |___|  C. 10-22LBS (A-K)  |___|  D. 22-44LBS (A-K)  |___|  E. >44LBS (A-K)  |___|  Lifting & lowering proportions  A. More time Lifting B. More time Lowering C. Equal time Lifting & Lowering  (How long were the loads in your hands NEAR, MID or FAR from you?)  A. NEAR (A-K) B. MID (A-K) C. FAR (A-K)  |___| |___| |___| |___|  140  PUSHING (Did you PUSH any items with your hands & how long?) 20. 21.  Push duration (A-K) A. Push LIGHT exertion (A-K) B. Push MODERATE exertion (A-K) C. Push HEAVY exertion (A-K)  |___| |___| |___| |___|  PULLING (Did you PULL any items with your hands & how long?) 22. 23.  Pull duration (A-K) A. Pull LIGHT exertion (A-K) B. Pull MODERATE exertion (A-K) C. Pull HEAVY exertion (A-K)  |___| |___| |___| |___|  PART D – VIBRATION WHOLE BODY VIBRATION (Did you OPERATE or RIDE any whole-body vibrating vehicle(s)/equipment?) 24. Whole body vibration exposure (Yes, No) |___| VEHICLE 1 A. NAME  _______________________________________  B. DURATION (A-K) C. ARM REST ADJUSTED FOR YOU (Yes, No, not Applicable) D. SEAT ADJUSTED FOR YOU (Yes, No, not Applicable) E. BACK REST ADJUSTED FOR YOU (Yes, No, not Applicable) F. GOOD BACK SUPPORT (Yes, No, not Applicable) (How long?) G1. SMOOTH PAVEMENT/CEMENT (A-K) G2. BROKEN PAVEMENT/CEMENT (A-K) G3. GRAVEL (A-K) G4. PACKED EARTH (A-K) G5. SOFT EARTH (A-K) G6. OFF-ROAD (A-K) G7. WATER (A-K) G8. AIR (A-K) G9. RAIL (A-K) H1. SMOOTHLY (A-K) H2. JERKY, acceleration/braking (A-K)  |___| |___| |___| |___| |___|  |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___|  141  (How long?)  I1. STATIONARY/IDLING (A-K) I2. <20KM/HR (A-K) I3. 20-40KM/HR (A-K) I4. 40-70KM/HR (A-K) I5. >70KM/HR (A-K)  VEHICLE 2 A. NAME  _______________________________________  B. DURATION (A-K) C. ARM REST ADJUSTED FOR YOU (Yes, No, not Applicable) D. SEAT ADJUSTED FOR YOU (Yes, No, not Applicable) E. BACK REST ADJUSTED FOR YOU (Yes, No, not Applicable) F. GOOD BACK SUPPORT (Yes, No, not Applicable) (How long?)  G1. SMOOTH PAVEMENT/CEMENT (A-K) G2. BROKEN PAVEMENT/CEMENT (A-K) G3. GRAVEL (A-K) G4. PACKED EARTH (A-K) G5. SOFT EARTH (A-K) G6. OFF-ROAD (A-K) G7. WATER (A-K) G8. AIR (A-K) G9. RAIL (A-K) H1. SMOOTHLY (A-K) H2. JERKY, acceleration/braking (A-K)  (How long?)  I1. STATIONARY/IDLING (A-K) I2. <20KM/HR (A-K) I3. 20-40KM/HR (A-K) I4. 40-70KM/HR (A-K) I5. >70KM/HR (A-K)  VEHICLE 3 A. NAME  |___| |___| |___| |___| |___|  |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___|  _______________________________________  B. DURATION (A-K) C. ARM REST ADJUSTED FOR YOU (Yes, No, not Applicable) D. SEAT ADJUSTED FOR YOU (Yes, No, not Applicable) E. BACK REST ADJUSTED FOR YOU (Yes, No, not Applicable) F. GOOD BACK SUPPORT (Yes, No, not Applicable)  |___| |___| |___| |___| |___|  142  (How long?)  G1. SMOOTH PAVEMENT/CEMENT (A-K) G2. BROKEN PAVEMENT/CEMENT (A-K) G3. GRAVEL (A-K) G4. PACKED EARTH (A-K) G5. SOFT EARTH (A-K) G6. OFF-ROAD (A-K) G7. WATER (A-K) G8. AIR (A-K) G9. RAIL (A-K) H1. SMOOTHLY (A-K) H2. JERKY, acceleration/braking (A-K)  (How long?)  I1. STATIONARY/IDLING (A-K) I2. <20KM/HR (A-K) I3. 20-40KM/HR (A-K) I4. 40-70KM/HR (A-K) I5. >70KM/HR (A-K)  VEHICLE 4 A. NAME  G1. SMOOTH PAVEMENT/CEMENT (A-K) G2. BROKEN PAVEMENT/CEMENT (A-K) G3. GRAVEL (A-K) G4. PACKED EARTH (A-K) G5. SOFT EARTH (A-K) G6. OFF-ROAD (A-K) G7. WATER (A-K) G8. AIR (A-K) G9. RAIL (A-K) H1. SMOOTHLY (A-K) H2. JERKY, acceleration/braking (A-K)  (How long?)  |___| |___| |___| |___| |___| |___| |___|  _______________________________________  B. DURATION (A-K) C. ARM REST ADJUSTED FOR YOU (Yes, No, not Applicable) D. SEAT ADJUSTED FOR YOU (Yes, No, not Applicable) E. BACK REST ADJUSTED FOR YOU (Yes, No, not Applicable) F. GOOD BACK SUPPORT (Yes, No, not Applicable) (How long?)  |___| |___| |___| |___| |___| |___| |___| |___| |___|  I1. STATIONARY/IDLING (A-K) I2. <20KM/HR (A-K) I3. 20-40KM/HR (A-K) I4. 40-70KM/HR (A-K) I5. >70KM/HR (A-K)  |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___| |___|  143  PART E – HEALTH HISTORY 25.  DID YOU EXPERIENCE ANY LOW BACK PAIN TODAY? (Yes, No)  |___|  (IF NO, SKIP TO 28) 26.  RATE LOW BACK PAIN TODAY (0-10)  27.  A. CHANGE WORK ACTIVITIES TODAY (Yes, No)  |___|___| |___|  B. IF YES, explain how __________________________________________ __________________________________________ __________________________________________ 28.  EXPERIENCE ANY LOW BACK PAIN LAST 6 MONTHS (Yes, No)  29.  WORST LOW BACK PAIN LAST 6 MONTHS (0-10)  |___|___|  30.  AVERAGE LOW BACK PAIN LAST 6 MONTHS (0-10)  |___|___|  31.  NUMBER OF DISABILITY DAYS IN LAST 6 MONTHS  |___|___|___|  |___|  FROM USUAL ACTIVITIES (WORK, SCHOOL OR HOUSEWORK) 32.  INTERFERENCE WITH DAILY ACTIVITIES (0-10)  |___|___|  33.  LOW BACK PAIN CHANGING RECREATIONAL ACTIVITIES (0-10)  |___|___|  34.  CHANGE ABILITY TO WORK (0-10)  |___|___|  35.  HOW MANY DAYS A WEEK OF EXERCISE (30 MIN) (0-7)  |___|  CONCLUSION 1.  CONTACT IN FUTURE (Yes, No)  2.  OTHER THINGS TO KNOW  |___|  ___________________________________________________________________________ ___________________________________________________________________________ 3.  COMMENTS  ___________________________________________________________________________ ___________________________________________________________________________ ___________________________________________________________________________  144  145  -2.95µV offset applied to data from the P4 device  Analysis Phase  Built -in preamplifier  Disposable Ag-AgCl wet-gel Blue Sensor EMG 30mm x 22mm electrodes , 25mm apart  VC and EMG interpolated to match up 7.6 and 10Hz sampling rates  Data Collection Phase  EMG segments concatenated and matched in time with the VC data  Data imported into GAS software for analysis  Data logged at 10Hz  RMS applied  Raw signal band pass filtered (8500Hz)  Raw signal sampled at 1000Hz  Appendix F: EMG data collection an processing schematic  Breaks removed and any clear motion artifacts/noisy segments removed.  (Mega Electronics, Finland) Range of -/+ 6500 µV Total gain of 412, CMRR of 110 dB Input impedance > 10 GOhms.  Mega electronics ME3000 P4/P8  Appendix G: Detailed EMG data cleaning methods There are three sets of EMG/VC data files resulting from the Back Study data collection. This was done to maximize the length of analyzable data for each phase of analysis. For a study comparing the EMG and VC measurements, data was analyzed for the maximum amount of time EMG AND VC were collected simultaneously. However, one device often measured for longer than the other, and sometimes only one was used. In these cases, the data was re-analyzed to also record summary statistics for the longest EMG and VC recordings. Both EMG-only and VC-only files were used for the Back Study modeling. However, this thesis uses the EMG-only file as the only direct measure data source. Before being analyzed and used for modeling, data needed to be cleaned of any errors which occurred during measurement or analysis. Such errors could result from many sources: 1) Measurement phase a) One of the channels came completely loose or the cable broke and collected virtually nothing (results in very low values) b) Poor skin connection or excessive rubbing/ pressure on the electrode led to motion artifact (results in very high values) 2) Analysis phase a) One of the measurement units (code-named ‘Bubba’) had an offset on all data such that the baseline was -2.95 rather than 0. If this offset was not added back on to files collected by this device, it would result in a systematic -2.95uV offset and negative values. Checking the lowest percentile values should identify this, no percentile value should be below zero. b) In the GAS software, calibration multipliers or coefficients must be applied to the data in order to calibrate for compression or %RC. If incorrect calibration multipliers were applied, this would also lead to a systematic error and nonsensical data. This would be harder to identify, so one would need to check for outliers. Descriptive statistics (mean, sd, max, min, etc.) were used to identify errors in the data. The outcome measures that were targeted for data clean-up were: mean, sd, p01, p05, p10, p50, p90, p95, duration. The P99, p100, MPF, 10-90range (overlap with above), cumulative variables were not considered. Because 100th and 99th percentile are nearly always unreasonably high due to instantaneous noise spikes, these values were not considered in data cleaning nor included in any analyses. Distributions of exposure metrics were also graphed into a histogram to show the presence of outliers and the degree of skewness in the data. During this process, certain characteristics or ‘flags’ were used to identify measurement days which may have excessive noise or systematic errors in analysis: 1) High outliers – the highest value of nearly every metric was investigated. 2) Blank or zero values 3) Negative values 4) Disparate left and right values: more than 20% difference between left and right values 5) Very low values: if the p1, p5 and p10 were all the same AND p50-p95 were all low (<700) it was determined that the channel was faulty and not delivering data. In this case job and tasks are checked to see if super-low activity is plausible. This usually only happened on one channel at a time, indicating that it was not the work which led to low values, so it was determined that the cable had become detached and did not collect.  146  Once a measurement day was identified or ‘flagged,’ it was investigated more thoroughly by checking the original EMG text file to see if the strange values (i.e. negative values or high values) originated there. This helped determine if it was a measurement problem or an analysis problem. If it seemed to be a data issue, the observations made on that worker were checked for that day, along with observer’s notes and summaries. If the EMG values made sense given the types of activities, then the data was left as-is. If the original text file looked normal, the errors were thought to have originated in analysis. In this case, the GAS analysis record was checked to make sure the calibration multipliers matched those on the GAS output. In addition, the signals of the GAS-processed files were graphed to ensure that the duration was selected appropriately and any breaks/noise had been removed. During initial GAS analysis, each file was inspected to make sure that the baseline was not wandering up or down, and especially for traces that did not return to zero for several seconds. Decision rules for removing EMG data: 1) Very high values: if means were above 5000 or p90 above 8000, it was determined that the values were unrealistically high 2) Very low values: if the p1, p5 and p10 were all the same AND p50-p95 were all low (<700) it was determined that the channel was faulty and not delivering data. In this case job and tasks are checked to see if super-low activity is plausible. This usually only happened on one channel at a time, so it was determined that the cable had become detached and did not collect. 3) Disparate left and right values: more than 20% difference between left and right values 4) Negative values: any negative EMG %RC or CNEMG values are not possible and need to be reanalyzed 5) If one summary statistic (mean, 5th percentile, etc) is unacceptable it is assumed that all data for that measurement day is unacceptable. However, 100th and 99th percentile are nearly always unreasonably high due to noise. Therefore, these values are not considered in data cleaning and will not be included in any analyses. Once a problem was identified, changes were made to a copy of the original data file An SPSS syntax file with all changes in it was retained for any future data cleaning efforts, or if a new file was compiled form the original data. The full list of changes to the EMG data is found in Table G.1.  