Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Persistence and adherence with cardiovascular and lipid-lowering drugs following acute myocardial infarction… Pataky, Reka Elizabeth 2009

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

Item Metadata

Download

Media
24-ubc_2009_fall_pataky_reka.pdf [ 1.51MB ]
Metadata
JSON: 24-1.0070908.json
JSON-LD: 24-1.0070908-ld.json
RDF/XML (Pretty): 24-1.0070908-rdf.xml
RDF/JSON: 24-1.0070908-rdf.json
Turtle: 24-1.0070908-turtle.txt
N-Triples: 24-1.0070908-rdf-ntriples.txt
Original Record: 24-1.0070908-source.json
Full Text
24-1.0070908-fulltext.txt
Citation
24-1.0070908.ris

Full Text

PERSISTENCE AND ADHERENCE WITH CARDIOVASCULAR AND LIPID-LOWERING DRUGS FOLLOWING ACUTE MYOCARDIAL INFARCTION IN BRITISH COLUMBIA  by REKA ELIZABETH PATAKY B.Sc., The University of British Columbia, 2007  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Health Care and Epidemiology)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  September 2009  © Reka Elizabeth Pataky, 2009  Abstract BACKGROUND – Pharmaceutical use for the secondary prevention of cardiac events after acute myocardial infarction (AMI) is widespread, but there is uncertainty as to how (or if) patients use their medicines over long periods of time. Accordingly, the purpose of this thesis is 1) to measure persistence with and adherence to ACE inhibitors, beta blockers and statins, following AMI in BC, 2) to construct a conceptually-driven model of adherence and persistence, with patient demographic, socioeconomic, health status and pharmaceutical use variables, and 3) to determine whether regional variation in adherence and persistence rates exist. METHODS – Using administrative data from the BC Linked Health Database and PharmaNet, I studied a cohort of BC patients who were hospitalized for their first AMI between 2001 and 2005. I measured persistence as days to first 90-day gap in medication, and I measured adherence in two ways: the proportion of days covered (PDC) between first prescription and first 90-day gap, and PDC in the first year post-discharge (1-year PDC). Rates of persistence and adherence were analyzed using Cox proportional hazards models and multivariate logistic regression respectively. RESULTS – Patient persistence and adherence with medication was generally high, with 70% of ACE inhibitor users, 73% of beta blocker users and 78% of statin users persisting at one year. Nearly 89% of users of any class were persistent at 1 year, as opposed to 52% of users of all three concurrently. Factors consistently associated with high adherence and persistence were high income, private insurance, the use of more drug classes (both before and after AMI), and being in the mid-range of age (60-69 years). Sex had mixed effects between classes, with women having higher persistence and adherence with beta blockers and lower with ACE inhibitors. Some regional variation existed, but effects were small and inconsistent. CONCLUSION – Most AMI patients in BC use at least one drug for several years after AMI, but few persist with all three recommended classes. Important next steps include determining the clinical outcomes of adherence and persistence, especially with drug combinations, to more clearly define optimal secondary prevention practices following AMI.  ii  Table of Contents Abstract ........................................................................................................................................................ ii Table of Contents ...................................................................................................................................... iii List of Tables ............................................................................................................................................... v List of Figures............................................................................................................................................. vi List of Abbreviations ................................................................................................................................ vii Acknowledgements..................................................................................................................................viii Chapter 1: 1.1  Introduction ........................................................................................................................ 1  Research Objectives ................................................................................................................... 4  Chapter 2:  Literature Review ............................................................................................................... 5  2.1  Introduction ................................................................................................................................ 5  2.2  Cardiovascular Disease and Acute Myocardial Infarction ................................................... 5  2.2.1  Burden of disease ............................................................................................................... 5  2.2.2  Treatment guidelines ......................................................................................................... 7  2.2.3  Treatment uptake .............................................................................................................10  2.3  Adherence and Persistence .....................................................................................................11  2.3.1  Definitions ........................................................................................................................11  2.3.2  Measurement and validation ..........................................................................................12  2.4  Adherence and Persistence with Cardiovascular Drugs .....................................................21  2.4.1  Factors associated with adherence and persistence ....................................................27  2.4.2  Outcomes of poor adherence and persistence ............................................................34  2.5  Conceptual Frameworks .........................................................................................................36  Chapter 3: 3.1  Study Design .............................................................................................................................46  3.1.1 3.2  Methods.............................................................................................................................46 Data sources .....................................................................................................................46  Cohort Definition ....................................................................................................................47  3.2.1  Index AMI ........................................................................................................................47  3.2.2  Inclusion and exclusion criteria .....................................................................................48  3.3  Study Variables .........................................................................................................................49  3.3.1  Outcome variables ...........................................................................................................49  3.3.2  Covariates ..........................................................................................................................55  3.3.3  Censoring variables ..........................................................................................................60 iii  3.4  Analytic Plan .............................................................................................................................61  Chapter 4: 4.1  Results................................................................................................................................63  Study Cohort .............................................................................................................................63  4.1.1  Cohort construction ........................................................................................................63  4.1.2  Cohort characteristics ......................................................................................................63  4.2  Persistence and Adherence Rates ..........................................................................................67  4.2.1  Persistence.........................................................................................................................67  4.2.2  Adherence .........................................................................................................................70  4.2.3  Adherence as PDC in first year .....................................................................................71  4.3  Factors Associated with Persistence and Adherence ..........................................................72  4.3.1  Model construction..........................................................................................................72  4.3.2  Multivariate analyses of persistence ..............................................................................73  4.3.3  Multivariate analyses of adherence ................................................................................77  4.3.4  Multivariate analyses of adherence as PDC in first year ............................................81  4.4  Summary ....................................................................................................................................85  Chapter 5: 5.1  Discussion .........................................................................................................................89  Overall Persistence and Adherence .......................................................................................89  5.1.1  Any class vs. all 3 classes.................................................................................................90  5.1.2  Measurement techniques ................................................................................................92  5.2  Key Factors Influencing Persistence and Adherence .........................................................94  5.2.1  Drug class and sex ...........................................................................................................95  5.2.2  Health status and related variables ................................................................................96  5.2.3  Pharmaceutical use patterns ...........................................................................................98  5.2.4  Regional variation ......................................................................................................... 100  5.3  Strengths and Limitations .................................................................................................... 101  5.4  Conclusions ............................................................................................................................ 103  References ............................................................................................................................................... 105  iv  List of Tables Table 2.1: Summary of recent published studies of adherence and persistence with cardiovascular medicines ................................................................................................................24 Table 4.1: Characteristics of study cohort, according to drug use following AMI .........................66 Table 4.2: Age-adjusted use rates of any drug and all 3 drugs following AMI, by health services delivery area ......................................................................................................................................67 Table 4.3: Percent of drug users persistent with medication at 6 months and at 1-year time intervals, by drug class and combination, from Kaplan-Meier persistence estimates. ..........68 Table 4.4: Percent of drug users persistent at end of follow-up (i.e. censored by end of study period, Dec. 31, 2006) .....................................................................................................................70 Table 4.5: Adherence, measured as the proportion of days covered (PDC) from first prescription to first 90-day gap in treatment, by class and combination .......................................................71 Table 4.6: Adherence, measured as the proportion of days covered (PDC) in the first 365 days of follow-up, by class and combination........................................................................................72 Table 4.7: Factors associated with experiencing a gap in treatment lasting more than 90 days, by drug class ...........................................................................................................................................75 Table 4.8: Factors associated with experiencing a gap in treatment lasting more than 90 days, by drug combination .............................................................................................................................76 Table 4.9: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered from first prescription to first 90-day gap in therapy, by drug class ....................................................................................................................................................79 Table 4.10: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered from first prescription to first 90-day gap in therapy, by drug combination ......................................................................................................................................80 Table 4.11: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered in the 365 days following first prescription (1-year PDC), by drug class ...........................................................................................................................................83 Table 4.12: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered in the 365 days following first prescription (1-year PDC), by drug combination .............................................................................................................................84 Table 4.13: Summary of findings of multivariate analyses, by outcome measure, drug class and combination ......................................................................................................................................86 Table 4.14: Summary of regional variation in multivariate analyses, by outcome measure, drug class and combination .....................................................................................................................88  v  List of Figures Figure 2.1: Summary of published values of persistence with ACE inhibitors, beta blockers and statins .................................................................................................................................................23 Figure 2.2: The WHO's 5 dimensions of adherence [26]....................................................................37 Figure 2.3: Osterberg and Blaschke‟s barriers to adherence [2] .........................................................40 Figure 2.4: Andersen-Newman behavioural model of utilization, adapted by Phillips [57] ..........41 Figure 2.5: Conceptual framework of factors influencing adherence and persistence with medication, among a population with identified treatment needs............................................43 Figure 3.1: Inventory model of drug supply (in days) and daily drug availability ............................50 Figure 3.2: Persistence, measured as time to first gap in drug availability exceeding 90 days. ......52 Figure 3.3: Adherence, measured as proportion of days with drug available, from first prescription to first gap in drug availability exceeding 90 days .................................................53 Figure 3.4: Adherence, measured as the proportion of days with drug available in the 365 days following the first prescription (1-year PDC) ..............................................................................54 Figure 3.5: Map of Health Services Delivery Area (HSDA) boundaries in British Columbia, from BC Stats, July 2008 [66]. ..................................................................................................................56 Figure 4.1: Flowchart of cohort construction ......................................................................................64 Figure 4.2: Venn diagram of ACE inhibitor, beta blocker and statin use following AMI .............64 Figure 4.3: Kaplan-Meier (product-limit) estimate of persistence, measured as days to first 90day gap in treatment, by drug class ...............................................................................................69 Figure 4.4: Kaplan-Meier (product-limit) estimates of persistence, measured as days to first 90day gap in treatment, by drug combination .................................................................................69  vi  List of Abbreviations 4S  Scandinavian Simvastatin Survival Study  ADG  Ambulatory diagnostic group  AMI  Acute myocardial infarction  ACE  Angiotensin-converting enzyme  ARR  Absolute risk reduction  ATC  Anatomical Therapeutic Chemical  BC  British Columbia  BCLHD  BC Linked Health Database  CRA  Canada Revenue Agency  CI  Confidence interval  CIHI  Canadian Institute for Health Information  CCHS  Canadian Community Health Survey  CCORT  Canadian Cardiovascular Outcomes Research Team  CHSPR  UBC Centre for Health Services and Policy Research  HPS  Heart Protection Study  HR  Hazard ratio  HSDA  Health services delivery area  ICD  International Classification of Diseases  ISPOR  International Society for Pharmacoeconomics and Outcomes Research  LHA  Local health area  MEMS  Medication event monitoring system  MSP  Medical Services Plan of BC  NNT  Number needed to treat  OR  Odds ratio  PDC  Proportion of days covered  RCMP  Royal Canadian Mounted Police  RR  Relative risk  SD  Standard deviation  WHO  World Health Organization  vii  Acknowledgements I learned a great deal while carrying out this thesis research and participating in the MSc. program in the School of Population and Public Health, and I am very grateful to have had the opportunity to do this work. I would like to sincerely thank my supervisor, Dr. Steve Morgan, for his guidance, support and patience through every step of developing, planning, conducting and finally presenting my research project. I would also like to thank my committee members, Drs. Charlyn Black and Barbara Mintzes, for their valuable insight, helping me to clarify my ideas while expanding my perspective on the project. Working at CHSPR, in a continuously supportive environment, has been a pleasure. I would like to thank the members of the Pharmaceutical Policy group, in particular Gillian Hanley, Lixiang Yan and Dr. Mike Law, for their help with this project. I am especially indebted to Gillian, my classmate and mentor for the past two years. Her ability to make sense of administrative datasets and her perseverance with seeming endless access requests were invaluable to my project. I would like to thank my peers in SPPH and the WRTC for making this a fantastic two years. Being surrounded by a network of friends who are always happy to tutor, commiserate, or just go for beers is the most educational part of higher education. Finally, special thanks to Tim, thesis or no thesis. Funding for my graduate studies was provided by the Canadian Institutes for Health Research and the Western Regional Training Centre for Health Services Research.  viii  Chapter 1: Introduction The use of pharmaceuticals in the outpatient setting is rising, especially for management of chronic diseases such as hypertension, high cholesterol, diabetes and asthma. From 1998 to 2007, spending on antihypertensive drugs in Canada rose 5.1% per year, to a total of $2.7 billion, while spending on statins rose 10% per year to $1.9 billion, with most of this growth being driven by the increased volume of drug use [1]. Although the rapid growth in drug use and expenditure is often framed as a problem and a threat to public insurance programs, it is not necessarily bad: if used appropriately, pharmaceutical treatments can effectively manage chronic diseases, improving the health of patients and reducing the burden on other aspects of the health care system. However, if these pharmaceutical therapies are not being used consistently or for sufficient periods of time, then they cannot work as intended. This may drive pharmaceutical expenditure, without a corresponding improvement in health outcomes or reduction in demand for other components of care. It is therefore important to understand how patients use their medicines, and to understand the outcomes of such use. A patient‟s ongoing use of a prescribed drug regimen can be quantified using two measures: adherence and persistence. Adherence describes the extent to which patients take their medicines as prescribed, with correct timing, dosage and frequency, and it is often measured as the proportion of days, in a specified interval, for which a patient has medication available to them [2, 3]. Persistence describes the duration of therapy, measured as the number of days before a patient discontinues therapy (or experiences an extended gap in therapy). It is  1  often reported in units of time or as the proportion of patients who have not discontinued therapy at a given point in time [3, 4]. In the case of acute myocardial infarction (AMI), evidence-based practice guidelines support the use of aspirin, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors and statins for the secondary prevention of cardiac events [5]. Use of these four drug classes after discharge from hospital is generally high and has increased rapidly over time. Among Ontario seniors from 1992 to 2005, for example, the post-discharge use of ACE inhibitors and beta blockers nearly doubled, from around 40% to 80%, and the use of statins increased nearly 20-fold, from around 4% to 80% [6]. However, there is uncertainty as to the degree to which AMI patients take medications appropriately in the months and years after discharge, and it is known that many patients frequently miss prescribed doses or discontinue therapy altogether. Rates of patient persistence with therapy for secondary prevention after AMI have been reported to drop quickly over time, with only 50% of ACE inhibitor users and 53% of beta blocker users continuing treatment for 2 years [7]. Published rates of persistence and adherence however vary across treatment categories and population groups, and comparisons between studies are difficult due to differences in the way that adherence and persistence are measured. Thus, my first research question is: 1. What are the rates of adherence and persistence with ACE inhibitors, beta blockers and statins following AMI in BC, and how do these rates vary using different measurement techniques? Many studies investigating individual-level variables, such as age, sex, income, and health status, have found moderate but often inconsistent associations with medication persistence and 2  adherence. The factors that appear most strongly associated with good adherence and persistence are those related to socioeconomic status or patterns of other prescription drug use [8, 9], but few studies have included such variables. Accordingly, in order to more thoroughly describe the context of adherence and persistence, my second research question is: 2. What individual-level factors, either demographic, socioeconomic, or those related to health status or past prescription drug use, are associated with improved adherence and persistence with these drug classes following AMI? Finally, previous work at UBC‟s Centre for Health Services and Policy Research (CHSPR) has indicated that utilization of pharmaceuticals, measured as age-specific prescription rates, varies across regions of BC, and is a large contributor to variations in drug expenditures [10]. Regional variation in adherence and persistence with medicines over time may be a factor contributing to regional variation in overall use. My third research question is therefore: 3. After adjusting for individual-level factors, is there unexplained regional variation in adherence and persistence with drugs following AMI in BC? The following chapters contain the review of the literature I conducted to inform this project, the methods I used to address these questions, my results, and a discussion of my findings. It is my hope that by answering these questions I contribute to the state of knowledge on adherence and persistence with medication use following AMI, and potentially add to a more general understanding of the appropriateness of drug use for chronic disease management.  