UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Empirical studies in opioid dependence Nosyk, Bohdan 2010

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

Item Metadata

Download

Media
24-ubc_2010_spring_nosyk_bohdan.pdf [ 4.94MB ]
Metadata
JSON: 24-1.0069848.json
JSON-LD: 24-1.0069848-ld.json
RDF/XML (Pretty): 24-1.0069848-rdf.xml
RDF/JSON: 24-1.0069848-rdf.json
Turtle: 24-1.0069848-turtle.txt
N-Triples: 24-1.0069848-rdf-ntriples.txt
Original Record: 24-1.0069848-source.json
Full Text
24-1.0069848-fulltext.txt
Citation
24-1.0069848.ris

Full Text

EMPIRICAL STUDIES IN OPIOID DEPENDENCE    by   Bohdan Nosyk   MA, University of British Columbia, 2003     A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF    DOCTOR OF PHILOSOPHY    in    THE FACULTY OF GRADUATE STUDIES   (Health Care and Epidemiology)     THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)    April 2010    © Bohdan Nosyk, 2010 ii  ABSTRACT  Background: Opioid dependence is a chronic disease characterized by periods of relapse and remission.  Methadone maintenance treatment (MMT) can be effective, though not all patients can be successfully maintained in treatment.  Further, treatment entrants often use multiple illicit drugs, a fact whose motive and effect are under-studied.  Rigorous evaluation at the aggregate- and patient-level is required to maximize the public health benefits of substance abuse treatment.  Objectives: (1) identify the determinants of the time to discontinuation of MMT across multiple treatment episodes; (2) determine the effect of price on drug consumption among polydrug users; (3) identify trends in compliance to MMT dosing guidelines in British Columbia, Canada from 1996-2007; (4) provide a comparative analysis of the psychometric properties of eight measures of health status for opioid users; (5) evaluate differences in trajectories of health status among opioid users following treatment enrolment.  Methods: We draw upon longitudinal data collected in administrative databases (study 1 and 3), experimental (study 4 and 5) and observational cohort studies (study 2) to shed light on the life course of treatment and relapse that opioid addicts experience, their decisions to consume addictive substances, and treatment evaluation at the aggregate- and patient-levels.  Results: This collection of studies found (1) patients experiencing multiple treatment episodes tended to stay in treatment for progressively longer periods in later episodes; (2) iii  While heroin and crack cocaine were price inelastic polydrug users treated crack cocaine as a substitute for heroin.  In terms of treatment evaluation at the aggregate-level, compliance to minimally effective dose guidelines, along with 12-month retention figures have fallen in the past 5 years in British Columbia (study 3).  At the patient-level each of the health status measures assessed in the North American Opiate Medication Initiative (NAOMI) trial showed limitations (study 4), however health status can be meaningfully improved through effective, patient-centered opioid substitution treatment (study 5). Conclusions: The results of these novel empirical studies suggest that while MMT can be an effective treatment option for opioid dependence, the individual and program- related factors that limit its effectiveness must be addressed to maximize the public health impact of this treatment. iv  TABLE OF CONTENTS  ABSTRACT........................................................................................................................ ii TABLE OF CONTENTS................................................................................................... iv LIST OF TABLES............................................................................................................. ix LIST OF FIGURES .......................................................................................................... xii ACKNOWLEDGEMENTS............................................................................................. xiv DEDICATION................................................................................................................ xvii CO-AUTHORSHIP STATEMENT................................................................................ xvii 1 INTRODUCTION ........................................................................................................... 1 1.1 Opioid Dependence: Epidemiology, Health and Economic Burden ........................ 1 1.2 Opioid Dependence as a Chronic, Recurrent Disease .............................................. 2 1.3 Available Treatment.................................................................................................. 4 1.4 Treatment Evaluation................................................................................................ 7 1.4.1 Evaluation in Observational Settings ................................................................ 8 1.4.2 Evaluation in Experimental Settings................................................................ 10 1.5 Prevalence and Effects of Polydrug Use................................................................. 13 1.6 Study Hypotheses, Objectives and Thesis Organization ........................................ 18 1.7 References............................................................................................................... 20 2 PROPORTIONAL HAZARDS FRAILTY MODELS FOR RECURRENT METHADONE MAINTENANCE TREATMENT.......................................................... 36 2.1 Introduction............................................................................................................. 36 2.2 Materials and Methods............................................................................................ 38 v  2.2.1 Cohort Definition ............................................................................................. 38 2.2.2 Determinants of Time to Discontinuation of MMT.......................................... 39 2.2.3 Statistical Analysis ........................................................................................... 40 2.3 Results..................................................................................................................... 44 2.4 Discussion ............................................................................................................... 48 2.5 Acknowledgements................................................................................................. 53 2.6 References............................................................................................................... 62 3 AN EMPIRICAL SPECIFICATION OF THE RATIONAL ADDICTION MODEL FOR ILLICIT DRUG USE............................................................................................... 69 3.1 Introduction............................................................................................................. 69 3.2 Theoretical Framework........................................................................................... 75 3.3 Empirical Framework ............................................................................................. 80 3.4 Data ......................................................................................................................... 84 3.4.1 Study Sample .................................................................................................... 84 3.4.2 Illicit Drug Prices ............................................................................................ 84 3.5 Empirical Implementation ...................................................................................... 87 3.6 Results..................................................................................................................... 91 3.7 Conclusions............................................................................................................. 95 3.8 Acknowledgements................................................................................................. 98 3.9 References............................................................................................................. 107 4 TRENDS IN METHADONE MAINTENANCE TREATMENT PARTICIPATION, RETENTION AND COMPLIANCE TO DOSING GUIDELINES IN BRITISH COLUMBIA, CANADA: 1996-2006............................................................................. 114 vi  4.1 Introduction........................................................................................................... 114 4.2 Methods................................................................................................................. 118 4.2.1Patient Population .......................................................................................... 118 4.2.2 Data Analysis ................................................................................................. 118 4.3 Results................................................................................................................... 121 4.4 Discussion ............................................................................................................. 124 4.5 Acknowledgements............................................................................................... 130 4.6 References............................................................................................................. 138 5 AN EMPIRICAL COMPARISON OF EIGHT HEALTH STATUS MEASURES FOR CHRONIC OPIOID DEPENDENCE ............................................................................ 146 5.1 Introduction........................................................................................................... 146 5.2 Methods................................................................................................................. 149 5.2.1 Population...................................................................................................... 149 5.2.2 Analysis .......................................................................................................... 149 5.3 Results................................................................................................................... 155 5.3.1 Content Validity ............................................................................................. 155 5.3.2 Floor or Ceiling Effects ................................................................................. 156 5.3.3 Internal Consistency ...................................................................................... 156 5.3.4 Construct Validity .......................................................................................... 157 5.3.5 Responsiveness............................................................................................... 158 5.3.6 Overall Ratings .............................................................................................. 158 5.4 Discussion ............................................................................................................. 160 5.4.1 Conclusions.................................................................................................... 165 vii  5.5 Acknowledgements............................................................................................... 166 5.6 References............................................................................................................. 176 6 HEALTH RELATED QUALITY OF LIFE GROWTH TRAJECTORIES AMONG PATIENTS IN OPIOID SUBSTITUTION TREATMENT........................................... 184 6.1 Introduction........................................................................................................... 184 6.2 Methods................................................................................................................. 187 6.2.1 Study Sample .................................................................................................. 187 6.2.2 Outcome Variable .......................................................................................... 188 6.2.3 Explanatory Variables ................................................................................... 188 6.2.4 Data Analysis ................................................................................................. 189 6.3 Results................................................................................................................... 192 6.3.1 Initial Growth Modeling Analysis.................................................................. 192 6.3.2 Latent Class Growth Analysis........................................................................ 193 6.3.3 Growth Modeling Analysis Accounting for Class Membership..................... 193 6.3.4 Factors Associated with HRQoL Trajectory Class Membership................... 194 6.3.5 Factors Associated with Changes in HRQoL ................................................ 194 6.4 Discussion ............................................................................................................. 196 6.5 References............................................................................................................. 207 7 CONCLUSIONS ......................................................................................................... 212 7.1 Summary of Findings and Implications................................................................ 212 7.2 Study Strengths and Limitations........................................................................... 215 7.3 Further Study ........................................................................................................ 221 7.4 Conclusions........................................................................................................... 222 viii  7.5 References............................................................................................................. 224 APPENDIX A:  MEDICAL PROFITEERING: THE ECONOMICS OF METHADONE DISPENSATION............................................................................................................ 225 A1. Editorial................................................................................................................ 225 A2. References............................................................................................................ 231 APPENDIX B: BC PHARMANET DATABASE CLEANING.................................... 233 B1. PharmaNet database: Methadone Dispensations.................................................. 233 B2. PharmaNet Database: Other Drugs ...................................................................... 239 APPENDIX C:  MATHEMATICAL DEFINITIONS OF TREATMENT PHASE TRANSITIONS FOR ANALYSIS ON COMPLIANCE TO DOSING GUIDELINES 241 APPENDIX D: ANALYSIS OF STRIDE DRUG PRICE DATA ................................. 247 D1. Constructing the Sample ...................................................................................... 248 D2. Preliminary Data Cleaning................................................................................... 248 D3. Additional Cleaning Related to Proper Model Estimation .................................. 249 D4. Define Key Variables........................................................................................... 250 D4.1 Dependent Variables...................................................................................... 250 D4.2 Independent Variables ................................................................................... 251 D5. Econometric Models for Price and Purity............................................................ 251 D5.1 Purity Model Specification............................................................................. 252 D5.2 Price Model Specification .............................................................................. 253 APPENDIX E: HEALTH STATUS MEASURES USED IN CHAPTERS 5 AND 6 ... 260 APPENDIX F: UBC BEHAVIOURAL RESEARCH ETHICS BOARD CERTIFICATES OF APPROVAL ............................................................................................................. 273 ix  LIST OF TABLES Table 2.1: Summary statistics on retention in methadone maintenance treatment........... 54 Table 2.2: Summary statistics of explanatory variables ................................................... 55 Table 2.3: Results from the proportional hazards gamma frailty model, with indicated determinants of time to MMT discontinuation ................................................................. 56 Table 3.1: Descriptive statistics of study sample.............................................................. 99 Table 3.2: Descriptive statistics of drug utilization ........................................................ 100 Table 3.3: Strength of instruments.................................................................................. 101 Table 3.4: Reduced-form demand equation results: Full sample (N=16,541), estimated via zero-inflated negative binomial regression (NB coefficients provided). .................. 102 Table 3.5:  Reduced form demand equation results: Polydrug users only (N=7,400), estimated via zero-inflated negative binomial regression (NB coefficients provided)... 103 Table 3.6:  Reduced-form demand equation results: 3-drug users only (N=1,697), estimated via negative binomial regression .................................................................... 104 Table 3.7: Reduced-form demand equation results: Age-stratified samples, polydrug users only (N=7,400), estimated via zero-inflated negative binomial regression (NB coefficients provided) ..................................................................................................... 105 Table 4.1: BC MMT dosing guidelines .......................................................................... 131 Table 4.2: MMT treatment retention: 1996-2006 ........................................................... 132 Table 4.3: Descriptive statistics of episodic dosing patterns .......................................... 133 Table 4.4: Patient and treatment delivery characteristics by duration of retention in treatment ......................................................................................................................... 134 Table 5.1: Description of health status measures............................................................ 167 x  Table 5.2: Summary statistics of index scores and items ............................................... 168 Table 5.3: Assessing content validity: conceptual coverage of health status measures . 170 Table 5.4: Results of exploratory and confirmatory factor analysis............................... 171 Table 5.5: Assessing criterion validity: ability of health status measures to discriminate between individuals with and without chronic medical conditions ................................ 172 Table 5.6: Responsiveness to improvements in illicit drug use at 12 months, from baseline ........................................................................................................................... 173 Table 5.7: Summary of empirical comparison of ASImed, ASIpsych, MAP-PHS, MAP- MHS, WHODAS-II, EQ-5D, EQ-VAS, SF-6D ............................................................. 175 Table 6.1: Parameter estimates for fixed and random effects components growth curve analysis before and after inclusion of latent class membership variables ...................... 202 Table 6.2: Adjusted associations of patient characteristics measured at trial entry with specific trajectories of HRQoL ....................................................................................... 203 Table 6.3: Factors associated with changes in EQ-5D index scores............................... 204 Table B.1: PharmaNet variables and constructed variables............................................ 234 Table B.2: Erroneous methadone dispensation records and resolutions......................... 236 Table B.3: Aggregated local health areas ....................................................................... 239 Table D.1: No. of purchase observations in STRIDE database by quarter, drug type ... 255 Table D.2: Geographical decomposition: STRIDE, CA datasets ................................... 256 Table D.3: Results on fixed effects from random coefficient regression models on the purity and real price of crack cocaine ............................................................................. 257 Table D.4: Results on fixed effects from random coefficient regression models on the purity and real price of powder cocaine.......................................................................... 258 xi  Table D.5: Results on fixed effects from random coefficient regression models on the purity and real price of heroin......................................................................................... 259   xii  LIST OF FIGURES  Figure 2.1: Estimated coefficient value β(t) (curved solid line) with 95% confidence interval (dashed lines) for treatment adherence (70%-90%) as a function of time from treatment initiation.  X-axis formatted using ‘kaplan-meier’ scale. ................................. 57 Figure 2.2: Estimated coefficient value β(t) (curved solid line) with 95% confidence interval (dashed lines) for carry privileges granted as a function of time from treatment initiation.  X-axis formatted using ‘kaplan-meier’ scale. ................................................. 58 Figure 2.3:  Kaplan-Meier curves for time to discontinuation of MMT by treatment episode: Full sample [N=17,005]. Points of censorship have been suppressed. .............. 59 Figure 2.4: Kaplan-Meier curves for time to discontinuation of MMT by treatment episode: Patients with ≥ 4 episodes [N=2,023].  Points of censorship have been suppressed. ........................................................................................................................ 60 Figure 2.5:  Hazard ratios on the episode count variable from multivariate models on time to MMT discontinuation, including patients with two episodes [N=7,898], three episodes [N=5,534], four episodes [N=3,660], five episodes [N=2,383] and six episodes [N=1,546].......................................................................................................................... 61 Figure 3.1: Estimated longitudinal drug price series ...................................................... 106 Figure 4.1: Compliance to starting dose, titration rate, carry dosing, and tapering rate guidelines, by calendar year............................................................................................ 135 Figure 4.2:  Compliance to minimum effective dose guideline vs. 12-month retention rate, by calendar year ...................................................................................................... 136 Figure 4.3: Episodic MMT dosing patterns .................................................................... 137 xiii  Figure 6.1: Distribution of changes in EQ-5D item responses: Baseline-12 months ..... 205 Figure 6.2: HRQoL Trajectories of patients enrolled in the NAOMI trial ..................... 206 Figure C.1: Definitions of treatment phase transitions ................................................... 246 xiv  ACKNOWLEDGEMENTS  A large debt of gratitude is owed to my supervisor Aslam Anis, and supervisory committee, including Dr. Atsushi Inoue, Dr. Ying MacNab, Dr. Martin Schechter, and Dr. Michael Krausz,   for providing me with invaluable mentorship and support.  From the outset, Dr. Anis et al provided sound advice an excellent opportunity to develop my ideas and skills in research.  I would have been lost without them.  The support of dedicated research assistants made this work possible: from the NAOMI study: Kurt Lock, Jill Chettiar, Nancy Laliberte, Douglas Ferris and Skye Matheson, Amanda Walker and Kara Sievewright, who collected research data used in studies 4 and 5; representatives from the Centre for Health Services and Policy research for extracting the data necessary for conducting studies outlined in chapters 1 and 3.  Representatives from the United States Drug Enforcement Administration (Katherine Myrick and Phyllis Drewery), who graciously allowed a foreign student access to sensitive data in the name of the advancement of science, and completion of study 2.  Special thanks to all other co-authors on the following studies, and others who shared their time and knowledge with me along the way, especially Huiying Sun, Daphne Guh and David Marsh.   I’ve learned more from these three individuals about statistical analysis and opioid dependence than any other, and the knowledge they have imparted upon me has shaped my approach to scientific research.  xv  Most of all, I want to thank my extremely supportive family, and in particular, my amazing wife Lianne for pushing me to follow my goals and for providing me with unyielding support from the beginning that gave me the necessary strength and motivation to begin and complete this work.  Without her countless compromises, her dedication and commitment, none of this would have been possible.  This research was generously supported by grants from the Canadian Institute of Health Research and the British Columbia Ministry of Health, Institute for Healthy Living and Sport.  Thanks to the Canadian Institutes of Health Research (co-sponsored by the British Columbia Mental Health Foundation), the CIHR Research in Addictions in Mental Health Policy and Services CIHR strategic training initiative and the Michael Smith Foundation for Health Research for their doctoral fellowship support.  xvi  DEDICATION    To my wife, Lianne xvii  CO-AUTHORSHIP STATEMENT  CHAPTER 2 The candidate is first author on this manuscript, developed the hypotheses, took part in cleaning the data, manipulated data, performed the statistical analyses, and wrote the final manuscript.  Co-authors included Drs. Ying Macnab and Huiying Sun, statisticians, Drs. David Marsh, Benedikt Fischer and Martin Schechter, who provided conceptual guidance and critiques, and Dr. Anis, chair of the supervisory committee.  CHAPTER 3 The candidate is first author on this manuscript, developed the hypotheses, cleaned and manipulated the data, performed all microeconomic and econometric analysis, and wrote the final manuscript.   Co-authors included Dr. Atsushi Inoue, who provided valuable advice and guidance throughout the analysis and preparation of the study, Dr. Dennis G. Fisher, who allowed access to the epidemiological dataset provided valuable advice on analysis and interpretation thereof, and Dr. Anis, chair of the supervisory committee.  CHAPTER 4 The candidate is first author on the manuscript, developed the hypothesis, took part in cleaning the data, manipulated data, performed statistical analysis and wrote the final manuscript.  Co-authors included Dr. David Marsh (addictions physician), Dr. Huiying Sun (statistician), Dr. Martin Schechter, each of whom provided conceptual guidance and critiques, and Dr. Aslam Anis, chair of the supervisory committee. xviii   CHAPTER 5 The candidate is first author on the manuscript, developed the design of the study, trained research staff to administer questionnaires, cleaned and manipulated data, performed statistical analysis and wrote the final manuscript.  Co-authors included Dr. Huiying Sun and Daphne Guh (statisticians), Drs. David Marsh, Eugenia Oviedo-Joekes, Suzanne Brissette and Martin Schechter, who were the primary and co-investigators of the NAOMI study, and Dr. Aslam Anis, chair of the supervisory committee.  CHAPTER 6 The candidate is first author on the manuscript, developed the design of the study, trained research staff to administer questionnaires, cleaned and manipulated data, performed statistical analysis and wrote the final manuscript.  Co-authors included Dr. Huiying Sun and Daphne Guh (statisticians), Drs. David Marsh, Eugenia Oviedo-Joekes, Suzanne Brissette and Martin Schechter, who were the primary and co-investigators of the NAOMI study, and Dr. Aslam Anis, chair of the supervisory committee.       1 1 
INTRODUCTION 1.1 Opioid Dependence: Epidemiology, Health and Economic Burden  Opioid addiction, commonly manifested as heroin addiction, is considered a chronic, recurrent disease (McLellan et al, 2001).  Opiates such as heroin activate opioid receptors in the brain, resulting in feelings of euphoria, subsequently releasing excess dopamine in the body.  This release of dopamine and the activation of the reward system can lead to addiction.  Once physically dependent, interruptions in use result in severe withdrawal symptoms including nausea, vomiting, sweating, abdominal pain and agitation.  Opioid dependence has profound effects on individuals’ physical and mental health, and is associated with psychiatric disorders, such as bipolar disorder, anxiety disorders and depression (Mason et al, 1998, Lelutiu-Weinberger et al, 2009; Mathers et al, 2009), elevated risk of infection and transmission of infectious diseases such as HIV/AIDS and Hepatitis C (Aceijas et al, 2004) and premature mortality as a result of overdose or other complications (Caplehorn,1994).  Chronic opioid use is also associated with poor psychosocial functioning, given that it is generally accompanied by unemployment, imprisonment, poor housing conditions, and illegal activity (Galai 2003).  Opioid dependence is a critical public health problem in Canada and most other parts of the world.  It is estimated that between 75,000 and 125,000 individuals inject drugs in Canada, with cocaine and heroin the most common substances injected (Single, 2001; Federal PATACOPH, 2000).  Institutional sources estimate approximately 80,000 regular  2 opioid users in Canada, with geographic differences in prevalence and trends.  Although availability of illicit prescription analgesic opioids such as hydromorphone and morphine has increased in recent years (Brands et al, 2004; Fischer et al, 2005), intravenous use of heroin prevails in Vancouver and Montreal (Popova, Rehm and Fischer, 2006; CHASE project team, 2005; Tyndall et al, 2006, De et al, 2008).  Cooper et al (2008) estimated the prevalence of injection drug use among black and white adults 95 U.S. Metropolitan Statistical Areas.  Their study reported a median of 156 injectors per 10,000 black adults, and 97 per 10,000 white adults.  The world health organization reports that the main problem drugs at the global level continue to be opiates (notably heroin) followed by cocaine and other stimulants.  Their World Drug Report further states that opiates accounted for 62% of all demand for treatment of substance dependence in 2003 (World Health Organization, 2005). 1.2 Opioid Dependence as a Chronic, Recurrent Disease  McLellan et al (2001) argued that the diagnosis, heritability, etiology and pathophysiology of opioid dependence are all comparable to other well-known chronic diseases, and made the case by comparing substance dependence with type 2 diabetes mellitus, hypertension and asthma.  Dependence is operationally defined in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, as a pathologic condition manifested by 3 or more of 7 criteria, two of which (tolerance and withdrawal) indicate physiologic dependence.  This diagnostic differentiation of use, abuse and dependence has been repeatedly shown to be reliable and valid (Diagnostic and Statistical  3 Manual of Mental Disorders, Fourth Edition, 1994; Schukit, Hesselbrock and Tipp, 1994).    Genetic heritability of heroin dependence has also been shown in studies in monozygotic and dizygotic twins, with heritability estimates within the range of those for the comparator chronic diseases (Tsuang et al, 1996; Kendler and Prescott, 1998; VandenBree et al, 1998; True and Xian, 1999).  The etiology of substance dependence suggests that environmental and individual factors predict initiation and eventual dependence.  While voluntary choice affects initiation and progression of opioid dependence, similar statements can be made in regards to hypertension.  Finally, studies on the pathophysiology of substance dependence suggest that there are lasting, and potentially permanent changes in neurological functioning as a result of continued use of heroin, cocaine and other substances – this literature is summarized succinctly in McLellan et al (2001), and updated by Galynker et al (2000) and Nestler (2005).  Longitudinal studies on individuals dependent on opioids and other drugs have generally identified patterns of chronic continued drug use, switching between periods of treatment and relapse, and full abstention (Bell et al, 2006; Galai et al, 2003; Dobler-Mikola et al, 2005; Siegal, Li and Rapp, 2002; Termorshuuizen et al, 2005).  Galai et al (2003) identified four patterns of injection drug use among a large prospectively followed cohort in Baltimore, MD (1988-2000): 1) persistent use (29%), 2) ceased injection (20%), 3) relapsed once (14%), and 4) multiple transitions (37%). Similar patterns of use were found among a cohort of primarily crack/cocaine abusers in Dayton, OH (1991-1994) (Siegal, Li and Rapp, 2002).   Furthermore, a study of drug abuse relapse following an episode of abstinence using the Amsterdam Cohort Study database found that over 85%  4 of episodes of abstinence were followed by relapse within five years of the start of the episode (Termorshuizen et al, 2005).  Evidence on the characteristics of opioid dependence and the observed longitudinal patterns of use suggest that treatment is required, and should be offered with a focus on long term maintenance and stabilization, rather than abstinence. 1.3 Available Treatment  Strategies to treat and stabilize the problems associated with illicit opioid use are many and varied.  Treatment ranges from abstinence to substitution treatment, with or without concomitant psychosocial support or psychiatric therapy.  Systematic reviews of experimental and observational studies suggest that substitution treatment is the most effective approach ((Mattick, Breen, Kimber, & Davoli, 2003; Amato et al., 2005; Van den Brink & Haasen, 2006)).  Several alternative forms of substitution treatment are currently available in different countries, including methadone, buprenorphine (Mattick, Kimber, Breen, & Davoli, 2008), morphine (Kraigher et al., 2005; Mitchell, White, Somogyi, & Bochner, 2006), dihydrocodeine (Krausz, Verthein, Degkwitz, Haasen, & Raschke, 1998; Robertson et al., 2006) and diacetylmorphine (Rehm et al., 2001; Haasen et al., 2007; Fischer et al., 2007), among others.  An alternative to oral opioid substitution therapy, levo-alpha acetylmethadol (LAAM), was approved for use in 1993, however this product has been removed from the market following case reports of life-threatening cardiac arrhythmia (Deamer, Wilson, Clark, & Prichard, 2001).  In an overview of systematic reviews of the effectiveness of opiate maintenance therapies, MMT was found  5 to be more effective in retaining patients (at one year) than Methadone Detoxification Treatment, Buprenorphine Maintenance Treatment, L-α-acetylmethadol (LAAM) maintenance treatment and heroin plus methadone maintenance treatment. MMT, however, was less effective than LAAM in preventing illicit heroin use, and was not significantly different from the other modes of treatment in terms of mortality and other criminal activity (Amato et al, 2003).  Methadone is a long-acting synthetic opioid agonist that is easily absorbed when taken orally and has a half-life of approximately 25 hours, allowing once-daily administration. Methadone Maintenance Treatment provides methadone orally on a regular (usually daily) basis to the patient (Seidenberg and Honegger, 1998).  Methadone occupies opioid receptors, reducing the cravings associated with heroin use while blocking the associated euphoria throughout the duration of its half-life, while not producing any of its own. (ONDCP, 2001).  Standard pharmacotherapy for opioid dependence in British Columbia and throughout Canada involves oral solution methadone, provided by specially-licensed physicians and, in most cases with ingestion witnessed daily by a pharmacist.  A treatment philosophy emphasizing indefinite maintenance has been shown to decrease the risk of relapse and mortality (Caplehorn et al, 1994) and has thus become widely accepted in Canada and elsewhere.  Determinants of treatment success (ie. longer duration of treatment, or successful abstention) include age at entry, average daily dose (Strain, 1999), socioeconomic status, social relations, employment, criminal history (Ward, Mattick and Hall, 1998), mental  6 health status, concurrent illicit drug use (Darke, 1998), and treatment motivation (Fletcher and Battjes, 1999).  Program-related factors, such as medical care (Lowinson et al, 1997), other substance abuse treatment (Stitzer and Chutuape, 1999; Best et al, 1999; Budney, Simpson and Joe, 1998) and mental health services (Mattick, Ward and Hall, 1998) have also been shown to influence treatment success.  The length of time in MMT is considered a predictor of MMT effectiveness both during, and post-treatment (Aszalos et al, 1999), however, evidence for post-treatment outcomes is scarce and mixed. Based on their review of the literature, Ward et al (1998) state that the benefits of MMT “continue only as long as patients remain on treatment”, suggesting that it may be necessary to retain patients in treatment indefinitely (Joseph, Stancliff and Langrod, 2000). This stands in direct contrast with the philosophy of MMT being prescribed as a bridge to abstinence, which supports limiting the dose and duration of methadone while promoting the gradual withdrawal of medication (Halliday, 1963). Evidence supporting the long-term effectiveness of these competing philosophies of care is scarce, however recent studies suggest that longer duration of exposure in treatment is associated with improved post-treatment outcomes such as reduced opioid use, reduced criminal activity and improved social productivity (Hubbard, Craddock and Anderson, 2003; Zhang, Friedmann and Gerstein, 2003; Dolan et al, 2005).  A Health Canada review of the literature on MMT (Health Canada, 2002) suggested that further research on MMT and its treatment outcomes is needed, since there are conflicting reports on reduction of illicit opioid and other drug use, and on physical and  7 mental health improvement (National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction, 1998), both during and after treatment. A recent comprehensive review of all experimental studies designed to estimate the effectiveness of MMT even suggested that rigorous and appropriate evaluation methods are rare, and that evidence for MMT’s effectiveness on primary treatment objectives is mixed and appears to be largely partial and short-term (Fischer and Rehm, 2005).  This critique highlighted two aspects of treatment that we attempt to address in this study. The first is in regards to the evaluation of MMT, which has been focused mainly on social outcomes (e.g. Criminal and economic activity) and has been operationalized by indicators that occur in, or impact on, societal levels, rather than focusing on a patient’s subjective health status or well-being. The second is that although many reviews state that methadone is also effective in reducing cocaine abuse, results supporting this claim are ambivalent at best, and defy biomedical rationale.  The former point criticizes the means by which MMT is evaluated, while the latter raises the issue of multiple drug use among individuals with opioid dependence, and strategies to treat these individuals.  We explore these two criticisms further in the next sections. 1.4 Treatment Evaluation  The primary outcome measure used in most observational and experimental studies evaluating the effectiveness of MMT has been retention in treatment, or duration of exposure of one year (Farre et al, 2002). A National Institute of Drug Abuse review found that the overall mean retention in MMT at one year was 39.8%, with a range of 25-  8 60% (Fischer, 2000). A study of patients in public and private MMT programs in Australia found a one-year retention rate of 38%, with the majority of discontinuations being dropouts rather than treatment completions. After discontinuation, 68% of patients re-entered MMT, underlining the disease’s chronic, recurrent nature (Bell et al, 2006).  Unfortunately, the vast majority of studies on the effectiveness of MMT have been conducted in the context of experimental studies with potentially selective participants and relatively short follow-up – most often one to two years.  The external validity of these studies is compromised by the potential for selection bias, while the short timeline provides only a snapshot of an individual’s drug use and drug treatment career (Hser, Longshore and Anglin, 2007).  Furthermore, the breadth of outcomes considered in experimental studies has generally been limited to treatment retention, decreases in illicit drug use and criminal activity.  We argue that the evaluation of MMT can be improved in both observational and experimental settings. 1.4.1 Evaluation in Observational Settings  As stated above, longitudinal studies of persons dependent on opioids and other drugs have generally identified patterns of chronic continued drug use, switching between periods of treatment/relapse and full abstention. As achievement of sustained drug abstinence has proven unattainable for the vast majority of heroin users (Termorshuizen et al, 2005), it is of interest to determine whether long-term maintenance on substitution treatment such as methadone can be achieved among patients experiencing multiple periods of treatment and relapse.  Such an evaluation could not feasibly be conducted in  9 the context of an experimental study, and thus should be considered in long-term observational studies.  Furthermore, there is a need to evaluate treatment at the program level in order to ensure public health benefits are being maximized, and to identify aspects of treatment delivery that can be improved upon.  It is well known that aspects of treatment delivery are predictors of treatment outcome (Magura, Nwakeze and Demsky, 1998; Anderson and Warren, 2004).   Numerous experimental and observational studies of opioid dependent patients in treatment have been conducted, however few have evaluated prescribing practices, and only one has done so at the national or provincial/state level.  D’Aunno and Pollack (2002) assessed the maintenance dose prescribed in a nationally representative sample of US methadone treatment programs and found that 35.5% of patients received less than the recommended 60mg daily dose.  Similarly, Strang et al. (2007) observed higher daily MMT doses after national guidelines were issued in the UK, but low overall adherence to daily dosing guidelines.  In British Columbia, MMT practice guidelines are supported by unique levers, including the requirement for a specific authorization to prescribe, mandatory training sponsored by the College of Physicians and Surgeons and a system of peer assessment for compliance with guidelines (Payte, 1995).  Included in the guidelines (released in 1995) are evidence- based strategies to safely and effectively adjust daily doses from treatment initiation (starting dose), titration and maintenance dosing, as well as the provision of carry, or take-home doses.  Formal guidelines on dose tapering were added in 2005 for  10 circumstances in which patients chose to voluntarily taper their dose with the goal of abstinence (College of Physicians and Surgeons of British Columbia, 2005).   Guidelines on starting doses and dose stabilization are designed to ensure patient safety, while maintenance and dose tapering guidelines were developed to maximize the odds of long- term maintenance, and successful tapering, respectively (Martin, Payte and Zweben, 1991; Health Canada, 2002; National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction, 1998) Conversely, guidelines on take-home (carry) doses were designed in part to ensure patient safety, but also to minimize public health risks through drug diversion (Ruel and Hickey, 1993). Evaluating changes in compliance to MMT treatment guidelines can both inform the extent to which the public health benefits of this treatment are being achieved. 1.4.2 Evaluation in Experimental Settings  Opioid dependence is a chronic disease with a documented impact on health-related quality of life.  As such, the need to integrate HRQoL data into treatment assessment is critical.  Engagement in opiate substitution treatment (OST) can potentially improve patients’ health status through the psychotherapeutic benefits of opiate agonist therapy itself, mediation of withdrawal symptoms and access to additional psychosocial and pharmacological treatment for co-morbid conditions.  HRQoL instruments which reflect societal preferences in assessing the burden of disease are recommended in conducting cost-utility analyses.  These analyses are critical for informing decisions regarding the allocation of health care resources.  Generic measures  11 of HRQoL are preferred, as they facilitate direct comparisons of the cost-effectiveness of treatments for substance abuse with alternative health interventions in different disease areas.  Evidence of improvement in HRQoL during OST can be used to argue for expansion of existing substance abuse treatment services or to evaluate the value of effective new interventions relative to current practice.  A systematic review of OST studies could not draw any conclusions on the effect of OST on HRQoL due to a lack of HRQoL evidence collected in past studies (Amato et al, 2003).  Several cross-sectional studies have identified factors such as age, duration and severity of drug use and presence of chronic disease as may partially explain variation in baseline HRQoL scores of opioid dependent patients entering treatment (Millson et al, 2006; Puigdollers et al, 2004; Astals et al, 2008).  Accurately measuring potential improvements in health in opioid maintenance treatment are important for evaluation of new health technologies and resource allocation decisions.  The use of quality adjusted life years (QALYs) is an attempted solution to incorporate both potential life prolongation and improvement in HRQoL (Neumann, Goldie and Weinstein, 2000).  Neumann et al. Stated that “QALYs represent the benefit of a health intervention in terms of the time in a series of quality-weighted  health states, in which the quality weights reflect the desirability of living in the state, typically from perfect health (weighted 1.0) to dead (weighted 0.0) (Neumann, Goldie and Weinstein, 2000). Therefore, once the quality weights are obtained for each health state experienced by an individual, they are multiplied by the duration of time spent in the health state.  