147  Table G.1: Record of Changes to the EMG data Measurement day  99 143 144  Error/notes October 2006 data reduction Too spiky to change. remove Reanalysis did not help, remove Left is high, use right only Right side is high, use left side only Very high left side compressions 149 high right side p95 High left and right sd Negative right side p1 and 5 Left comp p5 is too high Inclinometer data was not acceptable, Reanalyzed for “EMG only” file Jan 30 2007 Data reduction Left side too high Left side too high Right side is too high  153 195 196 214 156, 169 60  left side is too high left side is too high left side is too high Right side is too high High left side compressions Left too high, right too low Only partial data available  97, 98 215, 181 220 153 134 149, 94 57, 80 82 105, 208, 31, 48, 153  Change Remove Remove Remove left side Remove right side Remove all Remove all Keep it Re-analyze it (Robin’s file only) Remove left side Usable once reanalyzed Removed left, Kept right side Get rid of left side and averages get rid of right and average, keep left. Get rid of left side and averages Get rid of left side and averages Get rid of left side and averages Get rid of right side and averages Removed removed Removed all data  148  Appendix H: Detailed EMG analysis methods EMG Data Conditioning The goal of EMG data analysis/preparation was to produce a collection of summary statistics for each worker for use in determinants of exposure modeling. Raw EMG voltage is not very informative of physical loads. Because differences in electrode-skin interface change the absolute voltage between individuals and within individuals at different times, EMG voltage needs to be ‘normalized’ or transformed into other units. This study used percentage reference contraction (%RC) as described in Chapter 3. EMG data analysis The EMG summary statistics were generating using a LABVIEW-based analysis program called the ‘Goniometer Analysis Software’ or ‘GAS’. GAS was developed at the University of Washington by Dr. Peter Johnson and Jim Ploger for customized analysis of joint angle data collected via electrogoniometer. The similarity between the digitized electrical signals generated by EMG and electrogoniometers make GAS applicable to EMG data as well. For the UBC Back Study, the GAS software was used to analyze trunk inclinometer (called ‘VC’) and synchronize it with the EMG. For that reason, some inclinometer outputs will be discussed here as well. The GAS software calculates 18 different exposure metrics for each of 15 exposure channels. An exposure channel is the raw or modified signal from the VC or the EMG. These exposure channels are shown in Table H.1. Table H.1: Original Exposure channels, variable codes, and full variable name Source Data Channel Torso F/E 1st drv. Torso F/E Torso Lateral 1st drv. Torso Lateral L Erector SpinaeE R Erector Spinae (L ES)-(R ES) L VC predicted EMG R VC predicted EMG  Code TFE TF1 TL TL1 LES RES L_R LVC RVC  (L VC EMG )-(L ES)  L_L  (R VC EMG )-(R ES)  R_R  L Compression R Compression VC Compression  LC RC VCC  Avg EMG Compression  AC  Full name of variable Torso Flexion extension (in degrees) Torso Flexion extension speed (in degrees/sec) Torso Lateral flexion (in degrees) Lateral flexion speed (in degrees/sec) Left Erector Spinae muscle activity (in %RC) Right Erector Spinae muscle activity (in %RC) Difference between left and right muscle activity (in %RC) Left muscle activity predicted by VC (in %RC) Right muscle activity predicted by VC (in %RC) Difference between left VC-predicted muscle activity and measured EMG muscle activity (in %RC) Difference between right VC-predicted muscle activity and measured EMG muscle activity (in %RC) Spinal Compression estimated from Left EMG (Newtons) Spinal Compression estimated from Right EMG (Newtons) Average spinal Compression estimated from VC (Newtons) Spinal Compression estimated from average of left and Right EMG (Newtons)  149  Table H.2 shows exposure channels which were derived from the original exposure channels. For example, left and right side EMG were not always available to calculate an average EMG. The variable ‘best EMG’ is the average if available, or if not available, the exposure of the only channel which was available. Most of the exposure metrics had a relatively normal distribution. However, the distribution of the standard deviations was log-normal, therefore standard deviations were log-transformed (base e) before performing any parametric tests. Because there was a non-negligible number of 0 values, the 1st, 5th, and 10th percentiles distributions were leftcensored. To account for this, the 0 values of these exposure metrics were replaced by the limit of detection divided by √2 ; the limit of detection was the lowest resolution of the EMG recorder (2.95uV) divided by the individual’s reference contraction EMG measurement. Table H.2: Modified exposure channels, variable codes, and full variable name Derived exposure channel BEST erector spinae activity  Code BES  BEST compression  BC  Log-transformed SD  LN  LOD 1st percentile  LODp1  LOD 5th percentile  LODp5  LOD 10th percentile  LODp10  Full name of variable The right side muscle activity if it is available. If right side not available, then take the left side muscle activity. (%RC) The average of left and right compression if it is available. If not, either the left or right as available. (Newtons) Log-transformed version of the variable. Used for standard deviation of EMG values. st The APDF 1 percentile with zero values replaced by the limit of detection divided by root two. th The APDF 5 percentile with zero values replaced by the limit of detection divided by root two. th The APDF 10 percentile with zero values replaced by the limit of detection divided by root two.  