3  1.1 RESEARCH OBJECTIVES In summary, my objectives for this thesis are: 1. to measure adherence and persistence with ACE inhibitors, beta blockers and statins, alone and in combination, following first AMI for patients in BC, and to compare common measurement techniques, 2. to construct a conceptually-driven model of individual-level demographic, socioeconomic, health status, and pharmaceutical use variables to understand their relationships with adherence and persistence with these medicines, and 3. to determine whether regional variation in adherence and persistence rates exists, after adjusting for individual-level factors.  4  Chapter 2: Literature Review  2.1 INTRODUCTION In my thesis, my goal is to explore the appropriateness of pharmaceutical use for the long term management of chronic disease, specifically by investigating the factors associated with adherence and persistence with ACE inhibitors, beta blockers and statins following acute myocardial infarction (AMI) in BC. The purpose of this literature review is to clarify the reasoning behind my research question and hypotheses, and to place this thesis in context. To inform this study, I conducted a review of the literature in the areas of: 1) cardiovascular disease and AMI, 2) adherence and persistence definitions and measurement, and 3) adherence and persistence with pharmaceuticals following AMI. Lastly, I reviewed conceptual frameworks relating to adherence and health services utilization, and constructed a framework to underpin my study design and interpretation.  2.2  CARDIOVASCULAR DISEASE AND ACUTE MYOCARDIAL INFARCTION  2.2.1  Burden of disease Cardiovascular disease, including cerebrovascular disease, AMI and other ischaemic  disease, is the leading cause of death in Canada [11, 12]. There have been significant improvements in cardiovascular disease-related mortality over time – for example from 1950 to 1999, age-standardized death rates dropped from 702 to 288 per 100,000 men, and from 562 to 175 per 100,000 women – but the burden of cardiovascular disease remains high [11]. Approximately 5% of adult Canadians, or approximately 1.3 million people, reported living with 5  cardiovascular disease in 2001, ranging from 2.0% among Canadians 40-49 years old to as high as 21.9% in Canadians aged 70-79 and 26.8% among those over the age of 80 [13]. Cardiovascular disease is a significant contributor to the growing burden of chronic disease, and its appropriate management is a key concern for health care systems in Canada and abroad. In addition to the high prevalence of cardiovascular disease, the risk factors for cardiovascular disease, including smoking, obesity, hypertension and dyslipidemia, are also highly prevalent. While smoking has decreased over time as the target of many public health interventions, other risk factors are increasing. According to the 2000/01 cycle of the Canadian Community Health Survey (CCHS), 26% of Canadians are current smokers, 13% report being diagnosed with hypertension and 4% with diabetes, almost 15% are obese, and over half report being physically inactive [14]. These factors in turn significantly increase the likelihood of cardiovascular disease, and recurring cardiovascular events. Among cardiovascular diseases, AMI is of particular interest. In 1999, AMI alone accounted for 10% of all deaths in Canada [11]. According to data from the Canadian Institute for Health Information (CIHI), almost 140,000 Canadians were admitted to hospital for their first AMI between fiscal years 97/98 and 99/00. The overall 30-day in-hospital mortality rate for these patients was 12.3%, and ranges widely by age and sex. In-hospital mortality was 1.6% for men and 3.1% for women 20-50 years old, rising to 22.2% for men and 24.4% for women over 75 years [15]. Trends in cardiovascular disease suggest that the incidence of AMI has remained relatively steady over time, but mortality has been decreasing [16]. This has led to a growing population of AMI survivors. In the 2000/01 cycle of the CCHS, 2.7% of men and 1.5% of women reported previously having a heart attack [11]. These survivors are at especially high risk of subsequent AMI or other cardiovascular events, presenting a growing population 6  health challenge. Within one year after AMI, 12.5% of AMI survivors are readmitted to hospital for angina, 7.7% are readmitted for a second AMI, and 7.5% are readmitted for congestive heart failure [15].  2.2.2  Treatment guidelines Improvements in short-term survival after AMI can largely be attributed to improved  acute care in hospitals, through the use of clinically effective interventions [17]. Evidence-based guidelines recommend the initiation of reperfusion therapies as quickly as possible after presentation to hospital and diagnosis of AMI, including angioplasty or the use of fibrinolytic drugs [5]. Canadian data suggest that the use of reperfusion therapy in hospitals ranges from 6070%; however, in-hospital intervention is only one component of long-term survival following AMI [17]. An important aspect of treatment after AMI, and the focus of my thesis project, is the prevention of future cardiac events using outpatient pharmaceutical treatments. The goal of pharmaceutical treatment following AMI is to reduce the risk of subsequent cardiac events, especially by addressing the key risk factors of hypertension and dyslipidemia. Unless contraindicated, patients who survive their first AMI should be prescribed a beta-blocker, an angiotensin-converting enzyme (ACE) inhibitor, a statin, and low-dose aspirin, according to American College of Cardiology/American Heart Association guidelines, among others [5]. Each of these agents targets a different aspect of cardiac disease with a different mechanism of action, and appear to be most effective when used combination; however, in this review I will not focus on aspirin, because it is available over-the-counter in British Columbia and elsewhere, and cannot be measured using administrative prescription claims databases. It is therefore excluded from my analysis. 7  Randomized clinical trials have shown that beta-blockers, ACE inhibitors and statins reduce the risk of reinfarction or death after AMI [18]. A large meta-analysis of beta blocker clinical trials, incorporating almost 25,000 subjects, indicated that beta-blocker treatment reduces total mortality by 23% (OR 0.77, 95% CI: 0.69-0.85), with an absolute risk reduction (ARR) of 1.2% and number needed to treat (NNT) of 84 for one year, to avoid one death. Beta blockers also reduced the risk of non-fatal reinfarction, with an ARR of 0.9% and NNT of 107 for one year [19]. Trials of statins, beginning with the Scandinavian Simvastatin Survival Study (4S) in 1994, have shown reductions in mortality and reinfarction [18]. The 4S trial, with 4444 participants with a history of angina or AMI, reported a reduction in total mortality of 30% (RR 0.70, 95% CI: 0.58-0.85), with an ARR of 3.3% and NNT of 30 for five years. The trial also found a 37% reduction in the risk of non-fatal reinfarction (RR 0.63, 95% CI: 0.54-0.73), with an ARR of 6.7% and NNT of 15 for five years to prevent a non-fatal coronary event [20]. The more recent Heart Protection Study (HPS), which combined secondary and primary prevention, had 20,536 participants, only 41% of whom had a history of AMI. The HPS reported a 13% reduction in mortality (RR 0.87, 95% CI: 0.81-0.94), with an ARR of 1.8% and NNT of 56 for five years, and 38% reduction in non-fatal AMI (RR 0.62, 95% CI: 0.54-0.70), with an ARR of 2.1% and NNT of 48 for five years [21]. ACE inhibitors have also been shown to cause a reduction in mortality, ranging from 1630% in large clinical trials, however their effect on reinfarction appears smaller than the other classes, with reductions ranging from 7-25% [18]. A meta-analysis of three studies of long-term ACE inhibitor use after AMI, with a total of 5966 patients, found a 26% reduction in mortality (OR 0.74, 95% CI: 0.66-0.83), ARR 5.7% and NNT of 18 for 30 months. There was also a 8  statistically significant reduction in non-fatal reinfarction (OR 0.80, 95% CI: 0.69-0.94), with an ARR of 2.4% and NNT of 4, for 30 months, to prevent one event [22]. Adverse events most frequently associated with ACE inhibitor use include hypotension (14.7% of treatment group, OR 1.86, 95% CI: 1.65-2.10) and renal dysfunction (5.2% of treatment group, OR 1.49, 95% CI: 1.23-1.79) [22]. Hypotension is also frequently reported among beta blocker users, as is brachycardia [19]. Adverse events potentially associated with statin use include muscle damage (indicated by elevated creatine kinase, muscle pain, or weakness) and in serious cases rhabdomyolysis [20, 21]. On balance, however, the potential benefits of these drug classes for secondary prevention of cardiac events following AMI outweigh the potential harms. Subgroup analyses within many of these trials have indicated that the beneficial effects observed for each drug class are present regardless of concomitant drug treatments; however, few studies, and it appears no randomized controlled trials, have been done to explicitly test for improvements in mortality or reinfarction rates – or increases in adverse events – using all three drug classes in combination [18]. This is especially troubling considering how widely these drug classes are used in combination, and how widely this combination is advocated. One casecontrol study investigated the impact of the number of preventive drug classes used (beta blockers, ACE inhibitors, statins or antiplatelet agents) on reinfarction following first AMI. The authors found that the use of only one drug class resulted in a non-significant 6% reduction in the odds of experiencing a subsequent AMI, while using three drug classes resulted in a 41% reduction (95% CI: 6%-63%) [23].  9  2.2.3  Treatment uptake Studies by the Canadian Cardiovascular Outcomes Research Team (CCORT) have  indicated that use of these drugs for secondary prevention of cardiac events has been increasing rapidly over time. One such study measured rates of medication use within 90 days of discharge post-AMI, using administrative data from public drug insurance plans for seniors in Ontario [6]. The investigators found that in the years from 1992 to 2005 use of cardiovascular drugs for secondary prevention increased dramatically, and began to plateau around 2002 and 2003. The post-discharge use of ACE inhibitors increased from 42.0% to 78.4% from 1992 to 2005, beta blockers increased from 42.6% to 78.1%, and statins increased from only 4.2% to 79.2%, [6]. A follow-up study of the same cohort found that older patients were less likely to use these drugs after AMI, and these drugs were more likely to be prescribed by a cardiologist (as an attending physician) than another specialty [24]. A recent report by the Health Quality Council of Saskatchewan shows a similar trend in that province. From fiscal year ‟01/‟02 to ‟05/‟06, ACE inhibitor use increased from 56.6% to 67.8%, beta blocker use increased from 56.5% to 69.3%, and statin use increased from 39.7% to 62.5%, at 90 days post-discharge following AMI [25]. These use rates are approximately 10 percentage points lower than those cited above, however these rates are for all AMI patients 20 years and older, not only for seniors. The Saskatchewan report also investigated use of drug combinations, and found that in „05/‟06 86.7% of patients had filled at least one prescription for an ACE inhibitor, beta blocker or statin by 90 days post-discharge. Only 13.3% had filled no prescriptions, while 41.9% were using all three classes [25]. Overall, the use of ACE inhibitors, beta blockers and statins immediately following AMI appears quite high. Given the high burden of cardiovascular disease and the efficacy of these 10  drugs as shown in clinical trials, these drugs can be highly effective for preventing secondary events on a population scale. As with any ongoing treatment for chronic disease, however, they must be used consistently over long periods of time, which is where understanding rates of adherence and persistence is key.  2.3  ADHERENCE AND PERSISTENCE  2.3.1  Definitions Broadly, adherence has been defined by the World Health Organization as, “the extent  to which a person‟s behaviour – taking medication, following a diet, and/or executing lifestyle changes, corresponds with agreed recommendations from a health care provider” [26]. With respect to medicines, adherence is more narrowly defined as the extent to which a patient takes medication as prescribed, with respect to timing, dosage and frequency [2, 3]. At the population level, it is most often measured as the proportion of days in a specified time interval for which the subject has medication available [4, 27, 28]. The term compliance is commonly used interchangeably with adherence, however to some health care providers compliance suggests a passive obedience of doctors‟ orders, rather than a mutually agreed-upon treatment plan. As concepts of care have become more patient-centered, there has been an increased recognition that patients‟ decisions about whether or how to take medications are more complex than simply choosing to follow doctors‟ orders, therefore I will use the term adherence in this thesis [2, 29]. Another important measure of medication use over time, closely related to adherence, is persistence. The term persistence is used to describe the duration of drug therapy, from initiation to discontinuation. It is often reported in units of time, for example the number of  11  days from first prescription to first extended gap in treatment, or as the proportion of patients persistent with medication at a given point in time [4, 30]. A challenge I encountered while conducting background research and synthesizing the literature is that the terminology and measures used in the field are inconsistent. The International Society for Pharmacoeconomics and Outcomes Research (ISPOR) has recently attempted to define standardized terms, but it remains to be seen whether or not their definitions are widely adopted [3]. It is important to recognize that adherence and persistence can be thought of two distinct constructs, measuring related but different aspects of medication use. The reasons a patient has for stopping drug treatment entirely may be very different than those for missing doses or experiencing gaps between refills, therefore it is important to clearly define what is being measured in each study.  2.3.2  Measurement and validation  MEASUREMENT TECHNIQUES There are two broad classifications of techniques to measure medication adherence and persistence: direct and indirect. Direct measures include observation of drug-taking, and laboratory detection of the drug, a metabolite of the drug, or another biologic marker in body fluids [31]. These direct methods are the only way to determine whether a patient actually took their drugs or not, but are expensive, cumbersome, and not feasible for most studies, especially studies with large sample sizes or studies conducted in an outpatient setting. They also involve significant interaction between the subject and researcher, and are unlikely to be representative of a patient‟s normal drug-taking behaviour [31]. For these reasons, indirect measures of adherence and persistence are much more common.  12  Indirect methods to measure adherence and persistence include patient self-report, pill counts, electronic monitoring and measures of drug availability, as estimated using pharmacy records. These methods are considered indirect because they rely on patient-generated or third party information for data on drug use or distribution, but cannot provide proof that a patient consumed their medication or took it exactly as indicated by their physician [31]. Patient self-report of medication-taking behaviour can be achieved through diaries, interviews, or the use of patient questionnaires, such as the well-validated Medication Adherence Survey (MAS) [31]. These methods are simple and economical, but are susceptible to information bias, through social desirability or recall bias, and often overestimate adherence [2]. Another relatively simple method of measuring adherence is through pill counts. Patients are asked to bring their medication with them when meeting with their care provider or a researcher, and the number of pills remaining are counted and compared to the number of days since the drug was dispensed and the amount of drug dispensed. However, this method is also likely to overestimate adherence, because patients may neglect to bring in medication not stored in the original container, or may discard medication before the meeting to intentionally mask nonadherent behaviour [31]. For measuring adherence in particular, electronic monitoring of medication-taking is a highly reliable method, and is unique in that it can be used to collect detailed information on dose frequency, dose intervals and dose timing. Electronic monitoring is most often conducted using a Medication Event Monitoring System (MEMS) that consists of a medication bottle cap, specially equipped with a computer chip that records the date and time of bottle opening [31]. A drawback of electronic monitoring however is that the MEMS must be used appropriately to collect usable data – if a patient moves medication to another pill bottle or removes multiple 13  doses at once, the data will be misleading [2]. The devices can also be expensive and cumbersome, and patients may change their behaviour when they are aware they are being monitored [31]. The last indirect method of measuring adherence and persistence is through measures of medication availability derived from pharmacy records. In closed pharmacy systems, for example in health maintenance organizations or in jurisdictions with public drug coverage, where all dispensed prescriptions are entered into a single computerized claims database, the rate at which patients fill their prescriptions over time can be used to estimate adherence and persistence with medication. By knowing the amount of drug dispensed to a patient and the interval between fills, any short gaps in drug availability or extended periods of discontinuation can be identified, and measures of adherence and persistence can be calculated [2]. The strengths of this method lie mostly in the data source being used. Secondary administrative data are generally inexpensive and easy to access, and can be used to study entire target populations or very large samples thereof at the individual level. The data can be completely anonymized for research purposes and do not typically require patient consent, so there is no information bias associated with patients knowing their drug-taking behaviour is being observed [31]. The records are also often comprehensive and accurate, and a number of Canadian databases have been previously validated for pharmacoepidemiological research [32-34]. This method is especially valuable for determining adherence to ongoing drug treatments for chronic disease, but cannot be used effectively for short-term treatments. It also cannot detect cases of primary non-adherence, where a patient does not fill a prescription in the first place, or instances where a patient may have filled a prescription outside the pharmacy system [31]. An important consideration when using this method, is that pharmacy refill data measures how frequently a patient filled a prescription, not necessarily how frequently they took their drug, so results using 14  this method provide more of an upper limit to the potential amount of drug consumed, rather than its actual value [35]. A final limitation of this method is that it provides a very high level estimate of adherence, and it can be difficult to identify contextual influences on adherence using data that was collected originally for administrative and not research purposes. Because there is no gold standard for measuring adherence and persistence with prescription drugs, validation of measurement methods can be difficult. Measures derived from pharmacy records have therefore been compared to a number of other adherence measures, from both direct and indirect methods, in an attempt to validate their use. In general, these studies have found moderate, but statistically significant, correlations between rates calculated with pharmacy data and other measures [28, 