The  12 products of these individual calculations are then summed to obtain the total number of QALYs for that person through the formula: QALYi= t* sum(Ui(qj)Dj, where ui(qj) = the quality of life in period j (measured by utilities), t is the time interval of period in terms of years, Dj is the discount factor in period j.  Indirect preference or utility assessment techniques involve the use of generic health classification systems in the form of a questionnaire (Neumann, Goldie and Weinstein, 2000; Drummond et al, 1997).  Through completion of the health classification system, respondents are assigned a health state which, in turn, is valued using a scoring function that applies preference weights from another population (ie. society).  Due to their relative ease and low cost to administer when compared to directly elicited preference scores (collected through standard gamble, or time-trade-off exercises), these questionnaires are widely applied (Neumann, Goldie and Weinstein, 2000; Drummond et al, 1997).  These instruments commonly utilize multi-attribute utility theory (MAUT) to combine many attributes into a single utility value.  As it is beyond the scope of this chapter to describe MAUT in detail, the reader is referred to reviews for a complete description and assumptions involved with this theory (Neumann, Goldie and Weinstein, 2000; Drummond et al, 1997).  The most common examples of these questionnaires include the Health Utilities Index Mark2 and Mark3 (HUI2 and HUI3), the Euroqol (EQ-5D), and the Short-Form 6D (SF- 6D).  Each of these systems assesses different domains of health and relies on different scoring functions and methods to determine preference scores (Hawthorne, Richardson  13 and Day, 2001).  Other preference-based measures that have been less-commonly applied are the Quality of Well-being scale, the Finnish 15-D, and the Assessment of Quality of Life (AQoL) (Drummond et al, 1997; Hawthorne, Richardson and Day, 2001).  The field of treatment for substance dependence is evolving, with new forms of treatment being tested for both opioid and stimulant dependence.  Means with which to determine whether these new treatments are both effective and cost-effective are required.  Through the cost-utility analysis framework, preference-based measures of HRQoL can be used to inform resource allocation decisions in health care (Maetzel, Krahn and Naglie, 2003; Bae et al, 2003).  This is done through the calculation of the quality-adjusted life year (QALY) which is commonly used as the denominator of the incremental cost-utility ratio (ICER) calculation (Bae et al, 2003; Maetzel et al, 2003).  CUAs are used by administrative bodies such as the National Institute on Health and Clinical Excellence to make efficient health resource allocation decisions across disease areas, and thus serve an important role in healthcare delivery. 1.5 Prevalence and Effects of Polydrug Use  One of the key factors that confound the success of opioid substitution treatment, no matter how it is evaluated, is the use of other drugs.  Cocaine use in particular (either in powder form, as crack cocaine, or injected) is highly prevalent among opioid dependent patients.  Cocaine is a stimulant, a class of drugs which also includes methamphetamine, amphetamines, methylphenidate, nicotine, and MDMA.  Cocaine increases synaptic dopamine by blocking reuptake into presynaptic neurons; amphetamine produces  14 increased presynaptic release of dopamine, whereas opiates and alcohol disinhibit dopamine neurons, producing increasing firing rates (Koob and Bloom, 1998; Special Issue, Science, 1997; Series on Addiction, Lancet, 1996; National Academy of Sciences, 1995; Institute of Medicine, 1995; Wise and Bozarth, 1981).  At the moment, there are no proven medications for stimulant dependence.  Some of the earliest evaluations of MMT conducted by Ball and Ross (1991) reported that 47% of patients presenting for opioid substitution treatment were also using cocaine. This percentage is likely higher among those with more severe addiction, more chaotic lifestyles, who are harder to reach and meaningfully retain in treatment.  Among patients in the North American Opiate Medication Initiative, who had tried, and failed drug abuse treatment at least twice before, regular cocaine use was reported by 70%,of all participants.  Experimental studies on non-human subjects have shed light on the physiological aspects of heroin and cocaine polydrug use.  “Support for the idea that a history of opioid dependence plays a role in the interaction between heroin and cocaine intake comes from our recent studies in non-dependent and heroin-dependent rats.  Recent studies suggest that in rats previously made dependent on heroin, the self-administration of cocaine is enhanced.  Furthermore, the authors found that rats that experienced acute, spontaneous withdrawal from heroin showed a depression in locomotor activity that is reversed by cocaine administration.  Importantly, the greater the dose of heroin exposure, the more  15 cocaine is required to reverse the depression of locomotor activity observed during acute withdrawal” (Leri et al., 2003).  Use of cocaine during opioid substitution treatment (Methadone maintenance treatment (MMT)) has been shown to be associated with poor treatment outcome, increased use of heroin, and an attenuation of the effect of methadone (Magura et al., 1998; Hartel et al., 1995; Perez de los Cobos et al., 1997; Leri et al., 2003).  In a multivariate Cox regression model, Termorshuizen et al. (2005) showed that both daily and weekly/monthly use of cocaine during an episode of opioid abstinence was statistically significantly associated with relapse into frequent heroin use, controlling for other covariates.  More importantly, the authors found that there was no beneficial influence of dosage or attendance in MMT on relapse into cocaine.  The use of crack cocaine, widespread among methadone treatment patients and among samples of street users, is associated with high-risk behaviors for HIV and lower rates of heroin abstinence in methadone maintenance while reductions in opiate use are not always accompanied by reductions in the use of crack (Beswick et al., 2001).  Heroin- crack combined use has associations with psychological problems and may impact on treatment outcome.  The study of microeconomics can shed light on individual choice in consuming illicit drugs, and subsequently, the relationship between heroin and cocaine use.  The economic theory of rational addiction suggests illicit drug users behave rationally albeit short-  16 sightedly (Becker and Murphy, 1988).  That is, they make consumption choices that maximize their current-value “utility”, or well-being. While illicit drug use is harmful in the long-run, consuming an illicit drug can bring about positive utility in the short term, in the form of euphoria and avoidance of withdrawal symptoms. One empirical study found high rates of time preference (a measure of how present-oriented an individual is in their decision-making) among active illicit drug users (Bretteville-Jensen, 1999), suggesting that the long-term harms can be outweighed by the immediate benefit of drug consumption.  While methadone eliminates withdrawal symptoms and blocks the euphoric rush experienced with heroin, other illicit drugs can provide qualitatively different, though comparably intense chemical euphoria. In this sense, cocaine and heroin can be considered substitutes for one another.  For two goods to be strictly defined as economic substitutes an increase in the price of one good should lead to an increase in consumption of the other – this concept is captured empirically through estimation of the cross-price elasticity of demand.  A pair of goods can alternatively be classified as complements, where consumption of one decreases with the increase in price of another, or independent.  Further, the responsiveness of consumption of a good to changes in its own price is estimated with own-price elasticity of demand.  If the percentage decrease in consumption exceeds that of the decrease in price, the good is classified as price elastic. Illicit drugs and other addictive substances such as nicotine have been found to be price inelastic in most empirical studies, suggesting price increases decrease consumption less than proportionately.  17  While multiple drug use is common among illicit drug users, there have been a limited number of empirical studies providing the price elasticity of demand for illicit drugs, and fewer still that have considered multiple drug use.   Given the difficulty of observing illicit drug consumption in natural settings, research on the effect of drug prices on consumption has been conducted in experimental settings with hypothetical drug prices and purchase scenarios.  Behavioural economics has been indicated as a valuable conceptual framework for the study of drug dependence (Bickel et al. 1993).  Most notably, Petry and Bickel (1998) conducted a series of experiments on a small cohort of polydrug users in opioid maintenance treatment.  Varying the price of heroin and other drugs commonly used in the cohort, the authors found that heroin was price inelastic, and valium and cocaine purchases increased as heroin prices rose, suggesting these were economic substitutes.  The relationship between valium and heroin purchases were asymmetric, however, as heroin purchases were independent of valium prices.  This highlights the notion that polydrug users have an ordinal ranking of preferred substances that is based on past experience, including their level of dependence, as well as current prices and expected future prices.  Similar studies have been conducted to determine income elasticity among polydrug users (Petry, 2000), and relationships in the demand for alcohol and various stimulants (Sumnall et al, 2004).  Bickel, Degrandpre and Higgins (1993) provide an introduction to the methodology of these experiments and a review of early research in this field.   18 Despite the emerging pattern of multiple drug use among primary heroin users, treatment services are not always equipped to address multiple substance use (Beswick et al., 2001).  Suggestions have been made to target both treatment interventions and outcome assessment to multiple drugs, rather than a single drug, in clinical trials involving substance abusers (Rounsaville et al., 2003).  These sentiments have been echoed elsewhere, as cocaine use during MMT has predicted poorer treatment outcomes among patients recruited from the Australian Treatment Outcomes Study (ATOS) (Williamson et al, 2006). 1.6 Study Hypotheses, Objectives and Thesis Organization  The overall aim of this study was to shed light on the life course of treatment and relapse that opioid addict’s experience, their decisions to consume addictive substances, and the evaluation of treatment in observational and experimental contexts.  The first objective of the study was to to identify the determinants of the time to discontinuation of Methadone Maintenance Treatment (MMT) across multiple treatment episodes.  The second objective was to determine the responsiveness of consumption to price changes (cross- price elasticities of demand) of crack cocaine, powder cocaine and heroin among a cohort of chronic illicit users.  The third objective was to identify trends in compliance to MMT dosing guidelines at the population level in British Columbia, Canada from 1996-2007. The fourth objective was to provide a comparative analysis of the psychometric properties of eight discriminative and evaluative measures of health status among chronic opioid dependent patients.  The fifth objective of the study was to evaluate differences in trajectories of HRQoL among chronic opioid dependent patients and to identify factors  19 associated with improvement and deterioration in HRQoL following enrollment in opioid substitution treatment.  This thesis is composed of seven chapters, organized chronologically, addressing each of the objectives in order.  This first chapter provides a brief introduction to (1) the epidemiology of opioid dependence, (2) available treatment for opioid dependence, (3) methods of MMT evaluation, and (4) the prevalence and effects of polydrug use. Chapters two through six provide the contents of the research conducted to fulfill each of the above objectives. These chapters, along with appendix A are each stand-alone manuscripts, which have either been published, are in press, or are under review by a major, peer-reviewed, scholarly journal.  The work presented in this thesis was conceived, conducted and disseminated by the doctoral candidate as has been declared by the supervisor of the candidate.  The final chapter provides a summary of the research findings and outlines the strengths, limitations and the unique contributions and potential impact of the findings of these studies.  20 1.7 References  Aceijas C, Stimson G, Hickman M, Rhodes T (2004).  Global overview of injecting drug use and HIV infection among injecting drug users. AIDS 18:2295–303.  Amato L, Davoli M, Perucci CA, Ferri M, Faggiano F, Mattick RP (2003).  An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research.  Journal of Substance Abuse Treatment 28: 321-9.  Amato, L., Davoli, M., A Perucci, C., Ferri, M., Faggiano, F., & P Mattick, R. (2005). An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research. J Subst Abuse Treat, 28(4), 321-9.  Anderson JF, Warren LD (2004). Client retention in the British Columbia methadone program, 1996-1999.  Can J Public Health. 95(2):104-9.  Astals M, Domingo-Salvany A, Buenaventura CC, et al (2008).  Impact of substance dependence and dual diagnosis on the quality of life of heroin users seeking treatment. Subst Use Misuse. 43(5): 612-32.   21 Aszalos R, McDuff DR, Weintraub E, Montoya I, Schwartz R (1999).  Engaging hospitalized heroin-dependent patients into substance abuse treatment.  J Subst Abuse Treat. 19: 149-58.  Bae S-C, Cozillus M, Kuntz KM, Liang MH (2003).  Cost-effectiveness of low dose corticosteroids vs. non-steroidal anti-inflammatory drugs and COX-2 specific inhibitors in the long-term treatment of rheumatoid arthritis. Rheumatology 42: 46-53.  Ball JC, Ross A (1991).  The effectiveness of methadone maintenance treatment. Springer-Verlag, New York.  Becker GS, Murphy KM (1988).  A theory of rational addiction. Journal of Political Economy 96:675-700.  Bell J,  Burrell T, Indig D, Gilmour S (2006).  Cycling in and out of treatment; participation in methadone treatment in NSW, 1990-2002.  Drug and Alcohol Dependence 81: 55-61.  Best D, Gossop M, Greenwood J, Marsden J, Lehman P, Strang J (1999).  Cannabis use in relation to illicit drug use and health problems among opiate misusers in treatment. Drug and Alcohol Review 1999; 18: 31-8.   22 Beswick T, Best D, Rees S, Coomber R, Gossop M, Strang J (2001).  Multiple drug use: patterns and practices of heroin and crack use in a population of opiate addicts in treatment.  Drug and Alcohol Review, 20:201-4.  Bretteville-Jensen AL (1999).  Addiction and Discounting.  J Health Econ 18:393-407.  Bickel WK, DeGrandpre RJ, Higgins ST (1993).  Behavioural economics: a novel experimental approach to the study of drug dependence, 33(2): 173-92.  Brands B, Blake J, Sproule B, Gourlay D, Busto U (2004). Prescription opioid abuse in patients  presenting for methadone maintenance treatment. Drug Alcohol Depend 73:199- 207. Budney KM, Simpson DD, Joe GW (1998).  Marijuana use and treatment outcome among opioid dependent patients.  Addiction 93(4): 493-503.  Caplehorn JRM, Dalton MSYN, Cluff MC, Petrenas A-M (1994).  Retention in methadone maintenance and heroin addicts’ risk of death.  Addiction 89: 203-7.  CHASE Project Team (2005). Community Health And Safety Evaluation (CHASE). Vancouver Coastal Health Authority.   23 Clausen T, Anchersen K, Waal H (2008).  Mortality prior to, during and after opioid maintenance treatment (OMT): a national prospective cross-registry study.  Drug Alcohol Depend. 94(1-3):151-157.  College of Physicians and Surgeons of British Columbia (2005).  Methadone Maintenance Handbook.  http://www.bcguidelines.ca/gpac/guideline_methadone.html. (Accessed: June, 2008). Cooper HL, Brady JE, Friedman SR, Tempalski B, Gostnell K, Flom PL (2008). Estimating the prevalence of injection drug use among black and white adults in large U.S. metropolitan areas over time (1992--2002): estimation methods and prevalence trends.  J Urban Health. 85(6):826-56. D’Aunno T, Pollack HA (2002).  Changes in Methadone Treatment Practices: Results from a national panel study, 1988-2000.  JAMA 288(7): 850-6.  Darke S (1998).  The effectiveness of methadone maintenance treatment 3: Moderators of treatment outcome.  In J Ward, RP Mattick, W Hall (Eds.), Methadone Maintenance Treatment and Other Opioid Replacement Therapies (pp. 75-89).  Amsterdam: Overseas Publishers Association, Harwood Academic Publishers.  De P, Cox J, Boivin JF, Platt RW, Jolly AM (2008). Social network-related risk factors for bloodborne virus infections among injection drug users receiving syringes through secondary exchange. J Urban Health 85:77-89.  24 Deamer, R. L., Wilson, D. R., Clark, D. S., & Prichard, J. G. (2001). Torsades de pointes associated with high dose levomethadyl acetate (ORLAAM). J Addict Dis, 20(4), 7-14. Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (1994). Washington, DC: American Psychiatric Association 175-272.  Dobler-Mikola A, Hattenschwiller J, Meili D, et al (2005).  Patterns of heroin, cocaine and alcohol abuse during long-term methadone maintenance treatment.  J Subst Abuse Treat. 29:259-265.  Dolan KA, Shearer J, White B, Zhou J, Kaldor J, Wodak A (2005).  Four-year follow-up of imprisoned male heroin users and methadone treatment: mortality, re-incarceration and hepatitis C infection.  Addiction 100: 820-8.  Drummond MF, O’Brien B, Stoddart GL, Torrance GW (Eds.) (1997).  Methods for the economic evaluation of health care programmes.  2nd edition.  Oxford Medical Publications, Oxford.  Farre M, Mas A, Torrens M, Moreno V, Cami J (2002).  Retention rate and illicit opioid use during methadone maintenance interventions: a meta-analysis.  Drug and Alcohol Dependence 65: 283-90.  Federal PATACOPH (2001). Reducing the harm associated with injection drug use in Canada.  Ottawa: Federal, provincial and territorial advisory committee on population  25 health. Fischer B (2000).  Prescriptions, power and politics: The turbulent history of methadone maintenance in Canada.  Journal of Public Health Policy 21: 187-210. Fischer B, Rehm J, Brissette S, Brochu S, Bruneau J, El-Guebaly N, Noel L, Tyndall M, Wild C, Mun P, Baliunas D (2006).  Illicit opioid use in Canada: Comparing social, health, and drug use characteristics of untreated users in five cities (OPICAN study). Journal of Urban Health, 82(2): 250-66.  Fischer, B., Oviedo-Joekes, E., Blanken, P., Haasen, C., Rehm, J., Schechter, M. T. et al. (2007). Heroin-assisted Treatment (HAT) a Decade Later: A Brief Update on Science and Politics. J Urban Health, 84(4), 552-562.  Fletcher BW, Battjes RJ (1999).  Introduction to the special issue: treatment process in DATOS.  Drug and Alcohol Dependence 57: 81-7.  Galai N, Safaeian M, Vlahov D, Bolotin A, Celentano DD (2003).  Longitudinal patterns of drug injection behaviour in the ALIVE study cohort, 1988-2000: Description and Determinants.  Am J Epidemiol. 158: 695-704. Galynker II, Watras-Ganz S, Miner C, Rosenthal RN, Des Jarlais DC, Richman BL, London E (2000).  Cerebral metabolism in opiate-dependent subjects: effects of methadone maintenance. Mt. Sinai Journal of Medicine, 67(5): 381-7.  26 Hall W (1996).  Methadone maintenance treatment as a crime control measure [letter, see comments].  BMJ 312:1162.  Halliday R (1963).  Management of the Narcotic Addict.  British Columbia Medical Journal 5: 412-14.  Health Canada (2002).  Literature Review, Methadone Maintenance Treatment.  Prepared by Jamieson, Beals, Lalonde and Associates, Inc. for the Office of Canada’s Drug Strategy, Health Canada.  Health Canada (2002).  Best Practices, Methadone Maintenance Treatment.  Prepared by Jamieson, Beals, Lalonde and Associates, Inc. for the Office of Canada’s Drug Strategy, Health Canada.  Haasen, C., Verthein, U., Degkwitz, P., Berger, J., Krausz, M., & Naber, D. (2007). Heroin-assisted treatment for opioid dependence: Randomised controlled trial. Br J Psychiatry, 191, 55-62.  Hser Y-I, Longshore D, Anglin MD (2007).  The life course perspective on drug use: A conceptual framework for understanding drug use trajectories.  Eval Rev. 31:515-547.   27 Institute of Medicine, Fulco CE, Liverman CT, Earley LE, eds (1995).  Development of Medications for the treatment of opiate and cocaine addictions: Issues for the government and private sector.  Washington DC: National Academy Press.  Hubbard RL, Craddock SG, Anderson J (2003).  Overview of 5-year follow-up outcomes in the drug abuse treatment outcome studies (DATOS).  Substance Abuse Treatment 25(3): 125-34.  Joseph H, Stancliff S, Langrod J (2000).  Methadone maintenance treatment (MMT): A review of historical and clinical issues.  The Mount Sinai Journal of Medicine 67(5 and 6): 347-64.  Kendler KS, Prescott C (1998). Cannabis use, abuse and dependence in a population- based sample of female twins.  Am J Psychiatry. 155: 1016-22.  Koob GF, Bloom FE (1998).  Cellular and molecular mechanisms of drug dependence. Science 242: 715-23.  Kraigher, D., Jagsch, R., Gombas, W., Ortner, R., Eder, H., Primorac, A.  et al. (2005). Use of slow-release oral morphine for the treatment of opioid dependence. Eur Addict Res, 11(3), 145-51. Krausz, M., Verthein, U., Degkwitz, P., Haasen, C., & Raschke, P. (1998). Maintenance treatment of opiate addicts in Germany with medications containing codeine--results of a  28 follow-up study. Addiction, 93(8), 1161-7.  Leri F, Flores J, Rajabi H, Stewart J (2003).  Effects of cocaine in rats exposed to heroin. Neuropsychopharmacology, 28:2102-16.  Lowinson JH, Payte JT, Salsitz E, Joseph H, Marion IJ, Dole VP (1997).  Methadone Maintenance.  In JH Lowinson, JT Payte,  Salsitz E, Joseph H, Marion IJ, Dole VP (Eds.), Substance Abuse: A comprehensive Text (3rd ed., pp. 405-15).  Baltimore, Williams and Wilkins.  Maetzel A, Krahn M, Naglie G (2003).  The cost effectiveness of rofecoxib and celecoxib in patients with osteoarthritis or rheumatoid arthritis.  Arthitis and Rheumatism 49; 283- 92.  Maetzel A, Tugwell P, Boers M, Guillemin F, Coyle D, Drummond M, Wong JB, Gabriel SE on behalf of OMERACT 6 Economics Research Group (2003).  Economic evaluation of programs or interventions in the management of rheumatoid arthritis: defining a consensus-based reference case.  J Rheumatol. 30: 891-96. Martin J, Payte JT, Zweben JE (1991). Methadone maintenance treatment: a primer for physicians. J Psychoactive Drugs. 23(2):165-76. Mattick, R. P., Breen, C., Kimber, J., & Davoli, M. (2003). Methadone maintenance therapy versus no opioid replacement therapy for opioid dependence. Cochrane Database  29 Syst Rev, (4), CD002209.  Mattick RP, Kimber J, Breen C, et al (2008).  Buprenorphine maintenance versus placebo or methadone  maintenance for opioid dependence.  Cochrane Database Syst Rev. (2):CD002207.  Mattick RP, Ward J, Hall W (1988).  The role of counseling and psychological therapy. In Ward J, Mattick RP, Hall W (Eds.), Methadone Maintenance Treatment and Other Opioid Replacement Therapies (pp. 265-304).  Amsterdam: Overseas Publishers Association, Harwood Academic Publishers.  Magura S, Nwakeze PC, Demsky S (1998).  Pre- and in-treatment predictors of retention in methadone treatment using survival analysis.  Addiction. 93(1): 51-60.  McLellan AT, Lewis DC, O’Brien CP, et al (2000).  Drug dependence, a chronic medical illness: Implications for treatment, insurance, and outcomes evaluation.  JAMA 284(13): 1689-1695.  Millson P, Challacombe L, Villeneuve PJ, et al (2006).  Determinants of health-related quality of life of opiate users at entry to low-threshold methadone programs.  Eur Addict Res. 12(2): 74-82.   30 Mitchell, T. B., White, J. M., Somogyi, A. A., & Bochner, F. (2006). Switching between Methadone and Morphine for Maintenance Treatment of Opioid Dependence: Impact on Pain Sensitivity and Mood Status. Am J Addict, 15(4), 311-5.  National Academy of Sciences, Institute of Medicine (1995).  Dispelling the myths about addiction.  Washington, DC: National Academy Press.  National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction (1998).  Effective medical treatment of opiate addiction.  Journal of the American Medical Association 280(22): 1936-43. Nestler EJ (2005).  The neurobiology of cocaine addiction.  Sci Pract Perspect. 3(1):4-10. Neumann PJ, Goldie SJ, Weinstein MC (2000).  Preference-based measures in economic evaluation in health care.  Annu Rev Public Health 21: 587-611.  Payte JT (1995).  Opioid Dependence: A practical Guide to Pharmacotherapy. Vancouver, British Columbia.  Puigdollers E, Domingo-Salvany A, Brugal MT, et al (2004).  Characteristics of heroin addicts entering methadone maintenance treatment: quality of life and gender.  Subst Use Misuse. 39(9): 1353-68.  Popova S, Rehm J, Fischer B (2006). An overview of illegal opioid use and health  31 services utilization in Canada. Public Health 120:320-8. Petry NM (2000).  Effects of increasing income on polydrug use: a comparison of heroin, cocaine and alcohol abusers.  Addiction, 95(5): 705-17.  Petry NM, Bickel WK (1998)  Polydrug abuse in heroin addicts: a behavioural economic analysis.  Addiction 93(3):323-35.  Regier DA, Farmer ME, Rae DS, et al (1990).  Comorbidity of mental disorders with alcohol and other drug abuse: Results from the Epidemiologic Catchment Area (ECA) study.  JAMA 1990; 264: 2511-18.  Rehm, J., Gschwend, P., Steffen, T., Gutzwiller, F., Dobler-Mikola, A., & Uchtenhagen, A. (2001). Feasibility, safety, and efficacy of injectable heroin prescription for refractory opioid addicts: a follow-up study. Lancet, 358(9291), 1417-23. Robertson, J. R., Raab, G. M., Bruce, M., McKenzie, J. S., Storkey, H. R., & Salter, A. (2006). Addressing the efficacy of dihydrocodeine versus methadone as an alternative maintenance treatment for opiate dependence: A randomized controlled trial. Addiction, 101(12), 1752-9. Rounsaville BJ, Petry NM, Carroll KM (2003).  Single versus multiple drug focus in substance abuse clinical trials research.  Drug and Alcohol Dependence 70:117-25.  Ruel J-M, Hickey CP (1993). Diversion of Methadone: What are the risks?  BC Medical  32 Journal. 35(6):420-1.  Schukit MA, Hesselbrock, Tipp J (1994).  A comparison of DSM-III, DSM-IV and ICD- 10 substance use disorders diagnoses in 1992 men and women subjects in the COGA study.  Addiction, 89: 1629-38.  Siegal HA, Li L, Rapp RC (2002).  Abstinence trajectories among treated crack cocaine users.  Addictive Behaviours 27: 437-49.  Single E (1997). A sociodemographic profile of drug users in Canada . Ontario, Canada: HIV/AIDS prevention and community action programs of Health Canada., 2000. Special Issue.  Science. 278: 1-211.  Series on Addiction, Lancet.  1996; 347: 31-36, 97-100, 162-166, 237-240, 301-305, 373- 376.  Stitzer ML, Chutuape MA (1999).  Other substance use disorders in methadone treatment: prevalence, consequences, detection and management.  In EC Strain and ML Stitzer (Eds.), Methadone treatment for opioid dependence (pp. 86-117).  Baltimore: The Johns Hopkins University Press.   33 Strain EC (1999).  Methadone dose during maintenance treatment.  In EC Strain and ML Stitzer (Eds.), Methadone treatment for opioid dependence (pp. 62-85).  Baltimore: The Johns Hopkins University Press, 1999.  Strang J, Manning V, Mayet S, Ridge G, Best D, Sheridan J (2007).  Does prescribing for opiate addiction change after national guidelines?  Methadone and buprenorphine prescribing to opiate addicts by general practitioners and hospital doctors in England, 1995-2005.  Addiction 102; 761-770.  Sumnall HR, Tyler E, Wagstaff GF, Cole JC (2004).  A behavioural economic analysis of alcohol, amphetamine, cocaine and ecstasy purchases by polysubstance misusers.  Drug and Alcohol Dependence 76: 93-9.  Termorshuizen F, Krol A, Prins M, Geskus R, van den Brink W, van Ameijden EJC (2005).  Prediction of relapse to frequent heroin use and the role of methadone prescription: An analysis of the Amsterdam Cohort Study among drug users.  Drug and Alcohol Dependence 79: 231-40.  True WR, Xian H (1999).  Common genetic vulnerability for nicotine and alcohol dependence in men.  Arch Gen Psychiatry. 56: 655-61.  Tsuang MT, Lyons MJ, Eisen S, et al (1996).  Genetic influences on DSM-III-R drug abuse and dependence: a study of 3,372 twin pairs. Am J Med Genet. 67: 473-7.  34  Tyndall MW, Kerr T, Zhang R, King E, Montaner JG, Wood E (2006). Attendance, drug use patterns, and referrals made from North America's first supervised injection facility. Drug Alcohol Depend. 83:193-8.  Van den Brink, W., & Haasen, C. (2006). Evidenced-based treatment of opioid- dependent patients. Can J Psychiatry, 51(10), 635-46 .  VandenBree M, Johnson E, Neale M, Pickens R (1998).  Genetic and environmental influences on drug use and abuse/dependence in male and female twins.  Drug Alcohol Depend. 52: 231-41.  Walters JP (2000).  Methadone: Drug Policy Information Clearinghouse Fact Sheet. Executive office of the President, Office of National Drug Control Policy. www.whitehousedrugpolicy.gov  Ward J, Mattick RP, Hall W (1998). Methadone Maintenance Treatment and Other Opioid Replacement Therapies.  Amsterdam: Overseas Publishers Association, Harwood Academic Publishers.  Ward J, Mattick RP, and Hall W (1998).  How long is long enough?  Answers to questions about the duration of methadone maintenance treatment.  In J Ward, RP Mattick, and W Hall (Eds.), Methadone Maintenance Treatment and other Opioid  35 Replacement Therapies (pp. 305-336).  Amsterdam: Overseas Publishers Association, Harwood Academic Publishers.  Wise RA, Bozarth MA (1981).  Brain substrates for reinforcement and drug-self- administration.  Prog Nueropsychopharmacol. 5: 467-474.  World Health Organization (2005): World Drug Report.  www.who.org  Zhang Z, Friedmann PD, Gerstein DR (2003).  Does retention matter?  Treatment duration and improvement in drug use.  Addiction 98: 673-84.     36 2 PROPORTIONAL HAZARDS FRAILTY MODELS FOR RECURRENT METHADONE MAINTENANCE TREATMENT1  2.1 Introduction  Over forty years since its introduction as a form of opioid substitution treatment, methadone remains the most effective form of treatment for opioid dependence (Amato et al, 2003; Mattick et al, 2008).  In British Columbia, Canada, methadone is recommended to be delivered with a maintenance-oriented, rather than abstinence-oriented philosophy, thus acknowledging opioid dependence as a chronic disease (McLellan et al, 2000). This form of low-threshold treatment is provided by specially-licensed physicians and, in most cases, is dispensed and consumed at community pharmacies under direct supervision. The ineffectiveness of detoxification treatment is recognized (Caplehorn et al, 1994), and penalization of patients who relapse into illicit opioid use during maintenance treatment or voluntary withdrawal is discouraged (College of Physicians and Surgeons of British Columbia, 2005).  While in methadone maintenance treatment, decreases in illicit drug use, criminal activity and the rate of mortality have all been documented (Caplehorn et al, 1994; Hall, 1996; Clausen, Anchersen and Waal, 2008), however longitudinal studies on individuals dependent on opioids and other drugs have generally identified patterns of chronic  1 A version of this chapter has been published: Nosyk B, MacNab YC, Sun H, Marsh DC, Fischer B, Schechter MT, Anis AH. Proportional hazards frailty models for recurrent methadone maintenance treatment.  American Journal of Epidemiology, 2009; 170(6):783-92.  37 continued drug use, switching between periods of treatment and relapse, and full abstention (Bell et al, 2006; Galai et al, 2003; Dobler-Mikola et al, 2005; Bovasso and Cacciola, 2003; Termorshuizen et al, 2005; Hser, Longshore and Anglin, 2007).  As achievement of sustained drug abstinence has proven unattainable for the vast majority of heroin users (Dobler-Mikola et al, 2005), it is of interest to determine whether long-term maintenance on substitution treatment such as methadone can be achieved among patients experiencing multiple periods of treatment and relapse.  A number of studies have focused on identifying determinants of the time to discontinuation of MMT (Magura, Nwakeze and Demsky, 1998; Peles, Schreiber and Adelson, 2006) however none have appropriately modeled the time to discontinuation of successive treatment episodes as recurrent event data (Therneau and Grambsch, 2000). Doing so may shed light on the drug treatment careers of opioid-dependent patients, and provide evidence for health policy decisions on appropriate treatment for patients who had failed to benefit from methadone treatment in the past.  The objectives of our study were to identify the determinants of the time to discontinuation of Methadone Maintenance Treatment (MMT) among opioid-dependent individuals receiving MMT in one or more treatment episodes in British Columbia. Historically, approximately one-half of all Canadian methadone treatment slots have been located in the province of British Columbia (Fischer, 2000). To fulfill our objectives, we take advantage of a population-level administrative database that has collected information on methadone dispensations since 1996.  38 2.2 Materials and Methods  2.2.1 Cohort Definition  The data source for this study was the British Columbia PharmaNet database, which records all prescription drug dispensation in the entire province of British Columbia. Data fields available included a de-identified patient ID, date of birth, gender, drug identification number, the date of the prescription (date), the length of the prescription (number of days supplied, or days), drug dosage (quantity), de-identified prescriber code and pharmacy code, as well as a geographical identifier, aggregated by local health area. The study was approved by the University of British Columbia/Providence Health Care Behavioural Research Ethics Board.  The cohort included all individuals receiving MMT over an 11-year period: January 1st, 1996 to December 31st 2006.  MMT episodes were constructed using the service date and days supplied fields; a treatment episode length was calculated as the difference between the last and first days of dispensed medication ((datet1 + dayst1) – datet0) within a period of continuous retention in treatment, where continuous treatment entailed no interruptions in prescribed doses lasting longer than 30 days.  As such, we considered all treatment episodes beginning after January 30th, 1996, to ensure consistency in our calculation of episode lengths and eliminate left-censored observations. Treatment episodes that were ongoing as of the end of the follow-up period were (right) censored.  Efforts were made to correct any misclassification in the dataset in regards to prescribed doses, lengths and  39 dates of prescriptions.  We found that less than 1% of identified errors could not be corrected.  2.2.2 Determinants of Time to Discontinuation of MMT  From the date, days and quantity fields, we constructed variables indicating the mean and maximum dosage over the course of a treatment episode, whether or not carries were typically prescribed (mode of days > 1), and the percentage of days in which doses were missed, or treatment adherence (calculated as the percentage of days over the duration of an episode in which methadone was dispensed).  It should be noted that this measure of adherence only considers dose administration and consequently reflects only one aspect of performance-adherence.  The physician code was used to construct variables indicating methadone patient load, as it is plausible that prescribers with excessively large (or small) MMT patient loads may provide sub-optimal care for their patients through either lack of individualized treatment or inexperience.  Furthermore, variables indicating switches in prescribing physician and pharmacy within a treatment episode were constructed.  Information on other drugs dispensed in the six months prior to treatment initiation was used to construct the Clark Chronic Disease Score (CDS-1) (Von Korff, Wagner, Saunders, 1992).  CDS scores have performed well as predictors of hospitalization (Putnam et al, 2002) and mortality (Schneeweiss et al, 2001) in epidemiological research with administrative databases, suggesting they are useful indicators of baseline co-  40 morbidity for control of confounding.  A SAS Macro program calculating these empirically-derived co-morbidity scores for patients at the time of treatment entry was modified to accommodate the Canadian Drug Product Database (DPD) coding system (Pettus, 2005; Health Canada Drug Product Database).  Dispensations of the majority of antiretroviral medications for HIV-positive individuals in the province of British Columbia were held outside of the PharmaNet database; identification of the AIDS virus was therefore constructed from dispensations of prophylactic drugs.  Finally, local health areas were considered urban if they included an urban centre with a population > 50,000, and rural otherwise (British Columbia Census, 2001).  Geographical data was matched with 2001 census data on income (incidence of low income), unemployment (labour force participation rate) and education (percentage of 18 year olds without high school diplomas). Low neighbourhood socioeconomic status has been shown to be associated with a variety of adverse health outcomes (Diez Roux, 2004).  2.2.3 Statistical Analysis  The determinants of retention were analyzed using time-to-event data statistical techniques.  Mantel-Haenszel tests were conducted for each independent variable considered in the analyses.  Kaplan-Meier curves were plotted with respect to the key independent variables.  The assumption of proportionality for each of the study variables was assessed by visual inspection of Schoenfeld residual plots and weighted residuals score tests (Grambsch and Therneau, 1994).  Non-proportional hazards were accepted if the variation in ß(t) was not large in comparison to the parameter estimate, and the  41 variable did not have a circular relationship with the outcome.  The resulting parameter estimates were therefore time-averaged regression effects.  Given the simple structure of the regression model (ie. no random effects on covariates), non-proportionality of the covariates did not bias the estimates of the variance components (Xu and Gamst, 2007).  Proportional Hazards (PH) gamma frailty models were fitted to account for the dependence in the length of individuals’ repeated treatment episodes.  This specification was preferred to accelerated failure time models due to the minimal distributional assumptions required.  The PH gamma frailty model can be written as: , where h0(t) is the unspecified baseline hazard, Zij is the fixed effect covariate matrix for episode i=1, …, nj among individuals j=1,…, m, and ß is a vector of regression coefficients.  The unobserved random effect, or frailty, for the jth individual (vj) was assumed to follow a gamma distribution (vj ~ gamma (1/θ, 1/θ). Frailties are unobservable random variables corresponding to each individual’s underlying modification of the baseline hazard function.  They represent unobserved covariates common across multiple treatment episodes for each individual.  Furthermore, the PH gamma frailty model captures the correlation in episode lengths within an individual; conditional on the frailty terms, the episode lengths are independent (Sargent, 1998).  It should be noted that times between acute events, or ‘gap times’ have been assessed in other frailty model applications, our analysis includes only durations in which patients were enrolled in continuous methadone maintenance treatment, as described above.  42 While the choice of time scale differs, the PH gamma frailty model can be applied in either application (Cook and Lawless, 2008). The covariates included in the final regression model included baseline predictors and in-treatment variables (daily dose, treatment adherence) averaged over the duration of a treatment episode; relationships on the latter category should be interpreted as associations, as temporality cannot be established.  The sensitivity of parameter estimates was tested using several alternative model formulations.  First, prior studies have employed different treatment discontinuation rules, and the relatively liberal definition used here may be subject to some degree of misclassification.  The model was refitted to determine the sensitivity of parameter estimates to the length of the treatment discontinuation rule (from 30 days to 14, 60 days).  Second, individuals with numerous treatment episodes may have had different patterns of use than those experienced by MMT clients with only a single attempt; it is possible that pooling these clients may result in information loss on what may be considered distinct client types with different patterns of treatment response.  The PH frailty model was therefore refitted on cohorts of patients with 2 to 6 treatment episodes. Third, over 10% of the study cohort died during the follow-up period; inclusion of these cases may impact the interpretation of our results.  We thus re-ran our multivariate model after removing these cases to determine if coefficient estimates remained consistent with those found in the baseline model.  Finally, the outcome of later treatment attempts may have been dependent upon past treatment experiences (mean dose, adherence in episode t-1 and the means by which episode t-1 was discontinued – by dropout or successful  43 taper) or by the duration of time between treatment episodes.  Our final sub-analysis thus considered 2nd and subsequent treatment episodes of individuals experiencing multiple treatment episodes. The statistical analyses were executed using SAS version 9.1 and freeware [R] version 2.5.1. In particular, the PH gamma frailty models were fitted using [R], in which the EM algorithm was used for maximizing the partial log-likelihood function.   44 2.3 Results  A total of 18,160 individuals experienced 34,725 methadone maintenance treatment episodes in the 11-year study period.  Removing ongoing episodes at the start of follow- up (n=1,715, 4.9%), as well as episodes in which the mean dose was <10mg or >400mg (Indicating a high probability of coding errors) (n=353, 1.0%), our analysis included 17,005 patients, receiving a median of 2 (IQR: 1, 3; min: 1, max: 15) treatment episodes for a total of 32,656 treatment episodes included in our study.  Summary statistics on the length of time in treatment were presented in table 2.1.  The overall median episode length was 205 days (IQR: 50, 666), with 21% of all episodes ongoing at the end of follow-up.  Summary statistics on all explanatory variables were generated from individuals’ initial treatment episodes captured in the database, and presented in table 2. Nearly 34% of all patients entering treatment were female, with the majority (65%) between the ages of 20 and 40.  Most notably, just over half of all patients received mean daily doses below 60mg in their initial attempt, and 31% had < 90% adherence.  In testing the PH assumption, the null hypothesis of proportionality was rejected for most variables, as can be expected with large datasets featuring highly dispersed time-to-event lengths.  Closer inspection of Schoenfeld residual plots suggested that the variation over time in most cases was small relative to the size of the time-averaged coefficient value. The parameter estimate on treatment adherence, for instance, showed a great deal of  45 variation over time, however the variation was relatively small compared to the time- averaged parameter estimate, and at no point did ß(t) cross zero (Figure 2.1). Three variables (carry doses prescribed, prescriber transfer and pharmacy transfer) could be considered both predictors of, and influenced by the outcome. For instance, carry doses may be more convenient for patients, and thus enhance retention, but only those successfully maintained for at least 12 weeks were recommended to be considered for carry doses (College of Physicians and Surgeons of British Columbia, 2005).  The parameter estimate for carry privileges accordingly twice changes direction over the length of follow-up (Figure 2.2).  These variables were therefore omitted from the model.  Kaplan-Meier curves for time to discontinuation of MMT stratified by treatment episode for the entire cohort, and for patients experiencing at least 4 treatment episodes (generated by pooling all observations, thus not controlling for correlation between successive episodes) are displayed figures 3 and 4. While later treatment episodes had shorter median time to discontinuation overall (Figure 2.3), among patients with multiple treatment attempts, time to discontinuation was longer in later episodes (Figure 2.4).  Table 2.3 presents the results of the PH gamma frailty model.  The median frailty parameter hazard ratio for individual j was vj = 0.997 (Interquartile range: 0.833, 1.162; minimum: 0.221; maximum: 2.500); 50.5%.  The variance of the frailty term was relatively high (0.24), and its effect was statistically significantly greater than zero (P < 0.001), indicating that significant between-subject heterogeneity was present. Conditional on the individual-specific random effect, and controlling for all other covariates, later  46 attempts at methadone maintenance treatment had statistically significantly longer time to discontinuation [Episode 2: Hazard Ratio: 0.87 95% Confidence Interval: (0.84, 0.90); Episode 3: 0.84 (0.80, 0.88); Episode 4: 0.80 (0.75, 0.85); Episode 5: 0.83 (0.77, 0.90); ≥Episode 6: 0.79 (0.73, 0.86)].  Furthermore, episodes featuring mean daily doses over 60mg per day were associated with longer retention in treatment.  Poor adherence to treatment had the most profound effect on retention; missing 10-30% of prescribed doses over the course of an episode was associated with discontinuing treatment nearly 2.5 times sooner; those missing > 30% discontinued nearly 7 times sooner than those missing <10% of prescribed doses.  