The same 17 summary statistics were developed for each exposure channel, including APDF values, mean, standard deviation, and cumulative exposure. Two additional summary statistics were developed for spinal compression: percent time above 3400N and 6800N. The summary statistic variables are detailed in Table H.3.  150  Table H.3: Summary Statistics, variable codes, and full variable name Summary Statistic (+/-) Median (+/-) Mean (+/-) S.D. (+/-) Cumulative (+/-) RCMS (+/-) RCM (+/-) p=0.01 (+/-) p=0.05 (+/-) p=0.10 (+/-) p=0.50 (+/-) p=0.90 (+/-) p=0.95 (+/-) p=0.99 (+/-) p=1.00 (+/-) p=0.95-0.05 (+/-) Res (Hz) (+/-) (0-5Hz)MPF(0.016) Threshold 0.5 Threshold 1.0  Code med mean sd cum rcms rcm p1 p5 p10 p50 p90 p95 p99 p100 p955 res mpf thr5 thr1  Full name of variable th Median, 50 percentile (in units of channel) Arithmetic mean (in units of channel) Standard deviation (in units of channel) Cumulative (in units of channel x seconds) Rate of change metric/standard deviation (no units) Rate of change metric (in units of channel) st 1 percentile (in units of channel) th 5 percentile (in units of channel) 10th percentile (in units of channel) th 50 percentile (in units of channel) th 90 percentile (in units of channel) th 95 percentile (in units of channel) th 99 percentile (in units of channel) th Max 100 percentile (in units of channel) th th Range between 5 and 95 percentile (in units of channel) Resolution (units per second) Mean power frequency (units per second Percentage of time spent above 3400N Percentage of time spent above 6800N  151  Appendix I: Supplementary results: worker population The following tables describe different characteristics of the workers in each industry and or all industries combined. Table I.1: Participants’ personal and demographic factors by industry C  F  WP  W  T  Sex (% male) 100% 100% 97.2% 92.3 89.1 Identified as a 51.5% 26.3% 38.9% 53.8% 34.8% claimant in 2001 Mean height in cm 179 (6.1) 176 (7.3) 177.6 180.3 177.0 (sd) (8.9) (8.9) (7.3) Mean weight in kg 80.9 89.9 83.6 85.7 84.9 (sd) (11.4) (19.2) (15.9) (15.7) (16.2) Age in years on 43 7 (9.8) 48.5 (9.8) 38.9 (9.6) 38.8 43.0 sampling day (sd) (11.3) (13.2) C= construction, F = forestry, WP= wood products, W = warehousing, T = transporting  All industries 95.3 59.5 178.1 (7.9) 85.2 (16.1) 42.2 (12)  Table I.2: Participants’’ work hours and related factors C  F  WP  W  T  Mean hours per day 8.39 (.78) 7.97 (1.5) 9.6 (1.8) 8.3 (1.1) 8.3 (1.2) (sd) Mean days worked 5.14 (.36) 5.08 (.49) 4.4 (.77) 4.8 (.39) 4.5 (.73) per week (sd) Mean hours worked 43.0 (5.2) 40.7 (9.7) 42.2 (7.3) 49.6 (3.0) 37.2 (7.1) per week (sd) Mean Consecutive 2.4 (1.3) 3.5 (2.1) 2.6 (1.1) 3.0 (1.7) 3.0 (1.7) days worked at the time of measurement (sd) Total commuting 80.4 58.5 (36) 51.1 44.3 52.1 time per day in min (56.3) (32.2) (27.3) (42.4) (sd) C= construction, F = forestry, WP= wood products, W = warehousing, T = transporting  All industries 8.5 (1.4) 4.8 (.64) 40.3 (7.1) 2.9 (1.7)  55.9 (40)  152  Table I.3: Participants’ pain and activity reporting C  F  WP  W  T  All industries 41.6  % reporting back 39.5 52.6 47.2 35.9 34.1 pain on measurement day Mean back pain 3.6 (1.7) 3.3 (2.0) 3.2 (1.8) 3.2 (1.3) 2.9 (.79) 3.2 (1.6) rating for + measurement day (sd) % reporting change 9.1 25.0 11.8 21.4 40.0 75.1 in work activities due to back pain % reporting back 67.9 84.2 83.3 74.4 65.9 pain in the last 6 months Mean ‘worst’ back 5.8 (2.7) 6.7 (2.7) 6.4 (2.3) 6.7 (2.1) 5.6 (2.4) 6.3 (2.5) pain in last 6mo+ (sd) Mean ‘typical’ back 2.9 (2.0) 3.3 (2.0) 2.8 (1.5) 3.3 (1.3) 2.8 (1.6) 3.0 (1.7) + pain in last 6mo (sd) Mean number of 7.0 (27) 1.2 (3.8) 1.0 (2.1) 1.7 (4.5) 2.5 (7.1) 2.3 (11) disability days due to back pain (sd) Mean change in 2.0 (2.2) 2.9 (2.1) 2.3 (2.2) 2.6 (2.4) 1.9 (2.4) 2.4 (2.3) activity due to back pain in last 6 mo (sd) * Mean change in 1.8 (2.7) 2.4 (2.5) 1.7 (2.6) 2.4 (2.7) 2.0 (2.6) 2.1 (2.6) recreational activity due to back pain in last 6 mo (sd) * Mean change in 1.5 (2.1) 2.3 (2.4) 1.8 (2.4) 2.3 (2.6) 1.9 (2.2) 2.0 (2.4) work activity due to back pain in last 6 mo (sd) * Mean days per week 3.0 (2.6) 3.5 (2.6) 2.9 (2.1) 3.3 (2.3) 3.6 (2.5) 3.3 (2.4) with 30min or more exercise (sd) C= construction, F = forestry, WP= wood products, W = warehousing, T = transporting scale of 0 -10, 0 = no pain at all and 10 = worst pain imaginable scale of 0 -10, 0 = no change in activities and 10 = unable to perform listed activities  153  154  β (slope ) .386 .620 -.316 .602  R2  Mean p  .002 .007 .012 .117 .003 .210 .180 .145 .031 .034  .938 2.22 2.86 -.615 .126 2.15 5.36 2.0 -.753 .367  .030*  .057* .322 .197 *** .493 *** *** *** .038*  *** *** *** .057*  P  -7.02  43.1 70.4 66.1 -15.9 29.3 130.8 355 160.7 -23.1  β (slope ) 40 72.9 -14.4 63.3  .005  .002 .002 .002 .039 .090 .272 .258 .144 .014  .236 .101 .059 .030  R2  .419  .646 .604 .637 .020* *** *** *** *** .173  *** *** .004* .042*  p  .327 .002 .631 95.5 .056 .005* .155 0 .806 66.5 .030 .041* ** independent variables for cumulative are in total time,  .212 .101 .246 .026  β (slope ) .738 1.27 -.613 1.1  R2  Simple linear Regression Results 90th percentile %RC Cumulative %RC **  1.31 91.8 46.7 .207 *** % time standing 0 49.5 17.2 .087 *** % time walking 0 98.2 30.5 .232 *** % time sitting % time 0 69.1 3.74 .028 .050* Crouch/kneel/squatting 0 20.5 .18 .250 . .801 % time lying 0 6.57 1.33 .901 .004 .448 % time climbing 0 13.7 .37 1.19 .007 .312 % time in other postures o 3.7 94.1 44.5 -.352 136 *** % time with trunk 0-10 3.02 90.11 30.1 .105 .009 .278 % time with trunk 10-20° 0 42.2 12.0 1.109 .199 *** % time with trunk 20-45° 0 34.5 3.7 2.903 .188 *** % time with trunk 45-60g 0 58.3 4.76 1.057 .144 *** % time with trunk >60 % time with trunk 0 88.9 4.90 -.387 .029 .045* extended % time with trunk 0.51 98.2 21.6 -.192 .033 .033* supported 0 25.7 6.2 .260 .004 .471 % time with Lateral bend 0 30.3 5.55 -0.45 . .892 % time twisting * variables were significant at p<0.10 and offered to the multivariable models model not % time *** significant <0.0001  