36, 37]. A 1997 review by Steiner and Prochazka identified a number of studies that attempted to validate adherence defined with pharmacy data using direct measures of adherence, including three studies that compared adherence to measured serum or urine drug levels, and five studies that compared adherence rates to physiologic drug effects, such as changes in blood pressure or pulse rate [28]. All three measures of serum or urine drug level were significantly correlated to adherence calculated with pharmacy data, but the strength of the association was moderate (r = 0.21 to r = 0.47). Similarly, four of the five studies comparing adherence and physiologic drug effect found significant associations between adherence and their respective measures of drug effect [28]. The Steiner review also identified five studies that compared adherence defined with pharmacy data to other indirect measures of adherence: four studies compared it with selfreported adherence (one of which also included a comparison to pill counts), and the final study compared it with patient appointment-keeping behaviour. The correlations between calculated 15  and self-reported adherence to medicines were varied. Two studies, one investigating adherence to all drugs and one focused on anti-epileptics, found statistically significant correlations between the two measures, while a third, investigating psychiatric drugs, and fourth, looking at antihypertensive drugs, did not [28]. In the one study that also included pill count data, the association between pill count and refill adherence for all prescribed drugs was reported to be strong (r = 0.68) and statistically significant. The association between refill adherence and appointment-keeping in a group of arthritis patients was found to be weak (r = 0.22) but statistically significant [28]. More recently, a Canadian study, validating adherence calculated with Manitoba prescription claims data, compared adherence calculated with pharmacy data to pill counts and matched their analysis by drug class. The two measures were found to be highly concordant [37]. In the time since the 1997 Steiner review, MEMS has also been used to validate adherence calculated from pharmacy data [28, 36]. A frequently-cited study by Choo et al validated adherence from pharmacy data, patient report and pill count with MEMS [36]. They reported high levels of adherence, as measured by all four methods, but only moderate correlation between them. Adherence calculated from pharmacy records was more weakly correlated with MEMS than pill count adherence was (r = 0.32 vs. r = 0.52) but both were statistically significant [36]. Overall, the results of tests of validity for adherence calculated from pharmacy data are mixed, and the correlations between it and other methods of measuring adherence are moderate. It is challenging to compare validation studies for adherence calculated from prescription claims because many of the studies designed to test it use different comparators, and target different 16  drugs and populations of interest, making them difficult to compare directly. For example, some studies focus on a single class of drugs, while others look at adherence to all drugs among patients with at least one prescription drug. The type of drug or number of concurrent drugs being taken may both be factors associated with either adherent or non-adherent behaviour, and may also influence which measurement methods are most accurate and appropriate for validation purposes. It is also important to consider that all of the methods of measuring adherence are best suited to different contexts and study designs. Pharmacy data, for example, are best used to measure adherence over a long period of time in large population-based observational studies, while other measurement methods are better for short-term use on a smaller scale, such as in clinical trials. Understanding the limitations of every method is also important. Adherence calculated with pharmacy data measures the rate at which a patient refills prescriptions, not necessarily the rate at which they take them, and it cannot measure certain aspects of adherence, such as primary non-adherence or non-adherence to dose timing. CALCULATING ADHERENCE Even using pharmacy claims data, rates of adherence can be calculated many different ways. The studies above, validating the method, most commonly use a measure of daily medication availability but many others exist, and are interpreted different ways [28]. Steiner identifies three parameters for each adherence measurement: the distribution of the adherence variable, either continuous or dichotomous; the number of refill intervals evaluated, either single or multiple intervals; and whether the measure assesses availability of treatment or gaps in treatment [28].  17  For measures of availability the days‟ supply of medication is counted in the numerator, while for measures of gaps the numerator is the number of days in the interval of interest minus days‟ supply. The two measures are complementary, except in cases where oversupply occurs and a patient has more days‟ supply of drug than there are days in the interval or interest [28]. Continuous measures of adherence are often expressed as a percentage representing the proportion of days in which a patient had (or did not have, for gap measures) access to drug treatment. Dichotomous measures of adherence on the other hand classify subjects as either “adherent” or “non-adherent” by establishing cutoff values for continuous measures of adherence, although these cutoff values are often arbitrary and not based on any clinical significance. Some dichotomous measures may also classify patients as non-adherent if they experience gaps in their treatment exceeding a predefined length [28]. Lastly, the number of intervals included in the analysis may impact adherence measurements. Adherence can be measured for single refill intervals or for periods with multiple refill intervals combined. The most frequently used measures of adherence are continuous measures of medication availability, calculated as the number of days with medication available, divided by a defined time interval. This measure is often called the medication possession ratio or the proportion of days covered (PDC) [27]. A recent review compared these measures and others reported in studies of adherence, and used the study authors‟ methods to determine adherence of a test sample using data from an unrelated clinical trial. The reviewers found that despite slight differences between the measures, for example differences in the inclusion of the date of last dispensation, most measures provided roughly the same values for adherence as a straightforward PDC [27]. The measure of adherence to use may depend largely on the research question being answered, however due to the similar results found using simple PDC and other more complex measures 18  of continuous, multiple-interval adherence, it seems that the simplest calculations of adherence using pharmacy refill data may be the best [27]. Although there is no single superior measure, the simpler the calculation used, the fewer variables required and the simpler the interpretation of results. CALCULATING PERSISTENCE Persistence, the time to discontinuation of therapy, can also be measured in a number of ways using prescription claims from pharmacy data, including time to a fixed gap in treatment, time to a variable gap in treatment, or prescription anniversaries. The most common method used is time to a fixed gap in treatment; however the gap lengths used also vary by study. Prescription anniversary methods measure whether or not subjects are considered to be persistent with medication at a given point in time, by determining if they filled a prescription at or around that time [30]. For example, to measure persistence after one year of follow-up, any patient who filled a prescription between 305 and 365 days from the start of observation would be classified as persistent [38]. This method is simple to use and understand, but does not measure drug availability or use between the anniversary time points. It therefore provides no information as to whether a patient used their medication continuously or not. Methods measuring days to a gap in therapy on the other hand do provide this level of detail, and enable calculation of the length of continuous medication use. These methods use measures of daily medication availability (similar to adherence calculations, above) to determine the number of days before a patient experiences a gap in therapy [4]. Often these methods incorporate a grace period to account for short gaps in drug availability that would not normally be considered discontinuation of treatment. This is frequently called the maximum permissible 19  gap. Permissible gaps can be defined as fixed gaps, usually multiples of 30 days, or variable gaps, defined as a proportion of the duration of the previous prescription; however, fixed permissible gaps are most common [4, 30]. Persistence measured as days to gap in therapy can be used to estimate the probability of persistence with medication at any point of follow-up using survival analysis techniques. The length of permissible gap can have a significant impact on measured persistence. A recent study used pharmacy claims data from Quebec to compare different measures of adherence and persistence in a group of statin users [39]. When the permissible gap was defined as 7 days, only 41% of patients were persistent at one year, and 18% were persistent at 3 years. By contrast, when the permissible gap was 90 days, persistence increased to 89% at 1 year and 79% at three years, because far fewer patients experienced a gap that long [39]. There are advantages and disadvantages associated with any length of permissible gap selected. The shorter the gap the more sensitive the measure of persistence is to breaks in therapy, but the less likely that the gap will represent a true discontinuation of drug use. In the above study, 94% of patients who experienced a gap lasting longer than 7 days resumed statin use later in the followup period. This proportion dropped substantially, to 59%, when the permissible gap was extended to 90 days [39]. In summary, many potential methods of measuring adherence to medication exist, and although an indirect measure, with a substantial separation between time and place of the measurement and the health behaviour of interest, pharmacy refill data is a valid way to estimate adherence. Adherence measured using pharmacy claims databases is somewhat correlated with other measures of adherence, including direct measures of adherence and physiological outcomes, and is therefore an appropriate technique to use, provided that its limitations and 20  other threats to validity are recognized. Many different ways exist to calculate adherence, and should be carefully chosen to suit the research question. Overall, using pharmacy refill information from large, population-based databases can be an effective way to estimate adherence to medication, and potentially broader medication use patterns, in large populations over time. Based on the heavy use of pharmaceuticals, especially in the area of chronic disease management, pharmacy refill data can be a valuable resource to better understand the use and effectiveness of medication based treatments.  2.4  ADHERENCE AND PERSISTENCE WITH CARDIOVASCULAR DRUGS Many studies of adherence and persistence with cardiovascular drugs have been  conducted recently, seemingly recognizing the growing burden of cardiovascular disease and the potential for improved adherence and persistence to reduce morbidity and mortality. However, despite this emphasis, the literature has not been well-synthesized, and comparison of adherence and persistence rates across studies remains difficult due to varying methods and outcome measures. One review of adherence and persistence with treatments for hypertension, dyslipidemia and diabetes, found that one-year persistence rates ranged widely, from 35% to 92% [40]. The overall mean proportion of days with medication available in the first year was calculated to be 0.72 (SD: 0.18), with 59% (SD: 0.19) of subjects having a PDC ≥0.80. It is clear from the literature that long-term adherence and persistence with medicines is imperfect, but the degree of the problem, and the factors influencing medication-taking behaviour, are not necessarily well understood. I chose 17 studies for my review of the literature relating to adherence and persistence with medication for cardiovascular disease [7-9, 38, 41-53]. These studies are summarized in  21  Table 2.1. I did not conduct a systematic review to identify these studies; instead I selected studies I thought were particularly relevant to my research questions, and had clear and thorough methods. I searched Medline (with PubMed) and Web of Science using combinations of the keywords: adherence, persistence, compliance, ACE inhibitor, beta blocker, statin, acute myocardial infarction and secondary prevention. As I identified papers I also searched through references and citing articles. I prioritized papers that used administrative data, but included other methods if they had long follow-up periods and clear definitions of adherence and persistence. I also took care to find studies from Canada. Most studies (12) were conducted with cohorts of patients following AMI, as opposed to new drug users for primary prevention. All studies but 3 used pharmacy claims data to calculate adherence and persistence, while the others used patient self-report. The studies I selected also investigated a wide range of factors potentially associated with adherence and persistence, described in the following section. Persistence rates reported in these studies varied, but not nearly as widely as in the review described above. The proportion of the population persistent at 6 months after initiation of treatment was most frequently reported, and ranged from 70-80% for ACE inhibitors, 7488% for beta blockers, and 65-87% with statins. Persistence rates appeared to decline over time, decreasing to 58-66% for ACE inhibitors, 48-53% for beta blockers and 35-66% for statins, after at least three years of follow-up. These persistence values are summarized graphically in Figure 2.1; however, these ranges should be interpreted with caution, as the definitions of persistence vary. The highest rates, for example, are from a study by Eagle et al, which used patient self-report of medication use from a survey conducted 6-months following AMI [44]. This may be an overestimate of persistence if patients lost to follow-up were less likely to be adherent (selection bias), or if respondents overstated their drug use (information bias). More common were persistence measures using prescription claims, defining discontinuation in 22  treatment as a gap of 60 or 90 days, as in the studies of Akincigil, Gislason, Perreault and Rasmussen. Only one study measured persistence with combinations of drugs. Through follow-up surveys one month after AMI, Ho et al found that 88% of subjects reported being persistent with at least one of a statin, beta blocker or aspirin, while 66% of subjects were persistent with all 3 classes [46].  Figure 2.1: Summary of published values of persistence with ACE inhibitors, beta blockers and statins  Eight of the studies I identified measured adherence, most commonly defined as the proportion of the population with at least 80% of days with medication available (PDC ≥0.80) for a given follow-up period, usually one year. Only one study reported adherence with ACE inhibitors, with 69% of subjects adherent in the first year [38]. For beta blockers, adherence rates ranged from 45-74% at 1 year. Adherence to statins appeared to be slightly higher, ranging from 62-81% in the first year, but according to one study the statin adherence rate dropped to under 50% after 5 years [41]. Instead of only focusing on adherent patients, with PDC over 0.80, a study by Rasmussen et al also identified patients with especially low adherence, with PDC under 0.40 [51]. They reported that 9% of beta blocker users and 6% of stain users have particularly low adherence in the year following AMI.  23  Table 2.1: Summary of recent published studies of adherence and persistence with cardiovascular medicines Author Akincigil A [7]  Blackburn DF [41]  Year  Post-AMI  2007  Yes  2005  Yes  Outcome  Definition  Drugs  Persistence  Gap lasting >60 days  ACE inhibitors  78% persistent at 6 mo, 68% at 1 yr, 50% at 2 yrs  Beta blockers  82% persistent at 6 mo, 72% at 1 yr, 53% at 2 yrs;  Female sex  Fill frequency (# of 30-day Rx/month) over 0.80  Statins  61.8% adherent at 1 yr, 48.8% at 5 yrs  Increasing age  Adherence  Result  Positive association  More physician visits Higher chronic disease score  No  Persistence  Gap lasting >90 days  Statins  46.2% persistent (after up to 7 yrs follow-up)  Brookhart MA [43]  2007  No  Adherence  # of 30-day Rx in first year  Statins  10% filled one Rx, 50% filled >10 Rx  Healthy behaviours – including mammography, PSA testing, flu vaccine  Eagle KA [44]  2004  Yes  Persistence  Reported use in follow-up survey (6 months post-AMI)  ACE inhibitors  80% persistent at 6 mo  Male sex  Beta blockers  88% persistent at 6 mo  AMI (vs. angina) as event  Statins  87% persistent at 6 mo  Gap lasting >90 days  ACE inhibitors  65.6% persistent at 5 yrs  Concomitant statin use  Beta blockers  48.3% persistent at 5 yrs  Female sex  Yes  Persistence  Later statin initiation (after ACE/beta blocker)  Increased ACE/beta blocker adherence  2007  2006  Hospitalization >8 days  High income neighbourhood  Brookhart MA [42]  Gislason GH [45]  Negative association  Increasing age  Increasing age Concomitant ACE inhibitor or statin use  Statins  Ho PM [46]  2006  Yes  Persistence  Reported use in follow-up survey (1 month post-AMI)  Beta blocker, statin and aspirin  65.6% persistent at 5 yrs  Female sex  66% persistent with 3 drugs; 88% persistent with at least one  Above high school education  Increasing age  Concomitant ACE inhibitor or beta blocker use Increasing age (interaction with sex) Increased comorbidity  24  Table 2.1: Summary of recent published studies of adherence and persistence with cardiovascular medicines (cont.) Author  Year  Post-AMI  Kramer JM [47]  2006  Yes  Lachaine J [48]  2006  Newby LK [49]  2006  No  No  Outcome Adherence  Adherence  Persistence  Definition PDC ≥0.75  PDC ≥0.80  Reported use in ≥2 consecutive annual surveys*  Drugs  Result  Positive association  Beta blockers  69% adherent at 30 days post-discharge; 45% at 1 yr  statins  59.8% adherent at 2 yrs  other CLAs  43.3% adherent at 2 yrs  ACE inhibitors  39% persistent (with heart failure), 26% (without)  Male sex  Beta blocker  46% persistent  Revascularization  Increasing age  Use of >1 other therapy History of hypertension  History of diabetes or smoking  Male sex  Increasing age  Statins  43% persistent  Commercial insurance  Negative association  Increasing age  Interaction between age, sex and insurance (younger women with commercial insurance)  History of hypertension  Revascularization Use of >1 other therapy Perreault S [8]  2005  No Yes  Persistence Persistence  Gap >60 days  Statins  Gap >60 days  Statins  65% persistent at 6 mo, 35% at 3 years  Increasing age  71% persistent at 6 mo, 45% at 3 years  Diagnosis of diabetes, hypertension or respiratory disease  Male sex  Primary prevention (vs. post-AMI cohort) Increasing # of daily doses  Use of ≥3 drug classes  ≥2 dispensing pharmacies  Rural environment  ≥3 prescribers  Hospitalization in follow-up Perreault S [50]  2005  No  Persistence  Gap lasting >60 days  ACE inhibitors  71% persistent at 1 yr, 58% at 3 yrs  Combination therapy  Male sex  Switching (between classes)  Social assistance  Beta blocker  68% persistent at 1 yr, 57% at 3 yrs  Increasing age  Increasing # of daily doses  Diuretics  61% persistent at 1 yr, 48% at 3 yrs  Use of ≥3 drug classes  Rural environment, Diagnosis of diabetes, dyslipidemia or respiratory disease  ≥2 dispensing pharmacies ≥3 prescribers  Hospitalization in follow-up * Persistence expressed as % of respondents, not of drug users only  25  Table 2.1: Summary of recent published studies of adherence and persistence with cardiovascular medicines (cont.) Author  Year  Post-AMI  Rasmussen JN [9]  2007  Yes  Rasmussen JN [51]  2007  Simpson E [38]  2003  Yes  Yes  Outcome  Definition  Drugs  Persistence  Gap lasting >90 days  Beta blockers, statins  not stated  PDC ≥0.80 or <0.40  Beta blockers  74% with high adherence, 9% with low, at 1 yr  Statins  81% with high, 6% with low at 1 yr  ACE Inhibitors  69% adherent at 1 yr  Beta blockers  72% adherent at 1 yr  CLAs  80% adherent at 1 yr  ACE Inhibitors  70% persistent at 1 yr  Beta blockers  74% persistent at 1 yr  CLAs  84% persistent at 1 yr  Adherence  Adherence  Persistence  PDC ≥0.80  Rx filled in days 305365 postdischarge  Result  Positive association  Negative association  Higher income Higher education Prior beta blocker or statin use  Increasing age Psychiatric illness Recurrent admission to hospital  Wei L [52]  2002  Yes  Adherence  PDC ≥0.80  Statins  64% adherent at study end  Female sex  Wei L [53]  2004  Yes  Adherence  PDC ≥0.80  Beta blockers  59% adherent at first hospitalization (or study end)  Prior beta blocker use  Increasing age Increasing age Female sex  Rx = prescription, PDC = proportion of days covered  26  2.4.1  Factors associated with adherence and persistence As indicated in Table 2.1, many studies have investigated factors associated with  adherence and persistence. I have divided these factors into 6 categories for discussion: demographic variables, socioeconomic status, health status and comorbidities, cardiovascular events, prescription drug use, and interaction with the health care system and providers. DEMOGRAPHIC VARIABLES The demographic variables age and sex appear to have inconsistent associations with adherence and persistence. Increased age is positively associated with adherence and persistence in the studies by Blackburn and Perreault, while the opposite is true in studies by Ho and Rasmussen, and the effect varies by drug class in the study by Gislason. What appears to be happening in these studies is an inverted U-shape relationship, with both the youngest and oldest patients having the lowest values for adherence and persistence. The studies by Perreault et al include only patients under the age of 65, and report that hazard of ceasing treatment with either stains or beta blockers decreases (HR 0.98, 95% CI: 0.98-0.99) with every one year increase in age [8, 50]. On the other hand, the study by Rasmussen includes only patients over the age of 66, and reports that the odds of having poor adherence with statins or beta blockers increases (OR 1.03, 95% CI: 1.02-1.04 and OR 1.02, 95% CI: 1.01-1.02, respectively) for each additional year of age [51]. Gislason et al categorize age into intervals, and their findings support the inverted U-shaped relationship, with the oldest patients having lower persistence and younger seniors (60-69 years) having higher persistence. Subjects over the age of 80 were significantly more likely to discontinue therapy with ACE inhibitors (HR 1.16, 95% CI: 1.0512.8) or statins (HR 1.40, 95% CI: 1.16-1.68) than the reference category (age 30-59), while  27  subjects between 60-69 years were significantly less likely to discontinue therapy with these classes (HR 0.86, 95% CI: 0.79-0.93 and HR 0.88, 95% CI: 0.82-0.94 respectively) [45]. The effect of sex also varies in these studies, but it appears to change according to drug class. In general, women appear to be most adherent and persistent with beta blockers, men are most adherent and persistent with ACE inhibitors, and the effects are mixed for statins. For example, Gislason et al report that men are more likely to experience a break in treatment than women for beta blockers (HR 1.17, 95% CI: 1.12-1.22) and statins (HR 1.14, 95% CI: 1.071.22), but not for ACE inhibitors [45]. Unfortunately, none of the studies I found appear to discuss why these differences might exist. Only two studies reported on age-sex interaction. Ho et al, investigating the likelihood of discontinuing of all medications after AMI, found that the effect of increasing age was greater in women (OR 1.77, 95% CI: 1.34-2.34) than in men (OR 1.23, 95% CI: 1.02-1.47), per 10-year increment, but they did not suggest reasons why this effect might exist. Sex alone was found to have no effect [46]. In Kramer‟s study of commercial insurance and adherence with beta blockers, the investigators tested for interaction between age, sex, and insurance status. While sex alone did not have an effect, they reported that among subjects with commercial insurance, younger women (age 35-64 years) were less likely to be adherent than younger men (also 35-64 years). This trend was reversed among older women with commercial insurance, who were more likely to be adherent than older men (although this effect was not statistically significant) [47]. Women tend to be older than men when they experience their first AMI, and have worse outcomes than men of the same age, at least in the short term [15]. These differences may explain in part the strong relationship between these two variables.  