The statistical significance of the linear time trend variable suggested that episodes initiated later in the follow-up period tended to last slightly longer, controlling for all other covariates.  Patients receiving treatment from physicians with a patient load in the third quartile (89 - 182 patients) remained in treatment longer; however a patient load over this level was detrimental to treatment retention.  Patients with higher levels of treated co-morbidity had progressively longer time to discontinuation of treatment.  Finally, individuals residing in neighbourhoods with higher incidence of low income and low educational achievement had slightly shorter times to discontinuation.  In assessing the sensitivity of our results to different model structures, we found that coefficient estimates were robust to changes in the treatment discontinuation rule, and did not differ substantially when excluding patients who had died during follow-up (n=1,828 (10.1)).  Models executed on individuals with k=2 to 6 episodes showed that these cohorts showed distinct patterns of response to treatment, however each supported the  47 finding that later episodes tended to be progressively longer for treatment repeaters (Figure 2.5).  Finally, among individuals with at least 2 treatment episodes, the median time to treatment re-entry was 143 days (IQR: 63, 379), while 66.8% were classified as dropouts from their previous episode, and 2.4% had successfully completed treatment, according to our definitions. Both the dose and level of adherence of individuals’ previous treatment episode were predictors of treatment outcome, and those reaching abstinence in their previous episode had longer time to discontinuation; coefficient estimates on the episode count variable, however corresponded with those generated by the baseline model, suggesting that these factors did not confound our primary results.   48 2.4 Discussion  The primary finding of this study was that individuals with multiple methadone treatment attempts were maintained for successively longer periods in later treatment attempts. Results presented in figure 2.5 suggested that individuals may experience a number of failed attempts before being maintained long-term.  For instance, among clients with 4-6 treatment attempts, second attempts were, at most, only marginally longer than their initial treatment episode; however final attempts for each cohort were 85-90% longer. Parallels to treatment for other forms of substance dependence may be drawn; for instance, in the smoking cessation literature, studies suggests that re-treating relapsed smokers with the same medication may be of substantial benefit (Gourlay et al, 1995; Gonzales et al, 2001).  Alternative forms of treatment merit consideration as second-line options following relapse from MMT, and/or as a means to engage patients who choose not to re-enter MMT or cannot be productively maintained after numerous treatment attempts.  While other forms of pharmacological treatment have been implemented elsewhere (Mattick et al, 2008; Mitchell et al, 2006; Krausz et al, 1998; Rehm et al, 2001), these were unavailable in British Columbia during the study period.  A recent Cochrane review of dosing in MMT programs could not conclude that doses greater than 100mg per day produced better retention outcomes (Faggiano et al, 2003), however our population-level study found no such ceiling dose threshold.  Despite wide consensus on optimal dosing (Ward, Hall and Mattick, 1999; Blaney and Craig, 1999; Anderson and Warren, 2004), a large proportion of the individuals did not receive the  49 minimum optimal dose of 60mg per day. Higher daily doses may also improve treatment adherence, another important determinant of retention. Emphasizing to prescribers the merits of maintaining patients on higher doses can therefore have a profound public health impact.  Another key finding was that high physician patient load had an adverse effect on treatment retention.  MMT patient loads in the 3rd quartile had a significant positive effect on episode duration, which was robust to model formulation.  While our data did not include information on treatment setting, it is plausible that mid-size practices may produce better outcomes than practices with small MMT patient loads, and the largest clinics may not be offering responsive, individualized treatment.  We also found that progressively higher levels of treated co-morbidity predicted longer treatment episodes. Given that only treated co-morbidity is captured, the parameter estimates likely reflect an underlying characteristic indicating the desire and ability to acquire treatment.  Receiving other prescriptions may indicate attentive or comprehensive care and therefore positively influences retention.  Individuals’ health can be influenced by the socioeconomic characteristics of their neighbourhood, through features of the physical, social or service environment through a number of pathways (Diez Roux, 2004).  Using indicators that emphasized the extent of low socioeconomic status, rather than measures of central tendency, we found that individuals residing in neighbourhoods with lower high school completion rates were  50 retained in treatment for shorter durations.  These ecological variables have been used here as proxies of individual socioeconomic status; individuals of relatively lower socioeconomic status may have been self-selected for residence in poorer neighbourhoods.  The effect found may thus be the result of individual, rather than neighbourhood-level characteristics.  Application of the PH gamma frailty modeling technique allowed us to utilize all data available to us, rather than focusing on a single treatment episode.  Furthermore, application of PH gamma frailty models allowed for accurate inference into the effects of successive episodes through the use of the individual-specific frailty term, which takes into account the correlation between repeated episodes of treatment and captures unobserved heterogeneity across individuals.    Two prior studies have acknowledged the recurrent nature of methadone maintenance treatment, but have chosen to model these effects by calculating standard errors robust to the individual-level clusters, rather than acknowledging the true longitudinal nature of the data (Salter and Solomon, 1997).  It is crucial to note that these techniques cannot incorporate the correlation among successive observations for an individual entity, and thus may provide biased parameter estimates when used in applications with repeated measures across individuals. Given the effect of model formulation on coefficient estimates (specifically on the episode count variable, which was reversed in a previous study), this was an important innovation in the analysis of MMT data.  While frailty models have scarcely been used in applied studies, these methods may be useful in any study of diseases with periods of remission and relapse,  51 including cancer, autoimmune diseases and other forms of substance dependence (Cook and Lawless, 2008).   Our study was not without limitations.  First, our results primarily reflect the treatment careers of opioid dependent individuals with free access to primarily low-threshold, community-based methadone treatment programs.    Second, some proportion of the study sample may have been lost to follow-up due to intra-provincial out-migration; past studies of injection drug users in Vancouver, British Columbia suggest this proportion is small (1.2-5.5% per annum) (Rachlis et al, 2008).  This can be considered non- differential outcome misclassification, resulting in attenuation of measures of association towards the null, suggesting our hazard ratios may have been somewhat conservative. Third, retention in treatment is necessarily an intermediate outcome; while the benefits of MMT are only experienced while in treatment, other outcomes including opioid abstinence were not captured.  The data presented here relate only to the duration of medication administration.  Finally, there were both advantages and disadvantages in using administrative data to conduct this retrospective cohort study. The primary limitations were twofold: first, there was potential for misclassification of treatment episode lengths, dosage and adherence due to coding errors in the database.  As described earlier, we have taken steps to reduce this misclassification as much as possible.  Second, any non-experimental study may be subject to residual and/or unmeasured confounding (Fewell, Smith and Sterne, 2007). Other predictors of retention in treatment such as other illicit drug use, criminal activity,  52 motivational status and social supports were unavailable.  Though we cannot ascertain the individual effects of the unobserved factors, we can confidently state that their omission has not biased the coefficients on the existing fixed effects included in the analysis.  The advantages of using this dataset, however, were substantial: the use of this centralized drug dispensation database allowed for a population-level study unparalleled in size, which allowed us to observe not only the effects of several key determinants on individual episodes, but also the pattern of treatment, thus taking into account the chronic, recurrent nature of opioid dependence.  We believe our study adds considerable insight into the study of patients’ drug treatment careers (Hser et al, 1997) by capturing the natural patterns of remission and relapse exhibited by opioid dependent individuals.  In a population-level study of over 17,000 patients over an 11-year period, we found that methadone can be an effective form of treatment for opioid dependence even among those with multiple relapses. The prescription of higher daily doses of methadone and improvements in adherence can improve treatment retention substantially.   53 2.5 Acknowledgements  Author Affiliations: Centre for Health Evaluation and Outcome Sciences, Vancouver, British Columbia, Canada (Bohdan Nosyk, Huiying Sun, David C. Marsh, Martin T. Schechter, Aslam H. Anis), School of Population and Public Health, University of British Columbia, Vancouver, Canada (Ying C. MacNab, Martin T. Schechter, Aslam H. Anis), Centre for Healthcare Innovation and Improvement, Vancouver, British Columbia, Canada (Ying C. MacNab), Vancouver Coastal Health Authority (David C. Marsh), Department of Health Sciences, Simon Fraser University, Vancouver, British Columbia, Canada (Benedikt Fischer), Centre for Applied Research in Mental Health and Addictions Vancouver, British Columbia, Canada (Benedikt Fischer).  This study was funded by research grant from the Canadian Institute of Health Research (grant no. MOP-86476).  Bohdan Nosyk was supported in part by doctoral research awards from the Michael Smith Foundation for Health Research, the Canadian Institutes of Health Research, and the Research in Addictions and Mental Health Policy & Services CIHR strategic training initiative.     54  Table 2.1: Summary statistics on retention in methadone maintenance treatment   Treatment Episode Duration Treatment Retention, % Episode N Median IQR 6 months* 12 months^ 24 months# Censored, %** All 32,656 205 50  666 55.5 42.0 29.5 21.8 Episode 1 16,250 217 53  716 56.1 42.2 29.4 19.1 Episode 2 8,048 231 60  753 57.5 44.9 32.1 24.2 Episode 3 3,909 199 49  624 54.8 41.2 28.7 23.9 Episode 4 2,023 183 45  532 54.2  39.9 27.9 25.8 Episode 5 1,086 146 41  438 49.1 36.9 23.4 25.3 Episode ≥6 1,340 99  25  310 42.4 29.0 19.7 25.1 IQR: Interquartile range * Episodes initiated after (1,6,2006) were omitted. ^ Episodes initiated after (1,1,2006) were omitted. # Episodes initiated after (1,1,2005) were omitted. ** Retained as of December 1st, 2006.   55  Table 2.2: Summary statistics of explanatory variables  Explanatory Variables  N % Female  5,789 34.0 Age at treatment entry: < 20  601 3.5                                        20-30  5,049 29.7                                        30-40  5,738 33.7                                        40-50  4,244 25.0                                        50-60  1,080 6.7                                        ≥ 60  293 1.7 Optimal daily dose (60mg) ever prescribed  11,995 70.5 Overall mean daily dose: < 40mg  3,682 21.7                                           40-60mg  4,889 28.8                                           60-80mg  3,957 23.2                                           80-100mg  2,374 14.0                                           100-120mg  1,157 6.8                                           ≥ 120mg  946 5.6 Carry doses (most frequently prescribed)  3,755 22.1 Adherence to treatment: >90%  11,727 69.0                                         70% - 90%  4,053 23.8                                         < 70%  1,225 7.2 Ever transferred to a different prescriber  8,290 48.8 Ever transferred to a different pharmacy  8,715 51.3 Residing in urban area*  13,982 82.3  Median IQR Clark co-morbidity score** 1,014 381 1,977 Prescriber MMT patient load** 89 54 182 EC Income:   incidence of low income 14.0 10.9 22.6 EC Education:   % adults with high school diploma 27.7 20.9 34.5 EC Employment:    labour force participation rate 64.2 61.2 68.4 N=17,005; summary statistics on patients’ initial treatment episode captured in the dataset. IQR: Interquartile range. * Defined as LHAs with a city with > 50,000 individuals. **Model A (Clark et al, 1998). *** From a total of N=608 prescribing physicians.  56  Table 2.3: Results from the proportional hazards gamma frailty model, with indicated determinants of time to MMT discontinuation   Group Size HR 95% Confidence Interval P-value   N % Male   20,649  63.2 ref. -- -- -- Female  12,007  36.8 0.987 0.955 1.019 0.420 Age at treatment entry: <40 22,817  69.9 ref. -- -- --                                        ≥40 9,839  30.1 0.872 0.842 0.902 <0.001 Mean daily dose: < 40mg 6,844 21.0 1.207 1.161 1.255 <0.001                             40-60mg 9,396  28.8 ref. -- -- --                             60-80mg 7,468  22.9 0.701 0.674 0.729 <0.001                             80-100mg 4,643  14.2 0.539 0.514 0.564 <0.001                             100-120mg 2,272  7.0 0.441 0.414 0.469 <0.001                             ≥120mg 2,033  6.2 0.377 0.351 0.405 <0.001 Adherence: >90% 21,263  65.1 ref. -- -- --                    70-90% 8,335  25.5 2.289 2.213 2.366 <0.001                    <70% 3,058  9.4 6.835 6.534 7.150 <0.001 Episode 1 16,250  49.8 ref. -- -- -- Episode 2 8,048  24.6 0.871 0.843 0.901 <0.001 Episode 3 3,909  12.0 0.839 0.803 0.878 <0.001 Episode 4 2,023  6.2 0.801 0.754 0.850 <0.001 Episode 5 1,086  3.3 0.833 0.770 0.902 <0.001 Episode ≥ 6 1,340  4.1 0.791 0.732 0.856 <0.001 Patient load: Q1 8,346  25.6 ref. -- -- --                      Q2 7,995 24.5 0.976 0.938 1.015 0.230                      Q3 8,244  25.2 0.935 0.898 0.974 0.001                      Q4 8,071  24.7 1.074 1.030 1.121 0.001 Chronic Disease Score: Q1 6,767  20.7 ref. -- -- --                                        Q2 9,510  29.1 0.927 0.890 0.965 <0.001                                        Q3 8,147  25.0 0.901 0.863 0.940 <0.001                                        Q4 8,232  25.2 0.904 0.865 0.944 <0.001 Calendar year -- -- 0.980 0.974 0.985 <0.001 EC Education -- -- 1.004 1.002 1.005 <0.001 EC Employment -- -- 0.998 0.994 1.001 0.180 EC Income -- -- 1.003 1.001 1.005 0.010 HR: Hazard Ratio; ref.: reference group.  P-value: two-sided p-values reported.   57  Figure 2.1: Estimated coefficient value β(t) (curved solid line) with 95% confidence interval (dashed lines) for treatment adherence (70%-90%) as a function of time from treatment initiation.  X-axis formatted using ‘kaplan-meier’ scale.                  58 Figure 2.2: Estimated coefficient value β(t) (curved solid line) with 95% confidence interval (dashed lines) for carry privileges granted as a function of time from treatment initiation.  X-axis formatted using ‘kaplan-meier’ scale.                  59 Figure 2.3:  Kaplan-Meier curves for time to discontinuation of MMT by treatment episode: Full sample [N=17,005]. Points of censorship have been suppressed.                    60 Figure 2.4: Kaplan-Meier curves for time to discontinuation of MMT by treatment episode: Patients with ≥ 4 episodes [N=2,023].  Points of censorship have been suppressed.                    61  Figure 2.5:  Hazard ratios on the episode count variable from multivariate models on time to MMT discontinuation, including patients with two episodes [N=7,898], three episodes [N=5,534], four episodes [N=3,660], five episodes [N=2,383] and six episodes [N=1,546].  62 2.6 References

  Amato L, Davoli M, Perucci CA, et al (2003).  An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research.  J Subst Abuse Treat. 28:321-329.  Anderson JF, Warren LD (2004).  Client retention in the British Columbia Methadone Program, 1996-1999.  Can J Public Health  95(2):104-109.  Bell J, Burrell T, Indig D, et al (2006).  Cycling in and out of treatment; participation in methadone treatment in NSW, 1990-2002.  Drug Alcohol Depend. 81:55-61.  Blaney T, Craig RJ (1999). Methadone maintenance. Does dose determine differences in outcome? J Subst Abuse Treat. 16(3):221-228.  Bovasso G, Cacciola J (2003).  The long-term outcomes of drug use by methadone maintenance patients.  J Behav Health Serv Res. 30(3):290-303.  Caplehorn JR, Dalton MSYN, Cluff MC, et al (1994).  Retention in methadone maintenance and heroin addicts’ risk of death.  Addiction 89:203-207.  City Populations: British Columbia, 2001 Census estimates: http://www.citypopulation.de/Canada-BritishColumbia.html (Accessed June 1, 2008)   63 Clausen T, Anchersen K, Waal H (2008).  Mortality prior to, during and after opioid maintenance treatment (OMT): a national prospective cross-registry study.  Drug Alcohol Depend. 94(1-3):151-157.  College of Physicians and Surgeons of British Columbia (2005).  Methadone Maintenance Handbook.  http://www.bcguidelines.ca/gpac/guideline_methadone.html. (Accessed: June, 2008).  Cook RJ, Lawless JF (2008).  The statistical analysis of recurrent events.   New York, NY:  Springer-Verlag Publishers.  Diez Roux, AV (2004).  The study of group-level factors in epidemiology: Rethinking variables, study designs and analytical approaches.  Epidemiol Rev. 26:104-111.  Dobler-Mikola A, Hattenschwiller J, Meili D, et al (2005).  Patterns of heroin, cocaine and alcohol abuse during long-term methadone maintenance treatment.  J Subst Abuse Treat. 29:259-265.  Faggiano F, Vigna-Taglianti F, Versino E, et al (2003).  Methadone maintenance at different dosages for opioid dependence.  Cochrane Database Syst Rev. 3:CD002208.   64 Fewell Z, Smith D, Sterne JAC (2007).  The impact of residual and unmeasured confounding in epidemiologic studies: A simulation study.  Am J Epidemiol. 166(6);646- 655.  Fischer B (2000).  Prescriptions, power and politics: The turbulent history of methadone maintenance in Canada.  J Public Health Policy 21:187-210.  Galai N, Safaeian M, Vlahov D, et al (2003).  Longitudinal patterns of drug injection behaviour in the ALIVE study cohort, 1988-2000: Description and Determinants.  Am J Epidemiol. 158:695-704.  Gonzales DH, Nides MA, Ferry LH, et al (2001). Bupropion SR as an aid to smoking cessation in smokers treated previously with bupropion: a randomized placebo-controlled study. Clin Pharmacol Ther. 69:438–44.  Gourlay SG, Forbes A, Marriner T, et al (1995). Double blind trial of repeated treatment with transdermal nicotine for relapsed smokers.  BMJ 311:363– 6.  Grambsch P, Therneau T (1994). Proportional Hazards tests and diagnostics based on weighted residuals.  Biometrika 81:515-526.  Hall W (1996).  Methadone maintenance treatment as a crime control measure [letter, see comments].  BMJ 312:1162.  65  Health Canada Drug Product Database.  http://www.hc-sc.gc.ca/dhp- mps/prodpharma/databasdon/index-eng.php (Accessed April 1, 2008).  Hser Y-I, Longshore D, Anglin MD (2007).  The life course perspective on drug use: A conceptual framework for understanding drug use trajectories.  Eval Rev. 31:515-547.   Hser Y-I, Anglin MD, Grella C, et al (1997).  Drug Treatment Careers: A Conceptual Framework and Existing Research Findings.  J Subst Abuse Treat. 14(6):543-558.  Krausz M, Verthein U, Degkwitz P, et al (1998). Maintenance treatment of opiate addicts in Germany with medications containing codeine--results of a follow-up study. Addiction 93(8):1161-1167.  Magura S, Nwakeze PC, Demsky S (1998).  Pre- and in-treatment predictors of retention in methadone treatment using survival analysis.  Addiction 93(1):51-60.  Mattick RP, Kimber J, Breen C, et al (2008).  Buprenorphine maintenance versus placebo or methadone  maintenance for opioid dependence.  Cochrane Database Syst Rev. (2):CD002207.  McLellan AT, Lewis DC, O’Brien CP, et al (2000).  Drug dependence, a chronic medical illness: Implications for treatment, insurance, and outcomes evaluation.  JAMA 284(13):1689-1695.  66  Mitchell TB, White JM, Somogyi AA, et al (2006). Switching between Methadone and Morphine for Maintenance Treatment of Opioid Dependence: Impact on Pain Sensitivity and Mood Status. Am J Addict. 15(4):311-315.  Peles E, Schreiber S, Adelson M (2006).  Factors predicting retention in treatment: 10- year experience of a methadone maintenance treatment (MMT) clinic in Israel.  Drug Alcohol Depend. 82:211-7.  Pettus PT (2005).  Chronic Disease Score Utility.  http://harryguess.unc.edu/HMO Research Network/hmo research network.htm (Accessed April 1, 2008)  Putnam KG, Buist DSM, Fishman P, et al (2002).  Chronic disease score as a predictor of hospitalization.  Epidemiology 13(3):340-346.  Rachlis BS, Hogg RS, Wood E, et al (2008).  Factors associated with geographic migration among a cohort of injection drug users.  Health Place 14:536-543.  Rehm J, Gschwend P, Steffen T, et al (2001). Feasibility, safety, and efficacy of injectable heroin prescription for refractory opioid addicts: a follow-up study. Lancet 358(9291):1417-1423.   67 Salter A, Solomon PJ (1997).  Truncated recurrent event survival models for methadone data.  Biometrics 53:1293-1303.  Sargent DJ (1998).  A general framework for random effects survival analysis in the cox proportional hazards setting.  Biometrics, 54: 1486-97.  Schneeweiss S, Seeger SD, Maclure M, et al (2001).  Performance of comorbidity scores to control for confounding in epidemiologic studies using claims data.  Am J Epidemiol. 154(9):854-864.  Strike CJ, Gnam W, Urbanoski K, et al (2005).  Factors predicting 2-year retention in methadone maintenance treatment for opioid dependence.  Addict Behav. 30:1025-1028.  Termorshuizen F, Krol A, Prins M, et al (2005).  Prediction of relapse to frequent heroin use and the role of methadone prescription: An analysis of the Amsterdam Cohort Study among drug users.  Drug Alcohol Depend. 79:231-240.  Therneau TM, Grambsch PT (2000).  Modeling Survival Data: Extending the Cox Model. New York, NY:  Springer-Verlag Publishers.  Von Korff M, Wagner EH, Saunders K (1992).  A chronic disease score from automated pharmacy data.  J Clin Epidemiol. 45:197-203.   68 Ward J, Hall W, Mattick RP (1999).  Role of maintenance treatment in opioid dependence.  Lancet, 353:221-6.  Xu R, Gamst (2007).  On proportional hazards assumption under the random effects models.  Lifetime Data Anal,13:317-32.   69 3 AN EMPIRICAL SPECIFICATION OF THE RATIONAL ADDICTION MODEL FOR ILLICIT DRUG USE2  3.1 Introduction  The abuse of licit and illicit drugs carries a large social, health and economic burden on society.  Regulatory response to illicit drug use is comprised primarily of three components: enforcement, aimed at stemming the supply of illicit drugs; prevention, aimed at reducing drug use prevalence; and treatment, aimed at reducing the demand for illicit drugs among the addicted. The effectiveness of drug enforcement and treatment may be confounded by simple postulates of consumer behaviour.  Becker and Murphy (1988) introduced the theoretical model of rational addiction, which shows that drug consumption can be considered rational in an intertemporal utility optimization framework.  Drug users may trade-off the future disutility associated with drug abuse for the immediate benefits derived, including the experience of euphoria and mediation of withdrawal symptoms. This extension to classical microeconomic theory has facilitated observation of price elasticity of demand of illicit drugs; however the concept has not yet been extended in an empirical context to determine the effects of price changes of other illicit drugs which may be economic substitutes or complements.  We hypothesize that the immediate utility derived from consuming one illicit drug, for instance heroin, can also be had from  2 A version of this chapter has been submitted for publication: Nosyk B, Inoue A, Fisher DG, Anis AH.  An Empirical Test of Rational Polydrug Addiction.  70 consuming prescription opioids or stimulants such as cocaine; therefore, prices play a role in the consumption decisions of users of multiple illicit drugs. While previous studies have shown that illicit drug consumption does respond to price changes, enforcement efforts to eradicate the supply of heroin and cocaine have resulted in competing suppliers earning market share through alternate sources and trafficking routes (Ciccarone, 2009) and is further undermined by growing trends towards the use of alternatives produced domestically or obtained through different supply chains, such as prescription opioids, most notably oxycontin  (Rawson, Maxwell and Rutkowski, 2007; Maxwell and Rutkowski, 2008), synthetic stimulants such as methamphetamines and benzodiazepines (SAMHSA, 2002, 2003).  While these aggregate-level trends cannot be interpreted as evidence of substitution effects at the individual-level, it is clear that the market for illicit drugs in the U.S. and elsewhere has become more diversified, with a greater variety of products available to new and experienced consumers.  Furthermore, studies to date have shown that purity-adjusted prices of traditional drugs of abuse have been declining since the early 1990s (Caulkins et al, 2004). On the other hand, pharmacological treatment for drug abuse is often delivered with a mono-drug focus, even though it is well-established that many drug users presenting for treatment often abuse more than one drug.  Polydrug abuse within populations of heroin addicts has been observed within epidemiologic and clinical studies for some time, with evidence of cocaine use in over 47% of heroin dependant patients found as early as 1991 (Ball and Ross, 1991).  A more recent study of chronic heroin users conducted in Canada found that over 70% of participants used cocaine regularly (Oviedo-Joekes et al, 2008).  Experimental studies on non-human subjects have shed  71 light on the neurological and physiological effects of opioid and stimulant co- administration. Among rats previously made dependent on heroin, self-administration of cocaine is enhanced.  Furthermore, rats that experienced acute, spontaneous withdrawal from heroin showed a depression in locomotor activity that was reversed by cocaine administration.  Importantly, the greater the dose of heroin exposure, the more cocaine was required to reverse the depression of locomotor activity observed during acute withdrawal (Leri et al., 2004).  Economic theory expands understanding of the physiological and neurological effects of drug use to also consider how price changes may impact consumption decisions regarding various illicit drugs, and how drug consumption can be regarded in terms of a forward-looking plan to maximize intertemporal utility. While multiple drug abuse is common among illicit drug users, there have been a limited number of empirical studies providing estimates of price elasticity of demand for illicit drugs, and fewer still that have considered multiple drug use.  Van Ours (1995) used historical (aggregate) data on the opium trade in the Dutch East Indies from 1923 – 1938 in an application to the demand for opium using the rational addiction framework. He found support for the rational addiction hypothesis and found that opium was price inelastic (e = -0.7). Grossman and Chaloupka (1998) estimated the demand for cocaine among high school seniors from the ‘Monitoring the Future’ US National Survey.  They found that cocaine use in the past year, and frequency of use among participants was negatively related to the price of cocaine, with the long-run price elasticity of total consumption of - 1.35.  72 Using a pooled set of (cross-sectional) data from the National Household Surveys on Drug Abuse, Saffer and Chaloupka (1999) applied myopic models (assuming consumption is a function of price and income alone) in estimating the effect of price on the demand for alcohol, marijuana, cocaine and heroin and found that drug prices affected drug use or participation negatively.  Further, this study showed evidence of complementarity in the use of these drugs. Finally, Dave (2006) conducted an ecological study to estimate the relationship between cocaine and heroin prices and drug-related emergency department admissions in the United States.  Estimates of price elasticity of the probability of cocaine and heroin admissions were -0.27 and -0.10, respectively, suggesting that healthcare costs could be avoided as a result of drug price increases.  Like Grossman and Chaloupka (1998) and Saffer and Chaloupka (1999), Dave estimated illicit drug prices from the US Drug Enforcement Administration’s System to Retrieve Information from Drug Evidence (STRIDE) dataset. Parallel research on the effect of drug prices on consumption has been conducted in experimental settings with hypothetical drug price and purchase scenarios.  Most notably, Petry and Bickel (1998) conducted a series of experiments on a small cohort of polydrug users in opioid maintenance treatment.  Varying the price of heroin and other drugs commonly used by the cohort, the authors found that heroin was price inelastic, and valium and cocaine purchases increased as heroin prices rose, suggesting these were economic substitutes.  The relationship between valium and heroin purchases were asymmetric, however, as heroin purchases were independent of valium prices.  This highlights the notion that polydrug users have an ordinal ranking of preferred substances  73 that is based on past experience, including their level of dependence, as well as current prices and expected future prices.  Similar studies have been conducted to determine income elasticity among polydrug users (Petry, 2000), and relationships in the demand for alcohol and various stimulants (Sumnall et al, 2004).  Bickel, Degrandpre and Higgins (1993) provide an introduction to the methodology of these experiments and a review of early research in this field. The objective of our study is to determine the responsiveness of consumption to price changes (cross-price elasticities of demand) of crack cocaine, powder cocaine and heroin among a cohort of chronic illicit users.  We believe our analysis offers several advantages over previous studies published to date.  First, we focus our analysis on a cohort of illicit drug users recruited as part of a longitudinal epidemiological study. Homelessness, multiple drug use and acquisitive crime were highly prevalent among participants. The sample is thus representative of chronic illicit drug users which incur the highest societal economic burden through the healthcare and criminal justice systems. Further, it is likely that the study cohort in large part would not be captured by national/state survey methods, which have served as the basis for prior studies on the demand for illicit drugs.  Second, we use the most up to date statistical methods to calculate illicit drug price series, calculated at the city level where possible, and merged to chronic drug users assessed at different points in time at the individual level.  Finally, as multiple drug use is commonplace among chronic drug users presenting for treatment, explicitly modeling this behaviour provides a more accurate and complete account of the effects of changes in illicit drug prices on chronic drug users’ consumption behaviour.  74 The remainder of the study proceeds as follows.  We describe the theoretical framework for the empirical analysis in section 2, and then describe the data used in our analysis in section 3 and the empirical framework in section 4.  We describe our primary results and sensitivity analyses in section 5, and provide discussion of these results in section 6.  75 3.2 Theoretical Framework  We base our analysis of the demand for illicit drugs within the theory of rational addiction introduced by Becker and Murphy (1988).  The theory, built upon prior work by Stigler and Becker (1977) and Iannacone (1984) differed from past studies on the consumption of addictive goods, in that consumers of addictive substances were not assumed to behave purely myopically, wherein all future effects of the addiction are ignored (Houthakker and Taylor, 1970; Spinnewyn 1981; Mullahy 1985).   The theoretical model took into account three features of consumption of addictive goods which distinguished them from others.  Tolerance is conceptualized as having a given level of consumption yield lower utility as cumulative past consumption is higher; withdrawal refers to the disutility associated with cessation or interruption of consumption, and reinforcement represents a learned response to past consumption. Orphanides and Zervos (1995) extended the theory to better explain the dynamics of initiation of illicit drug use.  They argue that the essential feature lacking in the RA model is the recognition that inexperienced individuals are initially uncertain of the exact potential harm associated with consuming an addictive good.  The extension the authors propose is based on three fundamental points; first, that consumption of the addictive good is not equally harmful to all individuals, second, that each individual possesses a subjective belief structure concerning his/her potential to become addicted, and third, that this belief structure is optimally updated with information gained through consumption, via a Bayesian learning process.  As such, despite the rationality of their decisions, addicts regret their past consumption decisions and are not ’happy’.  They become  76 addicted through the unintended outcome of experimenting with an addictive good known to provide instant pleasure, and only probabilistic future harm. We focus on consumers with long histories of drug use, who regularly use, and may be physiologically addicted to more than one substance.  We therefore extend the classic rational addiction theoretical framework to account for the consumption of three illicit drugs.  Rational behaviour is defined as a consistent plan to maximize utility over time, where present-oriented behaviour is possible due to individuals’ high rates of time preference which may discount the future harms associated with illicit drug use (communicable disease, mental health conditions, premature mortality) in favour of current benefits derived from use (euphoria, mediation of withdrawal symptoms).  We follow notation and exposition as presented in Chaloupka (1991). At any moment in time, an individual’s utility can be expressed as a function of health, HTt, the euphoria produced by the consumption of addictive goods {crack cocaine (Rt), powder cocaine (Ct), heroin (Ht)}, Et, and a composite of other consumption commodities, Zt: (1) Ut = u[HTt, Et, Zt],  {UH, UE, UZ} > 0, {UHH, UEE, UZZ} < 0, where Ut is concave and has negative second derivatives for each of the arguments. Health can be considered a function of market goods and health promotion activities, Mt, as well as cumulative past drug use, or the “addictive stock” of crack cocaine RAt, powder cocaine, CAt, and heroin consumption, HAt. (2) HTt = HT[Mt, RAt, CAt, HAt], HTM > 0, HTMM < 0, {HTRA, HTCA, HTHA} < 0, {HTRARA, HTCACA, HTHAHA} < 0.  77              Crucially, euphoria (Et) (the short-term psychological and physiological effects of illicit drugs) is produced by any of the aforementioned addictive goods and their (independent) capital stocks.  The notion of withdrawal follows from the assumption that lower current consumption of illicit drugs leads to lower utility.  The notion of tolerance is incorporated into the model by the assumption that increased consumption has a positive effect on the production of euphoria, while greater past consumption has a negative effect.  Reinforcement is captured by the assumption that the marginal productivity if illicit drug consumption in the production of euphoria is larger with higher levels of addictive capital stock: (3) Et = E[Rt, Ct, Ht, RAt, CAt, HAt], {ER, EC, EH} > 0, {ERR, ECC, EHH} < 0, {ERRA, ECCA, EHHA} > 0. Assuming positive but diminishing marginal utility on the inputs used to produce the composite good, Xt, a derived instantaneous utility function is obtained as: (4) Ut = U[Rr, Ct, Ht, RAt, CAt, HAt, Yt], Where Yt is a vector including inputs into the production of the composite good and health.  At any time, the following relationships hold: (5) (6) (7) (8) (9) (10)  78 (11) (12) (13) (14) (15)               The characteristics of addictive consumption are reiterated in equations (5) to (14).  Withdrawal is captured in equations (5) - (7) since total utility decreases if consumption of any of the addictive substances is reduced.  Tolerance is captured through the negative marginal utility of the addictive capital stocks (equations (9) - (11)); greater past consumption lowers current utility, ceteris paribus.  Reinforcement is shown in (12) – (14), which state that the marginal utility of current consumption is larger as past consumption is greater for each drug.  A simple investment function describes the independent addictive capital stock accumulation processes: (16) for each substance D={C, R, H}, where δ is the constant rate of depreciation of the capital stock over time.  Under the assumption of a time-additive utility function, which ensures utility is separable in each of its arguments over time, a constant rate of time preference, σ, and an infinite lifetime, the lifetime utility function is: (17) .  79 Rational behaviour implies maximization of this function subject to a lifetime budget constraint.  Treating the price of the composite good Yt as the numeraire, ignoring the allocation of time over the life cycle and assuming perfect capital markets, the budget constraint can be expressed as: (18) , where R(0) is the discounted value of lifetime income and assets.  Maximizing intertemporal utility subject to the budget constraint provides the basis of the empirical framework, detailed below.

  80 3.3 Empirical Framework To obtain empirically estimable crack cocaine, powder cocaine and heroin demand as a function of Yt, Ct, Pt, Ht, CAt, PAt and HAt, a quadratic utility function is assumed, resulting in linear first order conditions for each of the above.  Assuming also the individual’s rate of time preference is equal to the market rate of interest, the resulting instantaneous utility function is as follows: (19)      Maximizing each equation out with respect to Yt and using results of the first order conditions for Ct, Pt, Ht, CAt, PAt and HAt, we derive the following demand equations: (20) (21) (22) In each equation, current price is predicted to be negatively related to consumption.  Further, current consumption is expected to be positively related to future consumption.  Opposing effects of reinforcement (positive) and an increase in the shadow price of the addictive capital stock (negative) suggest that the effect of the addictive capital stock may be positive or negative.  The features of rationality and addiction in the  81 RA model can be tested directly: if powder cocaine, crack cocaine and heroin are not addictive, then past consumption and price should have no effect on current consumption. If these drugs are addictive, but users are myopic, then future consumption and price should have no effect on current consumption. Of particular interest are the long-term price elasticities of demand for each substance, and the cross-price elasticities of demand with respect to each of the other substances.  To estimate this elasticity, we assume that in the long run, a steady-state level of consumption will be reached that serves to replace depreciation on the addictive stock (R*=δRA*; C*=δCA*; H*=δHA*, where RA*, CA* and HA* are the optimal level of the addictive capital stocks).  An anticipated permanent rise in price changes both consumption and the addictive capital stock in each period, and new steady-state equilibriums of crack cocaine, powder cocaine and heroin consumption are achieved. The resulting long-run elasticities are: (23) (24) (25) (26) (27) (28)  82 (29) (30) (31) Where  . Given the population under study (primarily crack cocaine and heroin users, with a smaller proportion using powder cocaine with lower frequency), hypotheses can be made regarding the signs and magnitudes of elasticity measures.  While own-price elasticities on illicit drugs should be price inelastic, secondary drugs of abuse (in this sample, powder cocaine) among polydrug users should be more elastic, given that they are less preferred, and therefore more responsive to price changes.  Further, we have already hypothesized that illicit drug users may treat different drugs producing euphoria (Et) as substitutes.  Complementarity in consumption is more likely to be observed with drugs less commonly used, when budgets allow for use. It should be noted that empirical estimation of Becker-Murphy rational addiction models with aggregate-level data has been criticized on both theoretical and empirical grounds.  Auld and Grootendorst (2004) demonstrate that the general conclusions found in such studies will be reached if consumption is serially correlated, but evidence for rational addiction is not present either because the good is not addictive or because  83 consumers cannot anticipate price movements. Furthermore, Ferguson (2000) argued that empirical specifications with aggregate-level data on consumption cannot reasonably be expected to display the dynamics predicted by models of individual optimizing behaviour.  We describe the individual-level consumption data used in this study below.  84 3.4 Data 3.4.1 Study Sample  Individual-level data on drug consumption, income and other factors was ascertained from a longitudinal epidemiological study on crack cocaine and heroin users. The Cooperative Agreement for AIDS Community-based outreach/intervention research program (COOP study) was conducted by National Institute of Drug Abuse researchers across the United States from 1992 to 1998.  The study recruited 31,088 respondents from 19 US cities and assessed drug use and other health and demographic information at baseline and at 3- or 6-months.   The study was among the first and largest undertaken to identify risk factors for HIV seroconversion among intravenous and other drug users. Baseline questionnaires in the COOP dataset included a field for drug use history, the number of days and the number of times injected in the past 30 days for a variety of drugs.  We selected patients who (i) completed the follow-up assessment, (ii) were not in drug treatment at the baseline study assessment and (iii) provided complete information on income and each of the other key covariates considered in our analysis. The reliability of self-reported data on drug consumption, employment and income were formally tested in this dataset, with favourable results (Dowling-Guyer et al, 1994; Needle et al, 1995; Johnson, Fisher and Reynolds, 1999). 3.4.2 Illicit Drug Prices  Data on crack cocaine, powder cocaine and heroin prices were computed from the System to Retrieve Information on Drug Evidence (STRIDE), maintained by the US  85 Department of Justice.  The STRIDE database collects information from purchases made by undercover drug enforcement agents, including cost, weight and purity. As our study focused on individual drug use, wholesale drug purchase observations from the STRIDE database were excluded.  Specifically, only powder cocaine purchases ≤ 10 grams, crack cocaine purchases ≤ 15 grams, and heroin purchases ≤10 grams (according to ‘quantity levels’ 1 and 2, as defined in Arkes et al, 2004), were included. Quarterly estimates of city-specific, purity-adjusted prices for heroin, powder cocaine and crack cocaine for the period (1990Q1 – 1998Q4) were constructed from the STRIDE database using a two step procedure to solve for expected purity, and purity adjusted price, respectively.  Expected purity is computed in the first stage using a linear mixed effect (random coefficient) regression model with purity regressed against the purchase quantity and a series of year-quarter dichotomous variables. Given the nested nature of the data, and the fact that specific clusters of observations are not necessarily independent, cities were treated as random effects in the model. In the second stage, the real drug price is regressed against expected pure amount (log(expected purity)+ log(amount) and the year/quarter indicator variables used previously.  The model is estimated using a generalized linear mixed regression model with a gamma distribution, log link function, and constant coefficient of variation.  This methodology is detailed in Arkes et al (2004).   Extreme outlier deletion within simple random intercept models and Laplace approximation techniques were used in order to achieve convergence in these highly parameterized models.  86 Best linear unbiased predictors of real, city-specific, purity-adjusted drug prices were calculated throughout the study follow-up period and matched to the individual- level drug consumption data within the epidemiological cohort study by year/quarter and city (Metropolitan Statistical Areas: Denver, Detroit, Houston, Los Angeles, Miami, New Orleans, New York, Philadelphia, Phoenix, Portland, San Antonio, San Francisco, St. Louis, Washington).  Study sites for which city-specific prices were not estimable were matched with regional drug prices (Anchorage (Pacific region), Lexington (East-South Central region), Raleigh-Durham/Wake County (South Atlantic region), Hartford (New England region)).  Further discussion on the use of STRIDE data in economic analyses can be found in Caulkins et al (2007).  