Postural variables  Observation Variables  Descriptive Results Min % Max % Mean time time  Table J.1: Descriptive statistics s and simple linear regression results for significant relationships between observed posture variables and EMG metrics by conceptual exposure groups  Appendix J: Supplementary results for Chapter 5: full bivariable results  155  Descriptive Results Min % Max % Mean time time β (slope ) .662  R  2  Mean p  .09 .101 .244 .108 .113 .144 .103 .029 .056 .251 .144 .099 .230  .317 5.57 .781 3.17 5.57 .781 2.34 2.5 .670 1.58 2.70 .892 1.25  ***  ***  ***  ***  *** *** *** *** *** .044* .005*  ***  ***  ***  P  69.6  39.3  109.8  82.0  41.8 170.8 243.3 42.4 71.9 72.8 37.9  97.3  203.6  β (slope ) 68.0  *** significant <0.0001  .102  2  β (slope ) .827  R  .234  .051  .070  .184  .194 .097 .082 .141 .024 .008 .067  .137  .138  .142  2  R  ***  .007*  .002*  ***  *** *** .001* *** .0718* .299 .002*  ***  ***  ***  p  Simple linear Regression Results 90th percentile %RC Cumulative %RC **  Manual materials handling variables % Time using power 0 78.6 2.05 .119 *** hand tool % time with Power hand 0 27.4 .45 1.91 .107 *** tool idling % time with Power hand 0 51.3 1.60 .942 .116 *** tool On 0 90.3 37.0 .437 .243 *** % time performing MMH 0 32.7 4.12 1.73 .115 *** % time Lifting 0 14 2.02 2.86 .106 *** % time Lowering 0 88.6 23.3 .428 .154 *** % time Holding 0 31.9 4.47 1.15 .089 *** % time Pushing 0 26.4 3.13 .881 .013 .184 % time Pulling % time handling loads at 0 82.3 18.0 .350 .054 .006* ‘near’ horiz. distance % time handling loads at 0 59.4 13.7 .864 .268 *** ‘middle’ horiz. distance % time handling loads at 0 28.7 5.2 1.3 .118 *** ‘extended’ horiz. distance % time handling loads 0 70.6 18.8 .468 .097 *** with 1 hand % time handling loads 0 64.4 18.1 .664 .232 *** with 2 hands * variables were significant at p<0.10 and offered to the multivariable models model ** independent variables for cumulative are in total time, not % time  Observation Variables  Table J.2. Descriptive statistics and simple linear regression results for significant relationships between observed MMH variables and EMG metric s by conceptual exposure groups  156  Descriptive Results Min % Max % Mean time time β (slope ) .161  R  2  Mean p  .083 .187 .071 0 .118 .01 .001  .814 2.01 1.3 .189 1.28 1.03 .736  .766  .248  ***  .847  .001*  ***  .001*  .300  P  *** significant <0.0001  .008  2  β (slope ) .316  R  20.9  83.0  33.1  -10.8  54.2  113.7  34.1  β (slope ) 12.9  .  .023  .019  .  .051  .239  .040  .004  2  R  .892  .076*  .105  .871  .008*  ***  .018*  .457  p  Simple linear Regression Results 90th percentile %RC Cumulative %RC **  Manual materials handling variables % time handling loads 0 68.6 9.53 .007 .318 <0.5kgs % time handling loads 0 65.1 14.2 .399 .071 .002* 0.5-4.5kgs % time handling loads 0 83.1 4.13 1.257 .260 *** 4.5-10kgs % time handling loads 0 74.6 3.21 .686 .068 .002* 10-20kgs % time handling loads 0 19.0 1.35 .009 . .986 >20kgs % time pushing or pulling 0 63.0 9.25 .644 .107 *** with ‘light’ force % time pushing or pulling 0 24.5 2.64 .438 .006 .355 with ‘moderate’ force % time pushing or pulling 0 10.2 .283 .294 . .823 with ‘heavy’ force * variables were significant at p<0.10 and offered to the multivariable models model ** independent variables for cumulative are in total time, not % time  Observation Variables  Table J.3. Descriptive statistics and simple linear regression results for significant relationships between additional observed MMH variables and EMG metric s by conceptual exposure groups  157  76.9 37.5 107.7 30 94.4 107.7 107.7 107.7 107.7 107.7 107.7 107.7  0 0 0 0 0 0 0 0 0 0 0  94 93 125  0  0 0 0  6.62  3.46  1.11  5.7  5.7  6.5  9.1  .43 3.7 .26 15.9  9.6  29.1 23.9 27  Descriptive Results Min % Max % Mean time time  .105  .210  .364  .197  .217  .194  .256  .006 .388 .520 .053  .408  β (slope) .059 .270 -.249  .008  .018  .029  .024  .021  .018  .043  . .041 .004 .002  .079  .006 .080 .144  .293  .115  .048*  .074  .089  .122  .015*  .991 .018* .449 .597  .381 .001* <0.000 1* .001*  4.21  8.69  16.6  6.89  11.8  7.47  11.6  -4.34 17.3 19.78 6.33  6.57  .002  .006  .010  .004  .011  .005  .015  0 .014 .001 0.05  .003  .004 .015 .031  .519  .392  .238  .462  .224  .403  .151  .913 .167 .705 .403  .514  .453 .155 .040*  p  1608  4805  8291  2309  4995  4427  5872  2884 7820 23043 1033  10555  .003  .015  .023  .005  .018  .014  .036  0 .026 .013 .001  .083  .015 .039 .075  R2  p  .525  .154  .075  .410  .123  .164  .028*  .828 .600 .184 .682  .001*  .160 .021* .001*  Cumulative %RC  β (slope) 2370 4773 -4543  R2  β (slope) .384 8.88 -8.82  R2 p  Simple Linear Regression Results 90th percentile %RC  Mean %RC  * variables were significant at p<0.05 and offered to the multivariable models  % time sitting % time Crouch/kneel/squatting % time lying % time climbing % time in other postures % time standing with trunk 0-10 deg % time standing with trunk 10-20 deg % time standing with trunk 20-45 deg % time standing with trunk 45-60 deg % time standing with trunk >60 deg % time standing with trunk extended % time standing with lateral bending % time standing with twisting  % time standing % time walking  Posture variables  Self-Report Variables  Table J.4. Descriptive statistics and simple linear regression results for significant relationships between self-reported posture variables and EMG metrics by conceptual groups  158  Descriptive Results Min % Max % Mean time time β (slope) .172 -1.07 6.08 28.5 34.5 35.1 35.3 35.2 -11.6 -3.2 -29.8 -13.0 -21.2 26.6  .059 .365 .372 .021 .009* .141 .449 .019* .002* .036* .229 .002* .021* .001*  .045  .029  .021  .005  .001  .026  .018  .003  .002  .019  .019  .002  0  .009  .013*  .048*  .093  .399  .702  .063  .119  .525  .631  .110  .116  .648  .928  .285  p  -9436  -7159  -6417  -12874  -4277  -4784  19045  3621  33258  1009  15862  4709  1631  .051  .030  .046  .009  .018  .039  .048  