28  Apart from age and sex, region of residence is another demographic variable that may be associated with adherence and persistence with medication, although few studies have investigated this. Kramer et al found that in the United States, adherence with beta blockers was poorest in the Southeast, compared to the Northeast (OR 0.70, 95% CI: 0.63-0.77). This effect was also significant when adjusting for commercial insurance vs. Medicare. Neither the West nor the Midwest were significantly different [47]. Unfortunately, the regional divisions used by Kramer et al are very coarse, and do not provide much information pertaining to why this regional variation might exist. The studies by Perreault et al also touch on regional differences in adherence and persistence, by investigating the effect of rural vs. urban residence on statin or antihypertensive use. They report that patients in rural environments are somewhat less likely to discontinue treatment (HR 0.89, 95% CI: 0.85-0.94) [8]. SOCIOECONOMIC STATUS Patients with higher socioeconomic status, measured with income, education and access to private insurance, tend to be more persistent and adherent with their medication. One of the studies by Perreault et al in Quebec found that hypertension patients receiving social assistance were 34% more likely to have discontinued therapy at 1 year (OR 1.34, 95% CI: 1.26-1.42) than those without social assistance [50]. Similarly, Akincigil et al reported that beta blocker users from high-income ZIP codes were 28% less likely to discontinue therapy (HR 0.72, 95% CI: 0.55-0.94) [7]. Studies incorporating education and access to private insurance also support this relationship between socioeconomic status, adherence and persistence. Patients with more than a high school education are significantly more likely to persist with at least one medication one month after AMI [46], while patients with commercial insurance (versus Medicare) are more likely to be adherent in the first year of follow-up after AMI [47].  29  One study, by Rasmussen et al, explicitly tested the effects of income and education on persistence with beta blockers and statins, but found conflicting results across classes [9]. The investigators found that among young statin users (aged 30-64), those in the highest income tercile were 27% less likely to experience a break in treatment than those in the lowest tercile (HR 0.73, 95% CI: 0.66-0.82). However, among beta blocker users this effect was reversed, with older patients (aged 65-74) in the highest income tercile being more likely to discontinue treatment (HR 1.11, 95% CI: 1.01-1.22). The authors suggest that one reason for this difference could be the cost of treatment, with statins being the more expensive of the two classes (the study was conducted in Denmark in a system partial drug reimbursement by public insurance, regardless of income) [9]. Variables measuring socioeconomic status are not often included in analyses of adherence and persistence because they can be difficult to acquire in administrative data, but it is clear that these factors can have a large impact on medication use. HEALTH STATUS AND COMORBIDITIES Variables pertaining to general health status or particular comorbid conditions have mixed influence on adherence and persistence. Both studies by Perreault et al indicate that patients diagnosed with diabetes and respiratory disease are more likely to persist with statin (HR 0.84 and 0.66, respectively) or antihypertensive treatment (OR 0.86 and 0.78 at one year) [8, 50]. These studies also show that patients are more likely to persist with statins if they also have a hypertension diagnosis, and are more likely to persist with antihypertensives if they also have a dyslipidemia diagnosis, suggesting a strong relationship between cardiovascular conditions. On the other hand, Newby et al reported that patients with comorbid diabetes were less likely to use beta blockers and aspirin consistently [49]. Measures of general health status also suggest that 30  patients with more comorbidities are more persistent or adherent with treatment. Blackburn et al reported that patients with a higher general chronic disease score were more likely to adhere to statins [41]. CARDIOVASCULAR EVENTS The studies I reviewed are studies of both cohorts of new drug users, and cohorts of users following their first AMI. A key factor in adherence and persistence with cardiovascular medication seems to be whether or not these drugs are used for primary prevention of cardiac events, or secondary prevention after an event has already occurred. Only one study, by Perreault et al, explicitly compared the use of statins for primary and secondary prevention, and found that statin users are 18% more likely to experience a gap in treatment (HR 1.18, 95% CI: 1.11-1.25) if they are using the drug for primary prevention, rather than after a diagnosis of coronary artery disease (as indicated by AMI, angina, or a revascularization procedure) [8]. The authors found that persistence with statins at 6 months was 65% for the primary prevention cohort and 71% for the secondary prevention cohort, and at 3 years was 35% and 45% respectively [8]. Even within the secondary prevention indication there appears to be some variation in adherence and persistence related to cardiovascular events. In the study by Eagle et al, patients whose index event was an AMI were significantly more likely to be adherent with beta blockers at 6 months than those whose index event was unstable angina (OR 1.33, 95% CI: 1.11-1.61) [44]. Cardiovascular events and procedures before or during the study period also appear to be positively associated consistent medication use. In a cohort of patients with coronary artery disease, subjects that had a history of AMI or a prior revascularization procedure, and those who experienced an AMI or who underwent coronary bypass surgery or angioplasty during follow31  up, were significantly more likely to report using aspirin and beta blockers consistently at any point during the follow-up period [49]. PRESCRIPTION DRUG USE A number of studies investigate concomitant drug use, and it appears that patients who use multiple drugs, or have a history of recent prescription drug use, are more likely to have better adherence and persistence. In Gislason‟s study of ACE inhibitor, beta blocker and statin use, persistence was highest for each drug class among patients who also used the other drug classes. For example, statins users were 13% less likely to experience a gap in therapy if they were also using ACE inhibitors (HR 0.87, 95% CI: 0.81-0.94) [45]. Newby and Perreault have also reported that using multiple drug classes concurrently is positively associated with persistence [8, 49, 50]. One study, by Rasmussen et al, additionally showed that the use of beta blockers or statins prior to AMI increases the likelihood that patients will be adherent with these respective treatments after an event as well [51, 53]. It appears that patients with evidence of more past or present drug use are also more likely to have higher adherence and persistence. While the use of more drug classes is positively associated with persistence and adherence, increased drug regimen complexity seems to have a negative effect. In addition to reporting that concomitant drug use increases persistence, Perreault et al also found that the higher the number of oral doses of drug per day (a measure of regimen complexity), the more likely patients were to discontinue statin treatment (HR 1.18, 95% CI: 1.15-1.20) or antihypertensive treatment (OR 1.07, 95% CI: 1.06-1.08, at one year) [8, 50].  32  INTERACTION WITH SYSTEM AND PROVIDERS Studies of adherence and persistence have used a wide range of variables to try to measure subjects‟ interaction with the health care system and providers, and subjects‟ propensity to use health services. One such study, by Brookhart et al, investigated the healthy user effect in adherence and persistence. They found that subjects who filled at least two statin prescriptions in their first year of use were significantly more likely to receive influenza and pneumonia vaccines, and undergo cancer screening with fecal occult blood tests, prostate-specific antigen tests (for men) and mammography (for women) [43]. These results suggest that adherence may be closely associated with a patient‟s health-seeking tendencies. Other studies also look at patients‟ contact with healthcare providers and the healthcare system using different variables. Blackburn et al reported that AMI patients with more physician service days per month were more adherent with statins, however differences in physician visits could be related to either poorer health status or a greater propensity to access care [41]. Hospitalization during follow-up has also been investigated, with mixed results. In Perreault‟s studies, at least one hospitalization in follow-up was associated with improved persistence with statins and antihypertensives [8, 50], while Rasmussen et al found that adherence decreased as the number of hospitalizations during follow-up increased [51]. In this case, recurrent hospitalization is likely an indicator of health status rather than of interaction with providers, and suggests more frequent interruption of outpatient drug use. Lastly, continuity of care may also be related to adherence and persistence with medication. Perreault‟s studies indicate that subjects with three or more prescribing physicians, or two or more dispensing pharmacies, are significantly more likely to discontinue treatment.  33  For example, among statin users, those with ≥3 prescribers are 77% more likely to experience a gap in treatment (HR 1.77, 95% CI: 1.66-1.89) [8].  2.4.2  Outcomes of poor adherence and persistence Many studies, in addition to measuring adherence and persistence rates with  cardiovascular medicines, go one step farther and try to determine the health outcomes of nonadherence or discontinuation of pharmaceutical treatment. Although the cardiovascular drugs in these studies have been proven effective in clinical trials, understanding the relationship between adherence and outcomes can provide insight into the real-world efficacy of these drugs, and the potential benefits of good drug-taking behaviour. In a case-control study of new statin users who experienced a non-fatal cardiovascular event (including AMI, angina or a revascularization procedure), adherence significantly reduced the likelihood of an event after at least one year of statin use. The odds of experiencing an event were 0.81 (95% CI: 0.67-0.97) among patients with a PDC of at least 0.90, as compared to those with PDC < 0.90 [54]. Interestingly, the effect of statin adherence was not statistically significant among case-control pairs with less than one year of statin use. A similar study, using a cohort of patients after their first AMI, found that patients who were adherent to statins (PDC ≥0.80) were significantly less likely experience a subsequent AMI during the follow-up period than those who were not adherent (PDC <0.60). The effect was especially pronounced in subjects younger than 65 years (HR 0.14, 95% CI: 0.04-0.46) [55] . Adherence and persistence with cardiovascular medications has also been associated with reduced mortality. A study by Ho et al reported that among patients who were discharged with beta blocker, statin and aspirin prescriptions, those who were not using any of the drugs after  34  one month had lower 1-year survival than those who persisted with at least one of the drugs (88.5% vs. 97.7%). This effect remained significant after adjustment for demographic and health status variables, with subjects who discontinued therapy being 3.81 times more likely to die during the first year of follow-up (95% CI: 1.88-7.72). When the three drug classes were evaluated separately, the effect also remained [46]. This relationship between drug use and mortality has also been shown by a study of adherence following AMI by Rasmussen et al. Among statin users, subjects with low adherence (PDC <0.40) or intermediate adherence (PDC 0.40-0.79) had higher mortality than those with high adherence (HR 1.12, 95% CI: 1.01-1.25, and HR 1.25, 95% CI: 1.09-1.42). This trend also existed for beta blockers, but the effects were smaller and not statistically significant [51]. In investigating the outcomes of adherence and persistence, there is a risk of bias due to the healthy user effect. As mentioned above, Brookhart et al found that patients who are adherent with medication are also very likely to display other health seeking behaviours, such as undergoing cancer screening and receiving influenza vaccinations [43]. A later study of the healthy user effect by Dormuth et al found that patients who were adherent to statins were less likely to experience events, like AMI, which statin use has been shown to prevent [56]. They also found, however, that patient who were adherent to statins were also less likely to experience events unrelated to statin use, such as bacterial infection, dental problems and accidents (burns, falls, motor vehicle accidents, etc.). These results indicate that not only do adherent patients engage in other behaviours to improve their health or prevent future illness, they also have better health outcomes overall, which may or may not be attributable to their drug use. It is therefore important to recognize that adherence and persistence with medication is only one factor contributing to improved health outcomes in the above studies.  35  In summary, adherence and persistence rates with cardiovascular medications vary somewhat according to the definitions used, but they tend to start high and decrease over time. Persistence, for example, at 6 months after initiation of treatment has been reported to range from 70-80% for ACE inhibitors, 74-88% for beta blockers, and 65-87% with statins, but drops to 58-66% for ACE inhibitors, 48-53% for beta blockers and 35-66% for statins, after at least three years of follow-up. The factors that tend to be positively associated with adherence and persistence are cardiovascular events (before drug initiation or during follow-up), higher socioeconomic status, access to private insurance, concomitant drug use, and greater propensity to use health services (as indicated by use of preventive health services, for example). Age and sex in particular have mixed results, and appear to interact strongly with one another. The findings that consistent use of statins and other drugs are associated with reduced cardiovascular morbidity and mortality, with the caveat of potential bias from the healthy-user effect, reinforce the need to understand patterns of adherence and persistence in these populations.  2.5  CONCEPTUAL FRAMEWORKS I considered three published conceptual frameworks when preparing this study of  adherence and persistence: the WHO‟s 5 dimensions of adherence, Osterberg and Blaschke‟s patient, provider and system interactions for adherence, and an adaptation of the AndersonNewman behavioural model of utilization [2, 26, 57]. The WHO and Osterberg frameworks are specifically related to adherence, while the third is more general and can be applied to all kinds of health services use. I incorporated aspects of each of these into the framework I used to conceptualize my study. The five dimensions of adherence identified by the WHO in their 2003 report, Adherence to Long-Term Therapies: Evidence for Action, are shown below in Figure 2.2. The 36  five identified are social and economic factors, health care team and system-related factors, condition-related factors, therapy-related factors, and patient-related factors [26]. As mentioned above, the WHO defines adherence broadly as the extent to which a person‟s behaviour corresponds with a health care provider‟s recommendations, and this framework seems to reflect that definition, as it does not explicitly address medication use.  Figure 2.2: The WHO's 5 dimensions of adherence [26] Copyright © 2003 World Health Organization.  The social and economic factors considered by the WHO framework are diverse, and include factors both at the individual and societal levels. Some individual-level factors identified that are negatively related to adherence are poor socioeconomic status, illiteracy, unemployment. Other individual-level factors, with mixed impact on adherence, include race and ethnicity, age (as in developmental age group; for example adolescents or the elderly), and marital or family status. Societal-level factors also included in this category are a lack of social support networks, high costs of medication, medical services or transportation, war and conflict, and social or environmental instability [26]. The breadth of these social and economic factors, especially at the societal level, appears to reflect the global mandate of the WHO, and although they are not  37  entirely relevant to my research questions, they clearly play an important role in adherence and provision of health services globally. The second factor in the WHO framework focuses on aspects of the health system or health care team. Research suggests that a good patient-provider relationship is positively associated with adherence, while system factors negatively associated with adherence include poorly developed health services and insurance systems, poor medication distribution and service availability, and lack of knowledge or resources for the management of chronic disease [26]. The last three factors work at an individual level, rather than at a societal or system level, to influence adherence. Condition-related and therapy-related factors are closely connected, and place demands on patients that can greatly influence their behaviour. Factors related to condition severity, disability and progression, and factors related to therapy availability, complexity, duration and effects, can significantly change individuals‟ perceptions of their disease and its treatment. This in turn changes the likelihood of adherence. Lastly, the patientrelated factors identified by the WHO are known to be closely associated with health behaviours. They include patients‟ knowledge and beliefs about their conditions, their selfefficacy in managing their condition, and their perceptions and expectations of treatment [26]. These factors, though clearly related to adherence, are especially difficult for researchers to quantify and for health care practitioners to change. For the purposes of my study, the WHO‟s 5 dimensions of adherence were an excellent starting point. However, with the research question I had chosen, focusing on pharmaceutical use among AMI patients, the therapy-related factors and condition-related factors would be more or less uniform in my study. The patients in my study would have similar health 38  conditions immediately following their first AMI, and should be receiving similar therapies. I therefore chose to further investigate conceptual frameworks that delved deeper into the aspects of health system factors, social/economic factors, and patient-related factors. The next relevant conceptual framework I found was one devised by Osterberg and Blaschke, focusing on aspects of patient, provider and system interactions that could present barriers to adherence with medication [2]. This framework is outlined in Figure 2.3. In the relationship between the patient and prescriber, poor or incomplete communication can lead to patients not understanding their disease and its management, the risks and benefits of treatment, or how their drug treatment should be used. Physicians are also at risk of prescribing overly complex regimens, and at risk of not considering patients‟ lifestyles, cost concerns, or attitudes towards treatment. In the relationship between patients and the broader healthcare system, patients may have difficulty accessing care, and may face restricted formularies and prohibitively high drug costs or copayments. The latter concerns also apply to the interactions between the provider and healthcare system, as providers may not understand the formulary and cost constraints in the system in which they and their patients operate. Patient, provider and system factors can greatly influence these interactions, in turn influencing adherence to medication.  