87 3.5 Empirical Implementation  In estimating demand functions for each drug, the number of times a drug was used in the past 30 days was assessed at both baseline and (3 or 6-month) follow-up. Given that the majority of crack, cocaine and heroin purchases by street users were for $10 (Golub and Johnson, 2004), we assumed a single time use was equivalent to one $10 purchase.  We then rescaled real prices per pure gram into real prices per single use using median purchase quantity data from $20 and $40 purchases from the STRIDE database ($10 purchases were uncommon).  One gram of crack was equivalent to 12.5 times used, one gram of powder cocaine was equivalent to 8 times used, and 1 gram of heroin was equivalent of 20 times used. As the dependent variables were based on count variables, and the study sample consisted of a heterogeneous mix of primarily crack cocaine and heroin users, as well as some primary powder cocaine users, the distributions of individual drug consumption data was skewed to the right and contained a large proportion of zeroes.  Using the total sample, each series was overdispersed and contained excess zeroes; each feature causes Poisson regression results to be inefficient (Cameron and Trivedi, 1998).  To deal with these features of our data, we used negative binomial (NB) and zero-inflated negative binomial (ZINB) regression models to fit equations (20) – (22) for (i) the complete CA study sample, (ii) polydrug users only, and (iii) 3-drug users only. (Cameron and Trivedi, 1986).  Similar techniques were recently used in estimating demand equations for cigarettes with individual-level data (Sheu et al, 2004), and the methods used are described more exhaustively therein.  88 Intuitively, the ZINB model puts additional weight on the probability of observing a zero through a mixing specification, dividing individuals into non-users and potential users (Jones, 2000). We chose a zero-inflated specification over a hurdle model specification due to the fact that those with zero drug consumption were not necessarily strictly non-users, as implicitly assumed with the contrary technique. The Vuong test (Vuong, 1989) was used to choose between negative binomial and zero-inflated negative binomial regression model formulations in cases (i) and (ii). Both NB and ZINB models were fitted with quadratic variance function, while the ZINB model was fitted with standard normal (probit) link function, which predicts the probability of being in the non-user group. The Vuong test statistic has an asymptotic standard normal distribution with large positive values (V > 1.96) favouring NB specification, and large negative values (V < -1.96) favouring models in ZINB specification. Self-reported drug use reported in baseline and follow-up questionnaires were used as period t and t+1 consumption.  To estimate addictive capital stocks for each illicit drug, once again following Chaloupka (1991), we used the following formula: (32) assuming the initial capital stocks for drugs j = {powder cocaine, crack cocaine, heroin} were equal to zero.  We therefore used mean consumption, the rate of depreciation and the number of years and individual had consumed each drug to construct the addictive capital stocks.  For never users, the stock took on a value of zero; for former users, mean consumption was equal to 1, while for current users it was the average of reported use at time t and t+1.  Contrary to cigarette smoking, relatively low rates of depreciation were  89 assumed on the basis of the evidence presented in numerous biological studies on humans and animals that show many of the neurological and physiological effects of heroin and cocaine use stay with the individual long after cessation, if this is ever achieved (Galynker et al, 2000; Nestler, 2005). Low depreciation rates do not imply more severe addiction, but rather that the consequences of addiction are longer-lived after cessation. Consistent with past studies, we included a set of exogenous covariates indicating patient sociodemographic characteristics in our reduced-form demand equations.  These included age, gender, ethnicity, an indicator of homelessness and income. Summary statistics on all covariates can be found in table 3.1.  Several additional modeling complexities arose due to the structure of the model, and the nature of the COOP dataset.  First, the addictive capital stock and lead of consumption for each drug in each demand function were endogenous, as shock to the error term will affect marginal utility in all periods.  Since current consumption is independent of other past or future prices, but related to past and future consumption, three lags and leads of drug prices were used as instruments for lagged and lead consumption, respectively.  Each of the exogenous covariates was also included as a regressor in these models.  The strength of the instruments was assessed by comparing F- statistics from the first-stage regression models on each instrument against critical values for the weak instrument test based on two-stage least squares bias (Stock and Yogo, 2002; Stock, Wright and Yogo, 2002). Second, the COOP study featured a non-randomized comparison of non- pharmacological interventions designed to reduce needle-sharing and risk-taking  90 behaviour.  Every participant entered into the study received a standard intervention, which consisted of street-based outreach and HIV prevention-based counselling, while others received enhanced interventions, including additional counselling services, educational videos, partook in social gatherings and attended support groups (Booth, Kwiatkowski and Stevens, 1998).  We assumed, and verified, that the standard intervention did not have a direct effect on drug consumption, while the enhanced interventions did.   As patients with higher levels of drug use may have been more likely to receive treatment, or receive higher intensity treatment, receipt of treatment at time (t+1) was endogenous.  We aimed to control for the effects of behavioural drug treatment by estimating the local average treatment effect (LATE) attributable to those receiving treatment at time (t+1).  Treatment allocation was used as an instrument for receipt of treatment at time (t+1).  This variable satisfies the conditions of existence, monotonicity and strength set out by Imbens and Angrist (1994).  Other exogenous covariates in the reduced-form demand equations were also included as regressors in the first-stage model, and the strength of the instrument was tested as described above. Long-run estimates of own- and cross-price elasticity of demand were evaluated at mean levels of baseline (time t) consumption and drug prices (real prices per 0.3, 0.75 and 0.4 grams for crack, powder cocaine and heroin, respectively).  Mean drug consumption for crack, powder cocaine and heroin was 51.32, 14.47 and 25.76 times used in the past 30 days, respectively, while real prices per pure gram at baseline were $288.38, $81.20 and $478.80.  Statistical Analysis Software (SAS) version 9.2 and [R] version 2.5.1 statistical software packages were used to conduct the above analysis.  91 3.6 Results  Descriptive statistics of the study sample were presented in table 3.1, while their drug consumption habits were presented in table 3.2.  Participants selected from the COOP study (N=16,451) were of extremely poor socioeconomic standing.  On average, participants were of age 38, 85% were non-white, and 65% were male.  The rate of unemployment was far higher than that of the general population (nearly 80%), only 59% had high school diplomas and 28% were homeless.  The majority of participants reported a monthly income below $500 per month.  Homelessness was slightly higher among current crack users, while employment was lower among heroin users.  Illicit drug use was also highly prevalent among respondents, with crack cocaine the most commonly used, and most frequently used among current users, followed by heroin and powder cocaine.  Interestingly, participants reported the longest durations of past use for powder cocaine, and shortest for crack cocaine.  In the early 1990’s crack cocaine was still a relatively new street drug, and the study period captured the years following the period of largest growth in the crack cocaine epidemic in the United States (Cornish and O’Brien, 1996).  City-specific real drug prices for a selection of the U.S. cities included in the study were presented in figure 1, while fixed effects estimates of first- and second- stage models on drug purity and prices were presented in tables A1-A2.  In general, real drug prices were stable throughout the 7-year study period however variation between cities  92 and over time within cities was high.  In general, cities with more drug purchase observations in STRIDE were more stable (eg. Los Angeles (heroin and crack cocaine), New York).  Consistent with prior studies with STRIDE data, drug prices tended to be lower in ports of entry for illicit drugs (in particular southern California and the Eastern Seaboard) and increased with greater distance from these locations, with the highest prices observed in the Northern and Central regions (Rueter et al, 1988; Caulkins and Padman, 1994; Caulkins, 1995).  Reduced-form demand functions for the full sample and for polydrug users only were estimated with ZINB regression.  Using the complete dataset, results indicated ZINB models provided better fit for each illicit drug (Powder cocaine: V= -10.36, p<0.001; Crack cocaine: V= -21.42, p<0.001; Heroin: V= -28.89, p<0.001). Vuong test results also supported the ZINB specification for the polydrug users sample.  Table 3.3 presents F-statistic values from the first stage-estimates of instruments for RAt, CAt and HAt, Rt+1, Ct+1, Ht+1 and the local average treatment effect, used in the future consumption instruments.  Based on Stock and Yogo (2002) F-statistic critical values, our powder cocaine consumption instruments were estimated with <10% bias, while all other instruments were estimated with < 5% bias.  Baseline results on NB coefficients from ZINB models for each drug, with assumed rates of depreciation of δ=0, 0.2 and 0.4 were presented in tables 4-6.  Crucially, in each model, own-drug addictive capital stock and future consumption were positive  93 and statistically significant, while coefficients on current prices were negative for each drug.  The effect of the addictive capital stock for each individual drug model was always significant at the 1% level (two-tailed t-test), suggesting that the effect of past consumption was larger than the opposing effect of an increase in the full price of drug use as the stock increased.  Coefficients on future drug consumption were similarly significant at at least the 10% level in all but the polydrug users model (Ht+1) indicating individuals were not behaving myopically.  Results were largely consistent within the full cohort and stratified (polydrug, multidrug) samples. Estimated long-run price elasticities of demand from each model were presented in table 3.4-3.6.  We found that both heroin and crack cocaine were highly price inelastic. As the patient cohort under study were primarily crack cocaine and heroin users, we hypothesized that powder cocaine was likely to be used only complementarily; our results confirm this hypothesis, with own-price effects indicating that demand for powder cocaine was highly elastic, though more inelastic among polydrug and 3-drug users. Further, negative cross-price elasticities on both heroin and crack cocaine (εrc, εhc) indicated that powder cocaine consumption decreased with increases in other drug prices – evidence of complementarity.  While cross-price effects between heroin and crack cocaine and heroin were small (as were all estimated cross-price effects), estimates were consistently positive, indicating that these drugs may have been treated as substitutes within the study sample. Results on cross-price elasticities were similar in the full cohort and restricted cohort models. Rational addiction theory implies that individuals with a greater preference for the present are potentially more subject to becoming addicted than those with a greater  94 preference for the future.  Older individuals are often assumed to be more future-oriented than younger individuals.  To explore the possibility of differences in behaviour based on different rates of time preference, we estimated demand equations on polydrug users stratified by age (<30 years vs. ≥ 30 years).  Results were shown in table 3.7.   While results on crack cocaine user were similar between cohorts, we found that effects on addictive capital stocks were larger for younger individuals, while effects on future consumption were smaller for both powder cocaine and heroin consumption, suggesting younger individuals behaved relatively more myopically in their consumption of these drugs.  95 3.7 Conclusions  This paper developed powder cocaine, crack cocaine and heroin demand equations derived from the Becker-Murphy (1988) model of rational addiction and estimates these demand equations using data on a cohort of illicit drug users from 19 U.S. cities.  In general, the estimates support the hypotheses that powder cocaine, crack cocaine and heroin use are addictive substances and their users do not behave myopically. Crack cocaine and heroin were price inelastic, while powder cocaine was price elastic in this cohort of primarily crack and heroin users, and acted as a complement to consumption of these other drugs. Further support for the Becker-Murphy model was found in the estimation of separate demand equations for subsamples based on age.  The strong effects of the addictive capital stock and relatively weak effects of future consumption among younger individuals supports the a priori hypotheses younger illicit drug users behave relatively more myopically than older users. The estimates presented above support the hypothesis that higher prices are independently associated with lower consumption of both heroin and crack within this cohort, however it is important to note that price increases in either crack cocaine or heroin alone would, at best, have no effect on consumption of the other drug, and may in fact increase other drug consumption, ceteris paribus.  This finding has important implications for drug enforcement efforts aimed at stemming the supply of individual drugs.  While national surveys and other data sources have observed increasing trends towards nonmedical prescription drug use and relatively new stimulants such as  96 methamphetamines at the aggregate level, we have shown that illicit drugs can be considered substitutes or complements for one another at the individual level, and thus may adjust consumption in response to temporary, and potentially also long-term supply disruptions.  This suggests that drug treatment may be a more effective means of reducing illicit drug use; however treatment programs must acknowledge and treat all forms of concurrent addiction/abuse – a sentiment echoed elsewhere in the community of drug treatment researchers and practitioners (Rounsaville, Petry and Carroll, 2003). Evidence of substitution effects between heroin and crack cocaine suggest that single- drug focused treatment such as methadone maintenance treatment, which acts as a substitute for heroin, eliminating withdrawal symptoms while blocking euphoric effects of other opioids, may not address concurrent crack cocaine use, and may indeed increase use among polydrug users in treatment. Specific challenges in the design of the study stem from the limitations of the particular datasets employed, and the use of city-level price data.  First, the COOP study had a follow-up rate of 52%, suggesting relatively poor external validity.  Furthermore, respondents provided information on the consumption of various substances based on the numbers of times in the past 30 that they have been used, which may be subject to recall bias.    Second, the STRIDE database has faced criticism and has been questioned in its usefulness in economic research given its non-random selection of drug purchases (see Horowitz et al, 2001, and related comments for a full discussion).  Recent innovations in the analytic methods used to construct these price series, implemented here, have attempted to address this criticism (Arkes et al, 2004).  Further misclassification on price data resulted from the use of city-level purity-adjusted prices.  We can only speculate that  97 the prices retrieved were representative of those faced by the users in the CA dataset, however any individual-level variation in purchase prices and purity within a given city and quarter could not be observed with the methodology employed. Due to ethical and logistic concerns it is unlikely that the ‘first-best’ design for this type of study, capturing drug consumption, nominal prices and drug purity at the individual level in a large, geographically dispersed sample of drug users, could ever be conducted. Measures of variance in drug prices across quartiles of self-reported drug use were similar.  As such, we considered the combination of misclassification from non-random sampling in STRIDE and the use of aggregate-level (city or regional) prices to be non-differential (ie. prices are not necessarily biased in either direction, simply more variable than in reality), resulting in coefficient estimates attenuated towards the null hypothesis (β=0).  While this is evident in the results of our reduced-form demand models, due to the large and geographically and temporally dispersed study sample, statistically significant effects were found in many cases for most key study variables. Despite these limitations, STRIDE remains a rich source of information to inform drug policy, and among the only sources for drug price and purity data available to researchers. The combination of these datasets provides a rare opportunity to study the demand for illicit drugs among chronic and multiple drug users.   98 3.8 Acknowledgements  Research for this paper was supported by Doctoral Research Awards from the Canadian Institutes of Health Research, British Columbia Mental Health Foundation, Michael Smith Foundation for Health Research and the Research in Addictions in Mental Health Policy and Services CIHR Strategic Training Initiative.  We would like to thank Drs. Robert Hogg, Steffanie Strathdee, Alex Kral and representatives at the Substance Abuse and Mental Health Services Administration (SAMHSA) for their help in identifying relevant data sources for this study.  We are extremely grateful to the Katherine L. Myrick and Phyllis Drewery at the United States Drug Enforcement Freedom of Information and Access Department for making the STRIDE database available to us. We are indebted to Jonathon P. Caulkins and Rosalie Pacula for helpful comments and suggestions.  Preliminary versions of this paper were presented at the 2009 International Health Economics Association World Congress in Beijing, China.  We thank the conference participants for helpful comments and suggestions.   99 Table 3.1: Descriptive statistics of study sample   Full Sample [N = 16,451] Current Heroin Users [N=7,178] Current Crack Users [11,639] Current Cocaine Users [N=6,309]  N % N % N % N % Age [Mean (SD)] 37.63 7.98 39.40 7.97 37.12 7.66 37.66 8.25 Female Gender 5698 34.64 2092 29.14 4420 37.98 1744 27.64 Married 2909 17.68 1606 22.37 1754 15.07 1179 18.69 Non-white 13909 84.55 5826 81.16 10330 88.75 5078 80.49 Employed, part-time or full-time 3516 21.37 1259 17.54 2539 21.81 1343 21.29 High school education 9731 59.15 4143 57.72 7050 60.57 3697 58.6 Monthly Income: <$500 9522 58.37 3722 52.46 6994 60.56 3492 55.88                               $500-$999 4293 26.32 1979 27.89 2976 25.77 1645 26.32                              $1000-$1999 1662 10.19 894 12.6 1078 9.33 710 11.36                              $2000-$3999 655 4.02 395 5.57 394 3.41 306 4.9                              ≥ $4000 180 1.1 105 1.48 106 0.92 96 1.54 Homeless 4553 27.68 1759 24.51 3541 30.42 1766 27.99 Receiving Drug treatment at t+1 10491 63.77 4161 57.97 7655 65.77 3949 62.59     Treatment Allocation: 1 8743 53.15 3932 54.78 5996 51.52 3469 54.98     Treatment Allocation: 2 5645 34.31 2390 33.3 4177 35.89 1972 31.26     Treatment Allocation: 3 842 5.12 539 7.51 495 4.25 440 6.97     Treatment Allocation: 4 198 1.2 111 1.55 113 0.97 65 1.03     Treatment Allocation: 5 1023 6.22 206 2.87 858 7.37 363 5.75    100  Table 3.2: Descriptive statistics of drug utilization   Full Sample [N=16,451] Current Heroin Users [N=7,178] Current Crack Users [11,639] Current Cocaine Users [N=6,309]   Mean SD Mean SD Mean SD Mean SD Heroin Use No. years since first use 10.95 11.47 18.12 9.91 8.59 10.84 13.02 11.35 No. times used, past 30 days 25.76 60.01 58.79 79.23 14.62 43.80 33.74 73.78 No. days used, past 30 days 8.93 12.68 20.30 11.61 5.30 10.52 10.68 13.12 No times used, past 30 days (t+1) 15.70 40.84 34.39 55.27 8.70 27.34 19.17 46.96 No. days used, past 30 days (t+1) 6.04 11.00 13.14 13.12 3.62 8.96 7.03 11.54 Crack Cocaine Use No. years since first use 6.66 5.75 5.56 5.83 7.99 5.43 6.16 5.91 No. times used, past 30 days 51.32 106.08 29.75 80.22 71.95 119.55 45.64 106.06 No. days used, past 30 days 12.27 11.85 7.48 10.97 16.92 10.75 10.32 11.74 No times used, past 30 days (t+1) 27.77 75.39 17.71 58.63 37.83 86.01 24.32 72.84 No. days used, past 30 days (t+1) 7.57 10.49 4.89 9.21 10.14 11.05 6.30 10.07 Cocaine Use No. years since first use 12.02 9.39 14.41 9.52 11.43 9.28 14.59 8.66 No. times used, past 30 days 14.47 55.17 20.05 64.61 10.88 48.57 37.56 83.85 No. days used, past 30 days 3.93 8.05 5.30 9.28 3.08 6.98 10.02 10.13 No times used, past 30 days (t+1) 7.35 38.60 10.95 46.53 4.56 27.17 15.53 55.47 No. days used, past 30 days (t+1) 2.09 6.00 3.14 7.22 1.49 5.01 4.24 8.13   101  Table 3.3: Strength of instruments  Instrumented Variable Degrees of  F-test statistic F-statistic critical values**  Freedom  5% bias 10% bias 20% bias RAt: Addictive Capital Stock: Crack Cocaine 12 73.77 19.40 10.78 6.22 CAt: Addictive Capital Stock: Powder Cocaine 12 17.83 19.40 10.78 6.22 HAt:  Addictive Capital Stock: Heroin 12 137.11 19.40 10.78 6.22 Rt+1: Crack Cocaine Consumption (t+1) 11 66.33 19.12 10.69 6.23 Ct+1: Powder Cocaine Consumption (t+1) 11 13.73 19.12 10.69 6.23 Ht+1: Heroin Consumption (t+1) 11 60.20 19.12 10.69 6.23 Local Average Treatment Effect* 13 71.73 19.64 10.84 6.21 Estimated from full sample, with δ=0.2. * entered into first-stage models on Rt+1, Ct+1 and Ht+1 ; ** Stock and Yogo (2002); based on regression models with 2 endogenous regressors.     102 Table 3.4: Reduced-form demand equation results: Full sample (N=16,541), estimated via zero-inflated negative binomial regression (NB coefficients provided).    δ=0 δ =0.2 δ =0.4  β  t-value   β t-value β  t-value RA t 0.001 (6.667)*** 0.001 (1.511) 0.004 (6.409)*** R t+1 0.003 (2.934)*** 0.008 (6.019)*** 0.004 (3.217)*** CA t 0.004 (10.718)*** 0.017 (16.307)*** 0.022 (10.287)*** C t+1 0.022 (3.429)*** -0.029 (-4.095)*** 0.020 (3.214)*** HA t 0.000 (1.918)* 0.001 (1.290) 0.003 (2.263)** H t+1 -0.034 (-11.699)*** -0.015 (-4.303)*** -0.034 (-11.692)***  -0.007 (-2.042)** -0.006 (-1.689)* -0.007 (-1.892)*  0.000 (-0.041) -0.001 (-0.371) 0.001 (0.176) CA t 0.001 (2.421)** 0.002 (2.326)** -0.003 (-2.879)*** C t+1 -0.009 (-4.095)*** -0.009 (-4.212)*** -0.011 (-5.346)*** RA t 0.003 (3.852)*** 0.006 (3.105)*** 0.019 (5.286)*** R t+1 0.075 (6.642)*** 0.076 (6.624)*** 0.073 (6.069)*** HA t 0.001 (2.129)** 0.004 (3.270)*** -0.001 (-0.544) H t+1 -0.030 (-4.536)*** -0.032 (-4.847)*** -0.021 (-3.497)***  -0.043 (-6.923)*** -0.044 (-6.956)*** -0.022 (-3.153)***  -0.024 (-5.153)*** -0.023 (-4.981)*** -0.015 (-2.872)*** HA t 0.000 (-1.856)* -0.001 (-2.062)** -0.001 (-1.713)* H t+1 -0.005 (-3.471)*** -0.005 (-3.428)*** -0.005 (-3.394)*** RA t 0.000 (0.919) 0.001 (0.946) 0.002 (0.628) R t+1 0.017 (1.719)* 0.017 (1.741)* 0.017 (1.663)* CA t 0.001 (5.399)*** 0.004 (5.681)*** 0.008 (6.069)*** C t 0.009 (1.860)* 0.007 (1.551) 0.007 (1.579)  -0.001 (-0.452) -0.001 (-0.726) -0.001 (-0.807)  0.002 (0.959) 0.001 (0.585) 0.001 (0.546) εC -5.139     -5.280     -2.965 εR -0.314     -0.329     -0.271 εH 0.075     -0.025     -0.044 εCR 0.010     0.002     0.020 εCH -0.001     0.000     0.001 εRC -0.037     -0.084     -0.064 εRH -0.001     0.000     0.001 εHC -0.051     -0.073     -0.035 εHR 0.003         0.006         0.004    103  Table 3.5:  Reduced form demand equation results: Polydrug users only (N=7,400), estimated via zero-inflated negative binomial regression (NB coefficients provided)    δ =0     δ =0.2    δ =0.4  β t-value β t-value β t-value RA t 0.001 (2.853)*** 0.001 (2.601)*** 0.002 (2.602)*** R t+1 0.007 (3.411)*** 0.007 (3.598)*** 0.007 (3.696)*** CA t 0.003 (8.092)*** 0.012 (7.779)*** 0.021 (7.459)*** C t+1 0.043 (3.099)*** 0.041 (2.957)*** 0.041 (2.902)*** HA t 0.000 (0.045) 0.001 (0.689) 0.002 (0.969) H t+1 -0.046 (-7.571)*** -0.046 (-7.624)*** -0.046 (-7.583)***  -0.014 (-2.703)*** -0.014 (-2.599)*** -0.014 (-2.605)***  -0.001 (-0.268) -0.001 (-0.234) -0.001 (-0.284) CA t 0.001 (4.335)*** 0.002 (4.195)*** 0.004 (4.340)*** C t+1 -0.009 (-5.600)*** -0.009 (-5.610)*** -0.009 (-5.544)*** RA t 0.001 (2.427)** 0.002 (2.053)** 0.004 (2.164)** R t+1 0.053 (9.782)*** 0.052 (9.589)*** 0.052 (9.445)*** HA t 0.001 (4.353)*** 0.004 (5.337)*** 0.008 (5.817)*** H t+1 -0.031 (-7.484)*** -0.032 (-7.783)*** -0.032 (-7.794)***  -0.030 (-5.144)*** -0.031 (-5.218)*** -0.030 (-5.181)***  -0.023 (-4.902)*** -0.022 (-4.681)*** -0.022 (-4.585)*** HA t 0.000 (0.833) 0.001 (1.233) 0.001 (1.604) H t+1 -0.005 (-2.813)*** -0.005 (-3.097)*** -0.005 (-3.114)*** RA t 0.001 (1.945)** 0.001 (1.516) 0.002 (1.373) R t+1 0.006 (0.653) 0.008 (0.930) 0.006 (0.655) CA t 0.001 (5.804)*** 0.004 (6.389)*** 0.009 (7.002)*** C t 0.008 (1.756)* 0.006 (1.296) 0.007 (1.519)  -0.002 (-0.923) -0.003 (-1.248) -0.003 (-1.281)  -0.001 (-0.529) -0.002 (-0.855) -0.002 (-0.910) εC -1.962     -1.962    -1.936 εR -0.649     -0.642    -0.658 εH -0.181     -0.267    -0.279 εCR 0.010     -0.002    0.000 εCH 0.001     0.001    0.001 εRC -0.049     -0.107    -0.099 εRH 0.006     0.009    0.009 εHC -0.008     -0.022    -0.016 εHR 0.004         0.002        0.002   104  Table 3.6:  Reduced-form demand equation results: 3-drug users only (N=1,697), estimated via negative binomial regression    δ = 0 δ = 0.2 δ = 0.4  Β t-value β t-value β t-value RAt 0.001 (5.273)*** 0.003 (5.547)*** 0.005 (5.508)*** Rt+1 0.006 (3.193)*** 0.007 (3.319)*** 0.007 (3.418)*** CAt 0.003 (5.298)*** 0.010 (4.381)*** 0.015 (3.762)*** Ct+1 0.002 (0.241) 0.000 (0.019) 0.000 (-0.033) HAt -0.001 (-2.636)*** -0.002 (-1.796)* -0.003 (-1.258) Ht+1 -0.009 (-1.408) -0.009 (-1.328) -0.009 (-1.390)  -0.009 (-0.989) -0.008 (-0.889) -0.009 (-0.976)  0.002 (0.240) 0.004 (0.474) 0.004 (0.465) CAt 0.001 (2.008)** 0.001 (2.013)** 0.002 (1.945)* Ct+1 -0.002 (-0.937) -0.002 (-0.932) -0.002 (-0.984) RAt 0.003 (3.945)*** 0.007 (3.079)*** 0.010 (2.624)*** Rt+1 0.036 (4.476)*** 0.034 (4.091)*** 0.034 (4.039)*** HAt 0.000 (-1.263) 0.000 (-0.058) 0.002 (0.702) Ht+1 -0.027 (-4.137)*** -0.027 (-4.065)*** -0.027 (-4.157)***  -0.036 (-3.131)*** -0.037 (-3.280)*** -0.038 (-3.344)***  -0.035 (-3.628)*** -0.033 (-3.412)*** -0.032 (-3.288)*** HAt 0.000 (1.474) 0.001 (1.623) 0.003 (3.706)*** Ht+1 -0.005 (-2.733)*** -0.005 (-2.849)*** -0.004 (-2.172)** RAt 0.000 (1.045) 0.003 (1.447) 0.009 (2.544)** Rt+1 0.010 (1.299) 0.008 (0.990) 0.012 (1.666)* CAt 0.000 (1.746)* 0.003 (2.131)** 0.005 (2.061)** Ct+1 0.010 (1.691)* 0.010 (1.720)* 0.006 (1.107)  -0.007 (-2.147)** -0.007 (-2.106)** -0.008 (-2.331)**  -0.001 (-0.365) -0.002 (-0.546) -0.001 (-0.408) εC -1.946     -2.000     -1.973 εR -0.229     -0.137     -0.165 εH -0.383     -0.406     -0.422 εCR 0.001     -0.001     -0.001 εCH 0.003     0.001     -0.002 εRC -0.005     -0.055     -0.043 εRH 0.003     0.007     0.006 εHC -0.015     -0.031     -0.051 εHR 0.001       0.000       -0.001   105  Table 3.7: Reduced-form demand equation results: Age-stratified samples, polydrug users only (N=7,400), estimated via zero-inflated negative binomial regression (NB coefficients provided)    Age > 30 Age < 30   Β        t-statistic Β       t-statistic RAt 0.001 (2.139)** 0.001 (0.959) Rt+1 0.004 (2.040)** 0.021 (3.801)*** CAt 0.013 (8.328)*** 0.004 (0.802) Ct+1 0.044 (2.953)*** 0.036 (0.903) HAt 0.000 (0.131) 0.000 (-0.025) Ht+1 -0.045 (-6.976)*** -0.051 (-3.005)***  -0.012 (-1.979)** -0.030 (-2.484)***  -0.004 (-0.708) 0.002 (0.199) CAt 0.002 (3.770)*** 0.001 (0.892) Ct+1 -0.009 (-5.313)*** -0.008 (-2.090)** RAt 0.001 (0.829) 0.010 (3.541)*** Rt+1 0.056 (9.438)*** 0.028 (1.981)** HAt 0.004 (4.427)*** 0.003 (1.653)* Ht+1 -0.031 (-6.855)*** -0.027 (-2.665)***  -0.030 (-4.719)*** -0.044 (-2.781)***  -0.017 (-3.373)*** -0.070 (-5.623)*** HAt 0.001 (1.150) 0.002 (1.509) Ht+1 -0.005 (-2.662)*** -0.012 (-1.902)* RAt 0.001 (0.583) 0.005 (1.339) Rt+1 0.007 (0.821) -0.023 (-0.809) CAt 0.004 (6.107)*** 0.007 (3.081)*** Ct+1 0.009 (1.698)* 0.013 (0.739)  -0.003 (-1.253) 0.000 (-0.016)  -0.002 (-0.817) 0.000 (-0.008)       106 Figure 3.1: Estimated longitudinal drug price series   107  3.9 References  Arkes J, Liccardo Pacula R, Paddock S, Caulkis JP, Reuter P.  Technical Report for the price and purity of illicit drugs: 1981 through the second quarter of 2003.  Executive office of National Drug Control Policy, Washington D.C, 2004. http://www.rand.org/multi/dprc/projects/drug_price_purity.html Auld MC, Grootendorst P.  An empirical analysis of milk addiction.  Journal of Health Economocs, 2004; 23: 1117-1133.  Ball JC, Ross A.  The effectiveness of methadone maintenance treatment.  Springer- Verlag, New York, 1991.  Becker GS, Murphy KM.  A theory of rational addiction.  The Journal of Political Economy, 1988; 96(4):675-700.  Bickel WK, DeGrandpre RJ, Higgins ST.  Behavioural economics: a novel experimental approach to the study of drug dependence, Drug and Alcohol Dependence, 1993; 33(2): 173-192. Booth RE, Kwiatkowski CF, Stephens RC.  Effectiveness of HIV/AIDS interventions on drug use and needle risk behaviors for out-of-treatment injection drug users. Journal of Psychoactive Drugs, 1998; 30(3):269-278.  108 Cameron AC, Trivedi PK.  Econometric models based on count data: comparisons and applications of some estimators.  Journal of Applied Econometrics, 1986; 1, 29-53. Cameron AC, Trivedi PK. Regression analysis of count data.  Cambridge University Press, New York, NY, 1998. Chaloupka FJ. Rational addictive behaviour and cigarette smoking.  Journal of Political Economy, 1991; 99(4): 722-742. Ciccarone D.  Heroin in brown, black and white: Structural factors and medical consequences in the US heroin market.  International Journal of Drug Policy 2009; 20: 277-282. Caulkins JP, Pacula RL, Arkes J, Reuter P, Paddock S, Iguchi M, Riley J. The Price and Purity of Illicit Drugs: 1981 Through the Second Quarter of 2003.  Executive Office of the President Office of National Drug Control Policy. Washington, D.C., 2004. http://www.rand.org/multi/dprc/projects/drug_price_purity.html Caulkins JP. Price and purity analysis for illicit drug: data and conceptual issues.  Drug and Alcohol Dependence, 2007; 90(1):S61-S68. Caulkins JP.  Domestic geographic variation in illicit drug prices.  Journal of Urban Economics. 1995; 37: 38-56. Caulkins JP, Padman R.  Quantity discounts and quality premia for illicit drugs.  The Journal of the American Statistical Association, 1993; 88: 748-757.  109 Cornish JW, O'Brien CP.  Crack cocaine abuse: an epidemic with many public health consequences.  Annual Review of Public Health, 1996; 17:259-273. Dave D.  The effects of cocaine and heroin price on drug-related emergency department visits.  Journal of Health Economics, 2006; 25(2):311-333. Dowling-Guyer S, Johnson ME, Fisher DG, Booth R, Needle R, Rhodes F, Weatherby N, Watters J, Estrada AL, Andersen M, Fleming D, Williams M, Kotransi L, Deren S, Tortu S.  Reliability of drug users’ self-reported HIV risk behaviours and validity of self- reported recent drug use.  Psychological Assessment, 1994; 1(4): 383-392.  Ferguson BS.  Interpreting the Rational Addiction Model.  Health Economics, 2000; 9:587-598. Galynker II, Watras-Ganz S, Miner C, Rosenthal RN, Des Jarlais DC, Richman BL, London E.  Cerebral metabolism in opiate-dependent subjects: effects of methadone maintenance. Mt. Sinai Journal of Medicine, 2000; 67(5): 381-387. Golub A, Johnson BD.  How much do Manhattan-arrestees spend on drugs?  Drug and Alcohol Dependence, 2004; 76: 235-246. Grossman M, Chaloupka FJ, Shim K.  Illegal drug use and public policy.  Health Affairs, 2002; 21(2):134-145. Houthakker HS, Taylor LD.  Consumer demand in the United States: Analyses and projections. 2nd edition Cambridge Mass: Harvard University Press, 1970.  110 Horowitz JL.  Should the DEA’s STRIDE data be used for economic analyses of markets for illegal drugs?  Journal of the American Statistical Association, 2001; 96(456): 1254- 1262. Iannaccone LR.  Consumption capital and habit formation with an application to religious participation.  Ph.D. dissertation, University of Chicago, 1984.  Imbens GW, Angrist JD.  Identification and estimation of local average treatment effects. Econometrica, 1994; 62(2): 467-475.  Johnson ME, Fisher DG, Reynolds G. Reliability of drug users’ self-report of economic variables.  Addiction Research, 1999; 17(3): 227-238.  Jones AM 2000.  Health econometrics.  In: Handbook of Health Economics (chapter 13), Culyer AJ, Newhouse JP (eds).  Elsevier: Amsterdam, 2000.  Leri F, Stewart J, Tremblay A, Bruneau J.  Heroin and cocaine co-use in a group of injection drug users in Montreal.  Journal of psychiatry & neuroscience, 2004; 29(1): 40- 47. Maxwell JC, Rutkowski BA.  The prevalence of methamphetamine and amphetamine abuse in North America: a review of the indicators, 1992-2007.  Drug & Alcohol Review, 2008; 27(3): 229-235.  111 Mullahy J.  Cigarrette smoking: Habits, health concerns and heterogeneous unobservables in a microeconometric analysis of consumer demand.  Ph.D. dissertation, Univ. Virginia, 1985.   Needle R, Fisher DG, Weatherby N, Chitwood D, Brown B, Cesari H, Booth R, Williams ML, Watters J, Andersen M, Braunstein M.  Reliability of self-reported HIV risk behaviours of drug users.  Psychology of addictive behaviours, 1995; 9(4): 242-250. Nestler EJ.  The neurobiology of cocaine addiction.  Science & practice perspectives, 2005; 3(1):4-10. Orphanides A, Zervos D.  Rational Addiction with learning and regret.  Journal of Political Economy, 1995; 103: 739-758. Oviedo-Joekes E, Nosyk B, Brissette S, Chettiar J, Schneeberger P, Marsh DC, Krausz M, Anis A, Schechter MT.  The North American Opiate Medication Initiative (NAOMI): Profile of Participants in North America's First Trial of Heroin-Assisted Treatment. Journal of Urban Health. 2008; 85(6): 812-825.  Petry NM.  Effects of increasing income on polydrug use: a comparison of heroin, cocaine and alcohol abusers.  Addiction, 2000; 95(5): 705-717. Petry NM, Bickel WK.  Polydrug abuse in heroin addicts: a behavioural economic analysis.  Addiction, 1998; 93(3): 323-335.  112 Rawson RA, Maxwell J, Rutkowski B.  Osycontin Abuse: Who are the users?  American Journal of Psychiatry, 2007; 164(11): 1634-1636. Rounsaville BJ, Petry NM, Carroll KM.  Single versus multiple drug focus in substance abuse clinical trials research.  Drug and Alcohol Dependence, 2003; 70:117-125. Rueter P, Crawford G, Cave J.  Sealing the borders: The effects of increased military participation in drug interdiction, RAND, Santa Monica, 1988. Saffer H, Chaloupka F. The demand for illicit drugs.  Economic Inquiry, 1999; 37(3): 401-411.  Sheu ML, Hu TW, Keeler TE, Ong M, Sung HY.  The effect of a major cigarette price change on smoking behavior in california: a zero-inflated negative binomial model. Health Economics, 2004; 13(8): 781-791.  Spinnewyn F.  Rational habit formation. European Economic Review, 1981; 15: 91-109.  Stigler GJ, Becker GS.  De Gustibus Non Est Disputandum.  American Economic Review, 1977;  67: 76-90. Stock JH, Yogo M.  Testing for weak instruments in linear IV regression.  National Bureau of Economic Research, Technical Working Paper No 284, 2002. http://www.nber.org/papers/t0284.pdf  113 Stock JH, Wright JH, Yogo M.  A survey of weak instruments and weak identification in generalized method of moments.  Journal of Business and Economic Statistics, 2002; 20(4): 518-528 Sumnall HR, Tyler E, Wagstaff GF, Cole JC.  A behavioural economic analysis of alcohol, amphetamine, cocaine and ecstasy purchases by polysubstance misusers.  Drug and Alcohol Dependence, 2004; 76: 93-99.  Substance Abuse and Mental Health Services Administration. Results from the 2001 National Household Survey on Drug Abuse: Summary of National Findings, vol. 1. Office of Applied Studies; Rockville, MD, 2002. http://www.oas.samhsa.gov/NHSDA/2k1NHSDA/vol1/toc.htm  Substance Abuse and Mental Health Services Administration. Emergency Department Trends from the Drug Abuse Warning Network, Final Estimates 1995–2002, DAWN Series: D-24, DHHS Publication No. (SMA) 03-3780. Office of Applied Studies; Rockville, MD, 2003.  Vuong Q. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica, 1989; 57:  307–334.  Van Ours J.  The price elasticity of hard drugs: The case of opium in the Dutch East Indies, 1923-1938.  Journal of Political Economy, 1995; 103(2): 261-279.  114 4 TRENDS IN METHADONE MAINTENANCE TREATMENT PARTICIPATION, RETENTION AND COMPLIANCE TO DOSING GUIDELINES IN BRITISH COLUMBIA, CANADA: 1996-20063  4.1 Introduction  Opioid Dependence is a devastating chronic medical illness (McLellan et al, 2000) associated with comorbid psychiatric disorders (Regier et al, 1990), elevated risk of infection and transmission of infectious diseases such as HIV/AIDS and Hepatitis C (Aceijas et al, 2004) and premature mortality (Caplehorn et al, 1994).  Dependence is operationally defined in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (1994) as a pathologic condition manifested by 3 or more of 7 criteria, including tolerance, withdrawal, and interference with social functioning.  These criteria are all considered for patients wishing to access treatment for opioid dependence.  Standard pharmacotherapy for opioid dependence in British Columbia and throughout Canada involves oral solution methadone, a long-acting synthetic opioid agonist prescribed by licensed physicians, with ingestion witnessed daily by a pharmacist. Buprenorphine and other forms of pharmacological treatment have only recently been indicated for opioid dependence in Canada.  A treatment philosophy emphasizing indefinite maintenance (Methadone maintenance treatment (MMT)) has shown to  3 A version of this chapter has been accepted for publication: Nosyk B, Marsh DC, Sun H, Schechter MT, Anis AH. Trends in methadone Maintenance treatment participation, retention and compliance to dosing guidelines in British Columbia, Canada: 1996-2006. Journal of Substance Abuse Treatment.  115 decrease the risk of relapse and mortality (Caplehorn et al, 1994), and has thus become widely accepted in Canada and elsewhere.  Systematic reviews have identified MMT as the most effective form of treatment for opioid dependence in terms of treatment retention and decreases in the use of illicit opioids (Amato et al, 2005); observational studies have shown decreased risk of overdose death, infectious-disease transmission and criminal activity (Ward, Hall and Mattick, 1999; Hall, Ward and Mattick, 1998).  Clinical practice guidelines are an important tool for integration of evidence- based care into daily practice, yet physician compliance with guidelines is poor (Austin et al, 2008). While some studies have shown improvements in practice consistent with guidelines over the course of 5 to 15 years (Van Dyke et al, 2009; Wirtzfeld et al, 2009), others have suggested substantial impacts on practice may rely on linking practice guidelines to reimbursement (Patrick and Hutchinson, 2009). In British Columbia, MMT practice guidelines are supported by unique levers, including the requirement for a specific authorization to prescribe, mandatory training sponsored by the College of Physicians and Surgeons and a system of peer assessment for compliance with guidelines (Payte, 1995).  Included in the guidelines (released in 1995) are evidence-based strategies to safely and effectively adjust daily doses from treatment initiation (starting dose), titration and maintenance dosing, as well as the provision of carry, or take-home doses. Formal guidelines on dose tapering were added in 2005 for circumstances in which patients chose to voluntarily taper their dose with the goal of abstinence (College of Physicians and Surgeons of British Columbia, 2005); guidelines have remained otherwise unchanged.  116  These aspects of treatment delivery are well-known predictors of treatment outcome (Magura, Nwakeze and Demsky, 1998; Anderson and Warren, 2004). Guidelines on starting doses and dose stabilization are designed to ensure patient safety, while maintenance and dose tapering guidelines were developed to maximize the odds of long- term maintenance, and successful tapering, respectively (Martin, Payte and Zweben, 1991; Health Canada, 2002; National Consensus Development Panel on Effective Medical Treatment for Opiate Addiction, 1998). Conversely, guidelines on take-home (carry) doses were designed in part to ensure patient safety, but also to minimize public health risks through drug diversion (Ruel and Hickey, 1993).  Despite the value in doing so and perhaps as a result of a dearth of centralized pharmacoepidemiological data, few studies have evaluated prescribing practices at the population level.  D’Aunno and Pollack (2002) assessed the maintenance dose prescribed in a nationally representative sample of US methadone treatment programs and found that 35.5% of patients received less than the recommended 60mg daily dose.  Similarly, Strang et al. (2007) observed higher daily MMT doses after national guidelines were issued in the UK, but low overall compliance to daily dosing guidelines.  British Columbia’s centralized provincial drug dispensation database provides a means with which to determine whether compliance with methadone dosing guidelines has improved over time. Evaluating changes in compliance to MMT treatment guidelines can both inform the extent to which the public health benefits of this treatment are being achieved and serve as a model for educational and peer-assessment supports to the  117 implementation of clinical practice guidelines in other areas.  We analyze dispensation records at the population-level to identify methadone dosing patterns, and determine the extent of compliance to clinical guidelines on starting doses, titration, maintenance, carry doses and tapering at the population level in British Columbia from 1996-2006.   118 4.2 Methods 4.2.1Patient Population The British Columbia PharmaNet database records all prescription drug dispensation to residents of British Columbia, Canada. The study cohort included each individual that received at least one dose of Methadone for opioid maintenance treatment over an 11-year period: January 1st, 1996 to December 31st 2006.  Analysis was performed on de-identified data.  As Methadone prescribed for opioid dependence was given a unique drug identification number in the BC PharmaNet database, prescriptions for pain indications were excluded from the analysis. The study was approved by the University of British Columbia / Providence Health Care Behavioural Research Ethics Board. 4.2.2 Data Analysis Compliance to each of the MMT dosing guidelines within each treatment episode was determined through analysis of drug dispensation data within distinct treatment episodes.  The BC College of Physicians and Surgeons methadone dosing guidelines were summarized in table 4.1.  The length of an MMT treatment episode was calculated as the difference between the last and first days of dispensed medication within a period of continuous retention in treatment (no interruptions in prescribed doses > 30 days).  As such, we considered all treatment episodes initiated after January 30th, 1996, eliminating ongoing episodes at the time of data capture. Database errors and misclassification were identified and corrected where possible; less than 1% of all dispensation records could not be  119 corrected.  Within each treatment episode the mean daily dose per week in treatment (weekly dose) was calculated. Periods of titration, maintenance dosing, and tapering were identified by examining changes in weekly dose over the course of each episode. Definitions of treatment phase transitions were presented in greater detail in an optional appendix. Summary statistics by treatment phase were calculated, as well as indicators of compliance to each of the guidelines stated above.  Specifically, compliance to the starting dose guideline was indicated by an initial daily dose below 40mg. Compliance to titration guidelines was indicated by average weekly dose increases ≤ 10mg during the titration phase. Compliance to the minimum effective dose guideline was achieved if the mean weekly dose was ever greater than, or equal to 60mg per day.  Carry guideline compliance was achieved if no carries longer than 4 days were ever prescribed during an episode, and no regular carries (longer than 2 days, not occurring more than once in a week) were prescribed within the first 12 weeks. Finally, dose tapering guideline compliance was indicated by an average rate of decrease in weekly dose no greater than 5%.  Annualized cohorts were created based on calendar year of MMT episode initiation.  The rate of compliance to each of the guidelines was then calculated and plotted over time.  Further, as retention in treatment is commonly used as an outcome measure in substance abuse treatment, we compared patient characteristics, and aspects of treatment  120 delivery among episodes lasting < 12 months, between 12 and 36 months, and greater than 36 months. In each case, univariate comparisons were conducted using two-sided chi-square tests.  Finally, overall episodic dosing patterns were also considered, classifying individual episodes into six different types based on titration, maintenance and tapering phases of treatment.   Examples of each of the dosing pattern types identified in the dataset were plotted, with rates of retention and successful tapering were presented for each episode type.  All data analysis was conducted using SAS version 9.1, while plots were created in [R] version 2.5.1.     121 4.3 Results  Methadone dispensations for opioid dependence totalled 10,916,956 in the 11- year study period. 