0  .014  0  .052  .008  .001  .013  R2  p  .008*  .045*  .012*  .274  .122  .021*  .011*  .845  .170  .855  .008*  .288  .677  .189  Cumulative %RC  β (slope) 3033  R2  β (slope) 7.43  R2 p  Simple Linear Regression Results 90th percentile %RC  Mean %RC  % time walking with trunk 0 93.3 18.3 .027 0-10 deg Time walking 10-20 deg 0 83.3 5.1 .140 .006 (% of working day % time walking with trunk 0 83.3 2.86 .156 .006 20-45 deg % time walking with trunk 0 50 1.54 .546 .040 45-60 deg % time walking with trunk 0 62.5 1.41 .564 .051 >60 deg % time walking with trunk 0 15 .30 1.41 .016 backwards % time walking with trunk 0 18.8 .45 .553 .004 sideways % time walking with trunk 0 50 1.10 .691 .041 twisting % time sitting with trunk 0 106 12.1 -.254 .071 upright % time sitting with trunk 0 87.5 8.8 -.229 .033 leaning forward % time sitting with trunk 0 37.5 .49 -.559 .011 backwards no support % time sitting with trunk 0 100 6.31 -.311 .069 backwards with support % time sitting with lateral 0 87.5 3.5 -.325 .039 bending % time sitting with 0 87.5 4.2 -.482 .085 twisting * variables were significant at p<0.05 and offered to the multivariable models  Posture variables  Self-Report Variables  Table J.5. Descriptive statistics and simple linear regression results for significant relationships between additional self-reported posture variables and EMG metrics by conceptual groups  159  Descriptive Results Min % Max % Mean time time β (slope ) .164  R2  Mean %RC  0 206 39.9 .123 % time performing MMH % time handling loads 0 87.5 7.1 .216 .022 <0.5kgs % time handling loads 0.50 87.5 10.3 .172 .019 4.5kgs % time handling loads 4.50 87.5 9.14 .317 .073 10kgs % time handling loads 100 106.3 7.24 .253 .047 20kgs % time handling loads 0 107.7 6.16 .212 .027 >20kgs % time handling loads at 0 87.5 11.7 .201 .034 ‘near’ horizontal distance % time handling loads at 0 87.5 11.8 .208 .032 ‘middle’ horizontal distance % time handling loads at 0 87.5 4.65 .644 .149 ‘extended’ horizontal distance * variables were significant at p<0.05 and offered to the multivariable models  Manual material handling factors  Self-Report Variables β (slope ) 8.05 4.8 5.19 17.6 16.1 6.93 11.5 5.9 38.0  .001* .086 .113 .001* .011* .056 .032 .037* <0.000 1*  p  .090  .005  .020  .004  .033  .039  .003  .002  .047  R2  <0.000 1*  .440  .106  .464  .034*  .021*  .533  .639  .012*  p  16849  3781  3621  2480  6927  4630  4228  2075  β (slope ) 2950  .159  .017  .017  .006  .055  .024  .018  .003  .063  R2  <0.0 001*  .127  .127  .006 * .380  .070  .125  .003 * .516  p  Simple Linear Regression Results 90th percentile %RC Cumulative %RC  Table J.6: Descriptive statistics and simple linear regression results for significant relationships between additional self-reported MMH variables and EMG metrics by conceptual groups  160  β (slope ) .148 .250  R  2  Mean %RC  Manual material handling factors 0 62.5 5.16 .005 % time pushing % time pushing with ‘light’ 0 62.5 2.7 .008 force % time pushing with ‘moderate’ 0 62.5 2.48 .091 .001 force % time pushing with ‘heavy’ 0 37.5 1.51 .057 0 force 0 87.5 7.94 -.066 .003 % time pulling 0 87.5 2.8 .027 0 % time puling with ‘light’ force % time puling with ‘moderate’ 0 83.3 3.85 -.073 -.006 force % time puling with ‘heavy’ 0 87.5 2.90 -.256 .003 force * variables were significant at p<0.05 and offered to the multivariable models  Self-Report Variables  Descriptive Results Min % Max % Mea time time n β (slope ) 8.5 5.51 2.68 11.3 -1.15 3.76 -2.35 -11.7  .429 .294 .702 .892 .558 .879 .650 .235  p  .004  0 .001 0  .001  0  .003 .001  2  R  .477  .893 .779 .848  .772  .883  .550 .761  p  -7679  -790 2209 -3207  -6180  2894  β (slope ) 4020 4528  .015  .001 .002 .005  .003  .002  .005 .004  2  R  .158  .782 .621 .430  .558  .632  .395 .453  p  Simple Linear Regression Results 90th percentile %RC Cumulative %RC  Table J.7. Descriptive statistics and simple linear regression results for significant relationships between self-reported pushing and pulling variables and EMG metrics by conceptual groups  161  -  Sex (percentage male) Height (cm) Weight (kg) Job variables Job title  -  68  -  42 (11.6) 95.3  -  -.007 -.001  β (slope ) -0.01  -  -.001 0  .004  R  2  Mean %RC  <.0001 * .385 .878 .809  .269 .780 .981  .439  p  .004 .001 .004 -  -  -  .010 .010  .001  2  R  -.119 3.67 -1.47  -  .495 .237  β (slope ) -.108  .011*  .468 .789 .449  .055*  .280 .232 .244  .699  p  -  3962 11428 -7452  -  12330 4583  β (slope ) -1283  -  .003 0 .001  -  .037 .021  .001  2  R  p  .043*  .502 .865 .776  .027*  .139 .025* .090  .732  Simple Linear Regression Results 90th percentile %RC Cumulative %RC  Days per week .036 2.32 .006 Hours per day -.001 .036 .000 Number consecutive days -.002 -.250 .000 worked Industry .003 • variables were significant at p<0.05 and offered to the multivariable models  18  Age  Personal factors  Self-Report Variables  Descriptive Results Min % Max % Mean time time  Table J.8. Descriptive statistics and simple linear regression results for significant relationships between additional self-reported posture variables and EMG metrics by conceptual groups  162  Mean  1.31 91.8 46.7 % time standing 0 49.5 17.2 % time walking 0 98.2 30.5 % time sitting % time 0 69.1 3.74 Crouch/kneel/squatting 0 20.5 .18 % time lying 0 6.57 1.33 % time climbing 0 13.7 .37 % time in other postures o 3.7 94.1 44.5 % time with trunk 0-10 3.02 90.11 30.1 % time with trunk 10-20 deg 0 42.2 12.0 % time with trunk 20-45 deg 0 34.5 3.7 % time with trunk 45-60 deg 0 58.3 4.76 % time with trunk >60 deg 0 88.9 4.90 % time with trunk extended 0.51 98.2 21.6 % time with trunk supported 0 25.7 6.2 % time with Lateral bend 0 30.3 5.55 % time twisting * variables were significant at p<0.10 and offered to the multivariable models model  Observation Variables Postural variables  Descriptive Results Min % time Max % time  8865 -1273 13403 -9292 5970 25091 62061 12801 -7784 -4091 14153 3496  .001 0 .001 .150 .044 .160 .135 .033 .019 .024 .018 .001  .723 .966 .651 <0.0001* .014 <0.0001* <0.0001* .031* .110 .071* .118 .678  Simple linear Regression Results Cumulative %RC β (slope) R2 p 7210 .114 <0.0001* 6970 .017 .123 -5447 .109 <0.0001* 12574 .019 .105  Table K.1: Descriptive statistics and simple linear regression results for significant relationships between observed postural variables and EMG outcomes by conceptual exposure groups  Appendix K: Supplementary results for Chapter 5: cumulative EMG using % time variables  163  Descriptive Results Max % time Mean  1.31 91.8 46.7 % time standing 0 49.5 17.2 % time walking 0 98.2 30.5 % time sitting % time 0 69.1 3.74 Crouch/kneel/squatting 0 20.5 .18 % time lying 0 6.57 1.33 % time climbing 0 13.7 .37 % time in other postures o 3.7 94.1 44.5 % time with trunk 0-10 3.02 90.11 30.1 % time with trunk 10-20 deg 0 42.2 12.0 % time with trunk 20-45 deg 0 34.5 3.7 % time with trunk 45-60 deg 0 58.3 4.76 % time with trunk >60 deg 0 88.9 4.90 % time with trunk extended 0.51 98.2 21.6 % time with trunk supported 0 25.7 6.2 % time with Lateral bend 0 30.3 5.55 % time twisting * variables were significant at p<0.10 and offered to the multivariable models model  Postural variables  Min % time  8865 -1273 13403 -9292 5970 25091 62061 12801 -7784 -4091 14153 3496  .001 0 .001 .150 .044 .160 .135 .033 .019 .024 .018 .001  .723 .966 .651 <0.0001* .014 <0.0001* <0.0001* .031* .110 .071* .118 .678  Simple linear Regression Results Cumulative %RC β (slope) R2 p 7210 .114 <0.0001* 6970 .017 .123 -5447 .109 <0.0001* 12574 .019 .105  Table K.2: Descriptive statistics and simple linear regression results for significant relationships between additional observed postural variables and EMG outcomes by conceptual exposure groups  164  Descriptive Results Max % time Mean  Manual materials handling variables % Time using power hand 0 78.6 2.05 tool % time with Power hand tool 0 27.4 .45 idling % time with Power hand tool 0 51.3 1.60 On 0 90.3 37.0 % time performing MMH 0 32.7 4.12 % time Lifting 0 14 2.02 % time Lowering 0 88.6 23.3 % time Holding 0 31.9 4.47 % time Pushing 0 26.4 3.13 % time Pulling % time handling loads at 0 82.3 18.0 ‘near’ horizontal distance % time handling loads at 0 59.4 13.7 ‘middle’ horizontal distance % time handling loads at 0 28.7 5.2 ‘extended’ horizontal distance % time handling loads with 1 0 70.6 18.8 hand % time handling loads with 2 0 64.4 18.1 hands * variables were significant at p<0.10 and offered to the multivariable models model  Min % time  .136 .134 .128 .090 .032 .052 .072 .013 .001 .030 .086 .033 .014 .129  17846 53650 24900 6719 22924 50516 7352 10900 4623 6575 12336 17313 4474 12506  <0.0001*  .166  .032*  <0.0001*  <0.0001* .036* .007* .001* .190 .783 .041*  <0.0001  <0.0001  <0.0001*  Simple linear Regression Results Cumulative %RC β (slope) R2 p  Table K.3: Descriptive statistics and simple linear regression results for significant relationships between observed MMH variables and EMG outcomes by conceptual exposure groups  165  Descriptive Results Max % time Mean  Manual materials handling variables % time handling loads 0 68.6 9.53 <0.5kgs % time handling loads 0.50 65.1 14.2 4.5kgs % time handling loads 4.50 83.1 4.13 10kgs % time handling loads 100 74.6 3.21 20kgs % time handling loads 0 19.0 1.35 >20kgs % time pushing or puling with 0 63.0 9.25 ‘light’ force % time pushing or puling with 0 24.5 2.64 ‘moderate’ force % time pushing or puling with 0 10.2 .283 ‘heavy’ force * variables were significant at p<0.10 and offered to the multivariable models model  Min % time  0 .013 .208 .044 .005 .011 .007 0  254 4352 28394 13799 -10429 5268 12085 -1519  .963  .311  .215  .425  .014*  <0.0001*  <0.0001*  .950  Simple linear Regression Results Cumulative %RC β (slope) R2 p  Table K.4: Descriptive statistics and simple linear regression results for significant relationships between additional observed MMH variables and EMG outcomes by conceptual exposure groups  166  Estimated proportion of variance explained by model  4.5-10kg load in hands (% time)  Trunk position 45-60 (% time)  o  Trunk position 20-45 (% time)  o  Trunk position 0-10 (% time)  o  Variable Observation Intercept (average for all subjects)  25164  11880  658164  β (slope)  Cumulative %RC  27.2%  <.0001  0.0218  p  Table K.5. Observations and self-reported ergonomic variables associated with median, 90th percentile, and cumulative EMG exposure in final linear regression equations.  167  .236 .022 .084  106 66.4 51.4  β (slope)  Mean %RC R2  .001  .045  <0.0001  p  105  131  180  .099  .032  .192  <0.0001  .036  <0.0001  90th percentile %RC β (slope) R2 p  590662  1204403  2402280  .019  .016  .205  p  .107  .136  <0.0001  Cumulative %RC β (slope) R2  ‘Any’ lifting above 10lbs looks like a better variable than the original lifting categories, but is similar to the variable ‘MMH lifting’. The trunk bending categories are far more consistently significant over all the EMG variables than ‘any trunk bending’. As a result, the trunk posture and MMH force variables were maintained as categories rather than as binary variables.  % time spent lifting more than 10lbs % time spent bending more than 45 % time spent bending more than 20  Observation Variables  Table L.1: Bivariable regression for derived MMH and trunk posture variables  Lifting and bending often coincide in occupational tasks, so the possibility of an interaction between these exposures was explored via an additional variable created by calculating the percentage of time spent handling a load greater than 4.5 kg while simultaneously flexing the trunk forward 20o or more. This variable was offered only when both a lifting variable and forward flexion variable were significant in the initial model, and was introduced as a final step. Null models (a model including only ‘subject’ or ‘company’ as a random effect and no fixed effects) were developed to determine if within-worker or within-company correlations had a significant impact on exposure estimations. Only ‘subject’ was included as a random effect term in the multivariate modeling.  