39  PATIENT  Patient and Provider • poor communication of treatment risks and benefits, and intended drug use • complex regimen  PROVIDER  Patient and System • poor access to clinic or pharmacy • formulary constraints • medication costs  HEALTHCARE SYSTEM  Provider and System • poor understanding of drug costs • formulary constraints  Figure 2.3: Osterberg and Blaschke’s barriers to adherence [2] Copyright © 2005 Massachusetts Medical Society. All rights reserved.  The Osterberg and Blaschke framework is highly focused on the use of medicines, and provides a level of detail not included in the WHO framework. It also emphasizes that adherence is subject to influence from multiple levels (patient, provider and system), and the interactions between them. However, this framework does not elaborate on what influence patient, provider or system factors can have on these interactions, and ultimately on adherence, which is a key part of my study. The last conceptual framework I investigated was the well-known Andersen-Newman behavioural model of utilization, adapted by Phillips to include environmental and providerrelated variables [57-59]. Phillips‟ version of the framework is shown in Figure 2.4 below, illustrating the relationship between environment, population characteristics, and health behaviours. The three individual determinants of health services use defined by Andersen and Newman, predisposing characteristics, enabling resources and need, are at the core of the Phillips model.  40  Figure 2.4: Andersen-Newman behavioural model of utilization, adapted by Phillips [57] Copyright © 1998 John Wiley & Sons, Inc.  The first individual-level component, predisposing characteristics, is comprised of the factors that can influence an individual‟s propensity to use health services, independently of need. These factors include demographics, social structure, and attitudes towards care. Demographics are included, because it is well acknowledged that individuals‟ patterns of health services use and risk of future disease vary by sex and age group. Related factors, such as marital status and past use of health services are included in this category. Anderson and Newman also recognized that an individual‟s status in society, measured by factors such as education, occupation, and ethnicity, can have a large influence on their predisposition to use health care services. Lastly, individuals‟ values regarding health, and their attitudes towards the health care system and providers, can influence the likelihood of their using health services. Enabling resources, the next determinant of health services use, are defined as conditions that permit a family to act on a value or satisfy a need regarding health services use [58]. This includes factors such as income, level of health insurance coverage, and whether individuals have access to a regular source of care. Phillips explicitly expands this determinant, by including provider-level and community-level enabling resources. In a similar fashion to the Osterberg model above, Phillips‟ provider-related factors emphasize the interaction between 41  patients and providers, and include variables such as provider specialty and the convenience of obtaining care. Community-level characteristics that could promote health services use include the availability of health care providers in the community, and the supply of health services [57]. The last determinant defined by Andersen and Newman, need, is the most immediate cause of health services use. It is divided into evaluated illness and perceived illness, where evaluated illness includes the clinical diagnosis and severity of illness, and perceived illness emphasizes patient and family perception. Perceived illness can be captured through variables such as self-reported health status or number of disability days, for example. An important aspect of these individual determinants of health services use is the feedback loop that exists, in the bottom right of Figure 2.4, between health behaviour and the characteristics of the population. Interactions with the health care system can significantly influence individuals‟ perceptions of the system and their attitudes towards care. Also, the outcomes from past health services use can influence both individuals‟ perceived and evaluated health status. In addition to the individual determinants of health services use, the Phillips adaptation expands upon the contextual variables in the model, including health care system factors and characteristics of the external environment. The health care system factors thought to potentially influence health services use are the policies, resources, organizations and finances of the system, which in turn influence the availability, accessibility and acceptability of health services. Characteristics of the external environment, not directly related to the health care system, which may also influence health services use are wealth, economic climate, politics, stability and prevailing norms [57].  42  The behavioural model of utilization can be applied to all kinds of health services use, and includes a wide range of potential environmental and individual-level determinants; however, for my project, I modified the Phillips framework to make it more specific to adherence and persistence with medication, in a population with a similar baseline need, and simplified it to focus on the variables of interest for my study (Figure 2.5).  REGIONAL LEVEL  INDIVIDUAL LEVEL  External Environment  • urban/rural • wealth, etc.  Predisposing Factors  Enabling Factors  Health Care System  • Age, sex • Health status • Rx use habits  • Income • Insurance • Regular Dr.  APPROPRIATE USE  • Adherence • Persistence  • availability and access  Figure 2.5: Conceptual framework of factors influencing adherence and persistence with medication, among a population with identified treatment needs.  I redefined health services use from the Anderson-Newman framework specifically as adherence and persistence with medication. In my framework I conceptualized adherence and persistence as the appropriate use of medicines by a patient, consistently over time. I intend the term appropriateness in this case to not refer to prescriber behaviour or the appropriateness of the prescription itself; instead, it is related to patient behaviour and the use of prescribed (and dispensed) medicines. The feedback loop between individual-level factors and appropriate use, suggests the potentially cyclical nature of this relationship with time. Any changes to predisposing or enabling factors (for example, a change in employment status) can potentially interrupt adherent behaviour. Alternatively, feedback from the appropriate (or inappropriate) use of medicines, such as changes in health status or attitudes towards care, can also influence 43  future adherence or persistence with medication. Although my research questions do not address the outcomes of adherence or persistence with medicines, or changes to predisposing or enabling factors over time, I thought it was important to highlight these potential interactions in my conceptual framework to help put results in context. Other changes I made to the Andersen-Newman framework included removing need as a determinant of use, because my study investigates adherence and persistence in a cohort of patients in which need is more-or-less consistent between individuals. For example, while patients in the cohort may have specific comorbidities or contraindications to drug use following AMI, their need for secondary prevention of cardiac events still exists. I also simplified the other two of Andersen and Newman‟s individual determinants, focusing on demographics, health status, and prescription drug use as predisposing factors, and socioeconomic status (income and insurance) and having a regular prescriber as enabling factors. In the original behavioural model of utilization, past illness and health services use is considered a strong predictor of future use. By explicitly incorporating past prescription drug use and concomitant use of multiple drugs, I hope to see if this relationship holds specifically for medication use as well in my study. As previously described, prior use of beta blockers or statins is associated with improved adherence with those classes following AMI [51], and use of multiple drug classes for secondary AMI prevention is also associated with improved persistence and adherence [41, 45]. In order to incorporate aspects of the Osterberg framework, which emphasized patient, provider and system interactions, I also included the number of prescribers as a potential enabling factor. Past studies suggest that having more prescribers may be associated with worse persistence with statins or antihypertensive drugs [8, 50].  44  A goal of this study is to investigate whether regional variation exists in adherence and persistence with medication therefore I chose to include regional level variables, analogous to Phillips‟ environmental factors. Although this study does not go beyond measuring regional variation to investigate regional factors in depth, the Phillips‟ framework outlines key factors which may contribute to variation at the regional level. These regional factors also tie in with the social/economic and health care system factors identified in the WHO framework, which have been shown to be closely related to adherence. With this framework in mind, and with information from past studies of adherence and persistence, I designed my study to evaluate factors related to adherence and persistence with ACE inhibitors, beta blockers and statins following AMI in BC.  45  Chapter 3: Methods  3.1 STUDY DESIGN In order to address my research questions, I used a historical (or retrospective) cohort study design [60], enabling me to follow individual prescription drug use over time. Using administrative data, I identified patients in BC who had their first AMI between 2001 and 2005, and followed their prescription drug use from the time of their AMI through to the end of 2006. A historical cohort study with administrative data was the most appropriate study design for addressing my research questions, because it enabled me to study nearly the entire population of BC over a long period of time, using existing well-validated data. My population of interest for this study, AMI patients, is well-suited to a cohort study, because of subjects‟ well-defined cohort entry (first AMI event). Also, using administrative data to measure my outcomes of interest, adherence and persistence, may introduce less information bias than other methods of measurement that involve contact between researchers and patients [28, 36].  3.1.1  Data sources This study was conducted using secondary data from two comprehensive, population-  wide health care databases in British Columbia: (1) the BC PharmaNet prescription drug claims database, and (2) the BC Linked Health Database (BCLHD). I used data from 1996 to 2006 to identify my cohort, to build measures of adherence and persistence, and to collect important covariates for my analysis, as described in sections 3.3 and 3.4 below. The BCLHD connects data describing hospital separations, through the Discharge Abstract Database; physicians visits, through the Medical Services Plan (MSP) fee-for-service 46  billing and registration records; and deaths, through records from the BC Vital Statistics Agency. These data sources are linked at the individual level and anonymized to remove all personally identifying information. PharmaNet contains records of all outpatient prescriptions filled in BC, with information on the drug and quantity dispensed, and limited information about the prescriber and payer. When combined, these databases are a rich source of information on demographics, health care services use and pharmaceutical use for nearly all of the 4.2 million residents of BC. Subpopulations that are not included in the BCLHD and PharmaNet are those whose health care is a federal responsibility, including First Nations peoples on-reserve, veterans and members of the armed forces, the RCMP, and federal inmates, consisting of approximately 4% of the BC population [10]. This study was approved by the Behavioural Research Ethics Board of the University of British Columbia, and permission to use data was obtained from the Ministry of Health Services and BC PharmaNet.  3.2 COHORT DEFINITION  3.2.1  Index AMI The cohort for my study consisted of all patients in BC who experienced their first acute  myocardial infarction between the years 2001 and 2005. The cohort was constructed by CHSPR analyst Lixiang Yan. All patients hospitalized for AMI in any of the years from 1996 to 2005 were identified using hospital separations data, by identifying subjects with a principal diagnosis of ICD-9 code 410: Acute Myocardial Infarction (for records from ‟96-‟00) or ICD-10 I21: Acute Myocardial Infarction, or I22: Subsequent Myocardial Infarction (for records from ‟01‟06). A broad diagnostic definition was used to maximize sensitivity, on the assumption that any 47  subjects who were inappropriately included in this first step would likely be excluded when considering other criteria, such as survival and use of drug therapy, as described later. Only subjects with an AMI between 2001 and 2005, without a previous AMI between 1996 and 2000, were included in the cohort for analysis. I used the date of hospital admission for a patient‟s index AMI as the date of admission to the cohort, and used the date of discharge to define a patient‟s index date. Hospital transfers were treated as separate visits in the data, and as a result many patients had multiple discharge dates in rapid succession. To accommodate potential hospital transfers, I defined index date as the last hospital discharge, with a diagnosis of AMI, within 30 days of the first admission date.  3.2.2  Inclusion and exclusion criteria In order to limit the cohort to cases that were most likely to be true AMIs, I applied a  number of exclusion criteria based on other published AMI cohort definitions [61]. These exclusion criteria were length of stay more than 30 days, and age less than 40 years or more than 100 years. I also required that cohort members were registered for MSP for at least 275 days in each of the two years before their first AMI. This restriction, combined with the requirement that no patient were hospitalized for AMI between 1996 and 2000, increased the likelihood that the index event identified was in fact subjects‟ first AMI, by ensuring between 2 and 5 years of prior observation. To be included in the analysis of adherence and persistence, two further restrictions were placed on the cohort: members had to survive at least 30 days following discharge for AMI, and members had to fill a prescription for any of ACE inhibitor, beta blocker or statin therapy within 30 days of discharge. Past research suggests that most AMI survivors who initiate drug  48  treatment for secondary prevention, do so within the first 30 days of discharge [45]. In fact, most patients begin treatment within 7 days, because they are discharged with prescriptions inhand [62]. By restricting the initial use period to 30 days, my hope is that this cohort captures those patients whose use of drug treatment is directly related to their AMI event. Many previous studies have also excluded patients with a length of stay less than 3 days, however these patients made up a large part of my cohort (~15%) so I chose to include them and flag them in my analyses (see Health Status variables, below). I also calculated age-adjusted drug use rates, for the 30 days following AMI, to verify that there were no substantial differences in drug use between the two groups.  3.3 STUDY VARIABLES  3.3.1  Outcome variables The outcomes of interest in this study, adherence and persistence, were both calculated  by first mapping out individuals‟ daily medication availability with a basic inventory model. This technique, and the SAS code on which I based my variable construction, were both generously provided by Dr. Michael Law of CHSPR [63]. By using the fields “date of service” and “days‟ supply dispensed” for each prescription recorded in PharmaNet for a given patient, I populated an array of variables, one variable for each day of follow-up (from discharge post-AMI to the end of 2006), with the amount of drug available in-hand to the patient on each day. On days when drugs were dispensed to the patient, their supply in-hand increased by the dispensed days‟ supply, as shown on the vertical axis in Figure 3.1. On each day between prescription fills, their supply in-hand decreased by one, to a minimum of 0. This array of days‟ supply in-hand was then translated to an array of daily drug availability, with availability coded as a binary variable 49  indicating whether or not a subject had drug supply in-hand for each day of follow-up. These  Drug Supply (days)  variables are illustrated as bars beneath the horizontal axis in Figure 3.1.  0  180  360  540  720  time (days)  Figure 3.1: Inventory model of drug supply (in days) and daily drug availability  These inventory models of drug supply were constructed separately for each drug class, mapping subjects‟ use of ACE inhibitors, beta blockers and statins. Switching between drugs of the same class was not considered an interruption in therapy. Individuals‟ arrays of drug availability for each class were also compared, to determine daily availability of any drug (from any of the three classes), or all three drug classes. These arrays were then used to calculate individual adherence and persistence variables, by drug class and by drug combination. To identify potential errors in the days‟ supply field of the PharmaNet records, a hypothetical daily dose was calculated using days‟ supply and quantity dispensed for each prescription filled. Any subject with a prescription where the calculated daily dose was less than 0.25 units per day or over 6 units per day was excluded from the analysis. I considered the lower limit, 0.25 units per day, to be the smallest possible unit that still allowed for potential pill splitting. I defined the upper limit as 6 units per day to include as many records as possible while still excluding prescription records with what appeared to be a 7-day supply that was mistakenly entered. There were relatively more prescriptions with a calculated daily dose of 7 50  units, versus 6 or 8 units, and these appeared to be errors on further examination. Subjects with these potential dosing errors were excluded entirely to avoid errors in their drug use arrays, which would in turn disrupt calculated adherence and persistence values. PERSISTENCE Duration of drug therapy was measured using persistence. Persistence was defined as the number of days to discontinuation of drug use, where discontinuation was defined as a gap in drug availability lasting greater than 90 days [3]. This definition is illustrated in Figure 3.2, using two hypothetical drug use profiles. As in Figure 3.1, the horizontal bars represent days with drug available; subject A is persistent for 720 days, while subject B is persistent for only 330 days. Short gaps in drug availability are not considered discontinuation. A gap of 90 days was deliberately chosen, because previous research has suggested that many patients reinitiate therapy after gaps in treatment, but the longer the gap, the less likely they are to reinitiate, as described in section 2.3.2 [45]. Also, BC PharmaCare, the public drug insurance plan in BC, and most private drug insurance plans do not pay for more than 90-100 days‟ supply of drug at a time, creating a de facto limit on prescription length. By setting the longest permissible gap at 90 days, patients who miss a full prescription “cycle,” approximately 90 days from dispensing to end, are considered to be no longer taking medication.  51  time (days)  start 180  360  540  720  A  B  720 days 330 days  Figure 3.2: Persistence, measured as time to first gap in drug availability exceeding 90 days.  Since persistence is measured as time-to-event, it is susceptible to right-censoring in subjects who do not experience a 90-day gap in medication before the end of the follow-up period, or before leaving the cohort for other reasons. The variables used to define censoring are described in detail below, in section 3.3.3. ADHERENCE I measured adherence in two ways for my study; both measure the proportion of days with medication available (also called the proportion of days covered, or PDC), with one measure using a variable denominator and the other using a fixed denominator. Both measures of adherence were calculated by drug class and by combination, as described above. For the first method, I counted the proportion of days covered (PDC) between a subject‟s first prescription and the end of their drug use, as defined by my measure of persistence. With this measure, I hoped to capture the intensity of a subject‟s medication use, while they were considered persistent with that medication. This measure is illustrated in Figure 3.3, using the same drug profiles as above. From their first prescription to discontinuation, subject A has a PDC of 0.917 while subject B has a PDC of 0.818. 52  Subjects who did not refill their prescriptions, i.e. subjects who filled only one prescription for the given class, were excluded from this analysis, because there was insufficient drug use to be able to infer a value for adherence. Any subject who filled only one prescription would have a PDC of 1.00 by default, due to the inventory-based definition of persistence I used. I identified prescription refills by counting the number of prescriptions dispensed to a patient, by drug class, during the time period from the first prescription to the first gap in therapy lasting more than 90 days. For my analysis of adherence with any class, I included patients who had refilled at least one drug class, while for my analysis of adherence with all three classes I included only patients who had refilled all drug classes.  