31,724 MMT episodes remained after removing episodes that were initiated in the last quarter of 2006, left-censored and erroneous cases.  Treatment retention improved from 1996-2001 to a high of 45.9% at 12 months, but decreased to 40.5% by 2005 (table 4.2).  Roughly 9% of all episodes initiated between 1996 and 1999 (873 episodes) were ongoing at the end of follow-up, while 15% of episodes initiated in 2000-2002 (1,608 episodes) were ongoing at the end of data capture.  Summary statistics on compliance to each of the dosing guidelines were presented in table 4.3.  Compliance to the starting dose guideline was 63.6%, and 19.6% for naive treatment episodes.  Excluding episodes lasting one week or less, we found that 67.2% of all episodes featured some dose titration, and 92.4% of these episodes were titrated within the acceptable range.  Titration lasted a median of 5 weeks (Interquartile range: 3, 8). The median length of the maintenance period was 14 weeks (4, 35); 69.2% of patients reached the minimally effective maintenance dose of 60mg per day.  Nearly 60% of all episodes had regular carries (longer than two days, or occurring more than once per week), and 54.2% of all episodes initiated carries after the first 3 months of treatment. Furthermore, 11.5% featured at least one instance of a carry longer than the recommended 4 days.  Finally, 46% of all episodes featured an attempted taper. The  122 median week of taper initiation was the 27th week (16, 53), with 30.4% of all episodes initiating tapers after at least one year in treatment.  While successful dose tapering was uncommon (11.4% of taper episodes; 5.3% overall), guidelines on the rate of decrease in daily doses during tapering was adhered to in 85.1% of all episodes that initiated taper.  Compliance to the starting dose, titration and carry guidelines improved throughout the study period (figure 4.1), while compliance to taper guidelines were relatively stable, but high throughout.  Compliance to carry guidelines showed the least improvement over the study period.  Figure 4.2 displays the trends in compliance to the minimum effective dose guideline and 12-month retention, respectively.  It is evident that from 2001 onward, both population-level indicators declined at a similar rate.  Univariate comparisons of characteristics of patients and treatment delivery were presented in table 4.4.  The mean age of individuals retained in treatment > 36 months was higher than among those retained < 36 months, while treatment adherence (calculated as the percentage of days during a treatment episode in which methadone was dispensed) was also higher.  A larger percentage of long-term treatment episodes had their dose titrated in the initial weeks, patients stabilized at considerably higher doses, were more likely to receive carry doses, and were also more likely to have attempted a taper (All p-values for univariate relationships: p<0.001).   123 Considering each of the phases of treatment, six dosing patterns were identified. Examples of each were plotted in figure 4.3. Over 96% of episodes with variable dosing (types E and F) lasted longer than 12 months.  It is also worth noting that episodes with dose titration (types B, D and F) had better retention outcomes, and were more likely to reach the point where a taper was even considered: 54.5% of all episodes with dose titration featured a taper or variable dosing period, compared to 31.9% of episodes without dose titration reaching this stage.  124 4.4 Discussion  Appropriate methadone dosing ensures patient safety and maximizes physicians’ chances of producing successful treatment outcomes with their patients, other factors notwithstanding.  MMT retention outcomes in British Columbia should be considered encouraging – that 20-25% of episodes lasted longer than three years suggests that MMT can be an effective treatment option.  Treatment outcomes observed in BC were comparable to that of US, UK and New South Wales, Australia (Burns et al, 2009; D’Aunno and Pollack, 2002; Strang et al, 2007), however higher retention figures have been reported elsewhere, including Ontario, Germany and Israel (Strike et al, 2005; Wittchen et al, 2008; Peles, Schreiber and Adelson, 2006).  While positive outcomes were identified, we also found that there was room for improvement in MMT initiation, titration, maintenance dosing and dose tapering practices.  In many cases the guideline compliance rates were increasing over the study period, however the rate of compliance to the maintenance dose guideline decreased between 2001 and 2006.  This was mirrored by a decline in the provincial 12-month retention rate.  Higher daily maintenance doses (specifically, over 60mg per day) have been shown to be independently associated with longer treatment duration in this dataset (Nosyk et al, 2009) and elsewhere (Brady et al, 2005; Faggiano et al, 2003; Maxwell and Shinderman, 2002).  Observation of episodic dosing patterns suggested that compliance to the various dosing guidelines were related, as compliance with titration guidelines was associated with better retention outcomes, and higher rates of reaching the point at which  125 tapering was considered.  One troubling finding was the low rate of compliance to guidelines on starting doses and dose titration. The risk of fatal methadone overdose during the first two weeks of treatment is estimated to be 6.7 times higher than that of heroin-dependent patients not in treatment, and 98 times higher than that of patients on maintenance doses of methadone (Caplehorn and Drummer, 1999).  The mortality rate of patients enrolled in methadone programs was 1.7 per 1000 in Ontario in 2000, with most of the deaths occurring in the first two weeks of treatment (Cairns, 2000).  A high risk of mortality early in treatment is evident in part because of methadone’s long half-life, as methadone blood levels rise for up to five days after starting or raising a dose until equilibrium is reached (College of Physicians and Surgeons of Ontario, 2005).  Furthermore, continued opioid and other drug use is common during transition into treatment.  Evidence of high starting doses is not confined to British Columbia; Dickinson et al reported that 47% of patients received a starting dose > 40mg following publication of National MMT guidelines in a UK clinic (Dickinson et al, 2006).  While MMT dosing guidelines in British Columbia and other Canadian provinces are primarily evidence-based, there are at least two aspects in which either limited information is available, or guidelines are influenced by external factors.  Policies on dose carries were formulated primarily to minimize the public health risks related to methadone diversion (Ruel and Hickey, 1993).  Treatment safety and effectiveness may be compromised among non-stabilized patients, however dose carries may allow  126 stabilized patients greater freedom in attending to family life, employment or other productive endeavours.  While cohort studies conducted in BC and elsewhere suggest that there is, in fact, a market for illicit methadone (Fischer et al, 2005; Davis and Johnson, 2008), the widespread availability and use of other inexpensive prescription opioids that have more desirable pharmacodynamic properties suggests that this concern is perhaps overstated.  In British Columbia, carries are more often prescribed in rural areas, and may be a necessary component in the reintegration of clients into society, as co-payments are required once an individual gains employment.  Several studies conducted in the United States have demonstrated success with ‘methadone medical maintenance’, prescribing long carries (up to 1 month) in stabilized, responsive patients (Harris et al, 2006). Evidence on the public safety issues regarding carry prescriptions should be balanced against evidence which supports its effectiveness.  Furthermore, the evidence base for methadone dose tapering for those wishing to achieve abstinence is relatively small, and suggests that while slower tapers are more effective, most patients either relapse or request to discontinue treatment (Senay et al, 1977; Calsyn, Malcy and Saxon, 2006).  These results were confirmed in this population- level study; even with a liberal definition of successful taper completion, the rate of completion in British Columbia was low – considering all episodes in the study period, only 5% were able to successfully taper off of methadone.  Further study into the determinants of taper initiation and identification of more successful tapering strategies for those wishing to withdraw from treatment are necessary to improve the evidence base for this less-studied aspect of methadone maintenance treatment.  127  While our study provides an accurate depiction of actual treatment delivery, clinical decision-making was not observed, leaving questions on the determinants of guideline unanswered.  In some cases sporadic treatment adherence limits physicians’ abilities to optimize dosing.  Among patients with > 90% adherence over a treatment episode, 68.9% had their dose titrated from the starting dose, while only 58.9% of those with adherence < 70% had their dose titrated (Fisher’s exact test: p < 0.01). Further, anecdotal reports from prescribing physicians suggest that patients may choose to stabilize at lower doses, thus intentionally not fully utilizing the pharmacodynamic properties of methadone.  Patient characteristics in the BC MMT program also changed over the timeframe. Comparing patient cohorts receiving treatment in the 1996 and 2006 calendar years, the 2006 cohort was older (mean age: 40 vs. 37 (1996)), a lower percentage were female (41% vs. 36%), had higher levels of treated comorbidity, according to the pharmacy- based Clark Chronic Disease score, and a lower proportion of patients were in their initial treatment attempt (39.4% vs. 50.5% in 2000).  These indicators point to a more complex population of users being treated in more recent years.  Many of these factors were independently associated with treatment retention in a separate analysis (Nosyk et al, 2009).  Nonetheless, only in the case of starting doses did the provincial dosing guidelines distinguish between patients with different clinical prognoses.  In fact, older, more experienced patients would likely require higher maintenance doses, running contrary to observed trends in compliance with minimum effective dose guidelines.  128  Thus, in the absence of information on physician characteristics and decision- making, we felt analyses comparing rates of guideline compliance between patient or physician populations or identifying aggregate-level determinants of guideline compliance would be better addressed in prospective studies designed specifically to address them.  As our objectives were to simply present population-level rates of compliance to dosing guidelines over time, statistical inference was kept to a minimum. This study should thus be placed in an appropriate context, and used as a basis for discussion on how to improve patient care for methadone clients.  One may question the generalizability of our results to other urban centres nationally and internationally.  Results of a national prospective cohort study suggest injected heroin is far more prevalent among opioid users in Vancouver compared to other major centres (Fischer et al, 2005), and the ominous health and social circumstances within Vancouver’s Downtown Eastside (DTES) have been described lucidly in editorial reports (Jones, 1998), observational (Spittal et al, 2006) and experimental studies (Oviedo-Joekes et al, 2008).  DTES residents accounted for nearly 15% of all treatment episodes during the study period, however their treatment outcomes were only slightly lower than that of the remainder of the treatment population (40.4% retention at 12 months within the DTES, vs. 42.1% elsewhere; p=0.04).  Opioid dependent patients in different locales may differ with respect to the degree and variety of other drugs used, which may necessitate additional psychological or pharmacological treatment, however methadone dosing guidelines are applicable in treating all opioid dependent patients.  129  In-migration of patients on MMT may have biased estimates of compliance to starting dose and titration guidelines.  Studies on migration patterns of injection drug users in Vancouver, British Columbia, suggest that the annual proportion of individuals migrating out of province is small (1.2-5.5% per year) (Rachlis et al, 2008), and may be lower still among stabilized patients in MMT in BC and elsewhere, thus limiting the impact of migration on our compliance figures.  Since our data was drawn from an administrative database, the accuracy and validity of the database may be questioned. We found that out of nearly 11 million individual methadone prescriptions, only 4.3% had errors in the “service date”, “days prescribed” and “quantity” fields, and less than 1% of the records could not be corrected.  Furthermore, we found that the numbers of patients in treatment were in line with figures reported elsewhere (Buxton, 2005).  Quantitative studies assessing trends in compliance to clinical practice guidelines can provide an evidence base for the maximization of public health and welfare. Prescription of appropriate starting doses and responsiveness to patients’ needs in effectively eliminating opioid withdrawal symptoms are critical factors in successfully maintaining opioid dependent patients in methadone treatment.   Physician training, specific authorization and peer-assessment may be important tools for improving physician compliance with clinical practice guidelines.  130 4.5 Acknowledgements  This study was funded by a Canadian Institutes of Health Research Operating Grant: MOP-86476.  Bohdan Nosyk was supported by doctoral research awards from the Michael Smith Foundation for Health Research, the Canadian Institutes of Health Research, and the Research in Addictions and Mental Health Policy & Services CIHR strategic training initiative.    131 Table 4.1: BC MMT dosing guidelines  Stage Recommendations 1.0 Starting Dose - Non-tolerant/opiate naïve: 5-10mg pd - Unknown Tolerance: 15-25 mg pd. - Known Tolerance: 20-40 mg pd. 2.0 Stabilization (Titration) Phase - Stabilization dose: (i) reduces/ eliminates withdrawal symptoms and drug craving, (ii) will not induce sedation or respiratory depression, (iii) blocks euphoric effects of illicit opioids. - Dose adjustments: 5-10mg range, not more frequently than every 3-5 days. 3.1 Maintenance Phase - Most patients will achieve stability on doses between 60- 100mg daily. 3.2 Carry Policy - Recommended that carries not exceed 4 days or 400mg. - Criteria: clinical stability (stable dose, social stability (incl. urine drug screens free of all mood-altering drugs for a min. of 12 week), and ability to store methadone safely. - Reasons for initiation of carry privileges must be documented by physician. 4.0 Tapering Phase - Maximum weekly reduction should be no more than 5% of total dose   132  Table 4.2: MMT treatment retention: 1996-2006      Duration in treatment (days) Retention in treatment (%)  N* median Q1 Q3 6 mo 12 mo 24 mo 36 mo censored 1996 1335 221 55 970 53.6 39.9 29.7 23.7 9.1 1997 2080 215 57 871 53.5 39.4 27.8 21.3 8.9 1998 2619 221 59 792 53.6 39.7 26.3 20.9 9.7 1999 3440 203 52 808 52.2 37.6 26.5 20.3 9.1 2000 3840 258 55 1126 56.0 43.5 31.6 25.4 14.1 2001 3576 298 70 1046 59.0 45.9 32.1 24.3 14.5 2002 3324 281 59 1014 58.3 44.5 31.0 23.2 16.5 2003 2408 273 60 992 57.4 43.9 30.3 22.9 21.3 2004 3123 231 42 762 54.7 41.2 27.7 -- 25.0 2005 3292 218 38 -- 53.1 40.5 -- -- 34.3 2006 3619 -- -- -- 53.5 -- -- -- 61.0 *Number of MMT episodes initiated per calendar year; mo: months; censored: percentage of right-censored cases, ongoing as of December 31st, 2006.   133 Table 4.3: Descriptive statistics of episodic dosing patterns  N (%) Median (IQR) Starting Dose*    Dose ≤ 40mg 20,772 (63.6)    Dose ≤ 25mg (MMT Naïve clients only)** 1,571 (19.6) Titration Phase†‡    Episodes in which titration was evident 19,835 (67.2)    Length of titration period (weeks)  5 (3, 8)    Average change in weekly dose ≤10mg 18,319 (92.4) Maintenance Phase†    Length of maintenance period (weeks)  14 (4, 35)    Maintenance dose (mg) ††  75 (50, 100)    Maintenance dose ≥ 60mg 21,951 (69.2) Dose Carries†    Episodes with regular dose carries 18,990 (59.9)    Episodes with regular dose carries within 3mo. 17,187 (54.2)    Episodes with dose carries ≥ 4 days or 400mg 3,655 (11.5) Tapering Phase†    Episodes in which dose tapering was evident 14,602 (46.0)    Week of dose tapering initiation  27 (16, 53)    Length of taper period (weeks): Uninterrupted tapers§   19 (8, 44)                                    Interrupted tapers  141 (79, 237)    Week of dose tapering initiation ≥ 52 weeks 4,435 (30.4)    Episodes achieving successful dose tapering 1,669 (11.4)    Average decrease in weekly dose ≤ 5% 12,429 (85.1) *All episodes included [N=32,656]; ** First-time treatment initiators included, 2001-2006 [N=8,015]; † Episodes initiated up to 2006Q3 included [N=31,724]; †† Calculated as the maximum weekly dose throughout the treatment episode; ‡ Episodes lasting ≤ 1 week [N=2,208] excluded; § N=8,766 (60.0%) of tapers were classified as uninterrupted.    134  Table 4.4: Patient and treatment delivery characteristics by duration of retention in treatment   Retained <12mo Retained 12- 36mo Retained >36mo P-value* N (%)  13086 (57.9) 4336 (19.3) 5170 (22.9) -- Female [N(%)] 4859 (37.1) 1507 (34.5) 1859 (36.0) 0.006 Age [Mean(SD)] 34.2 (9.9) 34.8 (9.5) 36.8 (9.4) <0.001 Adherence (%) [Mean(SD)] 86.9 (15.2)  92.8 (7.2) 96.0 (4.5) <0.001 Titration [N(%)] 6852 (52.4) 3229 (74.0) 3873 (74.9) <0.001 Maintenance Dose (mg) [Mean(SD)] 68.7 (62.2) 90.6 (51.6) 118.3 (69.3) <0.001 Regular Carries [N(%)] 2430 (18.6) 831 (19.0) 1151 (22.3) <0.001 Taper attempted [N(%)] 3327 (25.4) 3656 (83.7) 4506 (87.2) <0.001 *Analysis of episodes initiated 1996-2004 [N=22,622].  135 Figure 4.1: Compliance to starting dose, titration rate, carry dosing, and tapering rate guidelines, by calendar year     136  Figure 4.2:  Compliance to minimum effective dose guideline vs. 12-month retention rate, by calendar year     137 Figure 4.3: Episodic MMT dosing patterns  Type A: no titration, no taper, no variable dosing; Type B: titration, no taper, no variable dosing; Type C: no titration, taper; Type D: titration, taper; Type E: no titration, variable dosing; Type F: titration, variable dosing.  12mo: 12-month rate of retention; Tsucc: N(%) of successful tapers (final weekly dose ≤ 5mg. 138  4.6 References  Aceijas C, Stimson G, Hickman M, Rhodes T (2004). Global overview of injecting drug use and HIV infection among injecting drug users. AIDS, 18:2295–303.  Amato L, Davoli M, Perucci CA, Ferri M, Faggiano F, Mattick RP (2005).  An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research.  Journal of Substance Abuse Treatment, 28: 321-9.  Anderson JF, Warren LD (2004). Client retention in the British Columbia methadone program, 1996-1999.  Canadian Journal of Public Health, 95(2):104-9.  Austin PC, Tu JV, Ko DT, Alter DA (2008).  Use of evidence-based therapies after discharge among elderly patients with acute myocardial infarction.  Canadian Medical Association Journal, 179(9): 895-900.  Brady TM, Salvucci S, Sverdlov LS, Male A, Kyeyune H, Sikali E, Desale S, Yu P (2005).  Methadone dosage and retention: an examination of the 60mg/day threshold. Journal of Addictive Diseases, 24: 23-47.  Burns L, Randall D, Hall WD, Law M, Butler T, Bell J, Degenhardt L (2009).  Opioid agonist pharmacotherapy in New South Wales from 1985 to 2006: patient characteristics and patterns and predictors of treatment retention.  Addiction, 104: 1363-72. 139   Buxton J (2005).  Vancouver Drug Use Epidemiology Report: Vancouver site report for the Canadian Community Epidemiology Network on Drug Use. www.city.vancouver.bc.ca/fourpillars/pdf/report_vancouver_2005.pdf.  Accessed: September, 2008.  Cairns J (2000).  Methadone-related deaths in Ontario. Ontario Coroner’s Office: Toronto.  Calsyn DA, Malcy JA, Saxon AJ (2006).  Slow tapering from methadone maintenance in a program encouraging indefinite maintenance.  Journal of  Substance Abuse Treatment, 30: 159-63.  Caplehorn JRM, Dalton MSYN, Cluff MC, Petrenas A-M (1994).  Retention in methadone maintenance and heroin addicts’ risk of death.  Addiction, 89: 203-7.  Caplehorn JR, Drummer OH (1999).  Mortality associated with New South Wales methadone programs in 1994: lives lost and saved.  Medical Journal of Australia, 170(3): 104-9.  College of Physicians and Surgeons of British Columbia (2005).  Methadone Maintenance Handbook.  http://www.bcguidelines.ca/gpac/guideline_methadone.html. Physician License Required.  Accessed: June 1st, 2008. 140   College of Physicians and Surgeons of Ontario (2005). Methadone Maintenance Guidelines.  http://www.cpso.on.ca/policies/guidelines/default.aspx?id=1984 (Accessed: June 1st, 2008).  D’Aunno T, Pollack HA (2002).  Changes in Methadone Treatment Practices: Results from a national panel study, 1988-2000.  Journal of the American Medical Association, 288(7): 850-6.  Davis WR, Johnson BD (2008).  Prescription opioid use, misuse, and diversion among street drug users in New York City.  Drug and Alcohol Dependence, 92(1-3): 267-76.  Dickinson GL, Rostami-Hodjegan A, Lagundoye O, Seivewright N, Pratt P, Lennard MS (2006).  A six-year evaluation of methadone prescribing practices at a substance misuse treatment centre in the UK. Journal of Clinical Pharmacy and Therapeutics, 31: 477-84.  Faggiano F, Vigna-Taglianti F, Versino E, Lemma P (2003).  Methadone maintenance at different dosages for opioid dependence (Review).  Cochrane Database of Systematic Reviews, 3: CD002208.  Fischer B, Rehm J, Brissette S, Brochu S, Bruneau J, El-Guebaly N, Noël L, Tyndall M, Wild C, Mun P, Baliunas D (2005).  Illicit opioid use in Canada: comparing social, 141  health, and drug use characteristics of untreated users in five cities (OPICAN study). Journal of Urban Health, 82(2): 250-66.  Hall W, Ward J, Mattick RP (1998).  The effectiveness of methadone maintenance treatment: heroin use and other crime. In: Ward J, Mattick RP, Hall W, eds.  Methadone maintenance treatment and other opioid replacement therapies.  Amsterdam: Harwood Academic: 17-58. Harris KA Jr, Arnsten JH, Joseph H, Hecht J, Marion I, Juliana P, Gourevitch MN (2006). A 5-year evaluation of a methadone medical maintenance program.  Journal of Substance Abuse Treatment, 31(4):433-8. Health Canada (2002).  Best Practices, Methadone Maintenance Treatment.  Prepared by Jamieson, Beals, Lalonde and Associates, Inc. for the Office of Canada’s Drug Strategy, Health Canada.  Jones D (1998).  Vancouver’s “vision of hell” requires special type of MD.  Canadian Medical Association Journal, 159: 169-72.  Magura S, Nwakeze PC, Demsky S (1998).  Pre- and in-treatment predictors of retention in methadone treatment using survival analysis.  Addiction, 93(1): 51-60. Martin J, Payte JT, Zweben JE (1991). Methadone maintenance treatment: a primer for physicians. Journal of Psychoactive Drugs, 23(2):165-76. 142  McLellan AT, Lewis DC, O’Brien CP, Kleber HD (2000).  Drug dependence, a chronic medical illness: Implications for treatment, insurance, and outcomes evaluation.  Journal of the American Medical Association, 284(13): 1689-95.  National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction (1998).  Effective medical treatment of opiate addiction.  Journal of the American Medical Association, 280(22): 1936-43.  Nosyk B, MacNab YC, Sun H, Marsh DC, Fischer B, Schechter MT, Anis AH (2009). Proportional hazards frailty models for recurrent methadone maintenance treatment. American Journal of Epidemiology, 170(6):783-92. Oviedo-Joekes E, Nosyk B, Brissette S, Chettiar J, Schneeberger P, Marsh DC, Krausz M, Anis A, Schechter MT (2008).  The North American Opiate Medication Initiative (NAOMI): Profile of Participants in North America's First Trial of Heroin-Assisted Treatment. Journal of Urban Health, 85(6): 812-25.  Patrick DM, Hutchinson J (2009).  Antibiotic use and population ecology: how you can reduce your "resistance footprint".  Canadian Medical Association Journal, 180(4): 416- 421.  Payte JT (1995).  Opioid Dependence: A practical Guide to Pharmacotherapy. Vancouver, British Columbia. 143   Peles E, Schreiber S, Adelson M (2006).  Factors predicting retention in treatment: 10- year experience of a methadone maintenance treatment (MMT) clinic in Israel.  Drug and Alcohol Dependence, 82:211-7.  Regier DA, Farmer ME, Rae DS, Locke BZ, Keith SJ Judd LL, Goodwin FK (1990). Comorbidity of mental disorders with alcohol and other drug abuse: Results from the Epidemiologic Catchment Area (ECA) study.  Journal of the American Medical Association, 264: 2511-18.  Ruel J-M, Hickey CP (1993). Diversion of Methadone: What are the risks?  BC Medical Journal 35(6):420-1.  Senay EC, Dorus W, Goldberg F, Thornton W (1977).  Withdrawal from methadone maintenance.  Rate of withdrawal and expectation.  Archives of General Psychiatry, 34: 361-7.  Maxwell S, Shinderman MS (2002).  Optimizing long-term response to methadone maintenance treatment: a 152-week follow-up using higher-dose methadone.  Journal of Addictive Diseases, 21(3):1-12.  144  Spittal PM, Hogg RS, Li K, Craib KJ, Recsky M, Johnston C, Montaner JS, Schechter MT, Wood E (2006). Drastic elevations in mortality among female injection drug users in a Canadian setting.  AIDS Care,18(2):101-8.  Strang J, Manning V, Mayet S, Ridge G, Best D, Sheridan J (2007).  Does prescribing for opiate addiction change after national guidelines?  Methadone and buprenorphine prescribing to opiate addicts by general practitioners and hospital doctors in England, 1995-2005.  Addiction, 102; 761-770. Strike CJ, Gnam W, Urbanoski K, Fischer B, Marsh D, Millson M (2005).  Factors predicting 2-year retention in methadone maintenance treatment for opioid dependence. Addictive Behaviours, 30:1025-1028. Van Dyke MK, Phares CR, Lynfield R, Thomas AR, Arnold KE, Craig AS, Mohle- Boetani J, Gershman K, Schaffner W, Petit S, Zansky SM, Morin CA, Spina NL, Wymore K, Harrison LH, Shutt KA, Bareta J, Bulens SN, Zell ER, Schuchat A, Schrag SJ (2009).  Evaluation of universal antenatal screening for group B streptococcus.  New England Journal of Medicine, 360(25):2626-36. Ward J, Hall W, Mattick RP (1999).  Role of maintenance treatment in opioid dependence.  Lancet, 353: 221-26. Wirtzfeld DA, Mikula L, Gryfe R, Ravani P, Dicks EL, Parfrey P, Gallinger S, Pollett WG (2009). Concordance with clinical practice guidelines for adjuvant chemotherapy in 145  patients with stage I-III colon cancer: experience in 2 Canadian provinces.  Canadian Journal of Surgery, 52(2): 92-7. Wittchen HU, Apelt SM, Soyka M, Gastpar M, Backmund M, Gölz J, Kraus MR, Tretter F, Schäfer M, Siegert J, Scherbaum N, Rehm J, Bühringer G (2008). Feasibility and outcome of substitution treatment of heroin-dependent patients in specialized substitution centers and primary care facilities in Germany: a naturalistic study in 2694 patients. Drug and Alcohol Dependence, 95(3):245-57.   146   5 AN EMPIRICAL COMPARISON OF EIGHT HEALTH STATUS MEASURES FOR CHRONIC OPIOID DEPENDENCE 4
  5.1 Introduction  Opioid dependence is a chronic, recurrent condition that has profound effects on individuals’ physical and mental health, and is associated with a number of co-morbid conditions including HIV/AIDS, Hepatitis C, and mental health conditions such as bipolar disorder, anxiety disorders and depression (Mason et al, 1998; Lelutiu- Weinberger et al, 2009; Mathers et al, 2008).  Accurately measuring potential improvements in health in opioid substitution treatment, which may result from reductions in illicit drug use and subsequent mediation of withdrawal symptoms, improvements in mental health status and treatment of co-morbid conditions are important for evaluation of new health technologies and resource allocation decisions.  A number of different disease-specific and generic measures of health status have been used in studies on the treatment of opioid dependence.  Among the most widely- used measures is the Addiction Severity Index (ASI), which features composite scores summarizing mental and physical health status, among other domain scores.  The  4 A version of this chapter has been accepted for publication.  Nosyk B, Sun H, Guh D, Oviedo-Joekes E, Marsh DC, Brissette S, Schechter MT, Anis AH.  An empirical comparison of eight health status measures for chronic opioid dependence.  Journal of Clinical Epidemiology.  147  instrument was developed in 1984 and continues to be used today (McLellan et al, 2006), however additional instruments such as the Maudesley Addiction Profile, Euroqol EQ-5D (Puigdollers et al, 2004) among others may also be employed.  While there are obvious differences between these measures, each has been used to describe and discriminate between individuals with different levels of health status as well as facilitate evaluation of change in health status over time – that is, they were meant to be used as both discriminative and evaluative measures (Kirschner and Guyatt, 1985).  A discriminative measure is used to distinguish between individuals or groups on an underlying dimension when no external criterion or gold standard exists, while an evaluative measure can determine the effect of treatment on individuals’ health.  The commonality in the objectives of these measures facilitates a common approach to assessing their quality in use with populations of opioid-dependent patients.  To inform the decision of which measure to employ and report, agreement on what constitutes ‘quality’ is a requisite initial step.  Though a wide variety of methods covering a number of measurement characteristics have, and continue to be used, there exists no consensus or formal guideline for determining quality in health outcome measurement.  Terwee et al (2007) have proposed working criteria for the assessment of the quality of measures of health status, which includes establishing content validity and internal consistency, identifying floor or ceiling effects, criterion validity, and importantly, responsiveness, or the ability of an instrument to measure clinically important change (Terwee et al, 2007).  Responsiveness is a measurement characteristic that has been largely overlooked in studies assessing the validity of health status 148  measures in substance abuse; a recent review of studies evaluating the validity of the ASI, for instance, indicated that evidence of responsiveness of this instrument was absent (Makela, 2004).  More responsive measures will provide a means of more accurately tracking changes in health status, and indicate to providers what aspects of individuals’ health status can be meaningfully improved through treatment.  The North American Opiate Medication Initiative (NAOMI) randomized controlled trial employed a collection of eight measures of health status.  Our objective was to provide a comparative empirical analysis of the psychometric properties of each these measures to determine their quality in populations of chronic opioid dependent patients re-entering treatment.    149  5.2 Methods 5.2.1 Population  The North American Opiate Medication Initiative (NAOMI) is North America’s first randomized controlled trial comparing heroin assisted treatment (HAT) to methadone maintenance (MMT) as a second-line treatment of chronic, treatment- refractory heroin-dependant patients. Long-term injection opioid addicts with at least two prior attempts at MMT were recruited from Vancouver, British Columbia, Canada, and Montreal, Quebec, Canada, and randomly allocated to HAT or MMT.  Paper-based questionnaires were administered by trained research staff through face-to-face interviews at baseline and quarterly intervals until trial completion. The study recruited 251 subjects and had a 95.6% follow-up rate on an ‘intent to treat’ basis up to the 12- month trial endpoint. Details of trial methodology (Oviedo-Joekes et al, 2009a), baseline patient characteristics (Oviedo-Joekes et al, 2008) and results (Oviedo-Joekes et al, 2009b) were described at length elsewhere. 5.2.2 Analysis  Our analysis aims to describe and compare the psychometric properties of each of the health status measures described in table 5.1 on five key indicators of quality, including content validity, evidence of floor and ceiling effects, internal consistency, criterion validity and responsiveness.  Scores of ‘+’ or ‘-’ were assigned to index scores based on performance within each measurement category, according to pre-defined standards described below. 150    5.2.2.1 Content Validity: Content validity examines the extent to which the concepts of interest for a given population are comprehensively represented by the items in the questionnaire (Guyatt, Kirschner and Jaeschke, 1992).  This includes the conceptual focus of the items (i.e. in terms of symptoms, functioning, general health perceptions or overall quality of life), whether or not the items included in the scale sufficiently describe the given concept, and the overall interpretability and applicability of the items. Positive ratings were given if the conceptual focus of the items was relevant, as indicated by moderate to low maximum-valued (ceiling) responses on individual items, and all items were clear and interpretable to respondents.  5.2.2.2 Floor and Ceiling Effects: Floor or ceiling effects on a summary score can be considered present if at least 15% of respondents achieved the lowest or highest score, respectively.  Large floor or ceiling effects on summary scores may suggest that extreme items on either end of the scale may be missing, thus limiting content validity (Terwee et al, 2007).  We considered floor and ceiling effects on summary scores at baseline assessment, and assigned positive ratings on this characteristic if fewer than 15% of all scores were valued at the floor and ceiling.  5.2.2.3 Internal Consistency: Internal consistency measures the extent to which items that comprise a summary or index score correlate, thus measuring the same construct. We conducted exploratory factor analysis (EFA) and confirmatory factor analyses (CFA) to 151  assess sufficient unidimensionality (Lai, Crane and Cella, 2006).  In EFA, the number of factors selected was based on results of Velicer’s minimum average partial (MAP) test, and parallel analysis (O’Connor, 2000).  In CFA, single-factor models were fitted, and fit statistics were assessed to determine whether a the single-factor (unidimensional) model was appropriate (Confirmatory fit index (CFI) > 0.90, root mean square error (RMSEA) < 0.08).  Typical cross-validation, involving random selection of observations for EFA and CFA from the same dataset was infeasible due to relatively small sample sizes.  CFA was thus conducted on 12-month follow-up data, while EFA was conducted on baseline data. Following factor analysis, we calculated standardized Cronbach coefficient alpha values (αc) for each identified factor within ASI Medical and Psychiatric composite scores, as well as the MAP-PHS and MAP-MHS. As internal consistency is important for summary scores that attempt to measure a single underlying construct using multiple items (multi-item scales), we do not present αc values for the WHODAS-II, EQ-5D, EQ- VAS or SF-6D, which were designed as multi-domain measures.  Single-domain index scores that were deemed sufficiently unidimensional and had αc values from 0.70-0.95 were given positive ratings, while those showing evidence of more than one interpretable factor were given negative ratings.  5.2.2.4 Construct Validity: Construct validity refers to the extent to which scores on a particular instrument relate to other measures in a manner that is consistent with theoretically-derived hypotheses concerning the concepts being measured (Kirschner and Guyatt, 1985; Streiner and Nornam, 2003).  We hypothesized that chronic medical conditions would impact both physical and mental health negatively (Verhaak et al, 152  2005).  To test this hypothesis, we calculated the area under the receiver operating characteristic curve (AUC) to determine whether summary scores discriminated between patients with and without chronic medical conditions.  The outcome was self-reported, dichotomous, and described as ‘chronic medical conditions that interfered with patients’ lives’ at baseline assessment.  Receiver operating curves plot a function of the true positive rate (sensitivity) vs. the false positive rate (1-specificity) -- in this application, the ability of health status scores to discriminate between those with and without a chronic condition. The values retrieved from this calculation, along with their confidence intervals, range from 0.5 to 1, where perfect discrimination is indicated by an AUC of 1, and an AUC score of 0.5 reflects discrimination no better than chance.  The 95% credibility interval was constructed by nonparametric bootstrapping (1000 iterations) (Gonen et al, 2007).  We considered index scores for which we were 95% certain that AUCs were above 0.5 to have a positive rating on construct validity.  5.2.2.5 Responsiveness:  Responsiveness can be defined as the ability of a measure to detect clinically important changes over time (Guyatt et al, 1989).  We consider responsiveness to be one aspect of the validity of an evaluative health outcome measure; however the data requirements and methods used are distinct from those used to assess other forms of validity (Hays and Hadorn, 1992; Krabbe et al, 2004).  An external criterion of change that instruments should theoretically respond to must be chosen a priori.  An appropriate criterion may be a clinical endpoint, patient-rated global improvement, change in other patient-reported outcome measure, or some combination of clinical and patient-based outcomes (Revicki et al, 2008).  Self-reported decreases in 153  illicit drug use are commonly used as clinical endpoints in substance abuse treatment studies (Amato et al, 2005).  We hypothesize that reductions in illicit drug use, indicative of stabilization and compliance to treatment, would impact health positively through mediation of withdrawal symptoms, improvements in mental health status and access to treatment of co-morbid conditions.  We chose a decrease of 20% in the ASI illicit drug use composite score at the 12- month trial endpoint, from baseline, as our external criterion of change.  The illicit drug use composite score (version 2) combines the days of use of a comprehensive list of substances, with added weight on drug injection and multiple drug use in the scoring algorithm (Kokkevi and Hartgers, 1995). Using this criterion, we constructed ‘event’ and ‘control’ cohorts of individuals decreasing, and not decreasing illicit drug use by the above stated threshold, respectively.  We identified medical events (N=34), psychiatric events (N=18) and changes in housing status (N=54 events) as time-dependent confounders that may alter changes in items/indices.  In order to isolate the effect of the decrease in illicit drug use on health status, individuals experiencing these events at baseline or within 30 days of the 12-month follow-up assessment were excluded from either cohort.  To determine the extent to which a change in a measure relates to corresponding change in an external criterion (Husted et al, 2000) we again employ the AUC technique, this time to compare changes in health status in the event and control cohorts.  AUCs were calculated for the index scores, as well as the individual items/domains to more 154  specifically identify which items were most affected by decreases in illicit drug use. Once again, we considered index scores, and items, for which we were 95% certain that AUCs were above 0.5 to have a positive rating on responsiveness.   All statistical analysis was conducted using SAS version 9.1. 155  5.3 Results 5.3.1 Content Validity  A comprehensive listing of items, index scores and their summary statistics were presented in table 5.2, including the frequency of responses at the floor (minimum) and ceiling (maximum) values, while table 5.3 provides a conceptual breakdown of the content of each measure.  At baseline, 40-60% of patients reported ceiling valued- responses for items capturing general health perceptions (ASImed, ASIpsych).  As a result, this concept may be less useful in measuring the health status of patients re- entering opioid substitution treatment.  Results on mental, physical and social functioning (WHODAS-12, EQ-5D, SF-6D) were mixed; items classified in the latter tended to have high levels of ceiling-value responses at baseline, while those capturing the former concepts performed somewhat better.  Adverse physical and mental health symptoms, on the other hand, were commonly reported (less than 25% of patients provided ceiling responses for 10 of 25 items).  Furthermore, the four generic measures of health status (WHODAS-12, EQ-VAS, EQ-5D and SF-6D) each had low levels of ceiling responses on items, suggesting that overall quality of life is a useful concept to measure in this patient population.  Thus, the conceptual coverage of the MAP physical and mental health scores may be preferred to that of ASImed and ASIpsych; however neither considers overall quality of life, best captured by the EQ-5D and EQ-VAS.  Complete responses were obtained from respondents at baseline to most instruments, however it should be noted that 40 patients (16%) could not provide a response to item W12 (trouble in day-to-day work) at baseline. 156  5.3.2 Floor or Ceiling Effects  Both the ASI medical and psychiatric composite scores of the items feature large ceiling effects (39.4% and 37.5% respectively), which may affect their ability to capture improvements in health among those at the ceiling level (table 5.2).  The MAP-PHS index score had no ceiling or floor-valued responses, while ceiling and floor effects on the MAP-MHS index score were negligible. Furthermore, the four generic measures of health status (WHODAS-12, EQ-VAS, EQ-5D and SF-6D) each had low or negligible ceiling effects. 5.3.3 Internal Consistency  Results of exploratory and confirmatory factor analyses on the ASI psychiatric score, as well as the MAP-PHS and MHS scores were presented in table 5.4. All data were treated as ordinal, and promax rotation was used to generate rotated factor loadings, which takes into account the correlation between derived factors.  While EFA of ASImed suggested a single factor, CFA was infeasible given the low number of items, making judgement on the unidimensionality of the measure inconclusive.   EFA results suggested ASIpsych was unidimensional: factor loadings were generally small on the second factor, and there were high cross-loadings.  MAP and parallel analysis tests supported this finding.  The CFI and RMSEA generated by the single-factor CFA suggested sufficient unidimensionality, though RMSEA values were slightly higher than the identified threshold.  ASIpsych had an acceptable level of internal consistency (αc=0.83).  While Velicer’s MAP test and parallel analysis also suggested a single factor for the MAP-PHS, rotated factor loadings suggested two distinct factors: gastrointestinal/cardiovascular 157  (MP3-6, MP10) and pain/sensation (MP7-9).  Poor appetite (MP1) and fatigue (MP2) did not load highly on either factor.  Cross loadings were minimal overall, and Cronbach’s alpha values were high on each factor (αc=0.75 and αc=0.78 respectively).  Results of the CFA did not support a single-factor model.  Finally, the MAP-MHS loaded on two distinct factors: Anxiety (MP1-5) and Depression (MP6-10); this was supported by Velicer’s MAP test and parallel analysis.  The single factor model showed poor fit, further supporting results of the EFA.  Cronbach’s alpha values were high on each factor (αc=0.84 and αc=0.89 respectively). 5.3.4 Construct Validity  Results of our assessment of construct validity can be found in table 5.5.  The ASI medical composite score was able to distinguish between those with and without the chosen gold standard criterion measure at baseline (AUC=0.67; 95% C.I.: (0.60, 0.73)), as was the psychiatric composite score (0.60 (0.53, 0.67)).  The MAP-PHS index score discriminated well between those with and without chronic medical conditions (0.62 (0.55, 0.69)),  however those with chronic conditions reported similar levels of the symptoms contained in the MAP-MHS, thus, the index score did not discriminate on the criterion.  The WHODAS-12 index score was able to discriminate between those with and without chronic medical conditions (0.58 (0.50, 0.67)).  Both the EQ-5D index score and visual analog scale were able to discriminate between those with and without chronic medical conditions.  The SF-6D index score was able to discriminate between those with and without chronic medical conditions (0.58 (0.52, 0.65)). 158  5.3.5 Responsiveness  Table 5.6 presented results on the responsiveness of items and scores to decreases in illicit drug use from baseline, at the 12-month follow-up period (N=240 respondents). ASImed was responsive to decreases in illicit drug use, with an AUC of (0.60 (0.50, 0.71)).   While ASIpsych was not responsive to decreases in illicit drug use (0.55 (0.47, 0.65)), items on depression, anxiety/tension, thoughts of suicide and suicide attempts were lower among those decreasing illicit drug use, and had AUCs significantly greater than 0.5.  Finally, the MAP-PHS index score was found to be responsive to decreases in illicit drug use as well as its items measuring tiredness/fatigue, chest pains and numbness/tingling. The MAP-MHS index score discriminated significantly between those decreasing illicit drug use and not (0.63 (0.50, 0.75)), as did six of ten items: spells of terror or panic (MM5), feeling hopeless about the future (MM6), feelings of worthlessness (MM7), feeling no interest in things (MM8), feeling lonely (MM9), and suicide ideation (MM10). The WHODAS-12 index score and all but one of its item were not responsive to decreases in illicit drug use.  AUCs for the EQ-5D index score and all but the anxiety/depression item (E5) featured credibility intervals that did not cross 0.5. Changes in SF-6D scores were similar in the patient cohorts (AUC: 0.54 (0.49, 0.62)), however items measuring social functioning (S3) and vitality (S6) were responsive to decreases in illicit drug use. 5.3.6 Overall Ratings  Table 5.7 summarizes the results of this evaluation for each measure across domains.  