Appendix L: Supplementary results for Chapter 5: prediction modeling findings for collapsed variables  Appendix M: Supplementary results for Chapter 7: exposure by post-hoc groups Table M.1: Development of Post-hoc groups for mean EMG Category All group 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 All group 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 All group 3  Industry  Job  Transportation Forestry Transportation Transportation Forestry Wood and Paper Products Transportation  Bus driver Truck driver Truck driver Heavy equipment operator Logging Machinery Operators Forklift operator Air Transport Ramp Attendants  Transportation Construction Forestry Construction Transportation Forestry  Storekeepers and Parts Clerks Forklift operator Boomman Construction trades, nes Automotive mechanic Helicopter pilot  Post hoc 3  Wood and Paper Products  Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 All group 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 All group 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5  Transportation Transportation Warehousing Wood and Paper Products Transportation  Papermaking and Coating Control Operator Bus cleaner Construction supervisor Forklift operator Cabinet maker Ferry worker  Wood and Paper Products Forestry Transportation Construction Wood and Paper Products Forestry Forestry Construction Warehousing  Lumber grader, puller Saw filer Warehouse person Asphalt worker Log chipper/grinder Heavy equipment operator Heavy-duty equipment mechanic Construction supervisor Warehouse person  Construction Construction Construction Construction Forestry Forestry  Cabinet maker Construction carpenter Construction labourer Bricklayers Faller Construction labourer  Mean EMG 23.5 14.3 18.0 19.7 20.4 25.3 25.8 28.3 32.2 31.0 31.8 31.9 32.4 32.9 33.2 36.0 34.3 34.5 35.0 36.0 36.8 38.1 40.2 38.5 38.6 39.1 39.3 39.7 40.4 41.8 42.0 42.3 58.4 45.5 50.0 56.6 56.9 67.9 113.8  168  Table M.2: Development of Post-hoc groups for peak EMG Category All group 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 All group 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 All group 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 All group 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 All group 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5 Post hoc 5  Industry  Job  Transportation Forestry Forestry Forestry Transportation Transportation Wood and Paper Products  Bus driver Truck driver Helicopter pilot Logging Machinery Operators Truck driver Heavy equipment operator Forklift operator  Construction Transportation Transportation Forestry Transportation Construction  Forklift operator Storekeepers and Parts Clerks Air Transport Ramp Attendants Boomman Ferry worker Construction trades, nes  Warehousing Wood and Paper Products Wood and Paper Products Forestry Wood and Paper Products Transportation Transportation Forestry  Forklift operator Papermaking and Coating Control Operator Cabinet maker Saw filer Lumber grader, puller Automotive mechanic Construction supervisor Heavy-duty equipment mechanic  Warehousing Construction Wood and Paper Products Transportation Transportation Construction Forestry  Warehouse person Construction supervisor Log chipper/grinder Warehouse person Bus cleaner Ashphalt worker Heavy equipment operator  Construction Construction Construction Construction Forestry Forestry  Cabinet maker Construction carpenter Bricklayers Construction labourer Faller Construction labourer  Peak EMG 47.9 35.6 45.5 47.2 49.9 50.2 52.8 54.2 70.2 65.6 68.8 70.5 71.4 72.0 73.0 79.2 77.1 77.7 77.9 79.9 80.2 80.2 80.5 80.5 88.9 84.5 86.6 87.6 89.5 90.7 91.0 92.0 122.3 102.8 106.4 107.6 115.6 141.8 159.4  169  Table M.3: Development of Post-hoc groups for cumulative EMG Category  Cumulative EMG 6366 3383 5503 6105 6408 6511 7249 7337 7363 7437 8866 7952 8666 8752 9066 9130 9245 9253 10462  Industry  Job  Transportation Transportation Forestry Transportation Construction Transportation Warehousing Transportation Construction  Warehouse person Automotive mechanic Truck driver Air Transport Ramp Attendants Cabinet maker Bus cleaner Warehouse person Heavy equipment operator Construction supervisor  Post hoc 2 Post hoc 2  Forestry Transportation Wood and Paper Products Warehousing Construction Wood and Paper Products Construction  Helicopter pilot Truck driver Lumber grader, puller Forklift operator Ashphalt worker Cabinet maker Construction carpenter  All group 3 Post hoc 3  Forestry  Saw filer  9411  Construction  9840  All group 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 Post hoc 1 All group 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2 Post hoc 2  Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 Post hoc 3 All group 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 Post hoc 4 All group 5 Post hoc 5  Forestry Transportation Wood and Paper Products Construction Construction Forestry  Construction trades, nes Papermaking and Coating Control Operator Boomman Bus driver Forklift operator Construction labourer Forklift operator Logging Machinery Operators  Wood and Paper Products Forestry Transportation Forestry Transportation  Log chipper/grinder Heavy equipment operator Ferry worker Faller Construction supervisor  Post hoc 5 Post hoc 5 Post hoc 5  Construction Transportation Forestry  Wood and Paper Products  Forestry  Heavy-duty equipment mechanic Bricklayers Storekeepers and Parts Clerks Construction labourer  9949 10242 10628 10801 11028 11124 11130 13485 12840 13152 13359 14011 14062 15806 14353 14655 17066 17151  170  Appendix N: Copy of research ethics certificate  171  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0066568/manifest

Comment

Related Items