time (days)  start 180  360  540  720  A  B  0.917 0.818  Figure 3.3: Adherence, measured as proportion of days with drug available, from first prescription to first gap in drug availability exceeding 90 days  The second measure of adherence, which I have termed 1-year PDC in this study, used a fixed time interval for analysis. I calculated 1-year PDC as the proportion of days with medication available in the first 365 days after a subject‟s first prescription. Figure 3.4 shows the fixed 365 day time interval applied to the two example drug use profiles. The 1-year PDC for subject A, 0.918, is nearly the same as the adherence value calculated in Figure 3.3; however the 1-year PDC for subject B, 0.740, is somewhat lower than the value above, because subject B  53  discontinues treatment within the first year. I excluded subjects who were censored within the first 365 days (described below, in section 3.3.3) to ensure that all subjects included had at least a full year of observation. I also did not define a maximum permissible gap for this measure, as I did for persistence. All drug available in-hand in the first 365 days, regardless of the length of gaps in treatment, was included in calculation of PDC. One-year PDC is very commonly used in the literature as a measure of adherence [28, 40], although it can be somewhat difficult to interpret, because it also captures aspects of persistence. For example, a subject with a 1-year PDC of 0.75 could have taken their medication daily for the first three quarters of the year and stopped, or could have taken their medication 3 out of every 4 days for the full year. I therefore included this measure to improve comparability with previous studies of adherence.  time (days)  start 180  360  540  720  A  B  0.918 0.740  Figure 3.4: Adherence, measured as the proportion of days with drug available in the 365 days following the first prescription (1-year PDC)  54  3.3.2  Covariates  DEMOGRAPHIC The demographic variables included in this study are age, sex and region of residence. All three variables are from MSP registration records in the BCLHD. Age has been divided into 6 10-year age categories. As described in section 2.4.1, the effect of age on adherence and persistence does not appear to be linear, and categorizing age in this way may better portray this relationship. Local health area (LHA) is the smallest unit of geographic analysis in the BCLHD. There are 89 LHAs in BC, with populations ranging widely from around 500 to 350,000. Although they do not represent true boundaries in the provision of health services (as do the 5 regional health authorities, for example) they are small areas that are generally geographically and demographically consistent. LHA is the preferred level of geographic analysis to accurately identify local trends in adherence or persistence rates; however, to maintain adequate sample size for analysis, especially in remote areas of the province, LHA was aggregated to health services delivery area (HSDA). The 89 LHAs are nested within the 16 HSDAs, which are further nesting within the 5 health authorities. HSDAs provide a more general overview of regional variation in the province, and their geographic boundaries are shown in Figure 3.5, below. SOCIOECONOMIC Socioeconomic status was measured using three variables: income percentile, an indicator for income assistance, and an indicator for private insurance. Income percentile had been previously calculated by Gillian Hanley and colleagues at CHSPR, by combining Fair PharmaCare registration records and Canadian census data. Fair PharmaCare is BC‟s incomebased public drug insurance program, and as part of the registration process, registrants must provide household income information, and consent to having their income verified by the 55  Canada Revenue Agency (CRA) [64]. Almost 80% of BC households have registered for Fair PharmaCare, meaning that for most of the BC population, household-level CRA-validated income information is available and was used to construct income percentiles [65]. For the remainder of the population, income was estimated based on average neighbourhood income, at the level of Census Dissemination Areas, from the Canadian census in 2002. The actual and estimated income values were combined and ordered into income percentiles. For my analysis, I further aggregated these percentiles into three groups: the top 30%, the middle 40%, and the bottom 30%.  Figure 3.5: Map of Health Services Delivery Area (HSDA) boundaries in British Columbia, from BC Stats, July 2008 [66]. Copyright © Province of British Columbia. All rights reserved. Reprinted with permission of the Province of British Columbia. www.ipp.gov.bc.ca  56  Subjects were flagged as receiving social assistance if at least half of their prescriptions for ACE inhibitors, beta blockers or statins during the follow-up period were paid for through PharmaCare plan C, as indicated by the field “plan code” in each PharmaNet record. Plan C is the PharmaCare plan that provides full prescription coverage to residents who receive income assistance and medical benefits through BC‟s Ministry of Housing and Social Development [64]. The last socioeconomic variable, an indicator for private insurance, had also been previously created by CHSPR analysts, by flagging whether or not a household‟s MSP premium was paid by an employer, paid by another organization (for example, for residents of long-term care), or self-paid. This variable is based on the assumption that individuals whose MSP premiums are paid for by an employer are more likely to have private medical insurance, with some form of pharmaceutical coverage. HEALTH STATUS Health status was measured in three ways for this analysis: a general measure built with Johns Hopkins case-mix adjusting software, a flag for recipients of long-term care, and a flag for patients with a short length of stay for their index AMI. Much like the flag for social assistance, described above, the indicator for long-term care was derived from PharmaNet records. PharmaCare Plan B covers prescription drug costs for patients in residential care facilities in the province [64]. Patients with at least 50% of their ACE inhibitor, beta blocker and statin prescriptions paid through Plan B were flagged as being residents in long-term care facilities. Residing in long-term care is not only an indicator of potentially poor health status; it may also impact medication-taking behaviour, due to increased contact between patients and their health care providers or other caregivers. 57  Diagnostic codes from MSP claims and hospital records from the BCLHD for the year prior to each individual‟s index AMI were used to estimate general health status, using the Johns Hopkins Ambulatory Care Group system. This system is designed to estimate an individual‟s future health services use, based on patterns of their past use, and generates a number of variables describing health status [67]. The specific variable used for my analysis was a count of collapsed ambulatory diagnostic groups (ADG‟s). Every ICD diagnostic code has been classified into one of 34 ADG‟s, based on a number of general parameters, including: whether the diagnosed condition is acute or chronic; the likelihood that it would recur; the likelihood that a patient with that diagnosis would require future medical visits, specialist referrals or hospitalization; and any potential disability or reduced life expectancy associated with that diagnosis [67]. ADG‟s are not condition-specific; instead they describe health status in terms of general severity or complexity. The 34 ADG‟s can be further aggregated into 12 collapsed ADG‟s, based on the parameters of acuity, severity, stability and recurrence of medical conditions. It was a count of these 12 collapsed ADG‟s, calculated using the diagnostic codes for each individual in the cohort for the 365 days preceding their first AMI, that was included in this analysis as the measure of general health status. Patients with a length of stay less than three days for their index AMI, calculated as the difference between their admission date and their index date, were flagged. Patients with a short length of stay may have had less severe AMIs, or may perceive their AMIs as being less severe, which may in turn influence medication use behaviours.  58  PHARMACEUTICAL USE A number of new variables were created to try to understand the relationship between adherence, persistence, and other patterns of pharmaceutical use, such as prior drug use, polypharmacy, and prescriber continuity. Prior drug use was measured with three variables flagging whether or not individuals filled a prescription for ACE inhibitors, beta blockers or statins in the 120 days before admission to hospital for their first AMI. Subjects who are familiar with these classes of drugs, and accustomed to taking them, may also use them more consistently after AMI. The use of multiple drugs, polypharmacy, was measured by counting the number of drug classes used by each patient in the 120 days following discharge after first AMI. Using individuals‟ PharmaNet records for all prescriptions filled during the 120-day window, drugs were classified according to the WHO‟s Anatomical Therapeutic Chemical (ATC) classification system‟s third level. This system classifies drugs according to the organ or physiological system that they target (level one), their therapeutic, pharmacological and chemical properties (levels two, three and four, respectively), and their specific chemical substance (level five) [68]. The number of classes used by each patient was counted, and categorized into four groups: 1-3 classes, 4-6 classes, 7-9 classes, and over 10 classes. Lastly, provider continuity, a potentially important factor influencing adherence and persistence according to the literature [42], was measured by counting the number of unique prescribers each subject had in the first 120 days after discharge for AMI. The PharmaNet data used in this study has a field for practitioner identifier, encrypted for privacy. These identifiers were counted from all prescriptions filled in the 120 day time window, and the number of 59  unique prescribers each individual had was categorized into four levels: 1 prescriber, 2 prescribers, 3 prescribers, and 4 or more.  3.3.3  Censoring variables For the assessment of persistence and 1-year PDC, patients could be right censored in  three ways during the follow-up period: by being persistent with treatment to the end of the study period, by no longer being registered for MSP, or by dying during follow-up. For subjects who died, only month and year of death was available in the data, so they were censored on the first day of the month in which they died. Subjects who were no longer registered for MSP were defined as those who were registered for fewer than 275 days in any year of follow-up. In the MSP registry data only number of days registered per year is available, as opposed to dates marking the start or end of the registration period; therefore patients were censored on January 1 of the year in which their registration ended. By rounding follow-up time down to the start of the month, for death, or the start of the year, for registration, I generated more conservative estimates of persistence. There is some overlap between these two censoring definitions: depending on how early in the year they died, many subjects were also flagged as also not being registered for MSP. I therefore calculated censoring due to death first, and excluded subjects who died from my calculations of censoring due to end of MSP registration. By prioritizing death over MSP registration in this way, I hoped to conservatively estimate of follow-up time, without losing too much precision in my analysis. Lastly, subjects who were still using medication at the end of 2006, who had not yet experienced a gap lasting more than 90 days, who did not die, and who were continuously enrolled for MSP were censored when the study period ended on December 31, 2006.  60  3.4 ANALYTIC PLAN I first investigated the distribution of study variables in the cohort of AMI patients using simple descriptive statistics and frequency distributions, across 6 drug use categories: users of ACE inhibitors, beta blockers, statins, any class, all 3 classes concurrently, and non-users. I calculated crude use rates, by drug class and combination, and also calculated age-adjusted use rates for comparison across HSDAs. For the analysis of persistence I estimated median time to discontinuation with KaplanMeier survival curves, using death and MSP registration information to account for right censoring of the data, as described above. Using life tables derived from the survival curves I also calculated the proportion of users persistent at 6 months and at 1 year intervals for the duration of follow-up. I first analyzed adherence measured as PDC from first prescription to first gap in therapy >90 days using simple descriptive statistics, with and without prescription refill restrictions, for sensitivity testing of this exclusion criterion. I also analyzed 1-year PDC with simple descriptive statistics. Both measures of adherence were highly skewed and bounded by 0 and 1, therefore I created two dichotomous variables indicating whether PDC (from start to first gap) or 1-year PDC was 0.80. To analyze the effect of the above covariates on persistence I used multivariate Cox proportional hazards models. For my analysis of adherence, I conducted logistic regressions with the two dichotomized adherence variables. Before building these models, I examined the Spearman rank correlations between all covariates, to ensure there was little correlation between the variables selected for inclusion. For all study outcomes – persistence and the two adherence 61  measures, for each drug class and for any/all drug classes – I first adjusted for age and sex alone, age, sex and age-sex interaction, and age, sex and each of the groups of covariates: socioeconomic status, health status, and pharmaceutical use variables. Lastly, in the full models I adjusted for all covariates, including HSDA. All statistical analyses were done using the SAS statistical software package, version 9.1 for Windows (SAS Institute Inc., Cary, NC).  62  Chapter 4: Results  4.1 STUDY COHORT  4.1.1  Cohort construction There were 23,663 patients in British Columbia who were hospitalized for their first  AMI between 2001 and 2005, and who were also registered for MSP for at least two years before the event. A flowchart illustrating the construction of the cohort and application of exclusion criteria is shown in Figure 4.1. Of these patients, 3465 (14.6%) died in-hospital or within the first 30 days of discharge. Few patients were outside the age range of 40-99 years (1.3%), had lengths of stay exceeding or equal to 30 days (1.7%), or had potential dosing errors in their PharmaNet records (1.8%). There were 19,267 patients remaining after application of these exclusion criteria. Of these, 764 (4.0%) had missing data, resulting in 18,503 complete observations included in this analysis.  4.1.2  Cohort characteristics Of the 18,503 AMI patients included in the analysis, all but 2512 (13.6%) filled at least  one prescription for an ACE inhibitor, beta blocker or statin within 30 days of discharge. There were 11,494 (62.1%) users of ACE inhibitors, 12,949 (70.0%) users of beta blockers and 11,397 (61.6%) users of statins. Most patients filled prescriptions for at least one of these drug classes, with 7102 patients (38.4% of all AMI patients) appearing to use all three (Figure 4.2).  63  All patients with first AMI in ‘01-’05, in MSP for ≥2 years prior to event  23,663 3,465 (14.6%) die within 30 days  Patients surviving ≥30 days after discharge  20,198 253 (1.3%) outside age range  Patients between 40 and 99 years old  19,945 337 (1.7%) hospitalized ≥30 days  Patients hospitalized less than 30 days  19,608 341 (1.8%) with daily dose <0.25 or >6  Patients without potential days’ supply errors  19,267 764 (4.0%) with missing data values  Patients with complete data available  18,503  Figure 4.1: Flowchart of cohort construction  Figure 4.2: Venn diagram of ACE inhibitor, beta blocker and statin use following AMI  64  The characteristics of the study cohort, according to patients‟ use of each drug class, are shown in Table 4.1. The majority of patients were male across all groups, around 65-70%, with slightly more women among the non-users, and slightly fewer among the users of all three classes. There is very little variation between users of ACE inhibitors and beta blockers across all variables; however, statin users appear to be younger (mean age of 65.9 years, vs. 67.3), have slightly higher incomes, and have fewer major ADGs. Users of all three classes concurrently are also very different than both non-users and users of any drug. They are much younger than non-users (mean age 64.7 years, vs. 74.4) or users of any drug (mean age 67.9 years). Users of all three drugs also tend to have higher socioeconomic status, with more of these patients having private insurance (43.5% vs. 40.7% among all users) and higher incomes (30.6% in the top income category, vs. 20.8% among non-users). They also tend to be healthier, with fewer major ADGs (2.7% with 5+ major ADGs, vs. 6.8% among non-users) and fewer residents in longterm care (1.0%, vs. 3.5% among non-users). Also, far fewer users of all three classes died during the follow-up period than non-users or users of any drug (9.5%, vs. 34.7% among nonusers), likely due to the large age difference described above.  65  Table 4.1: Characteristics of study cohort, according to drug use following AMI Users (Any Class) N = 15,991 Sex  Age category  N  %  N  %  N  %  N  %  Female  5504  34.4  898  35.7  3799  33.1  4316  33.3  3542  31.1  2059  29.0  mean years years years years years years  1465 3110 3750 4335 2830 501  9.2 19.4 23.5 27.1 17.7 3.1  128 249 352 726 895 162  5.1 9.9 14.0 28.9 35.6 6.4  1120 2372 2727 2976 1952 347  9.7 20.6 23.7 25.9 17.0 3.0  1273 2631 3096 3416 2190 343  9.8 20.3 23.9 26.4 16.9 2.6  1220 2540 2899 3042 1537 159  10.7 22.3 25.4 26.7 13.5 1.4  861 1729 1836 1736 857 77  12.1 24.4 25.9 24.5 12.1 1.1  6509  40.7  1073  42.7  4727  41.1  5316  41.1  4899  43.0  3087  43.5  5756 5776 4459  36.0 36.1 27.9  1080 909 523  43.0 36.2 20.8  4117 4157 3220  35.8 36.2 28.0  4587 4688 3674  35.4 36.2 28.4  3801 4199 3397  33.4 36.8 29.8  2329 2596 2171  32.8 36.6 30.6  763  4.8  79  3.1  541  4.7  607  4.7  559  4.9  332  4.7  1240 8020 4507 1619 605  7.8 50.2 28.2 10.1 3.8  139 1019 785 398 171  5.5 40.6 31.3 15.8 6.8  921 5868 3206 1091 408  8.0 51.1 27.9 9.5 3.5  1070 6551 3606 1261 461  8.3 50.6 27.8 9.7 3.6  943 6003 3125 995 331  8.3 52.7 27.4 8.7 2.9  639 3798 1910 559 190  9.0 53.5 26.9 7.9 2.7  405  2.5  88  3.5  269  2.3  292  2.3  143  1.3  69  1.0  2457  15.4  501  19.9  1757  15.3  1979  15.3  1920  16.8  1183  16.7  4098 3107 3088  25.6 19.4 19.3  596 537 476  23.7 21.4 18.9  2880 2087 2039  25.1 18.2 17.7  3155 2400 2432  24.4 18.5 18.8  2565 2014 2139  22.5 17.7 18.8  1515 1103 1192  21.4 15.5 16.8  7096  44.4  7096  61.7  7096  54.8  7096  62.2  1-3 4-6 7-9 10+  1613 6467 4925 2986  10.1 40.4 30.8 18.7  350 550 510 346  13.9 21.9 20.3 13.8  984 4863 3610 2037  8.6 42.3 31.4 17.7  1241 5371 4000 2337  9.6 41.5 30.9 18.0  1030 4872 3484 2011  9.0 42.7 30.6 17.6  488 3259 2209 1140  6.9 45.9 31.1 16.1  1 2 3 4+  2516 5616 4260 3599  15.7 35.1 26.6 22.5  471 575 379 346  18.8 22.9 15.1 13.8  1825 4117 3072 2480  15.9 35.8 26.7 21.6  2016 4586 3422 2925  15.6 35.4 26.4 22.6  1668 3994 3128 2607  14.6 35.0 27.4 22.9  1060 2587 1910 1539  14.9 36.5 26.9 21.7  2690  16.8  872  34.7  1832  15.9  1959  15.1  1286  11.3  673  9.5  1 2 3 4 5+  Sum of major ADGs  Long-Term Care Hospitalization < 3 days ACE Inhibitor Beta Blockers Statins  Use of all 3 classes after AMI  Death during follow up  All 3 Classes N = 7096*  %  bottom 30% middle 40% top 30%  Prescribers post-AMI  Statins N = 11,397  N  40-49 50-59 60-69 70-79 80-89 90-99  Drug classes used postAMI  Beta Blockers N = 12,949  %  Social Assistance  Prior drug use  ACE Inhibitors N = 11,494  N  Private Insurance Income percentile  Non-Users N = 2512  67.9  74.4  67.3  67.3  65.9  64.7  * 7102 patients used all three classes of drugs in the first 30 days post-discharge, but only 7096 of those had all three drugs in their possession at the same time within those 30 days, according to the inventory model constructed  66  There is some regional variation in age-adjusted use rates across health services delivery areas (HSDAs, Table 4.2). The population of each HSDA varies widely, and the number of AMI patients contributed to the cohort by each HSDA ranged from only 304 in the Northeast to 2635 in Fraser South. When adjusted to the age of the full post-AMI cohort, the use of any class (ACE inhibitor, beta blocker or statin) ranged from 79.8% in the Northwest to 89.9% in Vancouver. The use of all three drug classes ranged from a low of 31.2% in the Northwest to 46.9% in the Northeast. Table 4.2: Age-adjusted use rates of any drug and all 3 drugs following AMI, by health services delivery area Health Services Delivery Area East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Fraser South Richmond Vancouver North Shore/Coast Garibaldi South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  N  Any Class (%)  All 3 Classes (%)  574 559 1964 1265 1208 1938 2635 563 1565 1358 1428 1629 581 318 679 304  81.7 84.1 88.6 86.2 86.3 85.6 85.4 87.0 89.9 86.3 88.1 85.0 89.0 79.8 85.8 83.9  37.7 35.2 40.9 35.0 37.6 33.4 41.4 35.7 41.5 35.9 41.6 35.7 40.0 31.2 41.1 46.9  4.2 PERSISTENCE AND ADHERENCE RATES  4.2.1  Persistence Kaplan-Meier curves, showing unadjusted persistence over time by class (Figure 4.3) and  by combination (Figure 4.4) are below. Median time to discontinuation was 1291 days (approximately 3.5 years) for ACE inhibitors, 1380 days (3.8 years) for beta blockers and 1799 days (4.9 years) for statins. Median persistence with statins was significantly higher than 67  persistence with the other two classes. At six months of follow-up after a patient‟s first prescription fill, 78.7% of ACE inhibitor users, 82.5% of beta blocker users and 86.0% of statins users were considered persistent (Table 4.3). These values decreased with longer follow-up, with persistence rates of 69.3%, 73.3% and 78.3% respectively at 1 year, and 42.2%, 43.8% and 49.6% at 5 years. Persistence with any of the three classes was very high for the duration of follow-up, while persistence with all three drugs dropped off quickly. Median persistence with any drug was 2183 days (6 years, the maximum possible for the study period) and for all three drugs was only 397 days (Figure 4.4). At 6 months, 93.1% of all users were persistent with at least one drug, decreasing slightly to 88.5% at 1 year and 72.2% at 5 years. By contrast, only 65.3% of users of all 3 classes remained persistent with all 3 classes at 6 