ASImed and WHODAS-12 received negative ratings as a result of low 159  conceptual relevance and poor item interpretability, respectively.  ASImed and ASIpsych had large ceiling effects, earning negative ratings on this domain.  MAP-PHS and MAP- MHS were not unidimensional, thus showing poor internal consistency.  MAP-MHS could not distinguish between those with and without chronic medical conditions, evidence of poor construct validity, while ASIpsych, WHODAS-12 and SF-6D were not responsive to decreases in illicit drug use. 160  5.4 Discussion  The objective of our study was to assess the quality of eight different discriminative and evaluative measures of health status based on several key measurement characteristics, including content validity, internal consistency, criterion validity and responsiveness.   Our findings suggest that the WHODAS-12 may be less favourable than the other measures based on results of poor content validity and responsiveness. MAP-MHS was less favourable as a discriminative measure than the other indices; however the choice between the other measures depends largely on the desired conceptual coverage of health status measures.   Ultimately, the question of which concepts are of primary importance and interest is left to research investigators or treatment program leaders, however it is logical to focus on concepts capturing aspects of health that (i) are of importance to patients at the time of entry into treatment, and (ii) can realistically be affected, or improved upon, through substance abuse treatment.  Based on these criteria, we can judge the usefulness of items from each of the six distinct concepts identified in table 5.3.  We concluded that the concepts capturing general health perceptions were less relevant than health functioning, health symptoms and overall health-related quality of life.   Analysis on the responsiveness of these items showed that only one of six items capturing general health perceptions were responsive to decreases in illicit drug use.  Concepts capturing health symptoms and overall quality of life were responsive to the external criterion suggesting the MAP-PHS, MAP-MHS, EQ- 5D and EQ-VAS were good evaluative measures in this population.  It is common for 161  disease-specific health status measures to be more responsive than generic measures, however in this study it appears that poor conceptual coverage of ASIpsych reversed this relationship.  Among the disease-specific measures, the MAP-PHS, was preferred to ASImed in capturing physical health status. ASIpsych was a better discriminative measure of mental health status, but MAP-MHS was a better evaluative measure. Though the Maudesley Addiciton Profile measures were not unidimensional, their conceptual focus may be more relevant than that of the ASImed and ASIpsych for populations of opioid-dependent populations, and were better evaluative measures of health status.  It should be clear that there was no single instrument that performed uniformly as “best” or “worst”.  Omissions in clearly important concepts were identified in each of the measures.  Specifically, ‘vitality’ was captured in only the SF-6D, and had low levels of ceiling responses.  Further, comparisons of absolute summary scores were complicated by differences in conceptual focus and the weight assigned to the various items.  For instance, the MAP summary scores assigned equal weight to differences in the persistence of diverse symptoms such as ‘feeling tense’ (MM1) and thoughts of ending your life’ (MM10), and ‘poor appetite’ (MP1) and chest pains (MP6).  On the other hand, the ASI composite scores focused more on general health perceptions, and the generic measures on health functioning.  Responses to items in the indirect utility measures (EQ- 5D and SF-6D) were weighted by societal preferences – these are contrasted with the individual preferences implicitly expressed in the EQ-VAS.  Finally, the recall periods of the instruments differed, with the EQ-5D and SF-6D assessing health ‘today’ and each of 162  the others focusing on the past 30 days.  Each of these factors suggests further evaluation of existing instruments, and potentially development of disease-specific measures of health status of substance abusers is required as treatment options continue to be developed.  While our assessment focused on a disparate collection of health status measures, some of which defy direct comparison, the EQ-5D and SF-6D are both generic, preference-weighted, indirect measures of ‘health utility’, designed to calculate quality adjusted life years (QALYs) for use in cost-utility analyses (CUA).   We found that in this patient population, the EQ-5D outperformed the SF-6D in terms of its responsiveness to decreases in illicit drug use. The EQ-VAS was also responsive and had good construct validity; however it is not suitable for CUA, as it does not reflect societal health state preferences.  The simpler descriptive system of the EQ-5D may have produced responses with less random variability.  The brevity of the instrument has been considered both strength (Brooks, Robin and de Charro, 2003) and weakness (Conner-Spady and Suarez- Almazor, 2004); higher completion rates and potentially more reliable responses can be offset by potential omission of critical conceptual coverage, distributional gaps and poorer reliability, discriminative ability and responsiveness in some populations (Hollingsworth et al, 1995; Essink-Bot et al, 1997; Jenkinson, Stradling and Petersen, 1998; Longworth and Bryan, 2003; Marra et al, 2005).  While the SF-6D’s social functioning and vitality items (neither of which are captured in the EQ-5D) were responsive to improvements in illicit drug use, the 163  preference-weighted index score was not; these items were clearly important to this patient population, however the preference weights attributed to them were relatively small, and thus had a minimal affect on the preference-weighted index score. In contrast, both measures included an item measuring ‘pain’ (E4 and S4), however only the EQ- 5D’s rather simple, 3-level scale measuring pain was responsive to decreases in illicit drug use, while the SF-6D’s more complex 6-level scale was not. This difference is demonstrative of the advantages of brevity and ease of administration, realized by use of the EQ-5D, regardless of conceptual coverage and assigned preference weights.  The empirical evidence from this study suggests the EQ-5D is preferable to the SF-6D in populations of opioid-dependent patients.  It should be noted that patients selected for the NAOMI trial were older than the British Columbia population average of opioid-dependent patients entering treatment (Oviedo-Joekes et al, 2008; Nosyk et al, 2009), and had high levels of co-morbidity, including HIV, Hepatitis C and self-reported psychological problems.  The poor health status of patients at baseline allowed for improvements to be realized.  Our results suggest that improvements in self-reported health status can be achieved through opioid substitution treatment, and that this should be considered an endpoint in evaluations of treatment for opioids and other substances of abuse.  Our study is not without limitations.  First, as discussed above, our results may not be generalizable to populations of opioid dependent patients with lower disease severity. Second, the measurement properties assessed in this paper did not satisfy the 164  comprehensive list of quality characteristics according to those set out by Terwee et al (2006).  Specifically, we have not considered each of the measures` reliability and concurrent validity.  Furthermore, by only assessing change in health status around a criterion that was hypothesized to improve health status, we have provided only limited evidence of responsiveness.  The frequencies of medical and psychiatric events and changes in housing status, occurring independently from improvements in illicit drug use, were too low to accurately measure responsiveness using our methodological framework. Relapse into frequent illicit drug use (using a symmetric 20% increase in drug use) from baseline (or any time point), was also not frequently observed: most patients either remained in treatment, lowering their illicit drug use, or dropped out prior to their first follow-up HRQoL assessment.  While the chosen criterion cannot produce evidence that the instruments can detect small changes in the constructs being measured, and cannot simultaneously determine responsiveness to both improvement and deterioration in health (evidence which may be produced using an anchor-based approach), they nonetheless allow us to determine whether the measures can discriminate between patients experiencing and not experiencing an objective and clinically meaningful outcome; a characteristic that all HRQoL measures used within populations of opioid dependent patients should possess. Further, discrimination better than chance (AUC statistically significantly greater than 0.5) was chosen as the criteria with which we judged evidence of construct validity and responsiveness.  While the AUC results presented fall below thresholds considered ‘good’ responsiveness by Terwee et al (2006) among others, we nonetheless feel our criteria is objective and defensible on theoretical and statistical grounds. 165  Finally, our study has by no means assessed a comprehensive set of instruments used to measure health status in opioid-dependent patients.  Comprehensive comparative evaluations should be conducted to identify the best available health status measures or inform development of new measures for use in populations of opioid-dependent patients. 5.4.1 Conclusions  We have provided an empirical comparison of five key measurement characteristics of eight generic and disease-specific measures of mental and physical health status used in a randomized controlled trial of chronic, treatment-refractory opioid addicts.  None of the instruments assessed performed uniformly as ‘best’ or ‘worst’.  On the basis of the results of this study, the EQ-5D appeared to be the preferable generic, indirect utility measure for use in the study population.  Though the Maudesley Addiciton Profile indices were not unidimensional, their conceptual focus may be more relevant than that of the ASImed and ASIpsych for populations of opioid-dependent populations, and were better evaluative measures of health status.  Our results provide an evidence base to inform selection and development of health status measures in populations of opioid-dependent patients.  166  5.5 Acknowledgements  The NAOMI study is funded by the Canadian Institutes of Health Research (CIHR). Bohdan Nosyk was supported by doctoral research awards from the Michael Smith Foundation for Health Research, the Canadian Institutes of Health Research, and the Research in Addictions and Mental Health Policy & Services CIHR strategic training initiative.  Dr. Schechter is a tier I Canada Research Chair.  We thank two anonymous reviewers for their constructive feedback, which improved this manuscript substantially. 167  Table 5.1: Description of health status measures  Measure [Type] No. items Measurement scale of items, index scores  Recall period (days) Existing evidence of validity in opioid dependent populations ASImed [4,12] [Disease-specific]  3 Items: M1: 0-30; M2,M3: 0-4 Composite score: summed from equally-weighted responses to M1- M3.  Range: 0 – 1, from best to worst possible health state. 30 ASIpsych [4,12] [Disease-specific] 11 Items: P1-P8: 0/1; P9: 0-30; P10,P11: 0-4 Composite score: summed from equally-weighted responses for P1- P11.  Range: 0 – 1, from best to worst possible health state. 30 Evidence on the validity of the ASI medical and psychiatric composite scores mixed [8]. Discriminative validity: provide effective initial screening for patients with impaired functional status as measured by the SF-36 component summary scores [13] MAP-PHS [14] [Disease-specific] 10 Items: MP1-MP10: 0-4 Index score: Summed responses from MP1-10.  Range: 0-40, from best to worst possible health state. 30 MAP-MHS [Disease-specific] 10 Items: MM1-MM10: 0-4. Index score: Summed responses from MM1-10.  Range: 0-40, from best to worst possible health state. 30 Concurrent validity: correlations with ASI medical, psychiatric items on days of medical/ psychological problems in the past 30 averaged ρ=0.72; perceived as evidence of strong concurrent validity [15]. WHODAS-12 [Generic] [16-20] 12 Items: W1-W12: 0-4. Index score: Summed responses from W1-12.  Range: 0-48, from best to worst possible health state. 30 None. EQ-5D [21-23] [Generic] 5 Items: E1-E5: 1-3.  Index score: Empirically-derived societal preference weights generated from a representative general population sample using the time trade-off valuation method [24].  Range: -0.11 – 1.0, from health state worse than death, to perfect health. Today EQ-VAS [Generic] 1 EQ-VAS score: 0-100; worst- to best-imaginable health state. Today Concurrent validity: EQ-5D items correlated well with selected items from the SCL- 90, MAP-HSS and EuropASI [25]. EQ-VAS was not assessed in this study. SF-6D [26-28] [Generic] 6 Items: S1,S4: 0-5; S2: 0-3; S3,S5,S6: 0-4. Index Score: Empirically-derived societal preference weights generated from a representative general population sample using the standard gamble valuation method [26].  Range: -0.3 – 1.0, from health state worse than death, to perfect health. Today None.    168   Table 5.2: Summary statistics of index scores and items       Floor Value Ceiling Value Measure Mean (SD) N (%) N (%) ASI Medical Composite Score 0.37 (0.35) 7 (2.8)  99 (39.4)*   M1: No. days experienced med. problems 8.92 (11.52) 45 (17.9) 113 (45.0)   M2: How troubled/bothered by med. problems 1.55 (1.52) 30 (12.0) 107 (42.6)   M3: Importance of treatment for med. problems 1.69 (1.74) 65 (25.9) 117 (46.6) ASI Psychiatric Composite Score:  0.20 (0.20) 0 (0.0) 94 (37.5)*   P1: Serious depression 0.23 (0.42) 57 (22.7) 194 (77.3)   P2: Serious anxiety/tension 0.37 (0.48) 92 (36.7) 159 (63.4)   P3: Trouble understanding/concentration 0.34 (0.48) 86 (34.3) 165 (65.7)   P4: Hallucinations 0.02 (0.14) 5 (2.0) 246 (98.0)   P5: Trouble controlling violent behaviour 0.08 (0.28) 21 (8.4) 230 (91.6)   P6: Took prescribed psychiatric medication  0.12 (0.33) 30 (12.0) 221 (88.1)   P7: Serious thoughts of suicide 0.12 (0.32) 29 (11.6) 222 (88.5)   P8: Attempted suicide 0.02 (0.13) 4 (1.6) 247 (98.4)   P9: No days troubled/bothered by psych. problems 8.00 (11.49) 45 (17.9) 102 (40.6)   P10: How troubled or bothered by psych. problems 1.34 (1.49) 28 (11.2) 123 (49.0)   P11: Importance of treatment for psych. problems 1.09 (1.48) 27 (10.8) 148 (59.0) MAP Physical Health Score 15.22 (7.17) 0 (0.0) 0 (0.0)   MP1. Persistence of Poor Appetite 1.96 (1.25) 28 (11.2) 41 (16.4)   MP2. Persistence of Tiredness/Fatigue 2.60 (0.89) 33 (13.2) 4 (1.6)   MP3. Persistence of Nausea (feeling sick) 1.44 (1.11) 11 (4.4) 58 (23.2)   MP4. Persistence of Stomach Pains 1.38 (1.21) 14 (5.6) 79 (31.6)   MP5. Persistence of Difficulty breathing 1.04 (1.21) 11 (4.4) 118 (47.2)   MP6. Persistence of Chest pains 0.78 (1.02) 3 (1.2) 140 (56.0)   MP7. Persistence of Joint/bone pains 2.07 (1.33) 45 (18.0) 42 (16.8)   MP8. Persistence of Muscle pains 1.80 (1.24) 24 (9.6) 51 (20.4)   MP9. Persistence of Numbness/tingling 1.40 (1.33) 24 (9.6) 91 (36.4)   MP10. Persistence of Tremors/shakes 0.76 (1.06) 5 (2.0) 142 (56.8) MAP Mental Health Score 14.27 (8.40) 0 (0.0) 11 (4.4)   MM1. Persistence of Feeling tense 1.92 (1.11) 19 (7.6) 31 (12.4)   MM2. Persistence of Suddenly scared for no reason 0.96 (1.06) 5 (2.0) 114 (45.6)   MM3. Persistence of Feeling fearful 1.06 (1.10) 6 (2.4) 103 (41.2)   MM4. Persistence of Nervousness or shakiness  1.42 (1.13) 7 (2.8) 67 (26.8)   MM5. Persistence of Spells of terror or panic 0.77 (1.03) 4 (1.6) 136 (54.4)   MM6. Persistence of Feeling hopeless about future 1.82 (1.28) 24 (9.6) 54 (21.6)   MM7. Persistence of Feelings of worthlessness 1.65 (1.34) 23 (9.2) 74 (29.6)   MM8. Persistence of Feeling no interest in things 1.83 (1.22) 18 (7.2) 50 (20.0)   MM9. Persistence of Feeling lonely 2.16 (1.28) 40 (16.0) 40 (16.0)   MM10. Persistence of Thoughts of ending your life 0.68 (1.05) 6 (2.4) 155 (62.0) WHO Disability Assessment Schedule-12* 23.55 (8.93) 0 (0.0) 11 (4.4)   W1. Difficulty in standing for long periods  2.18 (1.27) 14 (5.6) 111 (44.2)   W2. Difficulty in tending to h’hold responsibilities 2.24 (1.23) 9 (3.9) 95 (40.8)   W3. Difficulty in learning a new task 1.49 (0.85) 3 (1.2) 170 (68.3)   W4. Difficulty joining in community activities 2.10 (1.37) 15 (6.6) 123 (54.4)   W5. Extent emotionally affected by health condition 2.82 (1.25) 22 (8.8) 51 (20.4)   W6. Difficulty in short-term concentration 1.94 (1.18) 10 (4.0) 130 (51.8) 169    W7. Difficulty in walking a long distance 2.02 (1.35) 19 (7.6) 140 (55.8)   W8. Difficulty in washing your whole body 1.42 (0.89) 4 (1.6) 194 (77.3)   W9. Difficulty in getting dressed 1.33 (0.75) 2 (0.8) 201 (80.1)   W10. Difficulty in dealing with strangers 1.82 (1.06) 5 (2.0) 135 (53.8)   W11. Difficulty in maintaining a friendship  1.89 (1.21) 10 (4.0) 144 (57.8)   W12. Difficulty in your day to day work 2.03 (1.25) 13 (5.2) 126 (50.4) Euroqol EQ5D index score 0.70 (0.21) 0 (0.0) 35 (13.9) Euroqol Visual Analog Scale 58.16 (21.02) 1 (0.4) 1 (0.4)   E1. Mobility: difficulty walking about 1.35 (0.49) 1 (0.4) 165 (65.7)   E2. Self-Care: difficulty washing/dressing 1.21 (0.43) 2 (0.8) 201 (80.1)   E3. Usual Activities: work/study, family, leisure 1.49 (0.55) 7 (2.8) 135 (53.8)   E4. Pain/Discomfort: degree of pain or discomfort 1.90 (0.66) 43 (17.1) 67 (26.7)   E5. Anxiety/Depression: degree of anx./depression 1.78 (0.63) 28 (11.2) 83 (33.1) SF-6D index score 0.67 (0.14) 0 (0.0) 1 (0.4)   S1. Physical Functioning 2.43 (1.20) 6 (2.4) 58 (23.2)   S2. Role Limitations 2.33 (1.19) 67 (26.8) 81 (32.4)   S3. Social Functioning 2.57 (1.22) 17 (6.8) 67 (26.8)   S4. Pain 3.25 (1.59) 23 (9.2) 47 (18.8)   S5. Mental Health 2.71 (1.07) 11 (4.4) 37 (14.8)   S6. Vitality 2.93 (1.05) 21 (8.4)  20 (8.0) In all cases, floor and ceiling values represented by worst and best possible valued responses, respectively. * ceiling effect present in index score.  170   Table 5.3: Assessing content validity: conceptual coverage of health status measures   Health Health Functioning Health Overall  Symptoms Physical Mental Social Perceptions QoL ASImed     * [3] ASIpsych * [5]  * [3]  * [3] MAP-PHS * [10] MAP-MHS * [10] WHODAS-12  * [6] * [3] * [3] EQ-5D  * [4] * [1]   * [p.w score] EQ-VAS      * [scale] SF-6D  * [4] * [1] * [1]  * [p.w score] * Number of items within the scale pertaining to a given concept presented in square brackets; p.w: preference-weighted. 171   Table 5.4: Results of exploratory and confirmatory factor analysis   ASIpsych  MAP-PHS  MAP-MHS  Rotated Loadings  Rotated Loadings  Rotated Loadings  Factor 1 Factor 2  Factor 1 Factor 2  Factor 1 Factor 2 EFA    P10 0.89 0.35 D1F 0.54 -0.01 D2G 0.84 0.26    P9 0.88 0.43 D1D 0.53 0.06 D2F 0.80 0.30    P11 0.77 0.21 D1E 0.52 0.01 D2H 0.77 0.23    P2 0.71 0.27 D1J 0.46 0.03 D2I 0.68 0.26    P3 0.55 0.18 D1C 0.45 0.05 D2J 0.54 0.34    P1 0.56 0.32 D1B 0.27 0.14 D2B 0.21 0.76    P4 0.11 0.08 D1A 0.26 0.09 D2C 0.29 0.76    P7 0.38 0.61 D1G -0.05 0.64 D2E 0.23 0.71    P8 0.07 0.47 D1H 0.09 0.58 D2D 0.33 0.64    P5 0.24 0.33 D1I 0.16 0.42 D2A 0.37 0.41    P6 0.27 0.27 % Variance 0.91 0.14  0.93 0.17  0.87 0.18 Eigenvalues: Actual Expected*  Actual Expected*  Actual Expected*    Factor 1 4.08 1.44  3.93 1.41  5.23 1.41    Factor 2 1.30 1.31  1.17 1.29  1.37 1.29    Factor 3 1.08 1.23  1.10 1.20  0.76 1.20 MSC    0 0.11   0.11   0.23    1 0.02   0.03   0.07    2 0.04   0.04   0.04    3 0.05   0.05   0.06 CFA    Chi-square 128.42   219.91   318.28    p-value <0.01   <0.01   <0.01    CFI 0.94   0.78   0.80    RMSEA 0.0896   0.1493   0.1852 EFA: Exploratory Factor Analysis; MSC: Mean squared correlation; CFA: Confirmatory Factor Analysis; CFI: Bentler’s Confirmatory Fit Index; RMSEA: Root Mean Squared Error. * As generated by random data, according to parallel analysis. 172   Table 5.5: Assessing criterion validity: ability of health status measures to discriminate between individuals with and without chronic medical conditions   Health status Chronic Condition [N=134] No Chronic Condition [N=117]   Index score Mean (SD) Mean (SD) AUC (95% C.I.) ASImed 0.47 (0.35) 0.25 (0.32) 0.67 (0.60,0.73)* ASIpsych 0.23 (0.21) 0.16 (0.20) 0.60 (0.53,0.67)* MAP-PHS 16.68 (7.16) 13.53 (6.84) 0.62 (0.55,0.69)* MAP-MHS 14.33 (8.53) 14.21 (8.29) 0.53 (0.49,0.59) WHODAS-12 24.63 (8.94) 22.26 (8.79) 0.58 (0.50,0.67)* EQ-5D 0.67 (0.21) 0.74 (0.21) 0.59 (0.52,0.66)* EQ-VAS 55.75 (21.68) 60.93 (19.97) 0.58 (0.51,0.65)* SF-6D 0.65 (0.14) 0.69 (0.14) 0.58 (0.52, 0.65)* * p < 0.05 173   Table 5.6: Responsiveness to improvements in illicit drug use at 12 months, from baseline      Event Cohort   [N=108]    Control Cohort   [N=26]   Item/Score   Mean Difference (12mo-BL) (SD)   Mean Difference (12mo-BL) (SD)  AUC (95% C.I.) ASImed -0.12 (0.42) 0.04 (0.45)  0.60 (0.50, 0.71)*   M1 -2.82 (12.91) 2.12 (14.24)  0.58 (0.49, 0.70)   M2 -0.41 (1.89) 0.23 (1.95)  0.58 (0.49, 0.69)   M3 -0.69 (2.13) -0.04 (2.46)  0.58 (0.50, 0.69)* ASIpsych -0.05 (0.24) 0.00 (0.14)  0.55 (0.47, 0.65)   P1 -0.06 (0.44) 0.08 (0.27)  0.57 (0.51, 0.63)*   P2 -0.11 (0.60) 0.00 (0.49)  0.56 (0.50, 0.65)*   P3 -0.08 (0.58) -0.15 (0.46)  0.54 (0.50, 0.61)*   P4 0.01 (0.22) 0.04 (0.20)  0.52 (0.48, 0.58)   P5 -0.03 (0.35) 0.04 (0.20)  0.53 (0.48, 0.59)   P6 -0.04 (0.33) 0.04 (0.34)  0.54 (0.50, 0.62)*   P7 -0.05 (0.32) 0.08 (0.27)  0.56 (0.51, 0.62)*   P8 -0.02 (0.14) 0.00 (0.00)  0.51 (0.49, 0.54)   P9 -0.53 (13.77) -1.27 (8.52)  0.54 (0.47, 0.64)   P10 -0.38 (1.91) -0.23 (1.42)  0.55 (0.47, 0.65)   P11 -0.42 (1.95) -0.19 (0.98)  0.56 (0.48, 0.65) MAP-PHS -2.90 (6.56) 0.56 (6.64)  0.65 (0.53, 0.77)*   MP1 -0.39 (1.41) -0.04 (0.98)  0.61 (0.49, 0.71)   MP2 -0.47 (1.10) 0.00 (0.96)  0.66 (0.55, 0.76)*   MP3 -0.38 (1.29) -0.24 (1.39)  0.56 (0.49, 0.66)   MP4 -0.19 (1.28) -0.04 (1.46)  0.56 (0.49, 0.67)   MP5 -0.21 (1.18) 0.28 (1.37)  0.64 (0.49, 0.76)   MP6 -0.05 (1.02) 0.36 (0.91)  0.60 (0.51, 0.72)*   MP7 -0.45 (1.30) -0.04 (1.72)  0.59 (0.49, 0.74)   MP8 -0.36 (1.44) -0.16 (1.37)  0.56 (0.49, 0.67)   MP9 -0.27 (1.40) 0.24 (1.05)  0.61 (0.51, 0.72)*   MP10 -0.14 (1.07) 0.20 (1.08)  0.55 (0.45, 0.64) MAP-MHS -4.33 (7.31) 0.04 (7.95)  0.63 (0.50, 0.75)*   MM1 -0.22 (1.18) -0.08 (1.04)  0.55 (0.48, 0.66)   MM2 -0.29 (1.07) 0.08 (1.26)  0.57 (0.49, 0.68)   MM3 -0.21 (1.11) 0.04 (1.17)  0.58 (0.49, 0.71)   MM4 -0.36 (1.26) 0.00 (1.08)  0.58 (0.49, 0.70)   MM5 -0.26 (0.98) 0.04 (0.68)  0.60 (0.50, 0.71)*   MM6 -0.73 (1.23) -0.04 (1.37)  0.63 (0.51, 0.76)*   MM7 -0.64 (1.28) 0.00 (1.26)  0.63 (0.51, 0.74)*   MM8 -0.65 (1.17) 0.04 (1.43)  0.64 (0.52, 0.77)*   MM9 -0.55 (1.35) -0.12 (1.27)  0.60 (0.51, 0.71)*   MM10 -0.39 (0.83) 0.08 (0.81)  0.64 (0.54, 0.73)* WHODAS-12 -2.60 (7.34) -1.67 (9.04)  0.57 (0.44, 0.73)   W1 -0.45 (1.23) -0.19 (1.60)  0.57 (0.49, 0.68)   W2 -0.30 (1.23) -0.52 (1.04)  0.57 (0.49, 0.69)   W3 -0.15 (0.86) -0.04 (0.72)  0.55 (0.49, 0.65) 174    W4 -0.20 (1.42) -0.26 (1.37)  0.57 (0.46, 0.70)   W5 -0.49 (1.53) -0.44 (1.33)  0.55 (0.48, 0.65)   W6 -0.31 (1.30) -0.04 (1.15)  0.56 (0.48, 0.66)   W7 -0.12 (1.22) 0.23 (1.37)  0.60 (0.48, 0.72)   W8 -0.05 (0.81) -0.08 (1.38)  0.54 (0.48, 0.64)   W9 -0.02 (0.85) 0.04 (1.04)  0.53 (0.45, 0.62)   W10 -0.37 (1.11) -0.38 (0.94)  0.54 (0.47, 0.64)   W11 -0.29 (1.10) 0.31 (0.74)  0.66 (0.56, 0.74)*   W12 -0.42 (1.15) -0.38 (1.60)  0.55 (0.48, 0.66) EQ5D 0.13 (0.24) -0.01 (0.21)  0.61 (0.51, 0.72)* EQ-VAS 10.75 (25.77) 1.81 (19.11)  0.66 (0.54, 0.76)*   E1 -0.16 (0.61) 0.12 (0.52)  0.61 (0.52, 0.71)*   E2 -0.07 (0.45) 0.15 (0.46)  0.60 (0.52, 0.70)*   E3 -0.26 (0.70) 0.00 (0.69)  0.60 (0.51, 0.71)*   E4 -0.32 (0.81) -0.04 (0.66)  0.60 (0.51, 0.70)*   E5 -0.32 (0.69) -0.31 (0.62)  0.54 (0.49, 0.62) SF6D 0.06 (0.15) 0.03 (0.11)  0.54 (0.49, 0.62)   S1 -0.27 (1.51) 0.04 (1.31)  0.56 (0.49, 0.67)   S2 -0.32 (1.35) -0.35 (1.06)  0.54 (0.49, 0.63)   S3 -0.58 (1.60) -0.12 (1.18)  0.59 (0.50, 0.69)*   S4 -0.27 (1.51) 0.04 (1.31)  0.56 (0.48, 0.66)   S5 -0.32 (1.35) -0.35 (1.06)  0.54 (0.49, 0.62)   S6 -0.58 (1.60) -0.12 (1.18)  0.59 (0.50, 0.69)* SD: Standard Deviation; AUC (95% CI): Area under the receiver operating characteristic curve and 95% credibility interval. Event and Control cohorts indicate individuals who decreased, and did not decrease illicit drug use according to the pre-defined threshold, respectively. Abbreviations of items/scores according to table 1. *p < 0.05  175   Table 5.7: Summary of empirical comparison of ASImed, ASIpsych, MAP-PHS, MAP-MHS, WHODAS-II, EQ-5D, EQ-VAS, SF-6D    Scale Content Floor/ Internal Construct Responsiveness   Validity Ceiling Effect Consistency Validity  ASImed - - inc. + +  ASIpsych + - + + -  MAP-PHS + + - + +  MAP-MHS + + - - +  WHODAS-12 - + N/A + -  EQ-5D + + N/A + +  EQ-VAS + + N/A + +  SF-6D + + N/A + - Poor performance: ‘-‘ ; Adequate performance: ‘+’.  inc: inconclusive; N/A: not applicable.  176  5.6 References Amato L, Davoli M, Perucci CA, Ferri M, Faggiano F, Mattick RP (2005).  An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research.  J Subst Abuse Treat. 28(4):321-9. Brazier J, Roberts J, Deverill M (2002).  The estimation of a preference-based measure of health from the SF-36.  J Health Econ 21: 271-92.  Brazier J, Roberts J, Tsuchiya A, Busschbach J (2004).  A comparison of the EQ-5D and SF-6D across seven patient groups.  Health Economics 13(9):873-84.  Brooks R. EuroQol: the current state of play. Health Policy 1996; 37(1); 53-72.  Brooks R, Robin R, de Charro F (Eds.) (2003).   The measurement and valuation of health status using EQ-5D: a European perspective (evidence from the Euroqol BIOMED research programme).  Netherlands: Kluwer Academic Publishers.  Calsyn DA, Saxon AJ, Bush KR, et al (2004). The addiction severity index medical and psychiatric composite scores measure similar domains as the SF-36 in substance- dependent veterans: Concurrent and discriminant validity. Drug Alcohol Depend. 76:165- 171.  177  Chisolm TH, Abrams HB, McArdle R, Wilson RH, Doyle PJ (2005).  The WHO-DAS II: psychometric properties in the measurement of functional health status in adults with acquired hearing loss.  Trends Amplif. 9(3):111-26.  Coons SJ, Rao S, Keininger DL, Hays RD (2000). A comparative review of generic quality- of life instruments. PharmacEconomics. 17: 13-35. Darke S, Hall W, Wodak A, Heather N, Ward J (1992).  Development and validation of a multi-dimensional instrument for assessing outcome of treatment among opiate users: the Opiate Treatment Index.  Br J Addict. 87(5):733-42. Conner-Spady B, Suarez-Almazor E (2004).  Variation in the estimation of quality- adjusted life years by different preference-based instruments.  Med Care. 41: 791-801.  Essink-Bot ML, Krabbe PFM, Bonsel GJ, Aaronson, NK (1997).  An empirical comparison of four generic health status measures: the Nottingham Health Profile, the medical outcomes study 36-item Short Form Health Survey, the COOP/WONCA charts and the EuroQol instrument.  Med Care. 35: 522-37.  Euroqol Group (1990).  Euroqol – a new facility for the measurement of health-related quality of life.  Health Policy 16: 199-208.  Gonen M (2007).  Analyzing Receiver Operating Characteristic Curves with SAS.  SAS Press Series, Cary, NC, USA. 178  Guyatt GH, Deyo RA, Charlson M, Levine MN, Mitchell A (1989).  Responsiveness and validity in health status measurement: a clarification.  J Clin Epidemiol. 42(5): 403-8. Guyatt GH, Kirshner B, Jaeschke R (1992).  Measuring health status: what are the necessary measurement properties?  J Clin Epidemiol. 45(12):1341-5. Hays RD, Hadorn D (1992).  Responsiveness to change: an aspect of validity, not a separate dimension.  Qual Life Res. 1(1): 73-5. Hollingsworth W, Mackenzie R, Todd CJ, Dixon AK (1995).  Measuring changes in quality of life following magnetic resonance imaging of the knee: SF-36, EuroQol, or Rosser Index.  Qual Life Res. 4: 325-34.  Home page for the World Health Organization Disability Assessment Schedule II (WHODAS II). 2001. URL: http://www.who.int/icidh/whodas/index.html.  Hudson M, Steele R, Taillefer S, Baron M; Canadian Scleroderma Research Group (2008).  Quality of life in systemic sclerosis: Psychometric properties of the World Health Organization Disability Assessment Schedule II.  Arthritis Rheum. 59(2):270-278  Husted JA, Cook RJ, Farewell VT, Gladman DD (2000).  Methods for assessing responsiveness: a critical review and recommendations.  J Clin Epidemiol. 53: 459-68.  179  Jenkinson C, Stradling J and Petersen S.  How should we evaluate health status? A comparison of three methods in patients presenting with obstructive sleep apnoea.  Qual Life Res. 7: 95-100.  Kirschner B, Guyatt G (1998).  A methodological framework for assessing health indices. J Chronic Dis 1985; 38: 27-36.  Kokkevi A, Hartgers C (1995). EuropASI: European adaptation of a multidimensional assessment instrument for drug and alcohol dependence. European Addiction Research 1: 208– 210.  Krabbe PFM, Peerenboom L, Langenhoff BS, Ruers TJM (2004).  Responsiveness of the generic EQ-5D summary measure compared to the disease-specific EORTC QLQ C-30. Quality of Life Research 13: 1247-53.  Lai JS, Crane PK, Cella D (2006).  Factor analysis for assessing sufficient unidimensionality of cancer related fatigue.  Qual Life Res. 15: 1179-90.  Lelutiu-Weinberger C, Pouget ER, Des Jarlais DD, Cooper HL, Scheinmann R, Stern R, Strauss SM, Hagan H (2009).  A meta-analysis of the hepatitis C virus distribution in diverse racial/ethnic drug injector groups.  Soc Sci Med. 68(3):579-90.  180  Longworth L, Bryan S (2003).  An empirical comparison of the EQ-5D and SF-6D in liver transplant patients.  Health Economics, 12: 1061-7.  Makela K (2004). Studies of the reliability and validity of the addiction severity index. Addiction. 99:398-410; discussion 411-8.  Marra CA, Rashidi AA, Guh D, Kopec JA, Abrahamowicz M, Esdaile JM, Brazier JE, Fortin PR, Anis AH (2005).  Are indirect utility measures reliable and responsive in rheumatoid arthritis patients?  Qual Life Res. 14(5):1333-44. Marsden J, Gossop G, Stewart D, Best D, Farrell M, Lehmann P, Edwards C, Strang J (1998).  The Maudsley Addiction Profile (MAP): A brief instrument for assessing treatment outcome.  Addiction, 93(12): 1857-67. Mason BJ, Kocsis JH, Melia D, Khuri ET, Sweeney J, Wells A, Borg L, Millman RB, Kreek MJ (1998).  Psychiatric comorbidity in methadone maintained patients.  J Addict Dis. 17(3): 75-89. Mathers BM, Degenhardt L, Phillips B, Wiessing L, Hickman M, Strathdee SA, Wodak A, Panda S, Tyndall M, Toufik A, Mattick RP; 2007 Reference Group to the UN on HIV and Injecting Drug Use (2008).  Global epidemiology of injecting drug use and HIV among people who inject drugs: a systematic review.  Lancet. 15;372(9651):1733-45. 181  McLellan AT, Cacciola JC, Alterman AI, Rikoon SH, Carise D (2006).  The Addiction Severity Index at 25: Origins, Contributions and Transitions.  Am J Addict, 15(2): 113- 24. McLellan AT, Kushner H, Metzger D, Peters R, Smith I, Grissom G, Pettinati H, Argeriou M (1992).  The Fifth Edition of the Addiction Severity Index.  J Subst Abuse Treat. 9(3):199-213. Nosyk B, MacNab YC, Sun H, Marsh DC, Fischer B, Schechter MT, Anis AH (2009). Proportional hazards frailty models for recurrent methadone maintenance treatment.  Am J Epidemiol, 170(6):783-92.  O’Connor BP (2000).  SPSS and SAS programs for determining the number of components using parallel analysis and Velicer’s MAP test.  Behav Res Meth Instrum Comput. 32(3): 396-402.  Oviedo-Joekes E, Nosyk B, Marsh DC, Guh D, Brissette S, Gartry C, Krausz M, Anis A, Schechter MT (2009).  Scientific and political challenges in North America's first randomized controlled trial of heroin-assisted treatment for severe heroin addiction: Rationale and design of the NAOMI study.  Clin Trials. 6(3):261-71.  Oviedo-Joekes E, Nosyk B, Brissette S, Chettiar J, Schneeberger P, Marsh DC, Krausz M, Anis A, Schechter MT (2008).  The North American Opiate Medication Initiative 182  (NAOMI): Profile of Participants in North America's First Trial of Heroin-Assisted Treatment. J Urban Health. 85(6): 812-25.  Oviedo-Joekes E, Brissette S, Marsh DC, Lauzon P, Guh D, Anis A, Schechter MT (2009).  A Randomized Trial of Diacetylmorphine vs. Methadone for Treatment- Refractory Opioid Addiction.  N Engl J Med. 361(8): 777-86.  Pösl M, Cieza A, Stucki G (2007).  Psychometric properties of the WHODASII in rehabilitation patients.  Qual Life Res. 16(9):1521-31.  Puigdollers E, Domingo-Salvany A, Brugal MT, Torrens M, Alvaro J, Castillo C, Magrı, Martın S, Vazquez JM (2004).  Characteristics of Heroin Addicts Entering Methadone Maintenance Treatment: Quality of Life and Gender. Subst Use Misuse 39 (9): 1353–68.  Revicki D, Hays RD, Cella D, Sloan J (2008).  Recommended methods for determining responsiveness and minimally important differences for patient-reported outcomes.  J Clin Epidemiol 61: 102-109.  Shaw JW, Johnson JA, Coons SJ (2005).  US valuation of the EQ-5D health states: development and testing of the D1 valuation model.  Med Care. 43(3):203-20  Streiner DL, Norman GR (2003).  Health measurement scales.  A practical guide to their development and use.  New York: Oxford University Press. 183   Terwee CB, Bot SDM, de Boer MR, van der Windt DAWM, Knol DL, Dekker J, Bouter LM, de Vet HCW (2007).  Quality criteria were proposed for measurement properties of heath status questionnaires.  J Clin Epidemiol 60: 34-42.  Van der Zanden BP, Dijkgraaf MGW, Blanken P, de Borgie CAJM, van Ree JM, van den Brink W (2006).  Validity of the EQ-5D as a generic health outcome instrument in a heroin-dependent population.  Drug Alcohol Depend. 82(2):111-8.  Verhaak PFM, Heijmans MJWM, Peters L, Rijken M (2005).  Chronic disease and mental disorder.  Soc Sci Med 60(4): 789-97.  Ware JE, Sherbourne D (1992).  The MOS 36-item short form health survey (SF-36).  1. Conceptual framework and item selection.  Med Care 30: 473-83.  World Health Organization. International Classification of Functioning, Disability and Health (Icf). Geneva: World Health Organisation, 2001.   184  6 HEALTH RELATED QUALITY OF LIFE GROWTH TRAJECTORIES AMONG PATIENTS IN OPIOID SUBSTITUTION TREATMENT5  6.1 Introduction  Opioid dependence carries a significant health burden not only on the duration of life, but also health-related quality of life (HRQoL) (Millson et al, 2006).  Engagement in opiate substitution treatment (OST) can potentially improve patients’ health status through mediation of withdrawal symptoms, decreased need for drug-seeking behaviour and the potential to access additional psychosocial and pharmacological treatment for co- morbid conditions, as recommended in best practices guidelines (National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction, 1998; Health Canada, 2002).  Improvements in health status among patients successfully reaching study endpoints have been limited in some contexts.  A systematic review of OST studies could not draw any conclusions on the effect of OST on HRQoL due to a lack of HRQoL evidence collected in past studies (Amato et al, 2005).  One study reported that patients responding to treatment (ie. decreasing opioid use) did not differ significantly from non- responders on a variety of domains including mental and physical health status (Belding et al, 1998).  5 A version of this manuscript has been submitted for publication.  Nosyk B, Guh D, Sun H, Oviedo-Joekes E, Marsh DC, Brissette S, Schechter Anis AH.  Longitudinal predictors of health related quality of life (HRQoL) among patients in opioid substitution treatment.  185   Several cross-sectional studies have identified factors such as age, duration and severity of drug use and presence of chronic disease as may partially explain variation in baseline HRQoL scores of opioid dependent patients entering treatment (Millson et al, 2006; Puigdollers et al, 2004; Astals et al, 2008).  Prior longitudinal studies assessing HRQoL among out-of-treatment illicit drug users found little evidence of improvement or deterioration.  In a longitudinal study of out-of treatment crack cocaine users, frequency of use was negatively associated with physical functioning, social functioning and mental health; however changes in HRQoL were uncommon over the study’s 2-year follow-up period (Falck et al, 2000).  Another study reported HRQoL improvements in a cohort of drug users over time, however only in Short-Form SF-36 mental component scores, and only among those transitioning out of homelessness (Kertesz et al, 2005).  The effect of drug treatment was not assessed in either study, and may improve patients’ HRQoL regardless of baseline health status.  HRQoL instruments which reflect societal preferences in assessing the burden of disease are recommended in conducting cost-utility analyses.  These analyses are critical for informing decisions regarding the allocation of health care resources.  Generic measures of HRQoL are preferred, as they facilitate direct comparisons of the cost- effectiveness of treatments for substance abuse with alternative health interventions in different disease areas.  Evidence of improvement in HRQoL during OST can be used to argue for expansion of existing substance abuse treatment services or to evaluate the value of effective new interventions relative to current practice.  We analyzed 186  longitudinal HRQoL data from the North American Opiate Medication Initiative (NAOMI) randomized controlled trial to evaluate differences in trajectories of HRQoL among chronic opioid-dependent patients and to identify factors associated with improvement and deterioration following enrolment in OST.  A secondary objective was to evaluate differences in HRQoL growth between study treatment allocations: injectable diacetylmorphine and optimized methadone maintenance treatment. 187  6.2 Methods 6.2.1 Study Sample  The North American Opiate Medication Initiative (NAOMI) randomized controlled trial compared injectable diacetylmorphine (injection) to optimized methadone maintenance (oral) for those with long-standing opioid dependence.  Subjects were recruited from Vancouver, British Columbia, and Montreal, Quebec, Canada between March 2005 and June 2007.  Trial eligibility criteria included Opioid Dependence by DSM-IV criteria, a minimum age of 25 years, 5 years of opioid use and one year of residence in site/city location, regular opioid injection, two previous opiate addiction treatment attempts, but no enrolment in OST in the past 6 months and written and informed consent.  Exclusion criteria included pregnancy, severe medical or psychiatric conditions that are contraindicated for OST, justice system involvement likely to result in extended incarceration, serum bilirubin > 2.5x normal, stage II or greater hepatic encephalopathy or severe chronic respiratory disease.  The study recruited 251 subjects and had a 95.6% follow-up rate on an ‘intent to treat’ basis up to the 12-month trial endpoint.  The mean age of NAOMI participants was 40 years, 39% were female and over 53% had an additional chronic condition.  Stable housing (residence in a house or apartment) was uncommon (27%), and the cohort used cocaine an average of 16 of the past 30 days at baseline.  Further details of trial methodology (Oviedo-Joekes et al, 2009a), baseline patient characteristics (Oviedo- Joekes et al, 2008) and results (Oviedo-Joekes et al, 2009b) are described at length elsewhere. 188   6.2.2 Outcome Variable  Our primary outcome was patients’ self-reported HRQoL, measured by the Euroqol EQ-5D at quarterly follow-up intervals.  The EQ-5D is a generic HRQoL measurement tool where respondents choose 1 of 3 descriptors in 5 domains of their health (mobility, personal care, usual activities, pain/discomfort, and anxiety/depression) followed by a visual analog scale (VAS) of their overall current health state (Euroqol Group, 1990; Brooks, 1996).  This tool was selected to measure HRQoL because it has been shown to have good concurrent and discriminant validity (van der Zanden et al, 2006), as well as responsiveness (Nosyk et al, 2010) in populations of opioid-dependent patients, it can be easily converted to a health utility (Coons et al, 2006) and has minimal burden for patients during questionnaire completion.  We focused on the preference- weighted index score and the associated domains in this analysis. 6.2.3 Explanatory Variables  We sought to identify baseline and longitudinal determinants of changes in HRQoL.  Baseline characteristics included age, gender, presence of a chronic medical condition, HIV or Hepatitis C infection, treatment allocation, current and lifetime years of cocaine use.  Time-varying covariates included stable housing status, medical events (inpatient care, emergency room visits and drug overdoses), psychiatric events (in/outpatient psychiatric care or suicide attempts) and treatment adherence.  Decreases in illicit drug use were captured with the European Addiction Severity Index (EuropASI) 189  (Kokkevi and Hartgers, 1995) drug use composite score, which measures the use of illicit drugs over a 30-day time horizon.  Score ranged from 0 (least) to 1 (greatest severity) on a continuous scale, and were weighted more heavily for multiple and injection drug use. 6.2.4 Data Analysis  The analysis had three steps.  The first step involved applying individual and latent class growth models to the HRQoL data.  A random coefficient regression model (individual growth model) with subject-level random effect terms on the intercept and slope variables was fitted with linear and quadratic time trends to describe the overall trajectory of HRQoL and to examine between-subject variability in HRQoL trajectories (Raudenbusch and Bryk, 2002).  The random coefficient model can be expressed as follows:  where for l=(1,2,3), and . An unstructured covariance matrix was specified for .  Semi-parametric group-based latent class growth analysis was then used to identify major classes of HRQoL trajectories (Nagin, 1999; Jones, Nagin and Roeder, 2001).  As the EQ-5D index score is bounded at 1, and 13.9% (n=35) patients achieved the maximum score at baseline assessment, we apply a censored normal model to HRQoL data, with censorship point = 1.  Jones, Nagin and Roeder (2001) express the model as follows: 190   where . We considered models with up to six underlying latent classes (k=1, …, 6), and up to a cubic polynomial function for each class.  Model selection was based on the Bayesian Information Criterion (BIC) as a measure of goodness-of-fit (Schwartz, 1978; Yang, 1998).  The initial individual growth model was then re-fitted to include covariates indicating trajectory class, as identified by the latent class growth analysis, and product terms between trajectory class and growth parameters.  This allowed us to quantify latent growth trajectories and evaluate differences in between-subject heterogeneity after their inclusion into the growth model.  Second, in order to identify the characteristics of patients assigned to each of the latent classes of HRQoL growth, associations between trajectory class and baseline covariates were examined using polytomous logistic regression (Agresti, 1990). Unadjusted comparisons between patients in either of the study’s allocated treatments at each assessment were first tested using two-sided non-parametric Kruskal-Wallis tests. Treatment allocation was then included as a covariate in multivariate models on trajectory class, as described above.  Third, in order to identify determinants of changes in HRQoL from one assessment to another, differences in EQ-5D index scores between consecutive 191  assessments (assessmentt – assessmentt-1) were regressed against first-differences in each of the time-varying covariates in a multivariate linear mixed effect (random intercept) model (Wu, Gray and Brookmeyer, 1999).  Treatment allocation was added to this model as a fixed covariate to determine whether changes in HRQoL between follow-up assessments differed between cohorts, controlling for other time-varying covariates. Statistical Analysis was conducted using Statistical Analysis Software (SAS) version 9.1 and [R] freeware version 2.5.1.  192  6.3 Results  The majority of study participants (N=132, 55.0%) reported improvements in HRQoL, with an overall mean improvement of 23.9% at 12 months (from baseline). Mean EQ-5D scores improved from 0.70 (SD: 0.21) at baseline, 0.78 (0.20) at 3 months, 0.77 (0.20) at 6 months, 0.78 (0.19) at 9 months and 0.79 (0.19) at 12 months. Unadjusted comparison of EQ-5D scores between treatment allocation determined HRQoL was not statistically significantly different at baseline (injection: 0.71 vs. oral: 0.69, p=0.54). Within the individual domains, improvements at 12 months, from baseline were most common in pain / discomfort (33.2%), anxiety / depression (31.5%) and usual activities (30.3%) (Figure 6.1). 6.3.1 Individual Growth Modeling Analysis  The fixed and random effects estimates of parameters in the initial growth model are presented in table 6.1.  The model shows that, on average, subjects entered the trial with an EQ-5D index score of 0.71.  HRQoL then improved at a rate of 0.06 at baseline, with the rate of increase gradually decreasing over time (-0.01) until the 12-month endpoint.  Examination of random-effects parameter estimates for the growth curve model shows some degree of between-subject variation (estimated variance: 0.016), however less between-subject variation in the linear and quadratic components of the trajectories (estimated variances of 0.001 and 0.000, respectively).  A plot of the overall HRQoL trajectory for the entire cohort is shown in the top panel of figure 6.2.  193  6.3.2 Latent Class Growth Analysis  Based on BIC results we selected a growth mixture model with three HRQoL trajectory classes.  The shape of the trajectories for classes 1 and 3 was linear, while a quadratic polynomial function for class 2 provided the lowest BIC value among the candidate models.  Plots of the fitted HRQoL for each of the three trajectories (along with their 95% confidence intervals) are shown in the bottom panel of figure 2. Class 1, 2 and 3 included 56 (22.3%), 144 (57.4%) and 51 (20.3%) subjects, respectively.   Patients were assigned to the latent trajectory classes for which the posterior probability of latent class membership was highest.  Average values of latent class assignment probability were 0.870, 0.894 and 0.850 for classes 1 to 3, respectively.   This analysis pooled patients in the two treatment arms due to similarities in functional form and class structure.  Further, small sample sizes prevented inference on stratified samples. 6.3.3 Individual Growth Modeling Analysis Accounting for Class Membership  When the growth model was re-fitted with the class membership variable derived from the latent class growth analysis, random effect parameters for between-subject variances in intercept and slope were negligible.  Subjects in class 1 (stably low) were characterized by low HRQoL at baseline (0.538), and HRQoL change over time was not statistically significantly different from zero (-0.009; p=0.739).  Subjects in class 2 (moderate and increasing) had higher baseline HRQoL values (0.538 + 0.171 = 0.709; p< 0.001), and their HRQoL trajectory improved at a rate of 0.089 (-0.009 + 0.098 = 0.089; p < 0.001) at baseline, with the rate of increase gradually decreasing over time (0.004 – 0.021= -0.017; p < 0.001).  Subjects in class 3 (stably high) began with relatively high 194  HRQoL values (0.538 + 0.364 = 0.902; p < 0.001) and did not improve significantly over time (-0.009 + 0.038 = 0.029; p = 0.26).    While the inter- and intra-individual variance in HRQoL growth was low either model formulation due to the bounds of the index score, comparisons in the final model with those from the initial model (table 1), showed that classification lead to a reduction in population heterogeneity in HRQoL growth. 6.3.4 Factors Associated with HRQoL Trajectory Class Membership  Table 6.2 shows adjusted associations between subjects’ characteristics measured at trial entry and trajectory class; class 1 served as the reference category.  Female gender had a strong negative association with ‘moderate and increasing’ class membership (p=0.035) as well as ‘stably high’ class membership, however the latter were not statistically significantly different after controlling for other factors.  ‘Stably high’ class members were younger (p=0.046), more likely to be of aboriginal ethnicity (p=0.039), less likely to have a chronic condition (0.036) or reside in unstable housing conditions (0.044) than ‘stably low’ class members, and had lower illicit drug use severity (p=0.022).  Study treatment allocation was not significantly associated with ‘moderate and increasing’ or ‘stably high’ class membership.  Univariate relationships between class membership and baseline medical/psychiatric events, indication of Hepatitis C, HIV and current and lifetime cocaine use were removed due to statistical non-significance. 6.3.5 Factors Associated with Changes in HRQoL  Finally, changes in a number of time-varying covariates were associated with changes in HRQoL from one assessment to the next.  A change from unstable to stable housing status was associated with an improvement in HRQoL of 0.041 (p=0.023), while 195  a medical event decreased HRQoL by 0.036 (p=0.015).  Psychiatric events had a negative effect on HRQoL; however this effect was not statistically significant (p=0.626).  While discontinuation in treatment did not have a statistically significantly negative effect on HRQoL, decreases in illicit drug use had the largest effects on HRQoL.  An improvement in individuals’ ASI illicit drug use score of 20-50% lead to an improvement in HRQoL of 0.050 (p=0.005), while those improving their illicit drug use composite scores greater than 50% experienced an HRQoL improvement of 0.055 (p=0.002).    Finally, changes in HRQoL among patients in the injection and oral treatment arms were not statistically significantly different (p=0.750), controlling for the above time-varying covariates (table 6.3). 196  6.4 Discussion  The objectives of this study were to identify longitudinal trajectories of HRQoL improvement among chronic opioid dependent patients in OST, identify predictors of trajectory class and factors associated with changes in HRQoL.  The primary finding of the study was that HRQoL could, in fact, be improved through effective treatment which decreases individuals’ illicit drug use, with 55% of patients reporting higher EQ-5D index scores during study follow-up than at baseline.  Improvements were most common in the ‘pain/discomfort’, ‘anxiety/depression’ and ‘usual activities’ domains.  This may be interpreted as indicating decreased pain or discomfort as a result of opiate withdrawal, lower anxiety and depression from the demands of daily drug acquisition and a subsequent return to usual activities in absence of these demands.  HRQoL levels were higher among patients in the injection arm, however not significantly so, and patients were distributed evenly across latent HRQoL growth trajectories.  Further mixed-effects multivariate models on levels of HRQoL confirmed that the effect of treatment allocation was not statistically significantly different from zero, controlling for both the baseline and time-varying covariates described herein (results not presented).  Three latent classes of HRQoL growth trajectories were identified, grouping patients with stably low, moderate and increasing, and stably high HRQoL.  The majority of study participants fell in the moderate and increasing class.  Clients in the high- 197  HRQoL class did not have statistically significant HRQoL improvement over time, however high baseline EQ-5D scores suggest that these patients, who had relatively lower baseline illicit drug use severity and were less likely to have a chronic medical condition, considered themselves in relatively good health, therefore  room for improvement was minimal.  It is of concern that patients in the low-HRQoL class did not improve as a result of OST.  These patients were more likely to be female than class 2 members, had higher drug use severity, were more likely to have a chronic medical condition and reside in unstable housing conditions than class 3 members.  Targeted interventions may be required to help low-HRQoL patients realize benefits from treatment.  While individual growth curve models have been used in longitudinal studies on HRQoL growth (Zee, 1998; Chen and Cohen, 2006), our latent class growth modeling approach was adapted from other epidemiological applications (Karp et al, 2005; Croudace et al, 2003).  This novel application to longitudinal HRQoL data has shed new light on understanding HRQoL trajectories in opioid dependent patients in treatment.  We extended our analysis to also identify factors occurring during treatment that were associated with changes in HRQoL.  While retention in treatment at 12 months is a commonly used primary outcome measure in phase-III clinical trials in substance abuse, we found that decreases in illicit drug use, rather than retention in treatment, had the largest effect on self-reported HRQoL.   The positive effect of transition to stable housing was supported by Kertesz et al (2005).  Finally while both medical and psychiatric events had a negative impact on changes in HRQoL, only the former was statistically 198  significantly different from zero.  These findings can be attributed to their lower incidence during follow-up and high psychiatric severity among study clients reported at baseline (Oviedo-Joekes et al, 2008).  Our findings indicating improvements in HRQoL among opioid-dependent patients in treatment are sparsely supported in the literature thus far.  One study comparing patients responsive to treatment (as indicated by urine screening results indicating no illicit heroin use) to non-responders found no differences in health status (rather than HRQoL), as measured by the ASI medical and psychiatric composite scores. In fact, neither group demonstrated significant improvement in these domains over the study’s 6-month timeframe (Belding et al, 1998).  Another large observational study of patients entering OST reported no significant improvements in mental health and ‘quality of life’, measured within the WHO Composite International Diagnostic Interview (Wittchen et al, 2008).   Similarly-designed trials of injectable diacetylmorphine reported modest improvement in HRQoL during treatment (March et al, 2006; Makela, 2004); however patients included in these studies were permitted to be receiving treatment at trial entry, thus attenuating HRQoL improvements.  We believe this contrast in results may have been observed for at least four reasons.   The first was the analytic approach chosen in our analysis.  Presentation of overall trends in HRQoL may obscure distinctions between improving and non- improving patients, as we have uncovered with latent class growth analysis.  Second, the difference in results may have been an artefact of the measures of health status employed. 199   Among the most commonly used measures of health status in evaluations of substance abuse treatment interventions has been the Addiction Severity Index Medical and Psychiatric composite scores, however accumulated evidence on the validity of these scores was questioned recently.  A review of 37 studies on the psychometric performance of the ASI indicated that although the medical and psychiatric composite scores had high internal consistency (indicating unidimensionality), estimates of reliability were highly variable, little evidence of criterion validity was found, and assessment of the responsiveness of these measures in comparison to an external change criterion had not been formally tested (Makela, 2004).  Furthermore, the conceptual content of the ASI composite differ from that of the EQ-5D and other measures of HRQoL. The former measures focus on what can be considered ‘general health perceptions’, with items assessing the presence of medical/psychiatric problems, and the extent to which these problems have troubled or bothered patients and the perceived need for treatment. Psychiatric symptoms are also assessed with the ASI psychiatric composite score. Similar to other generic measures of HRQoL, the EQ-5D on the other hand captures, indirectly, overall health-related quality of life through direct assessment of physical, mental and social functioning (Brooks, 1996).  Another plausible explanation for the difference in results was likely the patient population under study.  NAOMI participants were selected based on the chronic nature of their drug abuse and unsuccessful prior treatment attempts.  Patients had high levels of illicit drug use at baseline and a high level of medical and psychiatric co-morbidity at trial entry (Oviedo-Joekes et al, 2008).  It is likely that the study selected a lower 200  proportion of stably high HRQoL clients than seen in populations of earlier-stage opioid dependent patients, however further study is required to confirm this hypothesis.  Finally, the form of treatment delivery, and the availability of other ancillary services likely had an effect on the extent to which patients’ HRQoL improved. Treatment was optimized in keeping with Health Canada Best Practices (2002). Psychosocial care was available to patients in either treatment arm.  Higher doses of methadone were delivered to clients (an average maintenance dose of 96 mg per day) (Oviedo-Joekes et al, 2009b) in comparison to those receiving treatment at the population level (Nosyk et al, 2009), and efforts were made to link patients with treatment for other medical and psychiatric conditions where possible.  Our study was not without limitations.  First, while a relatively small proportion of study participants were lost to follow-up, in some cases these losses were informative. Mixed effects regression models do not require the assumption that observations were missing completely at random and control for the effect of unmeasured subject-level heterogeneity. They are therefore recommended to provide unbiased coefficient estimates in analyzing longitudinal HRQoL data with missing observations (Fairclough et al, 2008).  Second, we relied on self-reported measures of drug use, which may be subject to underreporting.  Random urine tests administered at the treatment clinic supported self- 201  reported heroin use (Oviedo-Joekes et al, 2009b), suggesting that our measure of current drug use was, in fact, valid.  Finally, the NAOMI study recruited a highly selective cohort of opioid users who were relatively homogeneous in their current opioid use and treatment history.   The internal validity of our results are therefore strong, however it is unclear whether they extended beyond the study population of chronic, treatment-resistant opioid-dependent patients.  Further study in earlier-stage populations of opioid users are required to determine whether similar improvements in valid and responsive HRQoL measures can be achieved through treatment.  In summary, this study has demonstrated that HRQoL among chronic, treatment resistant opioid dependent patients can be meaningfully improved through effective, patient-centered treatment.  HRQoL should be considered an endpoint in evaluations of treatment for Opioid Dependence and other Substance Dependence.  202  Table 6.1: Parameter estimates for fixed and random effects components growth curve analysis before and after inclusion of latent class membership variables    Initial Model  Final Model Parameter Estimate (SE) p-value Estimate (SE) p-value Fixed Effects    Intercept 0.710 (0.012) <0.001 0.538 (0.019) <0.001    Time*  0.055 (0.011) <0.001 -0.009 (0.023) 0.739    Time2 -0.009 (0.003) <0.001 0.004 (0.006) 0.459    Trajectory class 1 -- -- (ref)    Trajectory class 2 -- -- 0.171 (0.022) <0.001    Trajectory class 3 -- -- 0.364 (0.027) <0.001    Time* Trajectory class 1 -- -- (ref)    Time* Trajectory class 2 -- -- 0.098 (0.027) <0.001    Time* Trajectory class 3 -- -- 0.038 (0.033) 0.296    Time2* Trajectory class 1 -- -- (ref)    Time2* Trajectory class 2 -- -- -0.021 (0.006) 0.002    Time2* Trajectory class 3 -- -- -0.007 (0.008) 0.401 Random Effects    Between-subject variance in intercepts 0.016 (0.128)  0.000 (0.000)    Between-subject variance in slopes for time 0.001 (0.010)  0.000 (0.000)    Between-subject variance in slopes for time2 0.000 (0.000)  0.000 (0.000)    Residual variance 0.023 (0.151)  0.022 (0.149) SE, standard error.   *Units for time: 3-month intervals from trial entry.  203  Table 6.2: Adjusted associations of patient characteristics measured at trial entry with specific trajectories of HRQoL     Class 2    Class 3  OR 95% C.I. p-value  OR 95% C.I. p-value Age 0.965 0.927 1.004 0.075   0.946 0.896 0.999 0.046 Gender: Male               Female 0.493 0.256 0.952 0.035  0.404 0.160 1.017 0.054 Ethnicity: Non-Aboriginal                  Aboriginal 1.392 0.621 3.120 0.422  3.020 1.057 8.632 0.039 Chronic medical condition: No                                             Yes 1.112 0.571 2.167 0.755  0.395 0.165 0.943 0.036 ASI Drug composite score 0.989 0.965 1.013 0.358  0.962 0.930 0.994 0.022 Housing stability: Stable                               Unstable 0.551 0.274 1.106 0.094  0.377 0.146 0.975 0.044 Treatment allocation: Oral                                     Injection 1.275 0.675 2.407 0.454   1.500 0.623 3.612 0.366  OR: Odds Ratio; CI: Confidence Interval.  * Polytomous logistic regression analysis using class 1 as the reference category.  204   Table 6.3: Factors associated with changes in EQ-5D index scores  first-differenced covariates Β SE p-value Intercept -0.002 0.012 0.876 Housing Status 0.041 0.018 0.023 Medical Event -0.036 0.015 0.015 Psychiatric Event -0.011 0.022 0.626 Treatment Discontinuation -0.020 0.017 0.249 Decrease in Illicit Drug Use: < 20% (ref)                                              20-50% 0.050 0.018 0.005                                              >50% 0.055 0.017 0.002 Treatment allocation: Oral (ref)                                     Injection 0.005 0.014 0.750 SE: standard error.  205   Figure 6.1: Distribution of changes in EQ-5D item responses: Baseline-12 months           2-level improvement  1-level improvement  No change  1-level deterioration  2-level deterioration      206  Figure 6.2: HRQoL Trajectories of patients enrolled in the NAOMI trial   207  6.5 References  Agresti A (1990).  Categorical data analysis.  New York, NY: John Wiley and Sons. Amato L, Davoli M, Perucci CA, et al (2005).  An overview of systematic reviews of the effectiveness of opiate maintenance therapies: available evidence to inform clinical practice and research.  J Subst Abuse Treat. 28(4):321-9. Astals M, Domingo-Salvany A, Buenaventura CC, et al (2008).  Impact of substance dependence and dual diagnosis on the quality of life of heroin users seeking treatment. Subst Use Misuse. 43(5): 612-32. Belding MA, McLellan AT, Zanis DA, et al (1998).  Characterizing “nonresponsive” methadone patients.  Journal of Substance Abuse Treatment. 15(6): 485-92. Brooks R (1996). EuroQol: the current state of play. Health Policy 37(1); 53-72. Chen H, Cohen P (2006).  Using individual growth model to analyze the change in quality of life from adolescence to adulthood.  Health Qual Life Outcome 4:10. Coons SJ, Rao S, Keininger DL, et al (2000). A comparative review of generic quality of life instruments. PharmacEconomics. 17: 13-35. Croudace TJ, Jarvelin M-R, Wadsworth MEJ, et al (2003).  Developmental typology of trajectories to nighttime bladder control: Epidemiological application of longitudinal latent class analysis.  Am J Epidemiol. 157: 834-42. 208  Dijkgraaf MGW, van der Zanden BP, de Borgie CAJM, et al (2005).  Cost utility analysis of co-prescribed heroin compared with methadone maintenance treatment in heroin addicts in two randomized trials.  BMJ. 330:1297-1302.  Euroqol Group (1990).  Euroqol – a new facility for the measurement of health-related quality of life.  Health Policy 16: 199-208. Fairclough DL, Thijs H, Huang I-C, et al (2008).  Handling missing quality of life data in HIV clinical trialsL what is practical?  Qual Life Res. 17: 61-73. Falck RS, Wang J, Siegal HA, et al (2000).  Longitudinal application of the Medical Outcomes Study 36-item Short Form Health Survey with not-in-treatment crack cocaine users.  Med Care. 38: 902-10. Health Canada.  Best Practices, Methadone Maintenance Treatment (2002).  Prepared by Jamieson, Beals, Lalonde and Associates, Inc. for the Office of Canada’s Drug Strategy, Health Canada. http://www.hc-sc.gc.ca/hc-ps/pubs/adp-apd/methadone-bp-mp/index- eng.php (Accessed: June 1, 2008) Jones BL, Nagin DS, Roeder K (2001).  A SAS procedure based on mixture models for estimating developmental trajectories.  Sociol Meth Res. 29: 374-93. Karp I, O’Laughlin J, Paradis G, et al (2005).  Smoking trajectories of adolescent novice smokers in a longitudinal study of tobacco use.  Ann Epidemiol. 15: 445-52. 209  Kertesz SG, Larson MJ, Horton NJ, et al (2005).  Homeless chronicity and health-related quality of life trajectories among adults with addictions.  Med Care. 43(6):574-85  Kokkevi A, Hartgers C (1995).  EuropASI: European adaptation of a multidimensional assessment instrument for drug and alcohol dependence, European Addiction Research 1: 208–210.  Makela K (2004). Studies of the reliability and validity of the addiction severity index. Addiction. 99:398-410; discussion 411-8. March JC, Oviedo-Joekes E, Perea-Milla E, et al (2006). Controlled trial of prescribed heroin in the treatment of opioid addiction.  J Subst Abuse Treat. 31(2):203-11. Millson P, Challacombe L, Villeneuve PJ, et al (2006).  Determinants of health-related quality of life of opiate users at entry to low-threshold methadone programs.  Eur Addict Res. 12(2): 74-82. National Consensus Development Panel on Effective Medical Treatment of Opiate Addiction (1998).  Effective medical treatment of opiate addiction.  Journal of the American Medical Association 280(22): 1936-43.  Nosyk B, MacNab YC, Sun H, et al (2009). Proportional hazards frailty models for recurrent methadone maintenance treatment.  American Journal of Epidemiology, 170(6): 783-92. 210  Nagin DS (1999).  Analyzing developmental trajectories: A semi-parametric, group- based approach.  Psychol Methods. 4: 139-57. Oviedo-Joekes E, Nosyk B, Marsh DC, et al (2009).  Scientific and political challenges in North America's first randomized controlled trial of heroin-assisted treatment for severe heroin addiction: Rationale and design of the NAOMI study.  Clin Trials. 6(3):261-71. Oviedo-Joekes E, Nosyk B, Brissette S, et al (2008). The North American Opiate Medication Initiative (NAOMI): profile of participants in North America's first trial of heroin-assisted treatment.  J Urban Health. 85(6):812-25. Oviedo-Joekes E, Brissette S, Marsh DC, et al (2009).  A Randomized Trial of Diacetylmorphine vs. Methadone for Treatment-Refractory Opioid Addiction.  N Engl J Med. 361(8):777-86. Puigdollers E, Domingo-Salvany A, Brugal MT, et al (2004).  Characteristics of heroin addicts entering methadone maintenance treatment: quality of life and gender.  Subst Use Misuse. 39(9): 1353-68. Raudenbusch SW, Bryk AS (2002).  Hierarchical linear models: Applications and data analysis methods.  Thousand Oaks, CA: Sage Publications.  Schwartz G (1978).  Estimating the dimension of a model.  Ann Stat. 6: 461-64.  van der Zanden BP, Dijkgraaf MG, Blanken P, et al (2006).  Validity of the EQ-5D as a generic health outcome instrument in a heroin-dependent population.  Drug Alcohol Depend. 82(2): 111-8. 211  Wittchen HU, Apelt SM, Soyka M, et al (2008).  Feasibility and outcome of substitution treatment of heroin-dependent patients in specialized substitution centers and primary care facilities in Germany: a naturalistic study in 2694 patients.  Drug Alcohol Depend. 95(3):245-57. Wu AW, Gray SM, Brookmeyer R (1993).  Application of random effects models and other methods to the analysis of multidimensional quality of life data in an AIDS clinical trial.  Med Care. 37(3):249-58. Yang C (1998).  Finite mixture model selection with psychometric applications.  Los Angeles, CA: University of California, Los Angeles. Zee BC (1998).  Growth curve model analysis for quality of life data.  Stat Med. 17: 757- 66. 212   7  CONCLUSIONS  7.1 Summary of Findings and Implications  The collection of studies presented above draw upon a number of large, unique and high-quality datasets, and feature up-to date analytic technique to fulfill their objectives.  Study (1) (chapter 2) drew upon provincial dispensation data in a longitudinal assessment of the determinants of retention in methadone maintenance treatment across multiple treatment episodes. The primary finding of study (1) was that patients experiencing multiple treatment episodes tended to stay in treatment for progressively longer periods in later episodes.  Results presented in figure 2.5 suggested that individuals may experience a number of failed attempts before being maintained long- term.  This study was among the first applications of repeated measures ‘frailty’ models in the medical literature, and revealed an integral finding on the patterns of rehabilitation in opioid dependent patients, which may hold for other substances of abuse such as nicotine dependence.  Study (2) (chapter 3) combined data on illicit drug prices, collected by undercover drug enforcement agents, with a prospective epidemiological study on illicit drug users to determine the effect of the change in drug prices on consumption. The results of study (2) suggested that polydrug users treated crack cocaine as a substitute for heroin, and powder 213  cocaine was complementary to heroin and crack cocaine use.  Heroin and crack cocaine were price inelastic, while powder cocaine consumption, within the population of primarily crack cocaine and heroin users under study, was price elastic. This is the first empirical test of the theory of rational addiction in illicit drug use since its development over 20 years ago.  While national surveys and other data sources have observed increasing trends towards nonmedical prescription drug use and relatively new stimulants such as crystal methamphetamine at the aggregate level, we have shown that illicit drugs can be considered substitutes or complements for one another at the individual level, and thus may adjust consumption in response to temporary, and potentially also long-term supply disruptions.  This suggests that drug treatment may be a more effective means of reducing illicit drug use than enforcement.  However, treatment programs must acknowledge and treat all forms of concurrent addiction/abuse.  Evidence of substitution effects between heroin and crack cocaine suggest that single-drug focused treatment such as methadone maintenance treatment, which acts as a substitute for heroin, eliminating withdrawal symptoms while blocking euphoric effects of other opioids, may not address concurrent crack cocaine use, and may indeed increase use among polydrug users in treatment.  Study (3) (chapter 4) again drew upon provincial dispensation data to identify trends in compliance to MMT dosing guidelines at the population level in British Columbia, Canada from 1996-2007.  In study (3), we found compliance to minimally effective dose guidelines, which are independently associated with treatment retention, fell from 2001 to 2006 in British Columbia.  This decline was mirrored by 12-month 214  retention figures over the same period.  This was the first comprehensive assessment of methadone dosing guideline compliance, and serves as a basis of evidence for policy reform in methadone prescribing.  We concluded that quantitative studies assessing trends in compliance to clinical practice guidelines can provide an evidence base for the maximization of public health and welfare.  Prescription of appropriate starting doses and responsiveness to patients’ needs in effectively eliminating opioid withdrawal symptoms are critical factors in successfully maintaining opioid dependent patients in methadone treatment.   Physician training, specific authorization and peer-assessment may be important tools for improving physician compliance with clinical practice guidelines.  Study 4 (chapter 5) used data collected alongside a randomized controlled trial to compare the psychometric properties of eight discriminative and evaluative measures of health status among chronic opioid dependent patients.  Study (4) revealed that none of the health status measures assessed in the North American Opiate Medication Initiative (NAOMI) trial performed uniformly as ‘best’ or ‘worst’, while the EQ-5D appeared to be the preferable generic, indirect utility measure.   This study is, to our knowledge, the first head-to-head comparison of health status measures in the field of substance abuse, and makes an important contribution in the knowledge gained on the performance of these measures.  We concluded that there was considerable need for further development of high quality disease-specific measures of health status for use in substance abuse.  Finally, study (5) demonstrated that that HRQoL could, in fact, be improved through effective treatment which decreases individuals’ illicit drug use, with 55% of 215  patients reporting higher EQ-5D index scores during study follow-up than at baseline. Improvements were most common in the ‘pain/discomfort’, ‘anxiety/depression’ and ‘usual activities’ domains.  This may be interpreted as indicating decreased pain or discomfort as a result of opiate withdrawal, lower anxiety and depression from the demands of daily drug acquisition and a subsequent return to usual activities in absence of these demands.  Latent class growth analysis identified three distinct HRQoL trajectories throughout the trial: low and constant, moderate and improving, high and constant.  Our latent class growth modeling approach was adapted from other epidemiological applications, and marks the first time, to our knowledge, that this methodology has been applied to HRQoL data.  This novel application to longitudinal HRQoL data has shed new light on understanding HRQoL trajectories in opioid dependent patients in treatment.  Together these studies provide a substantial contribution to the study of illicit drug addiction. 7.2 Study Strengths and Limitations  The individual studies presented herein were not without limitations.  Chapters 2 to 6 were based on four datasets, each with their own specific strengths and limitations. Chapters 2 and 4 were based on provincial administrative data on drug dispensation.  As all British Columbia residents were included, and all dispensations captured, the database is among the largest studies on methadone maintenance treatment delivery and outcomes, and provides a level of detail in treatment patterns not previously known.  There were, 216  however, several important limitations that require mention. First, our results in study (1) primarily reflect the treatment careers of opioid dependent individuals with free access to primarily low-threshold, community-based methadone treatment programs.    Second, some proportion of the study sample may have been lost to follow-up due to intra- provincial out-migration; past studies of injection drug users in Vancouver, British Columbia suggest this proportion is small (1.2-5.5% per annum).  This can be considered non-differential outcome misclassification, resulting in attenuation of measures of association towards the null, suggesting our hazard ratios may have been somewhat conservative.   Third, retention in treatment is necessarily an intermediate outcome; while the benefits of MMT are only experienced while in treatment, other outcomes including opioid abstinence were not captured.  The data presented here relate only to the duration of medication administration.  Finally, there were both advantages and disadvantages in using administrative data to conduct this retrospective cohort study. The primary limitations were twofold: first, there was potential for misclassification of treatment episode lengths, dosage and adherence due to coding errors in the database.  We have taken steps to reduce this misclassification as much as possible (see appendix B).  Second, any non-experimental study may be subject to residual and/or unmeasured confounding.  Other predictors of retention in treatment such as other illicit drug use, criminal activity, motivational status and social supports were unavailable. Though we cannot ascertain the individual effects of the unobserved factors, we can confidently state that their omission has not biased the coefficients on the existing fixed effects included in the analysis.  In regards to study (3), 217  while this study provided an accurate depiction of actual treatment delivery, clinical decision-making was not observed.  In many cases sporadic treatment adherence limits physicians’ abilities to optimize dosing, and anecdotal reports from prescribing physicians suggest that patients may choose to stabilize at lower doses, thus intentionally not fully utilizing the pharmacodynamic properties of methadone.  This study should thus be placed in an appropriate context, and used as a basis for discussion on how to better serve methadone clients.  The advantages of using this dataset were substantial: the use of this centralized drug dispensation database allowed for a population-level study unparalleled in size, which allowed us to observe not only the effects of several key determinants on individual episodes, but also the pattern of treatment, thus taking into account the chronic, recurrent nature of opioid dependence.  Study 2 (chapter 3) combined data from a prospective observational cohort study and a database capturing undercover purchases of illicit drugs by United States drug enforcement agency (DEA) field agents.  Both datasets had limitations. The observational cohort study had a very short, and non-uniform follow-up (only 1 follow-up assessment, 3 or 6 months following baseline assessment), losses to follow-up were great (nearly 50% of those observed at baseline were lost to follow-up), and there were several different interventions tested among participants.  These interventions were not randomly assigned, differed between study sites, and compliance to the treatment provided was 218  low.  Further, the study was completed over 10 years ago, and may not represent current drug use practices among street drug users.  The drug price dataset (STRIDE) has been criticized in the past (Horowitz et al) due to its non-random collection of drug purchases, which may not present an accurate depiction of prices for all purchases.  Some cities and regions are overrepresented in the dataset, while drug price estimates in other cities or regions are based on relatively small numbers of observations.  Recent innovations in the analytic methods used to construct these price series, implemented here, have attempted to address this criticism (Caulkins et al, 2004).  What’s more, while the relationship between drug consumption and purchase price requires assessment at the individual level, real, prices could only be acquired at the city level.  Ideally, information on drug purchase prices would be collected within the study, with samples purchased from users and tested for purity.  We can only speculate that the prices retrieved were representative of those faced by the users in the CA dataset. Any individual-level variation in purchase prices and purity within a given city and quarter could not be observed with the methodology employed. Due to ethical and logistic concerns it is unlikely that the ‘first-best’ design for this type of study, capturing drug consumption, nominal prices and drug purity at the individual level, could ever be conducted.  219  Nonetheless Despite these limitations, STRIDE remains a rich source of information to inform drug policy, and among the only sources for drug price and purity data available to researchers. The combination of these datasets provided a rare opportunity to study the demand for illicit drugs among chronic and multiple drug users. Previous studies on the demand for illicit drugs were based on national surveys, which likely could not reach the majority of chronic drug users.  Further, the observational cohort study offered two crucial advantages over other conventional cohort studies on drug users that made the analysis in chapter 3 possible.  The first was the fact that detailed information on drug consumption was obtained for a variety of different drugs. In most cases, only the number of days drugs were used in the past 30 days is assessed. In the COOP dataset, the number of times a drug was used was assessed, which allowed us to observe the quantity of substances consumed more accurately.  Studies (4) and (5) (chapters 5 and 6) were based on data collected prospectively alongside a randomized controlled trial.  Losses to follow-up were minimal, and appropriate measures were taken to ensure the quality of the data collected – the candidate trained individual interviewers to administer the questionnaires, and held bi- annual refresher sessions to resolve any problems or inconsistencies in data collection. These factors point to a high level of internal validity, however the trial was based on a highly selective cohort of individuals.  Second, in regards to study (4), the measurement properties assessed in this paper did not satisfy the comprehensive list of quality characteristics.  Specifically, we did not consider each of the measures` reliability and concurrent validity.  Furthermore, by only assessing change in health status around a 220  criterion that was hypothesized to improve health status, we provided only limited evidence of responsiveness.  The frequencies of medical and psychiatric events and changes in housing status, occurring independently from improvements in illicit drug use, were too low to accurately measure responsiveness using our methodological framework. Relapse into frequent illicit drug use (using a symmetric 20% increase in drug use) from baseline (or any time point), was also not frequently observed: most patients either remained in treatment, lowering their illicit drug use, or dropped out prior to their first follow-up HRQoL assessment.  While the chosen criterion cannot produce evidence that the instruments can detect small changes in the constructs being measured, and cannot simultaneously determine responsiveness to both improvement and deterioration in health (evidence which may be produced using an anchor-based approach), they nonetheless allow us to determine whether the measures can discriminate between patients experiencing and not experiencing an objective and clinically meaningful outcome; a characteristic that all HRQoL measures used within populations of opioid dependent patients should possess.  Further, discrimination better than chance (AUC statistically significantly greater than 0.5) was chosen as the criteria with which we judged evidence of construct validity and responsiveness.  While the AUC results presented fall below thresholds considered ‘good’ responsiveness by Terwee et al (2006) among others, we nonetheless feel our criteria is objective and defensible on theoretical and statistical grounds. Finally, our study has by no means assessed a comprehensive set of instruments used to measure health status in opioid-dependent patients.  Comprehensive comparative 221  evaluations should be conducted to identify the best available health status measures or inform development of new measures for use in populations of opioid-dependent patients. 7.3 Further Study  In many ways the studies described above are merely starting points for further in-depth research.  Some of this follow-up research is already underway, and more is planned. The datasets and analytic methods employed in chapters 2-4 will be used in a series of planned analyses that will comprise my postdoctoral training, scheduled to begin at the UCLA Integrated Substance Abuse Programs following degree completion.  The dataset used in chapters 2 and 4 was linked to data on health resource utilization (hospital separations, MSP files) and mortality (Vital Statistics).  Through this linkage, we aim to (i) study the specific effects of policy changes on aspects of treatment delivery and outcome, (ii) procure complete data on emergency department utilization and the capital and human resource burden, (iii) examine effects of benzodiazepine and other drug utilization on the health and treatment outcomes of methadone clients, (iv) determine differences in characteristics and outcomes of special subgroups including HIV- and Hepatitis C infected clients, aboriginals and those involved with the criminal justice system, (v) identify safe and effective dose tapering trajectories for patients voluntarily withdrawing from treatment, (vi) evaluate the economic costs and benefits of potential changes in policies regarding treatment provision, and (vii) analyse factors associated with and causing mortality.  222  Further analysis of the drug price data analyzed in study (3) could apply the latent class growth analysis methodology used in study (5) to identify trajectories of change in purity, nominal and real drug prices across geographic regions within the US, which could perhaps shed light on drug distribution patterns.  Further consideration of the implications of the rational addiction model, in particular, the rate of time preference, or an individual’s propensity to trade future for present utility, are being explored in a burgeoning new field called ‘neuroeconomics’.  This field aims to better understand decision-making behaviour at the biological level that economists have studied since the early stages of the discipline, and could only be conducted in applied settings, working alongside neurologists and other health professionals.   Longitudinal studies are required to not only identify associations of brain functioning with drug use severity and elicited rates of time preference, but also to understand how time preference changes from adolescence to adulthood, and from casual to chronic use.  Finally, through studies (4) and (5), we have made the case that (a) health status measures are important evaluative tools in measuring the benefits of substance abuse treatment, and (b) better disease-specific measures of health status need to be developed in the field of substance abuse.  The latter can be achieved through continued assessment and testing alongside clinical trials in substance abuse. 7.4 Conclusions            Taken together, the results of these novel empirical studies suggest that while methadone maintenance can be an effective treatment option for opioid dependence, the 223  individual and program-related factors that limit its effectiveness must be addressed in order to maximize the public health benefits of this treatment.  224  7.5 References  Caulkins JP, Pacula RL, Arkes J, Reuter P, Paddock S, Iguchi M, Riley J (2004). The Price and Purity of Illicit Drugs: 1981 Through the Second Quarter of 2003.  Executive Office of the President Office of National Drug Control Policy. Washington, D.C., 2004. http://www.rand.org/multi/dprc/projects/drug_price_purity.html Horowitz JL (2001).  Should the DEA’s STRIDE data be used for economic analyses of markets for illegal drugs?  Journal of the American Statistical Association, 96(456): 1254-1262.     225  APPENDIX A:  MEDICAL PROFITEERING: THE ECONOMICS OF METHADONE DISPENSATION6  A1. Editorial On September 9th, 2008, stories in Canadian print and television media revealed that certain Downtown Eastside Vancouver pharmacies were paying patients up to $10 per day to pick up their methadone prescriptions at their pharmacy.  While one media outlet enlisted the help of an addict with a hidden camera to break the story, physicians prescribing methadone in Vancouver have long complained that their patients were being manipulated by some pharmacies (Tomlinson, 2008).  In a neighbourhood with a high density of opioid addicts, it should come as no surprise that methadone dispensation has become a competitive market.  After all, a pharmacy can be viewed like any other competitive firm whose objective is to maximize profits.  In the case of methadone, the profit motive is particularly strong.  In Canada, methadone is usually dispensed daily in community pharmacies and ingested under pharmacist supervision (known as daily witnessed ingestion), unlike most other drugs which are dispensed monthly or quarterly.  While the same product is dispensed everywhere, competition occurs on several associated dimensions. Some differentiate their product by offering psychosocial care, most often in the context of a specialized treatment clinic, sometimes charging an additional monthly fee. Alternatively, some  6 A version of this chapter has been published.  Nosyk B, Anis AH.  Medical Profiteering: the case of methadone dispensation.  Canadian Medical Association Journal. 2009; 180(11):1093-4.  226  compete on price by sharing with the patient a portion of the $16.60 they receive in dispensing fees from the provincial government health plan.  Although geographic proximity increases competition, even a methadone-dispensing pharmacy with no natural competitors may find that offering monetary incentives increases their revenue.  There are both positive and negative consequences to this market-driven competitive behaviour.  On the positive side, financial incentives may improve retention in methadone treatment.  Voucher or prize-based contingency management strategies have been tested with some success in methadone maintenance treatment programs in the U.S. and are in essence based on the same principle (Stitzer and Petry, 2006).  What is troublesome, though, is how this financial incentive is spent. An observational study found that untreated opioid dependent patients can spend as much as 77% of their income on illicit drugs (Hutchinson et al, 2000), much of which is procured through drug dealing and other forms of acquisitive crime.  While methadone eliminates withdrawal symptoms and blocks the euphoric rush experienced with heroin, other illicit drugs can provide qualitatively different, though comparably intense chemical euphoria.  Seminal studies on opioid dependent patients entering methadone maintenance treatment suggested that 47% were also cocaine users (Ball and Ross, 1991).  In a large prospective cohort study of injection drug users in Vancouver, crack cocaine use was independently associated with enrolment in a methadone maintenance program (Kerr et al, 2005).  227  Just as firms behave rationally to maximize profits, the economic theory of rational addiction suggests illicit drug users behave rationally as well, albeit short- sightedly (Becker and Murphy, 1988).  That is, they make consumption choices that maximize their current-value “utility”, or well-being. While illicit drug use is harmful in the long-run, consuming an illicit drug can bring about positive utility in the short term, in the form of euphoria and avoidance of withdrawal symptoms. One empirical study found high rates of time preference (a measure of how present-oriented an individual is in their decision-making) among active illicit drug users (Bretteville-Jensen, 1999), suggesting that the long-term harms can be outweighed by the immediate benefit of drug consumption.  Therefore, polydrug-dependent individuals unmotivated for treatment may adopt a strategy of managing their opioid withdrawal with methadone and effectively substituting the desired euphoric effects of heroin with crack cocaine.  The income freed by eliminating the need to purchase illicit heroin can be used for crack or other illicit drugs.  Therefore, financial incentives offered by pharmacies may add to the amount of income available for illicit drugs and serve to enhance the appeal of this strategy.  This is not to say that providing any form of economic transfer to opioid addicts is necessarily negative. The first and best solution for providing economic transfers to any disadvantaged group is to provide unrestricted cash payments, much like current welfare payments.  However, if the secondary goal of providing transfers is also to encourage rehabilitative treatment, then, all else being equal, tying transfers to the actual consumption of methadone is the preferred solution.  If, on balance, the money paid to drug users is being used to fund the purchase of illicit cocaine, then the net effect of this 228  policy may be negative. These payments could produce more problem cocaine use and crime and would constitute an abuse of methadone maintenance treatment that, once publicly known, would undermine public support for providing and funding the treatment.  Further study is therefore needed to determine the effect of pharmacy payments as they occur in naturalistic settings.  