months, dropping to 51.8% at 1 year and 21.2% at 5 years (Table 4.3). Table 4.3: Percent of drug users persistent with medication at 6 months and at 1-year time intervals, by drug class and combination, from Kaplan-Meier persistence estimates. ACE Inhibitors  Beta Blockers  Statins  Any Class  All 3 Classes  6 months  78.7  82.8  86.0  93.1  65.3  1 year  69.3  73.3  78.3  88.5  51.8  2 years  59.0  61.4  67.0  81.8  37.6  3 years  52.6  54.4  60.0  76.7  30.9  4 years  47.2  48.8  54.5  72.2  25.0  5 years  42.2  43.8  49.6  68.1  21.2  68  Figure 4.3: Kaplan-Meier (product-limit) estimate of persistence, measured as days to first 90-day gap in treatment, by drug class  Figure 4.4: Kaplan-Meier (product-limit) estimates of persistence, measured as days to first 90-day gap in treatment, by drug combination  69  Persistence values were right-censored for many patients, most of whom were persistent to the end of the observation period, December 31, 2006. The proportion of censored patients was 53% among ACE inhibitor users, 56% among beta blockers users, and 61% among statin users. Of these censored observations, over 80% were censored by the end of the observation period (as opposed to death or end of MSP registration) as shown in Table 4.4. Among users of any of the three classes 77% were censored, and 82% of those were censored by the end of the study period. Overall, 62.9% of all users of any drug class remained persistent at the end of follow up. Far fewer users of all 3 classes were censored by any cause (32%), but the proportion of censored observations censored by the end of the study period remained high. Table 4.4: Percent of drug users persistent at end of follow-up (i.e. censored by end of study period, Dec. 31, 2006) Censored (all causes)  4.2.2  Censored by end of study period  N  % of users  N  % of censored  % of users  ACE Inhibitors  6093  53.0  4988  81.9  43.4  Beta blockers  7214  55.7  5869  81.4  45.3  Statins  6895  60.5  6020  87.3  52.8  Any Class  12,254  76.6  10,052  82.0  62.9  All 3 Classes  2284  32.2  1996  87.4  28.1  Adherence Adherence, measured as the proportion of days covered (PDC) while a patient is  considered persistent with medication (from first prescription to first gap lasting ≥90 days) is shown in Table 4.5. The mean PDC is very high for each drug class, ranging from 0.955 for ACE inhibitors to 0.938 for statins. PDC values are also highly skewed, with 95.1% of patients adherent with ACE inhibitors (PDC ≥0.80), and over 92% adherent with beta blockers or statins. The mean PDC for use of any drug is higher than for individual drug classes (0.968) and for use of all three drugs is much lower (0.906). Similarly, the proportion of patients adherent 70  with any drug class is 96.8% while for all three classes it is 84.5%. These adherence values exclude patients who filled only one prescription, and therefore have PDC of 1.00 by definition. Sensitivity tests indicated that excluding these individuals does not substantially change mean adherence, or the proportion of patients with PDC ≥0.80. Table 4.5: Adherence, measured as the proportion of days covered (PDC) from first prescription to first 90-day gap in treatment, by class and combination N  Mean PDC  % with PDC ≥0.80  ACE Inhibitors  10,202  0.955  95.1  Beta blockers  11,810  0.941  92.4  Statins  10,580  0.938  92.7  Any Class  15,503  0.968  96.8  5805  0.906  84.5  All 3 Classes  4.2.3  Adherence as PDC in first year The second definition of adherence, calculated as the PDC in the first 365 days  following a patient‟s first prescription, is shown in Table 4.6. This method, with a fixed denominator, gives lower values for adherence than above. The mean PDC in the first year ranges from 0.780 for ACE inhibitors to 0.833 for statins. The percent adherent, with PDC ≥0.80, follows the same order, with a low of 66.8% for ACE inhibitors and a high of 74.0% for statins. Compared to the adherence rates by drug class, the mean PDC and percent adherent are somewhat higher for users of any class (0.913 and 86.8%) and lower for users of all three classes (0.642 and 44.8%). These adherence values exclude patients who were censored in the first 365 days of follow-up and who therefore do not have a full year of observation available.  71  Table 4.6: Adherence, measured as the proportion of days covered (PDC) in the first 365 days of follow-up, by class and combination N  Mean PDC  % with PDC ≥0.80  ACE Inhibitors  10,975  0.780  66.8  Beta blockers  12,289  0.802  68.4  Statins  10,978  0.833  74.0  Any Class  15,033  0.913  86.8  6936  0.642  44.8  All 3 Classes  4.3  4.3.1  FACTORS ASSOCIATED WITH PERSISTENCE AND ADHERENCE  Model construction I constructed my model using the conceptual framework I defined, and grouped analysis  variables into demographic, health status, and drug use variables, which are predisposing factors in my framework, and socioeconomic variables, which are enabling factors. The initial calculation of Spearman correlation coefficients indicated that there were no strong relationships between analysis variables. The strongest correlations were between the number of unique prescribers post-AMI and the number of drug classes used (r = 0.45), between income group and private insurance (r = 0.44), and between age category and sex (r = 0.29). Preliminary models testing the effects of age, sex and their interaction, found that agesex interactions did not significantly affect persistence or adherence. The age-sex interaction term was therefore not included in further multivariate analyses. Multivariate models were constructed separately with demographic variables, socioeconomic variables, health status variables, and drug use variables. Results from these reduced models were nearly identical to results from the full models, which included all of the  72  analysis variables based on my conceptual framework, therefore only the full models are presented here. The models were also tested with and without HSDA. Adding HSDA did not alter the effect of any of the other analysis variables, therefore HSDA was included in full models. Tests of model fit indicated that all models had a statistically significant effect (p < 0.0001).  4.3.2  Multivariate analyses of persistence Cox proportional hazards modeling was used to test the effects of demographic,  socioeconomic, health status, and drug use variables on persistence. Table 4.7 shows results of the multivariate analysis, by drug class, with hazard ratios expressing the likelihood of experiencing a gap in therapy lasting at least 90 days. Across all drug classes, factors associated with decreased persistence (i.e. increased likelihood of a gap) are having very poor health status (indicated by 5 or more major ADGs, ranging from HR 1.27, 95% CI: 1.08-1.25, for ACE inhibitors to HR 1.53, 95% CI: 1.25-1.86, for statins) or hospitalization of less than 3 days for index AMI (HR 1.08, 95% CI: 1.00-1.17, for statins, to HR 1.16, 95% CI: 1.08-1.25, for ACE inhibitors). Factors associated with improved persistence across drug classes are having a private payer (ranging from HR 0.80, 95% CI: 0.75-0.86, for statins to HR 0.90, 95% CI: 0.840.95, for beta blockers), or using between 4 and 9 drug classes in total. Among users of ACE inhibitors, women were 10% more likely than men to experience a prolonged gap in treatment (HR 1.10, 95% CI: 1.03-1.17). Factors associated with improved ACE inhibitor persistence on the other hand were being between the age of 60-69 years, compared to the reference category of 40-49 years (HR 0.87, 95% CI: 0.79-0.96), being in the top income category (HR 0.86, 95% CI: 0.79-0.93), residing in long-term care (HR 0.79, 95% CI: 0.65-0.98) and using ACE inhibitors prior to the index AMI (HR 0.59, 95% CI: 0.55-0.63). 73  By contrast, among users of beta blockers, women were 15% less likely than men to experience a gap in treatment (HR 0.85, 95% CI: 0.80-0.90). Other factors associated with improved beta blocker persistence are increasing age, between 50 and 79 years, residing in longterm care (HR 0.79, 95% CI: 0.64-0.99), prior use of ACE inhibitors (HR 0.85, 95% CI: 0.800.91), beta blockers (HR 0.70, 95% CI: 0.64-0.75) or statins (HR 0.91, 95% CI: 0.84-0.98), and use of all three classes concurrently after AMI (HR 0.87, 95% CI: 0.83-0.92). Among statin users, persistence was not influenced by sex, but it was influenced similarly to the other two classes by age, income and prior drug use. Factors associated with improved statin persistence were increasing age, between 50 and 89 years, being in the top income category (HR 0.84, 95% CI: 0.77-0.86), prior statin use (HR 0.71, 95% CI: 0.65-0.78), and use of all three classes concurrently after AMI (HR 0.91, 95% CI: 0.86-0.98). In addition, statin persistence was also improved by having 2-3 unique prescribers. There was some regional variation in persistence across HSDAs, with the HSDA of Vancouver as a reference group, but it was not consistent across drug classes. Residing in Central Vancouver Island, for example, was associated with improved ACE inhibitor persistence (HR 0.88, 95% CI: 0.77-1.00), but worse beta blocker persistence (HR 1.38, 95% CI: 1.22-1.56).  74  Table 4.7: Factors associated with no longer being persistent with medication (experiencing a gap in treatment lasting more than 90 days), by drug class Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major ADGs* 4 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug classes 4-6 classes used post7-9 classes AMI* 10+ classes 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Garibaldi Area* South Vancouver Island Central Vancouver Island North Vancouver Island Northern Interior Northeast Northwest  HR 1.10 0.92 0.87 1.02 1.12 1.05 0.86 1.06 0.86 0.89 0.94 1.00 0.98 1.27 0.79 1.16 0.59 0.95 0.94 0.95 0.74 0.80 0.97 0.94 0.95 1.01 0.78 1.11 0.95 0.92 1.02 1.02 0.99 1.37 1.00 0.88 0.88 1.01 0.90 1.12 0.72  ACE Inhibitors 95% CI 1.03-1.17 0.83-1.02 0.79-0.96 0.92-1.13 1.00-1.26 0.86-1.28 0.81-0.92 0.99-1.14 0.79-0.93 0.77-1.01 0.85-1.04 0.90-1.12 0.86-1.12 1.07-1.50 0.65-0.98 1.08-1.25 0.55-0.63 0.88-1.02 0.87-1.02 0.89-1.00 0.67-0.82 0.72-0.89 0.86-1.10 0.87-1.02 0.87-1.04 0.91-1.11 0.65-0.95 0.93-1.32 0.84-1.07 0.80-1.06 0.89-1.17 0.90-1.15 0.89-1.11 1.15-1.63 0.87-1.14 0.77-1.01 0.77-1.00 0.85-1.21 0.76-1.06 0.90-1.39 0.56-0.93  p 0.003 0.112 0.007 0.751 0.051 0.648 0.000 0.081 0.000 0.079 0.220 0.984 0.741 0.006 0.029 0.000 0.000 0.150 0.139 0.057 0.000 0.000 0.659 0.161 0.263 0.873 0.012 0.235 0.403 0.243 0.723 0.765 0.917 0.000 0.986 0.062 0.044 0.873 0.207 0.317 0.012  HR 0.85 0.90 0.80 0.85 0.91 0.91 0.90 1.02 0.94 0.96 1.03 1.04 1.11 1.42 0.79 1.13 0.85 0.70 0.91 0.87 0.70 0.63 0.69 0.94 0.98 1.00 1.00 1.03 1.04 1.10 1.13 1.24 1.01 1.03 1.13 1.50 1.38 1.36 0.77 0.98 0.90  Beta Blockers 95% CI 0.80-0.90 0.82-0.99 0.73-0.88 0.77-0.94 0.82-1.02 0.74-1.12 0.84-0.95 0.95-1.09 0.87-1.01 0.84-1.09 0.94-1.14 0.94-1.15 0.98-1.27 1.20-1.68 0.64-0.99 1.05-1.21 0.80-0.91 0.64-0.75 0.84-0.98 0.83-0.92 0.64-0.76 0.57-0.69 0.62-0.78 0.87-1.02 0.90-1.07 0.91-1.10 0.83-1.20 0.87-1.23 0.92-1.17 0.96-1.25 0.99-1.29 1.10-1.40 0.90-1.14 0.87-1.23 0.99-1.29 1.32-1.70 1.22-1.56 1.15-1.60 0.64-0.92 0.78-1.22 0.72-1.13  p 0.000 0.026 0.000 0.001 0.111 0.368 0.000 0.564 0.097 0.530 0.532 0.445 0.095 0.000 0.037 0.001 0.000 0.000 0.009 0.000 0.000 0.000 0.000 0.112 0.628 0.946 0.967 0.714 0.536 0.026 0.077 0.000 0.824 0.726 0.070 0.000 0.000 0.000 0.004 0.860 0.354  HR 0.96 0.78 0.71 0.68 0.87 1.26 0.80 1.04 0.84 0.89 1.02 1.06 1.06 1.53 1.28 1.08 0.93 0.95 0.71 0.92 0.71 0.62 0.78 0.88 0.86 0.91 1.13 1.40 1.07 1.12 1.12 1.06 1.11 1.07 1.13 1.12 1.13 1.07 1.06 1.12 1.26  Statins 95% CI 0.90-1.03 0.71-0.87 0.64-0.78 0.61-0.76 0.77-0.99 0.98-1.62 0.75-0.86 0.97-1.12 0.77-0.92 0.77-1.04 0.91-1.13 0.94-1.19 0.91-1.23 1.25-1.86 0.98-1.67 1.00-1.17 0.86-1.01 0.87-1.03 0.65-0.78 0.86-0.98 0.65-0.79 0.55-0.69 0.68-0.89 0.81-0.97 0.78-0.95 0.82-1.02 0.92-1.39 1.14-1.71 0.93-1.23 0.96-1.31 0.96-1.31 0.93-1.22 0.97-1.26 0.88-1.30 0.97-1.32 0.97-1.30 0.98-1.30 0.88-1.29 0.88-1.29 0.89-1.42 0.98-1.62  p 0.303 0.000 0.000 0.000 0.028 0.074 0.000 0.280 0.000 0.134 0.786 0.351 0.438 0.000 0.071 0.048 0.070 0.213 0.000 0.007 0.000 0.000 0.000 0.007 0.003 0.097 0.241 0.001 0.336 0.157 0.143 0.379 0.119 0.514 0.108 0.119 0.099 0.497 0.537 0.346 0.074  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  75  Table 4.8: Factors associated with no longer being persistent with medication (experiencing a gap in treatment lasting more than 90 days), by drug combination Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major ADGs* 4 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug classes 4-6 classes used post7-9 classes AMI* 10+ classes 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Garibaldi Area* South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  HR 0.91 0.72 0.60 0.68 0.94 1.17 0.82 1.04 0.77 1.05 1.01 1.08 1.22 1.78 0.92 1.24 0.71 0.67 0.63 0.69 0.61 0.52 0.60 0.83 0.84 0.82 1.09 1.25 1.05 0.98 1.16 1.11 1.06 1.18 1.12 1.15 1.08 1.15 0.97 0.98 1.12  Any Class 95% CI 0.84-0.98 0.64-0.81 0.53-0.67 0.61-0.77 0.83-1.07 0.95-1.44 0.76-0.88 0.96-1.12 0.70-0.85 0.90-1.23 0.89-1.14 0.95-1.23 1.05-1.43 1.46-2.16 0.74-1.15 1.13-1.35 0.65-0.78 0.61-0.74 0.57-0.70 0.64-0.74 0.55-0.67 0.46-0.58 0.53-0.69 0.76-0.92 0.76-0.93 0.73-0.92 0.87-1.37 1.01-1.54 0.91-1.22 0.83-1.16 0.98-1.36 0.96-1.28 0.92-1.22 0.96-1.45 0.95-1.31 0.99-1.35 0.92-1.26 0.94-1.42 0.73-1.30 0.79-1.21 0.85-1.48  p 0.011 0.000 0.000 0.000 0.356 0.133 0.000 0.385 0.000 0.519 0.905 0.221 0.011 0.000 0.469 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.001 0.459 0.041 0.482 0.821 0.078 0.178 0.406 0.123 0.177 0.075 0.349 0.177 0.858 0.821 0.411  HR 1.03 0.97 0.88 0.96 1.04 1.40 0.87 1.06 0.92 0.78 1.01 1.05 1.01 1.34 0.84 1.06 0.72 0.87 0.94 0.67 0.68 0.82 1.02 1.07 1.17 0.97 1.01 0.99 1.02 1.09 1.14 0.98 1.20 1.04 1.16 1.24 1.15 1.04 0.85 1.02  All Three Classes 95% CI p 0.97-1.10 0.353 0.88-1.07 0.566 0.80-0.98 0.015 0.86-1.06 0.384 0.92-1.18 0.507 1.05-1.88 0.024 0.81-0.93 0.000 0.98-1.14 0.156 0.84-1.00 0.052 0.67-0.91 0.001 0.91-1.12 0.913 0.94-1.17 0.387 0.88-1.17 0.881 1.10-1.64 0.003 0.61-1.15 0.272 0.98-1.14 0.171 0.66-0.78 0.000 0.79-0.95 0.001 0.86-1.02 0.145 0.60-0.75 0.60-0.77 0.71-0.95 0.93-1.12 0.97-1.18 1.05-1.30 0.79-1.17 0.83-1.24 0.87-1.13 0.88-1.18 0.94-1.27 1.00-1.30 0.87-1.11 0.99-1.45 0.90-1.21 1.01-1.33 1.08-1.43 0.96-1.38 0.82-1.33 0.71-1.02 0.82-1.27  0.000 0.000 0.008 0.631 0.198 0.005 0.728 0.908 0.896 0.839 0.234 0.057 0.799 0.066 0.599 0.036 0.002 0.126 0.749 0.077 0.842  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  The multivariate analyses of persistence with any class of drug, or with all three classes, showed patterns similar to those in the analyses by drug class (Table 4.8). Private insurance, prior ACE inhibitor, beta blocker or statin use, and the use of more drug classes following AMI were associated with improved persistence, while very poor health status was associated with worse persistence. The age gradient was also pronounced in these two analyses. Patients aged  76  50-79 years were less likely to experience a gap with any class of treatment; however, for users of all three classes, patients aged 60-69 were 12% less likely to experience a gap in treatment (HR 0.88, 95% CI: 0.80-0.98) and patients aged 90-99 were 40% more likely to experience a gap (HR 1.40, 95% CI: 1.05-1.88). For users of any drug class, female sex was associated with improved persistence (HR 0.91, 95% CI: 0.84-0.98), as was being in the top income category (HR 0.77, 95% CI: 0.70-0.85), and having more than one prescriber. Having a short initial hospitalization was associated with worse persistence (HR 1.24, 95% CI: 1.13-1.35). For users of all three classes concurrently, receiving social assistance (HR 0.78, 95% CI: 0.67-0.91) decreased the likelihood of experiencing a gap in therapy. In contrast to persistence with any class of drug, patients with 4 or more prescribers were 17% more likely to experience a gap in therapy (HR 1.17, 95% CI: 1.05-1.30).  4.3.3  Multivariate analyses of adherence Multivariate analyses for the first measure of adherence, using PDC between first  prescription and first gap exceeding 90 days, are shown in Table 4.9 and Table 4.10. In these tables, odds ratios show the likelihood that a patient has as a PDC ≥0.80 and is considered adherent with treatment. Compared to the models of persistence, relatively few variables seem to affect adherence. Sex does not influence adherence with ACE inhibitors, beta blockers or statins, while increased age has a strong effect across all age categories, from 50 to 99 years. The effect is strongest for patients aged 80-89, who are 2.87 times more likely to be adherent with beta blockers (95% CI: 2.15-3.85), 3.83 times more likely to be adherent with ACE inhibitors (95% CI: 2.56-5.73), and 4.13 times more likely to be adherent with statins (95% CI: 2.95-5.79), as  77  compared to the reference category of patients aged 40-49 years. Other factors associated with having PDC ≥0.80 include being in the highest income category (ranging from OR 1.23, 95% CI: 1.01-1.58, for beta blockers to OR 1.43, 95% CI: 1.07-1.90 for ACE inhibitors), and using more classes of drugs following AMI. Among users of ACE inhibitors, having private insurance (OR 1.41, 95% CI: 1.14-1.74) and living in long-term care (OR 3.23, 95% CI: 1.01-10.4) are positively associated with a PDC ≥0.80. For users of beta blockers, long-term care shows much the same effect on adherence (OR 2.97, 95% CI: 1.29-6.82), and having 3 prescribers also influences adherence (OR 1.29, 95% CI: 1.02-1.63). Lastly, for statin users having private insurance (OR 1.26, 95% CI: 1.06-1.50) and using ACE inhibitors pre-AMI (OR 1.40, 95% CI: 1.12-1.76) both positively influence adherence. There is no regional variation in adherence to ACE inhibitors, and only residents of Central and North Vancouver Island appear to be significantly less adherent with beta blockers than the reference HSDA of Vancouver. However, for statins, six HSDAs show a decreased likelihood of adherence. The effect size ranges from a 38% odds reduction in Fraser East (OR 0.62, 95% CI: 0.41-0.96) to a 58% odds reduction in the Northern Interior (OR 0.42, 95% CI: 0.27-0.65).  78  Table 4.9: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered from first prescription to first 90-day gap in therapy, by drug class Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major ADGs* 4 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug classes 4-6 classes used post7-9 classes AMI* 10+ classes 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Garibaldi Area* South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  OR 1.06 1.35 2.29 2.43 3.83 3.82 1.41 0.84 1.43 1.16 0.99 0.80 1.05 0.67 3.24 0.83 0.98 1.33 1.07 1.18 1.36 1.52 1.56 0.93 0.92 0.83 2.12 0.85 1.03 1.13 0.68 1.14 0.90 2.46 1.21 1.22 1.26 1.27 1.43 0.83 0.83  ACE Inhibitors 95% CI 0.86-1.32 1.02-1.79 1.69-3.09 1.78-3.32 2.56-5.73 1.77-8.27 1.14-1.74 0.67-1.06 1.08-1.90 0.77-1.74 0.69-1.40 0.55-1.15 0.65-1.68 0.38-1.19 1.01-10.40 0.66-1.06 0.77-1.25 1.00-1.77 0.82-1.40 0.97-1.44 0.89-2.06 1.10-2.10 1.09-2.25 0.69-1.24 0.66-1.26 0.59-1.18 0.98-4.60 0.47-1.52 0.68-1.57 0.71-1.81 0.44-1.04 0.75-1.74 0.62-1.31 0.96-6.31 0.75-1.97 0.76-1.97 0.80-2.00 0.69-2.33 0.63-3.25 0.50-1.37 0.42-1.63  p 0.570 0.034 0.000 0.000 0.000 0.001 0.002 0.142 0.013 0.490 0.938 0.228 0.854 0.171 0.049 0.138 0.888 0.052 0.630 0.098 0.012 0.016 0.155 0.617 0.590 0.309 0.056 0.578 0.871 0.601 0.078 0.537 0.582 0.060 0.436 0.403 0.317 0.448 0.398 0.458 0.587  OR 1.16 1.42 2.23 2.41 2.87 1.90 1.16 1.06 1.24 1.41 0.88 0.77 0.81 1.06 2.97 0.99 1.17 1.17 0.96 1.08 1.34 1.37 1.32 1.14 1.29 1.02 1.16 0.72 1.04 1.08 0.77 0.96 1.03 1.16 1.29 0.90 0.65 0.64 0.88 0.73 0.83  Beta Blockers 95% CI 0.98-1.36 1.14-1.76 1.77-2.81 1.90-3.06 2.15-3.83 1.15-3.16 0.99-1.36 0.89-1.27 1.01-1.53 0.99-1.99 0.68-1.14 0.59-1.02 0.58-1.15 0.64-1.76 1.29-6.82 0.82-1.19 0.98-1.41 0.95-1.45 0.79-1.17 0.93-1.25 1.05-1.71 1.04-1.80 0.97-1.81 0.93-1.41 1.02-1.63 0.79-1.31 0.70-1.92 0.47-1.12 0.75-1.43 0.74-1.56 0.54-1.09 0.69-1.33 0.76-1.40 0.71-1.89 0.87-1.90 0.63-1.28 0.47-0.89 0.42-0.98 0.51-1.50 0.50-1.08 0.49-1.40  p 0.084 0.002 0.000 0.000 0.000 0.013 0.071 0.514 0.037 0.055 0.321 0.069 0.239 0.815 0.010 0.893 0.090 0.133 0.707 0.297 0.018 0.023 0.081 0.208 0.035 0.904 0.574 0.143 0.814 0.689 0.142 0.786 0.833 0.562 0.201 0.545 0.008 0.038 0.631 0.118 0.486  OR 0.96 1.70 2.85 3.77 4.13 3.71 1.26 0.95 1.26 1.26 1.03 1.02 1.01 1.12 0.70 1.04 1.40 1.13 0.94 1.10 1.34 1.63 2.29 1.01 0.94 0.97 0.73 0.46 0.77 0.59 0.62 0.76 0.57 0.81 0.85 0.81 0.70 0.59 0.61 0.42 0.70  Statins 95% CI 0.80-1.14 1.37-2.10 2.26-3.59 2.92-4.88 2.95-5.79 1.66-8.30 1.06-1.50 0.78-1.15 1.00-1.58 0.87-1.82 0.79-1.34 0.77-1.36 0.69-1.47 0.61-2.04 0.31-1.55 0.85-1.27 1.12-1.76 0.89-1.44 0.75-1.17 0.94-1.29 1.04-1.73 1.21-2.20 1.58-3.32 0.80-1.27 0.73-1.21 0.73-1.29 0.42-1.27 0.28-0.77 0.51-1.15 0.39-0.90 0.41-0.95 0.51-1.12 0.40-0.82 0.46-1.41 0.55-1.33 0.53-1.23 0.47-1.05 0.36-0.95 0.33-1.14 0.27-0.66 0.38-1.28  p 0.626 0.000 0.000 0.000 0.000 0.001 0.008 0.571 0.046 0.217 0.821 0.886 0.964 0.724 0.378 0.682 0.003 0.309 0.564 0.245 0.025 0.001 0.000 0.955 0.613 0.834 0.271 0.003 0.195 0.014 0.026 0.164 0.002 0.452 0.477 0.327 0.084 0.030 0.123 0.000 0.247  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  79  Table 4.10: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered from first prescription to first 90-day gap in therapy, by drug combination Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major 4 ADGs* 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug 4-6 classes classes 7-9 classes used post10+ classes AMI* 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Area* Garibaldi South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  OR 1.13 1.94 3.16 3.31 3.77 4.83 1.21 1.05 1.53 1.04 0.84 0.75 0.87 0.69 0.97 0.88 1.87 1.55 1.63 1.50 1.42 1.56  Any Class 95% CI 0.91-1.40 1.49-2.53 2.37-4.23 2.45-4.46 2.62-5.41 2.34-9.99 0.98-1.49 0.84-1.32 1.16-2.02 0.70-1.57 0.60-1.18 0.52-1.08 0.54-1.38 0.38-1.28 0.48-1.96 0.69-1.11 1.40-2.49 1.13-2.12 1.19-2.24 1.23-1.82 1.07-1.87 1.13-2.16  p 0.268 0.000 0.000 0.000 0.000 0.000 0.081 0.681 0.003 0.836 0.323 0.123 0.550 0.241 0.939 0.275 0.000 0.006 0.002 0.000 0.014 0.008  1.36  0.92-2.02  0.122  1.86  1.25-2.77  0.002  1.05 1.10 1.42 1.16 0.70 1.01 1.09 0.62 1.12 0.96 1.04  0.81-1.37 0.82-1.48 1.01-2.01 0.60-2.25 0.40-1.23 0.66-1.54 0.67-1.76 0.41-0.95 0.73-1.71 0.65-1.43 0.56-1.94  0.714 0.535 0.046 0.657 0.213 0.982 0.734 0.029 0.615 0.853 0.897  1.02 0.92 0.82 0.79 0.65 0.71 0.66 0.57 0.79 0.72 1.03  0.81-1.29 0.72-1.19 0.62-1.08 0.47-1.32 0.39-1.08 0.50-1.01 0.45-0.97 0.38-0.86 0.54-1.15 0.51-1.00 0.57-1.87  0.871 0.530 0.154 0.365 0.097 0.059 0.036 0.007 0.211 0.053 0.923  1.27  0.77-2.11  0.347  0.82  0.54-1.23  0.329  1.45 0.87 1.09 1.50 0.69 0.59  0.87-2.43 0.57-1.35 0.59-2.00 0.62-3.60 0.41-1.14 0.31-1.11  0.153 0.542 0.791 0.366 0.149 0.100  0.56 0.56 0.48 0.70 0.65 0.87  0.38-0.81 0.39-0.81 0.31-0.76 0.38-1.27 0.43-1.00 0.49-1.53  0.003 0.002 0.002 0.238 0.051 0.622  OR 1.24 1.20 1.57 1.91 3.80 1.86 1.13 0.94 1.12 1.73 0.99 0.86 0.71 1.45 1.55 0.95 1.01 1.12 0.85 1.51 1.52  All Three Classes 95% CI p 1.03-1.48 0.021 0.95-1.50 0.119 1.25-1.99 0.000 1.48-2.46 0.000 2.62-5.51 0.000 0.80-4.30 0.148 0.96-1.33 0.152 0.78-1.14 0.556 0.90-1.39 0.313 1.16-2.56 0.007 0.77-1.28 0.957 0.65-1.13 0.281 0.50-1.01 0.059 0.79-2.67 0.235 0.54-4.44 0.414 0.79-1.16 0.632 0.83-1.23 0.916 0.89-1.41 0.328 0.69-1.05 0.138 1.10-2.08 1.07-2.16  0.011 0.019  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  The models for users of any class or of all three classes show similar effects to the analyses by class, with some small differences. For users of any class of drug, increased age (all categories, 50-99 years), increased income (top 30%) and the use of more drug classes (up to 10) remain positively associated with a PDC ≥0.80. In addition to these factors, prior use of either ACE inhibitors (OR 1.87, 95% CI: 1.40-2.49), beta blockers (OR 1.55, 95% CI: 1.13-2.12) or 80  statins (OR 1.63, 95% CI: 1.19-2.24) increases the likelihood of adherence, as does the use of all three classes following AMI (OR 1.50, 95% CI: 1.23-1.82). Patients with at least four prescribers are also more likely to be adherent (OR 1.42, 95% CI: 1.01-2.01) while patients living in the HSDA of Fraser East are less likely to be adherent (OR 0.62, 95% CI: 0.41-0.95). Among users of all three classes the effect of increased age is reduced to only patients 60-89 years, and it is also the only model in which female sex is positively associated with adherence, increasing the likelihood of having PDC ≥0.80 by 24% (OR 1.24, 95% CI: 1.031.48). Social assistance is also positively associated with adherence for the first time, with an OR of 1.73 (95% CI: 1.16-2.56) among users of all three drugs. Residents in five HSDAs are less likely to be adherent to all three drugs, with odds reductions ranging from 34% in Thompson Cariboo (OR 0.66, 95% CI: 0.45-0.97) to 52% in North Vancouver Island (OR 0.42, 95% CI: 0.27-0.65).  