Regardless of the competitive behaviour displayed by some pharmacies, it must be emphasized that making methadone maintenance treatment as widely available as possible is a socially beneficial strategy. A methodologically rigorous simulation study found that an additional $1 spent on increasing access to treatment yields $76 in discounted lifetime benefits (Zarkin et al, 2005). Deregulation and increases in the number of prescribing physicians have coincided with a 240% increase in the number of patients in treatment since 1996. Thus, economic incentives for community pharmacies to continue to dispense methadone must remain in order to maintain widespread access  Alternative strategies for providing methadone maintenance treatment exist. Office-based treatment is provided in countries such as the UK and Switzerland. In the U.S., methadone is allowed only in opioid treatment programs, which face capacity constraints, shorter hours of operation and, crucially, high human resource demands, resulting in higher marginal costs.  If one accepts the assumption that the marginal cost of using physicians or opioid treatment programs to provide methadone under a daily witnessed ingestion program are higher, then both consumer surplus (utility gain) and 229  producer surplus (pharmacy profits), and thus social welfare, is maximized with a pharmacy-based daily witnessed ingestion program.  Our recommendation is therefore to continue with the current pharmacy-based model. Any policy changes that can limit or decrease access to methadone maintenance treatment should be discouraged. Recognition of the economic incentives of all parties involved is imperative in understanding the illicit drug use phenomenon and devising effective strategies to reduce the social burden it imposes. 230  KEY POINTS: • Recent media reports have exposed pharmacists providing kickbacks to methadone clients. • Price competition among methadone-dispensing pharmacies should be understood as natural behaviour of profit-maximizing firms operating in a competitive market. • The net effect of these payments need not be negative, given the potential improvements in treatment retention. • Given the societal benefits from increasing access to treatment, policy changes that can limit or decrease access to methadone maintenance treatment should be discouraged.   Declaration of Competing Interests: Neither of the authors has any relevant competing interests to declare.  231  A2. References  Ball JC, Ross A.  The effectiveness of methadone maintenance treatment.  New York (NY): Springer-Verlag; 1991.  Becker GS, Murphy KM.  A theory of rational addiction. Journal of Political Economy 1988;96:675-700.  Bretteville-Jensen AL.  Addiction and Discounting.  J Health Econ 1999;18:393-407.  Hutchinson SJ, Gore SM, Taylor A, Goldberg DJ, Frischer M.  Extent and contributing factors of drug expenditure of injectors in Glasgow. Multi-site city-wide cross-sectional study.  Br J Psychiatry 2000;176:166-72.  Kerr T, Marsh D, Li K, Montaner J, Wood E. Factors associated with methadone maintenance therapy use among a cohort of polysubstance using injection drug users in Vancouver.  Drug Alcohol Depend 2005;80:329-35.  Stitzer M, Petry N.  Contingency management for treatment of substance abuse.  Annu Rev Clin Psychol 2006;2:411-34.  232  Tomlinson K.  Pharmacy paid addict kickback: hidden camera.  Available at: http://www.cbc.ca/canada/british-columbia/story/2008/09/07/bc-prescription-for- profit.html?ref=rss&loomia_si=t0:a16:g2:r1:c0.135526  Zarkin GA, Dunlap LJ, Hicks KA, Mamo D. Benefits and costs of methadone treatment: results from a lifetime simulation model.  Health Econ 2005:14;1133-50.   233  APPENDIX B: BC PHARMANET DATABASE CLEANING  B1. PharmaNet database: Methadone Dispensations  Our retrospective cohort studies on the prescription of Methadone for the treatment of opioid addiction relied upon the British Columbia PharmaNet administrative database.  Maintained by the BC College of Physicians and Surgeons, this database tracks prescription drug utilization for all residents of British Columbia.  The cohort for which data was requested was defined as all patients who had ever received a prescription of Methadone for for the treatment of opioid addiction (Drug Identification Number: 999792 (prior to April, 2001) and 6999990/6999991/6999992/6999993 (following April, 2001)) between January 1st, 1996, and December 31st, 2006.  Despite the high level of control in procuring and maintaining MMT records in the PharmaNet database, errors may still arise.  In most cases, these errors can be considered non-differential outcome and/or exposure measurement error, which results in the attenuation of parameter estimates towards the null hypothesis (ie. odds ratios, or hazard ratios not statistically significantly different from one) (Copeland et al, 1977). We sought to identify and, where possible, correct these errors, to produce the most consistent and accurate methadone dispensation records possible.  As methadone is a controlled substance that can be diverted and sold in the illicit drug market, pharmacies dispensing methadone to patients for opioid addiction are 234  required to follow additional procedures in maintaining detailed records of all prescriptions dispensed (Pharmacy Methadone Maintenance Guide, 2007). A unique record is created for each drug dispensation. Our dataset consisted of ten variables, while an additional fourteen were constructed to facilitate the process of cleaning the dataset (table B1).  Table B.1: PharmaNet variables and constructed variables Variable Description Variable Name PharmaNet variables   1. De-identified ID variable based on Provincial Health Number id   2. Latest birth date birthdt   3. Latest sex sex   4. Date of service (ie. date of dispensation) dtt   5. Drug identification number  din   6. Days supply dayst   7. Quantity (mg/ml) qtyt   8. De-identified prescriber ID presid   9. De-identified pharmacy ID pharmid   10. Local Health Area* lha Constructed variables   11. Date of service (t-2) dtt-2   12. Date of service (t-1) dtt-1   13. Date of service (t+1) dtt+1   14. Date of service (t+2) dtt+2   15 Days supply (t-2) dayst-2   16. Days supply (t-1) dayst-1   17. Days supply (t+1) dayst+1   18. Days supply (t+2) dayst+2   19. Quantity (t-2) qtyt-2   20. Quantity (t-1) qtyt-1   21. Quantity (t+1) qtyt+1   22. Quantity (t+2) qtyt+2   23. Error indicator variable (0-5) error   24. Corrected date of service newdt   25. Corrected days supply  newdays   26. Corrected quantity newqty * LHA’s with fewer than 10 patients receiving treatment were merged with other lha’s based on geographic proximity and similarity in household income (BC stats. 2005).  235  In most cases, pharmacies maintain their own computer inventory systems for MMT, where patient records (id, date of birth, sex, prescriber id, pharmacy id) and service dates for each dispensation are generated automatically, and days and quantity fields are altered according to whether or not the patient appeared, and whether the entire dose was ingested. When a patient appears, (s)he signs in, receives his/her daily dose (or a take-home, carry dose, in special circumstances) then returns the cup and any remaining liquid.  Pharmacists observe the process, offer water to ensure that the patient has in fact ingested the drug, destroy any non-ingested liquid and enter the final quantity ingested either in a pharmacy database which is linked to the broader PharmaNet database, or into the PharmaNet database directly.  This suggests that only the days and quantity fields were susceptible to human error.  Our efforts therefore focused on correcting cases in which these fields appeared to be erroneous. All error checks and other data manipulation were executed in consultation with representatives of the BC Methadone program, of the BC College of Pharmacists.  Eight distinct types of human error were identified, pertaining to the days carried and quantity fields.  Two periods of lags and leads of the service date, days and quantity fields were constructed to aid in the identification and resolution of database errors. Comprehensive descriptions of the errors, along with resolutions were stated in table A2. A SAS Macro was created to capture and sequentially correct all errors described in table B2; following each error correction, the next error correction would be made on the newly corrected data.  All data cleaning was executed using SAS version 9.1. 236  Table B.2: Erroneous methadone dispensation records and resolutions Error Type Condition Resolution 1. dayst = 0 10 dayst-1  =  dtt  - dtt-1 11 0 < dayst-1 <  dtt  - dtt-1 (qtyt/(qtyt-1/dayst-1)) ≤  (dtt+1- dtt) newdays = (qtyt/(qtyt-1/dayst-1)) 12 dayst+1  =  dtt+2  - dtt+1 13 0 < dayst+1 < dtt+2  - dtt+1 (qtyt/(qtyt+1/dayst+1)) ≤  (dtt+1- dtt) newdays = (qtyt/(qtyt+1/dayst+1)) 14 (qtyt/(dtt+1 - dtt)) = (qtyt+1/(dtt+2-dtt+1))  15 (dtt+1 - dtt) ≤ 30 (qtyt/(dtt+1 - dtt)) ≥ 0.5 newdays = (dtt+1 - dtt) 2. dayst ≠  (dtt+1 – dtt) (21) (qtyt/(dtt+1 - dtt)) = (qtyt-1/dayst-1) (qtyt/(dtt+1 - dtt)) = (qtyt-2/dayst-2) = (qtyt+1/(dtt+2 -dtt+1)) (22) (qtyt/(dtt+1 - dtt)) = (qtyt-2/dayst-2) = (qtyt+1/dayst+1) (qtyt/(dtt+1 - dtt)) = (qtyt+1/(dtt+2-dtt+1)) = (qtyt+2/(dtt+3-dtt+2)) (qtyt/(dtt+1 - dtt)) = (qtyt+1/dayst+1) = (qtyt+2/ dayst+2) (qtyt/(dtt+1 - dtt)) = (qtyt+1/(dtt+2-dtt+1)) = (qtyt+2/ dayst+2) (23)  (qtyt/(dtt+1 - dtt)) = (qtyt+1/dayst+1) =(qtyt+2/(dtt+3-dtt+2)) (24) (dtt+1 – dtt) < dayst = qtyt (qtyt-1/dayst-1) ≥ 5 newdays = (dtt+1 - dtt) (qtyt-1/dayst-1) ≠ (qtyt/dayst) (41)  (dayst-1 + dayst) = dtt+1 - dtt-1  newdt = (dtt-1 + dayst-1) 3. (qtyt/(qtyt-1/dayst-1)) is an integer (qtyt-1/dayst-1) = (qtyt+1/dayst+1) (qtyt/(qtyt-1/dayst-1)) ≤30 (31) (qtyt/(qtyt-1/dayst-1)) is an integer  (qtyt-1/dayst-1) = (qtyt+1/(dtt+2 - dtt+1)) |qtyt/dayst - qtyt-1/dayst-1| > 10 |qtyt/dayst - qtyt+1/(dtt+2 - dtt+1)| > 10 (dtt-dtt-1) ≤ 30 (dtt+1-dtt) ≤ 30 (32) qtyt-1= qtyt newdays = (qtyt/(qtyt-1/dayst-1)) 237   (qtyt-1/dayst-1) =  (qtyt-2/dayst-2) (33) qtyt/(qtyt-1/dayst-1) is an integer (qtyt-1/dayst-1) =  (qtyt-2/dayst-2) (qtyt/(qtyt-1/dayst-1)) ≤30  qtyt = qtyt+1 (qtyt+1/(dtt+2 - dtt+1)) = (qtyt+2/(dtt+3-dtt+2)) (38) qtyt = qtyt+1 (qtyt+1/(dtt+2 - dtt+1)) = (qtyt+2 /dayst+2) (qtyt+1/(qtyt+1/dayst+1)) is an integer (qtyt+1/(dtt+2 - dtt+1)) = (qtyt+2/(dtt+3-dtt+2))  (39) (qtyt+1/(qtyt+1/dayst+1)) is an integer (qtyt+1/(dtt+2 - dtt+1)) = (qtyt+2 /dayst+2) newdays = qtyt/(qtyt+1/(dtt+2 -dtt+1))  4. dayst > dtt+1 – dtt (51) |(qtyt/dayst) – (qtyt-1/dayst-1)| > 10 |(qtyt/dayst) – (qtyt+1/(dtt+2 - dtt+1))| > 10 |(qtyt/(dtt+1 - dtt)) – (qtyt-1/dayst-1)| ≤10 |(qtyt/(dtt+1 - dtt))– (qtyt+1/(dtt+2 - dtt+1))| ≤ 10 newdays = (dtt+1 - dtt) After correction 5. dayst = 0  (60)   [none] 6. dayst > dtt+1 – dtt (61)  [none] (62) qtyt-1/dayst-1 = qtyt+1/dayst+1  [none] 7. |(qtyt/dayst)  –           (qtyt-1/dayst-1)| > 50     dtt  - dtt-1 ≤ 30  (63) |(qtyt/dayst) – (qtyt+1/dayst+1)| > 50 and [(qtyt-2/dayst-2) = (qtyt-1/dayst-1) OR (qtyt+2/dayst+2) = (qtyt+1/dayst+1)] [none] 8. (qtyt/dayst) > 500  (64)  [none]   238  A total of 11,031,480 MMT dispensations were recorded in the province of British Columbia between January 1st 1996 and December 31st 2006.  We found that only 4.3% of the records contained an error in the date, days carried and quantity fields.  Many of the error corrections depended on a stable dosing regimen (ie. preqty1/predays = qty/days = postqty/postdays1), therefore apparent errors in instances of variable dosing could not be corrected.  The remaining errors were corrected according to the resolutions described in table B2.  It is worthy to note that the majority of the errors occurred in < 5% of all the distinct patients in our cohort.  Lha’s with fewer than 10 patients receiving treatment were merged with other lha’s based on geographic proximity and similarity in household income.  239   Table B.3: Aggregated local health areas LHA Aggregate LHA code Local Health Areas 1 802 FERNIE / CRANBROOK / KIMBERLEY / WINDERMERE 2 802 FERNIE / CRANBROOK / KIMBERLEY / WINDERMERE 3 802 FERNIE / CRANBROOK / KIMBERLEY / WINDERMERE 4 802 FERNIE / CRANBROOK / KIMBERLEY / WINDERMERE 18 809 GOLDEN / REVELSTOKE 19 809 GOLDEN / REVELSTOKE 5 810 CRESTON / KOOTENAY / LAKE ARROW LAKES 6 810 CRESTON / KOOTENAY / LAKE ARROW LAKES 10 810 CRESTON / KOOTENAY / LAKE ARROW LAKES 12 812 GRAND FORKS / KETTLE VALLEY 13 812 GRAND FORKS / KETTLE VALLEY 23 823 CENTRAL OKANAGAN / SUMMERLAND 77 823 CENTRAL OKANAGAN / SUMMERLAND 25 825 100 MILE HOUSE / LILLOOET 29 825 100 MILE HOUSE / LILLOOET 28 828 QUESNEL / BURNS LAKE 55 828 QUESNEL / BURNS LAKE 30 831 SOUTH CARIBOO / MERRITT 31 831 SOUTH CARIBOO / MERRITT 49 849 BELLA COOLA VALLEY / UPPER SKEENA / NISGA'A 53 849 BELLA COOLA VALLEY / UPPER SKEENA / NISGA'A 92 849 BELLA COOLA VALLEY / UPPER SKEENA / NISGA'A 50 852 QUEEN CHARLOTTE / PRINCE RUPERT 52 852 QUEEN CHARLOTTE / PRINCE RUPERT 56 859 NECHAKO / PEACE RIVER SOUTH 59 859 NECHAKO / PEACE RIVER SOUTH 60 860 PEACE RIVER NORTH / FORT NELSON 81 860 PEACE RIVER NORTH / FORT NELSON 84 884 VANCOUVER ISLAND WEST / VANCOUVER ISLAND NORTH 85 884 VANCOUVER ISLAND WEST / VANCOUVER ISLAND NORTH 54 888 SMITHERS / TERRACE 88 888 SMITHERS / TERRACE  B2. PharmaNet Database: Other Drugs  Dispensations for drugs other than methadone for opioid dependence were not scrutinized and cleaned to the extent described above, as these fields were not used to create treatment episode lengths, mean dosage during treatment, etc.  The date of dispensation and drug identification number were used to identify whether or not a 240  specific drug was dispensed during a treatment episode.  The days prescribed field was used to include long prescriptions written before treatment initiation.  241  APPENDIX C:  MATHEMATICAL DEFINITIONS OF TREATMENT PHASE TRANSITIONS FOR ANALYSIS ON COMPLIANCE TO DOSING GUIDELINES  The BC Methadone dosing guidelines were summarized in table 5.1.  Methadone dispensation data was arranged in three ways to observe adherence to dosing guidelines and identify treatment patterns.  First, we calculated the daily dose per dispensation for all methadone dispensations in the dataset. Temporal trends in the distribution of daily doses were thus presented using all dispensation data.  Second, we arranged dispensation data by treatment episode – this distinction was used for the remainder of the analysis. Continuous retention in treatment was defined as no interruptions in prescribed doses lasting longer than 30 days. All treatment episodes beginning after January 31st, 1996 were considered in the analysis.  Episodes with mean doses <10mg or >400mg (indicating a high probability of coding errors; n=353 episodes (1.1%)) were excluded. The length of retention in treatment was calculated for each episode, and presented by calendar year of episode initiation.  Initial daily doses of all episodes were then summarized. First episodes with no dispensations in the past 5 years were considered episodes with unknown tolerance– the assessed date range for these attempts was therefore 2001-2006.  Furthermore, the length of each dispensation was considered in assessing adherence to carry dose guidelines by treatment episode.  242  Third, we arranged the data by week in treatment to consider treatment phase transitions among complete (non-censored) episodes.  The mean daily dose per week (weekly dose) was calculated (dt = (Σ drug quantity dispensed in week t) / (Σ days receiving treatment in week t)), with four lags and leads, covering a 9-week period.  Periods of titration, maintenance dosing, and tapering were identified by examining changes in weekly dose over the course of each non-censored episode. The end of the titration period was defined as the first week of a minimum of 4 weeks of non- increasing dosing. The maintenance period was defined as the time between the start date, or end of titration if observed, and the start of tapering (defined below), or episode end if no taper was observed.  Episodes with weekly doses of at least 60mg per day during the maintenance period were considered to reach the recommended maintenance dose stated in the BC guidelines.  The onset of the tapering period was then defined as the first instance after 12 weeks of treatment where the weekly dose decreased and remained at or below this lower level for at least four weeks.  We defined a taper interruption as the first instance after the start of a taper in which the weekly dose increased and remained at a higher level for at least 4 weeks.  Episodes with multiple taper attempts and reversals (variable dose episodes) were identified and summarized separately from episodes in which no more than one taper interruption was observed.  Taper periods in which the mean weekly dose decreased at ≤ 5% per week on average were considered in agreement with prescribing guidelines. Episodes with a final mean weekly dose ≤ 5mg were considered successful tapers.  243  In order to determine the adherence to prescription guidelines regarding the rate of dose titration, maintenance dosing, and the rate of dose tapering, we were required to explicitly define each of these phases of treatment.  Considering only non-censored treatment episodes observed in our database, the mean weekly dose was calculated (dt = (Σ drug quantity dispensed in week t) / (Σ days receiving treatment in week t)), with four lags (dt-1, dt-2, dt-3, dt-4) and leads (dt+1, dt+2, dt+3, dt+4), covering a 9-week period. Transitions across the three phases of treatment are defined explicitly below.  Titration End: The end of the titration period was defined as the first week of a minimum of 5 weeks of stable (non-increasing) dosing:  (1)  doset > doset-1; doset ≥ doset+1; doset ≥ doset+2; doset ≥ doset+3; doset ≥ doset+4.  This definition is displayed graphically in figure A1, caption (a).  Dose titration may also include downward adjustments in cases where too high a dose had been prescribed.  We thus constrained taper starts to begin only as early as week 12.  We then evaluated each case in which the taper start week was within 2 weeks of titration end to identify potentially misclassified titration end weeks; n=82 such cases were identified.  Of these cases, 13 decreased once and stabilized for at least four weeks thereafter (suggesting that 244  this later point of dose stabilization was the true end of titration).  Visual inspection of each case showed that either dose continued to decrease after four weeks, or the episode was discontinued at or before 4 weeks, suggesting that each of these cases was correctly classified.  Taper Start: The onset of the tapering period was defined as first instance, after the 12- week point, where the weekly dose decreased and remained at or below this lower level for at least four weeks:  (2)  doset < doset-1; doset < doset-2; doset < doset-3; doset < doset-4; doset ≥ doset+1; doset ≥ doset+2; doset ≥ doset+3; doset ≥ doset+4.; doset ≥ 12 weeks  This definition is displayed graphically in figure A1, caption b.  Taper Interruption: Often dose tapers were interrupted and reversed, increasing to a new stabilization dose. In some cases, subsequent taper attempts occurred.  Episodes with 245  multiple taper attempts and reversals, or a variable dosing pattern were identified and summarized separately from episodes in which no more than one taper interruption was observed.  We defined a taper interruption as the first instance after the start of a taper in which the weekly dose increased and remained at a higher level for at least 4 weeks:  (3)  doset > doset-1; doset-1 ≤ doset-2; doset-1 ≤ doset-3; doset-1 ≤ doset-4; doset-1 < doset+1; doset-1 < doset+2; doset-1 < doset+3; doset-1 < doset+4  246  Figure C.1: Definitions of treatment phase transitions         A)Titration End: doset > doset-1; doset ≥ doset+1; doset ≥ doset+2; doset ≥ doset+3; doset ≥ doset+4.  B)Taper Start: doset < doset-1; doset < doset-2; doset < doset-3; doset < doset-4; doset ≥ doset+1; doset ≥ doset+2; doset ≥ doset+3; doset ≥ doset+4.; doset ≥ 12 weeks  C)Taper Interruption: doset > doset-1; doset-1 ≤ doset-2; doset-1 ≤ doset-3; doset-1 ≤ doset-4; doset-1 < doset+1; doset-1 < doset+2; doset-1 < doset+3; doset-1 < doset+4  247  APPENDIX D: ANALYSIS OF STRIDE DRUG PRICE DATA  The key determinant of drug consumption, according to classical microeconomic theory, is price.  Drug price at the individual level were not available from the CA database, and adjustments have to be made for purity to obtain the price per pure gram. Therefore, purity adjusted drug prices will be obtained from the US Drug Enforcement Administration’s System To Retrieve Drug Evidence (STRIDE) database, which holds information on undercover drug purchases of DEA agents, as well as that of drug seizures.  Once retrieved, each individual drug purchase and seizure is tested for purity, and all data, including date of retrieval, drug identifiers and other fields, are entered into the STRIDE database. An estimated total of 55,947 observations including drug purities and purchase prices will be available for this analysis within the duration of data made in our request to the DEA (table 1). Given the geographic spread of the CA database (table 2), and the length of the study (1991-1997), sufficient spatial and temporal price variation should be achieved to produce a quarterly, city-specific, purity-adjusted price series for the majority (if not the entirety) of the CA database.  Sections A2.1 through A2.5 detail the sample construction, data cleaning, and price and purity econometric modeling procedures that will be employed in constructing this price series.  These procedures follow those prescribed by the authors of “The Price and Purity of Illicit Drugs: 1981 through the second quarter of 2003” Report, sponsored by the U.S. Office of National Drug Control Policy (2004).  248  D1. Constructing the Sample  a) Identify drug category: cocaine (drug category 620), heroin (610 & 905), methamphetamine (111), and marijuana (531). b) Check for duplicate records c) Restrict sample to include only observations from the US d) Drop observations in which information on state in which the acquisition occurred is missing e) Limit sample to observations where the method of acquisition was either a purchase (P), seizure (S), or lab seizure (L)  D2. Preliminary Data Cleaning  a) Delete observations with missing or zero amounts in ‘amount’ or ‘potency’ fields. b) Restrict sample to observations that were measured in grams c) Delete observations in which ‘purity’ field was missing, or value>100 d) Aggregate within drug categories drug forms that were physically similar, and, hence, likely to be indistinguishable to a buyer:  i) Cocaine: 9041L000 (crack), 9041L005 (powder), 9041L900 (cocaine, salt undetermined.  9041L900 reflects cocaine observations that are too small to analyze chemically (exclude 8.8%).  ii) Heroin: 9200.005 (heroin hydrochloride), heroin base (9200.000), Domestic Monitoring Program (9DMP.000), salt undetermined (9200.900).  Latter category: 249  Mexican black tar observations – these form a different series; also some of the DMP observations are dropped. e) Delete gross outliers:  i) Weight <0.1 g (7% of sample)  ii) Separate observations in which price was missing (75.3% decrease)  iii) Delete observations which were outside the distribution of realistic prices for 1 gram (not adjusted for purity).  Nominal price or inflation-adjusted (real) price. (0.5% decrease).  Cocaine: N: <$3.00 per gram; R: <$2.00 per gram, R: >$3,000.00 per gram. Heroin: N: <$3.00 per gram, R: <$7.50 per gram, R: >$10,000 per gram.  D3. Additional Cleaning Related to Proper Model Estimation  a) Delete all crack cocaine observations prior to 1986 (too sparse: only 137 observations). b) To limit the amount of artificial volatility on the time series, for each drug model, a cell (particular quarter or year) had to have at least five observations to be included in the model. c) Remove observations with zero purity.  It is possible to generate a predicted purity for all zero-purity observations, allowing them to be brought back into the analysis for estimating the final price model; omission of zero-purity observations influenced only the sample of observations predicting the purity model. d) Exclude extreme residuals (3.7% of sample).  Process described later.  250  Resulting sample of 131,184 observations: starting sample for estimation of price/purity models for Jan 1 1981- May 1 2003 dataset (23 years).  Our subset: Jan 1 1989 – Dec 31 1998: 10 years: 10/23= 0.43.  Therefore we expect a total sample of 0.43*131,184 = 56,409 (Actual: H, PC, CC: 55,947 (table 1)) - AMP: 5,023 - MJ: 1,874 - Heroin: 11,435 - Powder Cocaine: 18,900 - Crack Cocaine: 19,176.  Note: care needs to be taken in dealing with duplicate observations and correcting erroneous date codes (section 5, page 11, technical report).  D4. Define Key Variables D4.1 Dependent Variables  a) Price model: inflation-adjusted price.  Adjust for inflation using quarterly CPI for all urban consumers (1982-1984 base year). b) Purity model: drug purity.  For powder cocaine, crack and heroin: two-stage estimation technique: 1) purity of drug estimated, 2) predicted value of purity observations (expected purity) is entered into RHS of price model. 251   D4.2 Independent Variables  a) Amount of drug involved in purchase (weight in grams) o Separate observations for each drug into three/four quantity levels based on amount involved in transactions. o  Specify quantity ranges on amounts unadjusted for purity – this reduces the likelihood that valid rip-offs are misclassified. [table 3, page 15, technical report] b) Date seized (to identify year/quarter) construct 40 dichotomous indicators representing specific quarter/years from 1989Q1 to 1998Q4. c) Location: dichotomous indicators for 29 metropolitan areas and an ‘other’ category that encompassed the rest of the country outside the stated cities.  Other category is separated into 9 separate census subdivisions: Pacific, Mountain, North West Central, East North Central, West South Central, East South Central, South Atlantic, Mid Atlantic, New England (38 geographically distinct areas) D5. Econometric Models for Price and Purity  Expected Purity Hypothesis: The empirical model for price is based on the assumption that it is the buyer’s perception of purity at the time of the transaction (not actual purity of the drug) that determines the price he or she is willing to pay.  This is due to the fact that drugs are experience goods, in that purchasers cannot readily assay the purity of the 252  drug prior to first purchase.  Therefore it is not actual purity of the drug that governs the negotiated price at the time of transaction, but rather the expected purity.  This results in low-purity purchase observations being retained in the analysis.  Also, it results in a two-step procedure where expected purity rather than actual purity is included in the price regression model.  D5.1 Purity Model Specification  (1)    Where timeij is a vector of dummy variables representing a year/quarter, AMTijk is the raw weight of the ith observation in city k at time j.  The terms γ0, γ1, γ2 are the overall mean estimates for the intercept, time and amount effects.  This is estimated as a general linear random effects model.  The estimates of expected purity generated from this model are then used in the second stage price model.  For cases in which the predicted was <0 or >100, prediction was modified to equal 0.5 or 99.5, respectively. 253   D5.2 Price Model Specification  (2)   (3)  The real price for observation i in period j and city k is estimated as a function of time effects, city effects, and the natural logarithm of expected pure grams (ln(AMT) + ln(expected purity) = ln(AMT*expected purity)).  This model is estimated as a hierarchical generalized linear model with a log link function, a gamma error structure and a constant coefficient of variation.  Also, in (3), λ is the coefficient of variation of the real price.  To reduce the potential influence of gross outliers, the model is estimated, and the residuals from the model are kept, standardized and plotted.  Observations that fell beyond 3.09 were deleted (prob of deleting good data is set to 0.002).  Then continue 254  process of re-estimating model and deleting residuals until no gross outliers remained. (most models lost 3-5% of observations).  255   Table D.1: No. of purchase observations in STRIDE database by quarter, drug type    Powder Cocaine Crack Cocaine Heroin  I II III IV T I II III T I II III T 1989Q1 79 138 377 237 831 282 76 72 430 66 47 53 166 1989Q2 86 85 365 203 739 407 103 79 589 56 33 59 148 1989Q3 82 87 291 253 713 315 116 102 533 85 54 79 218 1989Q4 90 98 236 184 608 214 56 55 325 58 39 71 168 1990Q1 95 109 226 175 605 290 88 82 460 102 49 58 209 1990Q2 47 63 190 135 435 205 87 64 356 113 28 69 210 1990Q3 75 77 231 170 553 338 117 95 550 88 45 57 190 1990Q4 69 67 250 157 543 244 125 75 444 100 25 48 173 1991Q1 143 118 334 235 830 329 167 142 638 173 55 46 274 1991Q2 70 85 305 304 764 342 167 199 708 181 67 61 309 1991Q3 47 96 318 287 748 327 115 181 623 155 69 69 293 1991Q4 46 76 266 231 619 271 148 149 568 83 32 40 155 1992Q1 69 88 278 236 671 236 219 186 641 125 51 53 229 1992Q2 48 62 190 152 452 209 152 133 494 116 39 77 232 1992Q3 66 59 223 221 569 216 243 207 666 92 43 72 207 1992Q4 45 44 160 98 347 178 178 165 521 131 38 53 222 1993Q1 44 47 99 74 264 148 161 129 438 172 67 55 294 1993Q2 51 53 141 66 311 138 195 159 492 156 71 58 285 1993Q3 65 54 143 96 358 105 177 166 448 173 99 118 390 1993Q4 58 50 110 78 296 126 192 136 454 134 61 48 243 1994Q1 50 55 139 92 336 111 217 189 517 196 93 67 356 1994Q2 30 51 148 99 328 122 250 177 549 173 83 79 335 1994Q3 55 45 166 147 413 173 327 269 769 192 84 86 362 1994Q4 28 46 139 106 319 86 304 268 658 173 85 68 326 1995Q1 30 63 136 96 325 92 369 297 758 197 79 104 380 1995Q2 24 44 103 108 279 126 238 205 569 223 92 76 391 1995Q3 34 55 105 104 298 195 216 189 600 202 104 102 408 1995Q4 47 36 113 99 295 127 207 180 514 187 68 54 309 1996Q1 28 50 120 111 309 111 294 254 659 214 93 62 369 1996Q2 46 55 140 167 408 142 303 304 749 224 111 99 434 1996Q3 31 55 158 189 433 132 302 306 740 187 99 93 379 1996Q4 24 46 150 148 368 182 348 264 794 197 92 73 362 1997Q1 29 77 145 114 365 198 350 332 880 280 90 87 457 1997Q2 55 71 98 88 312 190 322 252 764 248 119 106 473 1997Q3 52 79 154 169 454 187 325 402 914 218 99 95 412 1997Q4 41 57 155 156 409 98 271 317 686 52 41 79 172 1998Q1 35 56 171 145 407 103 338 392 833 314 142 99 555 1998Q2 34 68 163 192 457 162 312 398 872 210 140 141 491 1998Q3 52 70 188 174 484 161 339 357 857 160 124 105 389 1998Q4 44 55 147 159 405 196 333 301 830 202 120 100 422 Total 2144 2690 7571 6255 18660 7814 8847 8229 24890 6408 2970 3019 12397   256   Table D.2: Geographical decomposition: STRIDE, CA datasets STRIDE database CA database Atlanta,GO (NA) Baltimore, MD (NA) Boston, MA (NA) Buffalo, NY (NA) Chicago, IL (NA) Cleveland, OH Dallas, TX (NA) Denver, CO Denver, CO Detroit, MI Detroit, MI Houston, TX Houston, TX Kansas City, MO (NA) Los Angeles, CA Long Beach, CA** Miami, FL Miami, FL, Collier County, FL*** Milwaukee (NA) Minneapolis-St. Paul, MN (NA) New Orleans, LO New Orleans, LO New York, NY New York, NY Newark, NJ (NA) Philadelphia, PA Philadelphia, PA Phoenix, AZ Tucson, AZ; Flagstaff, AZ# Pittsburgh, PA (NA) Portland, OR Portland, OR San Antonio, TX San Antonio, TX San Diego, CA (NA) San Francisco, CA Oakland-Richmond, CA## Seattle, WA (NA) Saint Louis, MO Saint Loius, MO Tampa, FL (NA) Washington DC (NA) Pacific Anchorage, AL Mountain (NA) West North Central (NA) East North Central Dayton-Columbus, OH* West South Central (NA) East South Central Lexington, KY South Atlantic Raleigh-Durham/Wake County, NC Mid Atlantic (NA) New England Hartford, CN^ * Approx 200km from Cleveland, OH ** Approx 35km from Los Angeles, CA (within LA county) *** Approx 100km from Miami, FL # Approx 200km from Phoenix, AZ ## Approx 20km from San Francisco, CA ^ Approx 150km from Boston, MA (different state) Pacific: Alaska, Hawaii, Washington, Oregon, California Mountain: Arizona, Idaho, Montana, Colorado, New Mexico, Utah, Nevada, Wyoming West North Central: N. Dakota, S. Dakota, Nebraska, Kansas, Minnesota, Iowa, Missouri East North Central: Wisconsin, Illinois, Michigan, Indiana, Ohio 257  Table D.3: Results on fixed effects from random coefficient regression models on the purity and real price of crack cocaine  Crack             Stage 1:  Purity Stage 2: Real price Cocaine: Estimate S.E. t-value Effect Estimate S.E. t-value Intercept 0.884 0.011 82.520 Intercept 4.459 0.181 24.676 Amount -0.007 0.000 -15.340 ln(EPA)* -0.256 0.048 -5.297 Y91Q3 -0.021 0.013 -1.570 Y91Q3 -0.047 0.140 -0.333  -0.003 0.009 -0.370  -0.130 0.181 -0.719 Y92Q1 -0.018 0.013 -1.380 Y92Q1 -0.033 0.130 -0.252  -0.076 0.015 -5.260  0.130 0.125 1.035  -0.074 0.017 -4.400  0.126 0.133 0.948  -0.023 0.015 -1.530  0.102 0.133 0.764 Y93Q1 -0.063 0.019 -3.270 Y93Q1 0.021 0.374 0.056  -0.065 0.015 -4.260  0.115 0.132 0.871  -0.072 0.014 -5.240  0.066 0.149 0.442  -0.063 0.012 -5.290  0.075 0.150 0.500 Y94Q1 -0.051 0.014 -3.550 Y94Q1 0.039 0.176 0.224  -0.071 0.014 -5.060  0.022 0.171 0.127  -0.070 0.012 -5.900  0.030 0.140 0.216  -0.062 0.010 -6.240  0.007 0.153 0.046 Y95Q1 -0.069 0.011 -6.080 Y95Q1 -0.005 0.125 -0.040  -0.104 0.011 -9.290  -0.005 0.127 -0.041  -0.162 0.022 -7.340  -0.022 0.138 -0.159  -0.129 0.016 -8.290  0.046 0.158 0.291 Y96Q1 -0.097 0.018 -5.430 Y96Q1 0.062 0.145 0.428  -0.134 0.014 -9.740  -0.022 0.148 -0.146  -0.124 0.017 -7.500  0.018 0.135 0.130  -0.093 0.017 -5.410  0.072 0.102 0.708 Y97Q1 -0.118 0.013 -8.870 Y97Q1 -0.023 0.143 -0.158  -0.207 0.027 -7.740  0.117 0.156 0.752  -0.159 0.018 -8.910  0.134 0.146 0.919  -0.120 0.014 -8.270  0.132 0.166 0.794 Y98Q1 -0.121 0.013 -9.220 Y98Q1 0.077 0.133 0.579  -0.125 0.012 -10.830  0.093 0.164 0.565  -0.136 0.016 -8.480  0.066 0.148 0.444   -0.168 0.013 -12.770   0.064 0.185 0.347 * ln(EPA)=natural log of expected pure amount: log(E(purity)*amount      258  Table D.4: Results on fixed effects from random coefficient regression models on the purity and real price of powder cocaine  Powder Stage 1: Purity   Stage 2: Real price Cocaine: Estimate S.E. t-value Effect Estimate S.E. t-value Intercept 0.654 0.021 31.180 Intercept 4.327 0.130 33.322 Amount 0.001 0.002 0.489 ln(EPA)* -0.278 0.087 -3.212 Y91Q3 -0.003 0.034 -0.099 Y91Q3 0.065 0.150 0.434  0.042 0.036 1.183  0.086 0.159 0.540 Y92Q1 0.009 0.039 0.230 Y92Q1 -0.083 0.161 -0.518  -0.019 0.037 -0.511  -0.006 0.175 -0.034  0.004 0.033 0.106  -0.018 0.159 -0.111  0.030 0.030 0.979  0.073 0.153 0.474 Y93Q1 0.008 0.035 0.220 Y93Q1 -0.082 0.198 -0.413  0.040 0.030 1.344  0.051 0.182 0.279  0.013 0.036 0.367  -0.091 0.169 -0.538  -0.002 0.034 -0.059  -0.004 0.188 -0.020 Y94Q1 0.052 0.038 1.353 Y94Q1 -0.048 0.174 -0.275  0.005 0.039 0.140  0.023 0.192 0.122  -0.004 0.027 -0.150  0.019 0.207 0.090  -0.019 0.045 -0.411  0.003 0.193 0.015 Y95Q1 0.066 0.027 2.416 Y95Q1 0.037 0.159 0.232  -0.032 0.031 -1.023  -0.083 0.168 -0.493  -0.047 0.033 -1.412  0.057 0.175 0.328  -0.041 0.031 -1.332  0.038 0.179 0.213 Y96Q1 0.025 0.028 0.906 Y96Q1 0.246 0.177 1.391  -0.002 0.026 -0.067  0.085 0.154 0.554  0.038 0.027 1.420  0.206 0.171 1.206  0.111 0.029 3.805  0.036 0.201 0.180 Y97Q1 0.017 0.038 0.463 Y97Q1 0.087 0.176 0.496  -0.035 0.032 -1.105  0.075 0.168 0.443  0.005 0.036 0.137  0.010 0.159 0.066  0.037 0.029 1.288  0.153 0.171 0.895 Y98Q1 0.060 0.037 1.615 Y98Q1 -0.045 0.164 -0.275  0.062 0.030 2.042  0.103 0.176 0.584  0.011 0.031 0.343  0.002 0.168 0.010   0.056 0.035 1.582   0.342 0.200 1.713 * ln(EPA)=natural log of expected pure amount: log(E(purity)*amount  259   Table D.5: Results on fixed effects from random coefficient regression models on the purity and real price of heroin  Heroin: Stage 1: Purity   Stage 2: Real price   Estimate S.E. t-value Effect Estimate S.E. t-value Intercept 0.263 0.027 9.769 Intercept 4.132 0.175 23.570 Amount 0.007 0.003 2.385 ln(EPA)* -0.250 0.060 -4.164 Y91Q3 0.006 0.027 0.222 Y91Q3 0.050 0.131 0.383  0.020 0.022 0.921  -0.050 0.155 -0.324 Y92Q1 0.051 0.028 1.783 Y92Q1 0.045 0.145 0.307  0.101 0.023 4.333  0.111 0.129 0.860  0.109 0.033 3.337  0.181 0.174 1.041  0.207 0.044 4.671  0.098 0.204 0.482 Y93Q1 0.129 0.043 2.980 Y93Q1 0.294 0.198 1.483  0.268 0.037 7.285  0.244 0.236 1.033  0.198 0.036 5.461  0.187 0.185 1.011  0.205 0.048 4.298  0.183 0.205 0.891 Y94Q1 0.147 0.037 3.941 Y94Q1 0.193 0.207 0.934  0.133 0.036 3.709  0.174 0.214 0.812  0.221 0.031 7.217  0.243 0.192 1.270  0.204 0.037 5.571  0.293 0.197 1.492 Y95Q1 0.208 0.027 7.633 Y95Q1 0.196 0.183 1.070  0.253 0.026 9.855  0.160 0.191 0.836  0.190 0.030 6.400  0.246 0.194 1.265  0.173 0.032 5.413  0.203 0.209 0.973 Y96Q1 0.119 0.033 3.570 Y96Q1 0.065 0.172 0.380  0.098 0.034 2.875  0.036 0.189 0.192  0.151 0.047 3.212  0.139 0.193 0.717  0.189 0.035 5.461  0.129 0.203 0.637 Y97Q1 0.219 0.031 7.011 Y97Q1 0.076 0.188 0.406  0.162 0.030 5.366  0.131 0.184 0.712  0.243 0.037 6.582  0.374 0.201 1.861  0.224 0.042 5.277  0.314 0.246 1.277 Y98Q1 0.245 0.039 6.216 Y98Q1 0.153 0.198 0.774  0.257 0.033 7.695  0.116 0.194 0.598  0.185 0.029 6.439  0.149 0.239 0.625   0.187 0.052 3.576   -0.287 0.242 -1.187 * ln(EPA)=natural log of expected pure amount: log(E(purity)*amount  260  APPENDIX E: HEALTH STATUS MEASURES USED IN CHAPTERS 5 AND 6  Health status measures collected longitudinally within the North American Opiate Medication Initiative are attached below in the following order:  1. Euroqol EQ-5D 2. Euroqol EQ-VAS 3. Short Form SF-6D 4. Maudesley Addiction Profile – Physical Health Score 5. Maudesley Addiction Profile – Mental Health Score 6. WHODAS -12 questionnaire 7. Addiction Severity Index: Medical status section (from with the ASI Medical Composite Score is constructed) 8. Addiction Severity Index: Psychiatric status section (from with the ASI Psychiatric Composite Score is constructed)  261   262   263   The Short Form 6D: This survey asks for your views about your health. This information will help you keep track of how you feel and how well you are able to do your usual activities. Answer every question by selecting the answer as indicated. If you are unsure about how to answer a question, please give the best answer you can. Please circle the level that you feel represents your health in the following categories (ie. Physical Functioning, Role Limitations, Social Functioning, Pain, Mental Health, Vitality Level Physical Functioning Level Pain 1 Your health does not limit you in vigorous activities 1 You have no pain 2 Your health limits you a little in vigorous activities 2 You have pain but it does not interfere with your normal work (both outside the home and housework) 3 Your health limits you a little in moderate activities  3 You have pain that interferes with your normal work (both outside the home and housework) a little bit 4 Your health limits you a lot in moderate activities  4 You have pain that interferes with your normal work (both outside the home and housework) moderately 5 Your health limits you a little in bathing and dressing  5 You have pain that interferes with your normal work (both outside the home and housework) quite a bit 6 Your health limits you a lot in bathing and dressing 6 You have pain that interferes with your normal work (both outside the home and housework) extremely  Role limitations  Mental health 1 You have no problems with your work or other regular daily activities as a result of your physical health or any emotional problems 1 You feel tense or downhearted and low none of the time 2 You are limited in the kind of work or other activities as a result of your physical health 2 You feel tense or downhearted and low a little of the time 3 You accomplish less than you would like as a result of emotional problems 3 You feel tense or downhearted and low some of the time 4 You are limited in the kind of work or other activities as a result of your physical health and accomplish less than you would like as a result of emotional problems 4 You feel tense or downhearted and low most of the time   5 You feel tense or downhearted and low all of the time   Social functioning  Vitality 1 Your health limits your social activities none of the time 1 You have a lot of energy all of the time 2 Your health limits your social activities a little of the time 2 You have a lot of energy most of the time 3 Your health limits your social activities some of the time 3 You have a lot of energy some of the time 4 Your health limits your social activities most of the time 4 You have a lot of energy a little of the time 5 Your health limits your social activities all of the time 5 You have a lot of energy none of the time   264        265                          WORLD HEALTH ORGANIZATION  DISABILITY ASSESSMENT SCHEDULE  WHODAS II         12-Item Interviewer Administered       266   This instrument was developed by the WHO’s Assessment, Classification and Epidemiology Group within the framework of the WHO/NIH Joint Project on Assessment and Classification of Disablements. The International Task Force members who contributed to the development of this instrument include: Elizabeth Badley Canada  Ron Kessler  USA Karen Ritchie France  Robert Trotter  USA Srinivasa Murthy India  Michael Von Korff USA Charles Pull  Luxembourg  Robert Battjes  NIDA Hans Hoek  Netherlands  Bennett Fletcher  NIDA Durk Wiersma  Netherlands Bridget Grant  NIAAA Martin Prince  UK  Cille Kennedy NIMH The WHO team: Somnath Chatterji  Jayne
Lux
 Shekhar Saxena Patrick Doyle Christopher Nelson T. Bedirhan Üstün JoAnne Epping-Jordan Jurgen Rehm Matilde Leonardi Ritu Sadana  Field Trial Centers:  Thomas Kugener Austria  Hans Hoek Netherlands Kruy Kim Hourn  Cambodia  Bisi Odejide Nigeria Yao Guizhong China  José Luis Segura García Peru Jesús Saíz Cuba  Radu Vrasti  Romania Venos Mavreas Greece  José Luis Vazquez Barquero Spain Srinivasa Murthy  India, Bangalore   Adel Chaker Tunisia Hemraj Pal  India, Delhi  Berna Ulug  Turkey Ugo Nocentini  Italy  Martin Prince  UK Miyako Tazaki  Japan  Ron Kessler  USA Elie Karam  Lebanon  Katherine McGonagle USA Charles Pull  Luxembourg Michael Von Korff  USA  The proper use of this instrument requires appropriate training of interviewers including use of the WHO-DAS II Interviewer’s Training Manual and Interview Guide.  The computerized version of the interview (I shell) is available for computer-assisted interviews or for data entry.  Informant (proxy) and self-administered versions of this instrument are available for field-testing.  Permission to translate this instrument into any language should be obtained from WHO.  All translations should be prepared according to the WHO translation guidelines.  For additional information, please contact: Dr T. Bedirhan Üstün Group Leader Assessment, Classification and Epidemiology Group World Health Organization CH – 1211 Geneva 27 Switzerland Tel:  + + 41 22  791 3609 Fax: + + 41 22 791 4885 Email: ustunb@who.ch   267   PREAMBLE  SAY TO RESPONDENT: The interview is about difficulties people have because of health conditions. (HAND FLASHCARD #1 TO RESPONDENT). By health condition I mean diseases or illnesses, other health problems that may be short or long lasting, injuries, mental or emotional problems and problems with alcohol or drugs.  I remind you to keep all of your health problems in mind as you answer the questions. When I ask you about difficulties in doing an activity think about (POINT TO FLASHCARD #1).  Increased effort Discomfort or pain Slowness Changes in the way you do the activity  (POINT TO FLASHCARD #1).  When answering, I’d like you to think back over the last 30 days.  I also would like you to answer these questions thinking about how much difficulty you have, on average over the past 30 days, while doing the activity as you usually do it.   (HAND FLASHCARD #2 TO RESPONDENT).  Use this scale when responding. (READ SCALE ALOUD): None, mild, moderate, severe, extreme or cannot do.   (FLASHCARDS #1 AND #2 SHOULD REMAIN VISIBLE TO THE RESPONDENT THROUGHOUT THE INTERVIEW. )    268  Screening questions in WHO DAS II  How much difficulty did you have in:   S1  Standing for long periods such as 30 minutes?   S2  Taking care of your household responsibilities?   S3  Learning a new task, for example, learning how to get to a new place?   S4  How much of a problem did you have joining in community activities (for example, festivities, religious or other activities) in the same way as anyone else can?   S5   How much have you been emotionally affected by your health condition?   S6   Concentrating on doing something for ten minutes?   S7  Walking a long distance such as a kilometre [or equivalent]?   S8   Washing your whole body?   S9   Getting dressed?  
S10
   Dealing with people you do not know?  
S11
   Maintaining a friendship?   S12  Your day-to-day work?    269   Health Conditions: Diseases, illnesses or other health problems Injuries Mental or emotional problems Problems with alcohol Problems with drugs    Having difficulty with an activity means:  Increased effort Discomfort or pain Slowness Changes in the way you do the activity    Think about the past 30 days only        270  Flashcard #1      1                                      2                                         3                                       4                                          5   None   Mild        Moderate   Severe        Extreme / Cannot Do  271           272    273  APPENDIX F: UBC BEHAVIOURAL RESEARCH ETHICS BOARD CERTIFICATES OF APPROVAL       UBC-Providence Health Care Research Institute Office of Research Services 11th Floor Hornby Site - SPH c/o 1081 Burrard St. Vancouver, BC V6Z 1Y6 Tel: (604) 806-8567 Fax: (604) 806-8568  ACKNOWLEDGEMENT LETTER  PRINCIPAL INVESTIGATOR: INSTITUTION / DEPARTMENT: UBC-PHC REB NUMBER: Aslam Anis  UBC/Medicine, Faculty of/School of Population and Public Health  H06-03372 SPONSORING AGENCIES: Canadian Institutes of Health Research (CIHR) PROJECT TITLE: An evaluation of long term health outcomes following discontinuation of Methadone Maintenance Treatment  This letter will acknowledge receipt of the following document(s) regarding the above study, which has/have been reviewed by the UBC-PHC Research Ethics Board Chair or Associate Chair: Study completion  DATE OF ACKNOWLEDGEMENT: January 28, 2010  Approval of the UBC-PHC Research Ethics Board or Associate Chair, verified by the signature of one of the following:    274  

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:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0069848/manifest

Comment

Related Items