4.3.4  Multivariate analyses of adherence as PDC in first year The multivariate models of adherence measured as 1-year PDC behave much like the  multivariate model of persistence (shown in Table 4.11 and Table 4.12). Many patient factors have a significant positive association with 1-year PDC ≥0.80 across all three drug classes, including having private insurance (ranging from OR 1.13, 95% CI: 1.04-1.24, for beta blockers to OR 1.35, 95% CI: 1.22-1.49, for statins), being in the top income category (OR 1.13, 95% CI: 1.01-1.27, for beta blockers to OR 1.21, 95% CI: 1.07-1.36, for ACE inhibitors), using ACE inhibitors before AMI (OR 1.15, 95% CI: 1.03-1.29, for statins to OR 1.91, 95% CI: 1.71-2.13, for ACE inhibitors), using all three classes concurrently post-AMI (OR 1.11, 95% CI: 1.02-1.21, for ACE inhibitors to OR 1.20, 95% CI: 1.10-1.30, for beta blockers), and using more drug classes of any kind post-AMI. Having worse health status, with 5 or more major ADGs, 81  decreased the likelihood of having 1-year PDC ≥0.80 (ranging from OR 0.74, 95% CI: 0.560.96, for ACE inhibitors to OR 0.59, 95% CI: 0.44-0.79, for ACE inhibitors). Female sex decreased the likelihood of having 1-year PDC ≥0.80 for users of ACE inhibitors by 16% (OR 0.84, 95% CI: 0.77-0.92), and having a short hospitalization for index AMI also decreased that likelihood, by 14% (OR 0.86, 95% CI: 0.77-0.96). By contrast, among beta blocker users female sex increased the odds of having 1-year PDC ≥0.80 by 33% (OR 1.33, 95% CI: 1.21-1.45). Beta blocker users were also more likely to be adherent if they had previously used beta blockers, before their AMI (OR 1.42, 95% CI: 1.27-1.60), or if they were above 60 years of age. Among statin users, patients between 50-89 years old were more likely to have 1-year PDC ≥0.80, as were prior statin users (OR 1.27, 95% CI: 1.11-1.44) and patients with 2 or 3 unique prescribing physicians. There was no consistent regional variation in 1-year PDC by drug class. For ACE inhibitors, residents of East Kootenay were slightly more likely than the reference HSDA of Vancouver to have 1-year PDC ≥0.80, while residents of Richmond were less likely. For beta blockers, all three HSDAs on Vancouver Island had a decreased likelihood of having 1-year PDC ≥0.80, while for statins only one HSDA, Kootenay Boundary, showed this effect.  82  Table 4.11: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered in the 365 days following first prescription (1-year PDC), by drug class Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major ADGs* 4 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug classes 4-6 classes used post7-9 classes AMI* 10+ classes 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Garibaldi Area* South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  OR 0.84 1.05 1.14 0.99 0.96 1.15 1.23 0.91 1.21 1.15 1.02 0.96 1.01 0.74 1.30 0.86 1.91 1.06 1.07 1.11 1.84 1.62 1.33 1.02 0.99 0.86 1.34 1.06 1.13 1.04 0.87 0.90 1.02 0.66 1.04 1.18 1.17 1.13 1.26 1.09 0.94  ACE Inhibitors 95% CI 0.77-0.92 0.90-1.22 0.97-1.33 0.84-1.15 0.81-1.15 0.85-1.55 1.12-1.35 0.82-1.00 1.07-1.36 0.93-1.40 0.88-1.20 0.81-1.13 0.83-1.24 0.56-0.96 0.94-1.78 0.77-0.96 1.71-2.13 0.94-1.18 0.95-1.20 1.02-1.21 1.58-2.14 1.37-1.92 1.10-1.61 0.90-1.16 0.86-1.14 0.74-1.00 1.01-1.79 0.80-1.40 0.94-1.35 0.85-1.28 0.71-1.07 0.74-1.08 0.86-1.21 0.50-0.87 0.85-1.28 0.97-1.45 0.96-1.42 0.87-1.48 0.88-1.80 0.85-1.40 0.67-1.30  p 0.000 0.553 0.106 0.858 0.662 0.373 0.000 0.062 0.002 0.191 0.773 0.620 0.905 0.024 0.110 0.009 0.000 0.364 0.281 0.020 0.000 0.000 0.003 0.741 0.928 0.057 0.042 0.675 0.206 0.704 0.196 0.239 0.836 0.004 0.679 0.105 0.124 0.367 0.203 0.505 0.696  OR 1.33 1.12 1.34 1.31 1.22 1.42 1.13 0.97 1.13 1.08 0.91 0.89 0.88 0.70 1.33 0.93 1.24 1.42 1.02 1.20 1.80 2.02 1.82 1.10 1.07 0.96 1.00 0.96 0.97 0.94 0.75 0.75 0.94 0.98 0.89 0.63 0.67 0.61 1.04 1.16 0.92  Beta Blockers 95% CI 1.21-1.45 0.97-1.29 1.16-1.55 1.13-1.51 1.03-1.44 1.03-1.95 1.04-1.24 0.88-1.08 1.01-1.27 0.88-1.31 0.78-1.05 0.76-1.04 0.72-1.06 0.54-0.91 0.96-1.83 0.84-1.04 1.12-1.37 1.27-1.60 0.91-1.13 1.10-1.30 1.57-2.07 1.73-2.35 1.53-2.17 0.98-1.24 0.94-1.22 0.83-1.11 0.76-1.31 0.74-1.26 0.81-1.16 0.77-1.16 0.61-0.91 0.62-0.89 0.79-1.11 0.75-1.28 0.72-1.08 0.52-0.77 0.56-0.82 0.48-0.79 0.75-1.44 0.90-1.49 0.67-1.27  p 0.000 0.118 0.000 0.000 0.021 0.031 0.007 0.612 0.034 0.460 0.197 0.137 0.180 0.007 0.082 0.188 0.000 0.000 0.783 0.000 0.000 0.000 0.000 0.111 0.318 0.578 0.985 0.772 0.705 0.569 0.005 0.001 0.445 0.894 0.237 0.000 0.000 0.000 0.806 0.253 0.624  OR 1.03 1.22 1.48 1.57 1.31 0.84 1.35 0.92 1.20 1.17 0.96 0.96 0.84 0.59 0.69 0.96 1.15 1.06 1.27 1.16 1.88 2.31 1.87 1.21 1.19 1.04 0.91 0.63 0.95 0.87 0.92 0.91 0.84 0.90 1.10 1.01 0.85 0.88 0.77 0.78 0.81  Statins 95% CI 0.93-1.14 1.05-1.42 1.27-1.73 1.34-1.84 1.09-1.58 0.57-1.23 1.22-1.49 0.83-1.03 1.06-1.37 0.94-1.45 0.81-1.13 0.81-1.15 0.67-1.04 0.44-0.79 0.46-1.03 0.86-1.08 1.03-1.29 0.93-1.20 1.11-1.44 1.06-1.27 1.62-2.20 1.94-2.75 1.54-2.29 1.05-1.38 1.03-1.38 0.89-1.23 0.67-1.24 0.47-0.85 0.77-1.17 0.69-1.09 0.73-1.16 0.74-1.12 0.69-1.01 0.67-1.21 0.87-1.40 0.81-1.26 0.68-1.04 0.67-1.17 0.53-1.12 0.59-1.02 0.57-1.14  p 0.589 0.010 0.000 0.000 0.005 0.362 0.000 0.165 0.005 0.153 0.605 0.662 0.111 0.000 0.072 0.526 0.016 0.387 0.000 0.002 0.000 0.000 0.000 0.006 0.022 0.613 0.554 0.003 0.644 0.225 0.499 0.378 0.067 0.486 0.408 0.939 0.119 0.379 0.172 0.065 0.227  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  83  Table 4.12: Factors associated with adherence to medication, measured as having ≥0.80 proportion of days covered in the 365 days following first prescription (1-year PDC), by drug combination Variable Sex Age Category*  Value Female 50-59 years 60-69 years 70-79 years 80-89 years 90-99 years  Private Payer Income middle 40% Group* top 30% Social Assistance 2 Sum of 3 major ADGs* 4 5+ Long-term care Hospitalized < 3 days ACE Inhibitor Prior drug Beta blocker use Statin Use of combination (all 3) post-AMI Drug classes 4-6 classes used post7-9 classes AMI* 10+ classes 2 prescribers Prescribers 3 prescribers post-AMI* 4+ prescribers East Kootenay Kootenay Boundary Okanagan Thompson Cariboo Fraser East Fraser North Health Fraser South Services Richmond Delivery North Shore/Coast Garibaldi Area* South Vancouver Island Central Vancouver Island North Vancouver Island Northwest Northern Interior Northeast  OR 1.07 1.32 1.63 1.49 1.18 1.11 1.38 0.90 1.37 0.80 0.93 0.85 0.69 0.46 0.85 0.83 1.73 1.62 1.83 1.83 2.12 2.70 2.39 1.32 1.36 1.30 0.95 0.78 0.94 0.94 0.76 0.74 0.87 0.71 0.96 0.91 0.86 0.76 0.89 0.84 0.70  Any Class 95% CI 0.95-1.19 1.10-1.57 1.36-1.95 1.24-1.78 0.97-1.44 0.80-1.53 1.23-1.55 0.80-1.02 1.18-1.59 0.64-1.00 0.77-1.13 0.69-1.03 0.54-0.87 0.34-0.62 0.62-1.18 0.73-0.95 1.51-1.99 1.38-1.89 1.55-2.16 1.65-2.04 1.83-2.45 2.28-3.22 1.94-2.94 1.15-1.52 1.16-1.60 1.09-1.55 0.67-1.36 0.56-1.08 0.74-1.18 0.73-1.22 0.59-0.97 0.60-0.93 0.70-1.08 0.51-0.97 0.74-1.24 0.71-1.17 0.67-1.09 0.55-1.05 0.58-1.36 0.62-1.16 0.47-1.06  p 0.268 0.002 0.000 0.000 0.103 0.522 0.000 0.092 0.000 0.055 0.478 0.101 0.002 0.000 0.341 0.005 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.004 0.780 0.138 0.584 0.661 0.030 0.009 0.205 0.030 0.768 0.452 0.209 0.091 0.587 0.297 0.091  OR 1.03 1.02 1.14 1.10 1.03 0.91 1.14 0.97 1.19 1.55 0.95 0.92 0.91 0.76 1.42 0.96 1.54 1.13 0.98 2.22 2.09 1.76 0.93 0.88 0.70 0.97 1.01 1.00 0.91 0.81 0.82 0.99 0.68 0.98 0.83 0.72 0.78 1.00 1.11 0.92  All Three Classes 95% CI p 0.92-1.15 0.616 0.86-1.20 0.845 0.96-1.35 0.142 0.92-1.31 0.313 0.83-1.28 0.783 0.55-1.52 0.728 1.02-1.28 0.020 0.85-1.10 0.596 1.03-1.37 0.018 1.22-1.98 0.000 0.80-1.13 0.553 0.76-1.12 0.404 0.72-1.17 0.467 0.54-1.07 0.119 0.84-2.41 0.194 0.84-1.10 0.564 1.36-1.76 0.000 0.98-1.30 0.103 0.85-1.12 0.733 1.79-2.76 1.65-2.65 1.35-2.29 0.80-1.09 0.74-1.04 0.58-0.84 0.70-1.34 0.72-1.41 0.80-1.24 0.71-1.17 0.63-1.05 0.65-1.03 0.81-1.21 0.49-0.96 0.76-1.25 0.65-1.05 0.57-0.92 0.57-1.07 0.67-1.50 0.84-1.48 0.64-1.32  0.000 0.000 0.000 0.385 0.128 0.000 0.869 0.951 0.969 0.483 0.113 0.088 0.908 0.028 0.855 0.121 0.008 0.122 0.994 0.457 0.658  *Reference categories are: Age=40-49 yrs; Income Group=bottom 30%; sum of ADGs=1; drug classes post-AMI=1-3 classes; prescribers post-AMI=1 prescriber; HSDA=Vancouver; bold text indicates p<0.05  For all users of ACE inhibitors, beta blockers or statins, many factors increased the likelihood of having 1-year PDC ≥0.80, including being older (between 50-89 years), having private insurance (OR 1.38, 95% CI: 1.23-1.55), being in the top income class (OR 1.37, 1.181.59), prior use of ACE inhibitors (OR 1.73, 95% CI: 1.51-1.99), beta blockers (OR 1.62, 95% CI: 1.38-1.89) or statins (OR 1.83, 95% CI: 1.55-2.16), and use of all three in combination following AMI (1.83, 95% CI: 1.65-2.04). Using more drug classes and having more prescribers 84  were also associated with higher 1-year PDC, while worse health status (4 or more major ADGs) and short hospitalization for index AMI were negatively associated with 1-year PDC ≥0.80. Far fewer factors were associated with use of all three drug classes concurrently. Higher income and private insurance remained significant positive predictors of 1-year PDC ≥0.80 (OR 1.19, 95% CI: 1.03-1.37, and OR 1.14, 95% CI: 1.02-1.28, respectively), while age, prior beta blocker or statin use, poor health status and shorter hospitalization no longer had an effect. The use of more drug classes post-AMI also increased the likelihood of 1-year PDC ≥0.80, but having 4 or more prescribers decreased the likelihood of being adherent (OR 0.70, 95% CI: 0.58-0.84). Lastly, being in receipt of social assistance increased the likelihood of adherence with all three drugs in combination (OR 1.55, 95% CI: 1.21-1.98). Patients living in the HSDA of Richmond were less likely to have 1-year PDC ≥0.80 both for any drug class and for all three classes in combination. No other HSDA had a consistent effect across both models.  4.4 SUMMARY Unadjusted values for persistence and the two measures of adherence, by drug class and combination, and the results of the multivariate analyses are summarized below (Table 4.13). In general, increasing age (above the reference group of 40-49 years), higher income, having private insurance and using more pharmaceuticals, both before and after AMI, increased the likelihood of adherent and/or persistent medication-taking behaviour. Very poor health status had the opposite relationship, and was negatively associated with adherence and persistence.  85  Table 4.13: Summary of findings of multivariate analyses, by outcome measure, drug class and combination Outcome Persistence  Class ACE inhibitors  Result Median: 1291 d 78.7% persistent at 6 mo; 42.2% at 5 yrs  Beta blockers  Median: 1380 d 82.8% persistent at 6 mo; 43.8% at 5 yrs  Statins  Median: 1799 d 86.0% persistent at 6 mo; 49.6% at 5 yrs  Any Class  Median: 2183 d 93.1% persistent at 6 mo; 68.1% at 5 yrs  All 3 Classes  Median: 397 d 65.3% persistent at 6 mo; 21.2% at 5 yrs  Adherence (PDC from start to first 90-day gap)  ACE inhibitors  Mean: 0.955  Beta blockers  Mean: 0.941  95.1% with PDC ≥0.80  92.4% with PDC ≥0.80 Statins  Mean: 0.938 92.7% with PDC ≥0.80  Any Class  Mean: 0.968 96.8% with PDC ≥0.80  All 3 Classes  Mean: 0.906 84.5% with PDC ≥0.80  Positive Effect Age 60-69 years Private insurance High income (top 30%) Long-term care Use of multiple drug classes (4-9) Prior ACE inhibitor use Female sex Age 50-79 years Private insurance Long-term care Prior ACE inhibitor, beta blocker or statin use Use of all 3 classes Use of multiple drug classes (≥4) Age 50-89 years Private insurance High income (top 30%) Prior statin use Use of all 3 classes Use of multiple drug classes (≥4) More prescribers (2-3) Female sex Age 50-79 years Private insurance High income (top 30%) Prior ACE inhibitor, beta blocker or statin use Use of all 3 classes Use of multiple drug classes (≥4) More prescribers (≥2) Age 60-69 years Private insurance Social assistance Prior ACE inhibitor or beta blocker use Use of multiple drug classes (≥4) Age 50-99 years Private insurance High income (top 30%) Long-term care Use of multiple drug classes (4-9) Age 50-99 years High income (top 30%) Long-term care Use of multiple drug classes (4-9) More prescribers (3) Age 50-99 years Private insurance High income (top 30%) Prior ACE inhibitor use Use of multiple drug classes (4-9) Age 50-99 years High income (top 30%) Prior ACE inhibitor, beta blocker or statin use Use of all 3 classes Use of multiple drug classes (4-9) More prescribers (≥4) Female sex Age 60-89 years Social assistance Use of multiple drug classes (≥4)  Negative Effect Female sex Poor health status (≥5 major ADGs) Short hospitalization Poor health status (≥5 major ADGs) Short hospitalization  Poor health status (≥5 major ADGs) Short hospitalization  Poor health status (≥4 major ADGs) Short hospitalization  Age 90-99 years More prescribers (≥4)  86  Table 4.13: Summary of findings of multivariate analyses, by outcome measure, drug class and combination (continued) Outcome Adherence (1-year PDC)  Class ACE inhibitors  Result Mean: 0.780 66.8% with PDC ≥0.80  Beta blockers  Mean: 0.802 68.4% with PDC ≥0.80  Statins  Mean: 0.833 74.0% with PDC ≥0.80  Any Class  Mean: 0.913 86.8% with PDC ≥0.80  All 3 Classes  Mean: 0.642 44.8% with PDC ≥0.80  Positive Effect Private insurance High income (top 30%) Use of multiple drug classes (≥4) Prior ACE inhibitor use Use of all 3 classes Use of multiple drug classes (≥4) Female sex Age 60-99 years Private insurance Prior ACE inhibitor or beta blocker use Use of all 3 classes Use of multiple drug classes (≥4) Age 50-89 years Private insurance High income (top 30%) Prior ACE inhibitor or statin use Use of all 3 classes Use of multiple drug classes (≥4) More prescribers (2-3) Age 50-79 years Private insurance High income (top 30%) Prior ACE inhibitor, beta blocker or statin use Use of all 3 classes Use of multiple drug classes (≥4) More prescribers (≥2) Private insurance High income (top 30%) Social assistance Prior ACE inhibitor use Use of multiple drug classes (≥4)  Negative Effect Female sex Poor health status (≥5 major ADGs) Short hospitalization Poor health status (≥5 major ADGs)  Poor health status (≥5 major ADGs)  Poor health status (≥4 major ADGs) Short hospitalization  More prescribers (≥4)  After adjusting for demographic, socioeconomic, health status and pharmaceutical use variables, there remained some unexplained variation in adherence and persistence across HSDAs (Table 4.14). However, no region had consistently better or worse persistence and/or adherence than the reference HSDA of Vancouver.  87  Table 4.14: Summary of regional variation in multivariate analyses (after adjusting for individual-level factors), by outcome measure, drug class and combination  +  All 3 Classes  Any Class  Statins  Beta Blockers  ACE Inhibitors  Adherence (1-year PDC)  All 3 Classes  Any Class  Statins  Beta Blockers  –  ACE Inhibitors  Any Class  –  Adherence (PDC from start to first 90-day gap) All 3 Classes  Statins  East Kootenay  Beta Blockers  Health Services Delivery Area  ACE Inhibitors  Persistence  +  Kootenay Boundary  –  –  Okanagan  – –  Thompson Cariboo Fraser East  – –  –  Fraser North  –  Fraser South Richmond  –  – –  –  –  – – –  –  North Shore/Coast Garibaldi South Vancouver Island Central Vancouver Island  +  North Vancouver Island Northwest Northern Interior  +  – – – +  – –  – –  –  – – –  – – –  –  –  Northeast Reference HSDA is Vancouver; + indicates a positive effect and - indicates a negative effect (p < 0.05)  88  Chapter 5: Discussion  5.1  OVERALL PERSISTENCE AND ADHERENCE The persistence and adherence rates for ACE inhibitors, beta blockers and statins  following acute myocardial infarction in my study are generally high, and appear to be on par with rates reported in the literature. I found that over 85% of AMI patients filled a prescription for at least one of these drugs in the 30 days following their discharge from hospital between 2001 and 2005, with around 62% of patients using ACE inhibitors, 70% using beta blockers, 62% using statins, and nearly 40% using all three in combination. These rates are somewhat lower than those reported in Ontario for 2005, which were for seniors only, but nearly identical to rates in Saskatchewan for that same year [6, 25]. At 6 months post-AMI, the proportion of the population persistent with medication was 86% for statins, 83% for beta blockers and 79% for ACE inhibitors, values that are towards the top of the ranges identified in the literature (Figure 2.1). Persistence rates also decreased over time as expected. In this study, I found that patients persisted with statins significantly longer than with the other two classes, with a median time of almost 5 years between start and first 90-day gap, compared to a median of less than 4 years for ACE inhibitors and beta blockers. This trend has also appeared in studies by Simpson and Gislason, using similar definitions of persistence [38, 45]. Increased unadjusted persistence with statins is likely due to differences in the composition of the population of statin users: compared to the other classes, more statin users had private insurance and high incomes, used a combination of all 3 drug classes, and had fewer major ADGs. These factors were subsequently found in multivariate analysis to be associated with improved persistence. 89  Mean adherence, measured as the proportion of days covered (PDC) between first prescription and first 90-day gap, was very high for all classes, suggesting that while patients are persistent with therapy they experienced very few gaps in drug availability. Among the three drug classes, adherence was highest for users of ACE inhibitors, with 95% of them being classified as adherent (PDC ≥0.80) for the duration of their ACE inhibitor use, compared to under 93% for beta blockers and statins. Adherence in the first year of follow-up, 1-year PDC, was lower than adherence calculated with previous definition, as expected. The proportion of patients classified as adherent for the first year was 67% for ACE inhibitors, 68% for beta blockers, and 74% for statins. These findings are well within the range established in the literature [41, 47, 51].  5.1.1  Any class vs. all 3 classes Of the studies I reviewed in preparation for this thesis, only one attempted to calculate  adherence and persistence rates with any drug, regardless of class, or multiple drugs in combination [46]. Although it appears that drug combinations are not frequently evaluated, I chose to calculate persistence and adherence with any class and all three classes, to better understand the upper and lower limits of persistence and adherence rates. It is important to recognize that these drugs are most frequently prescribed in combination, and users of ACE inhibitors, beta blockers and statins are not independent. As Figure 4.2 indicates, only around 20% of all drug users were using a single class. By calculating persistence with any drug class, this study measured the highest proportion of the population with a nearly continuous (allowing up to a 90-day gap) exposure to medication. This was the least restrictive definition of drug use in the study, counting the availability in-hand of any cardiovascular drug from any of the three classes. Persistence with 90  any drug class was 93% at 6 months following AMI, and only decreased to 68% at 5 years, suggesting that of the AMI patients who use medication after their event, a majority of them continue to use some kind of cardiovascular medication for years after their initial event. Adherence among these patients was also very high, with almost 97% adherent (PDC ≥0.80) from first prescription to first gap over 90 days, and 87% adherent in the first year following discharge (1-year PDC ≥0.80). Users of all three drug classes faced the most restrictive definition of drug availability, which is to have all three classes in-hand to count towards adherence and persistence measures. By investigating adherence and persistence with this combination, it is possible to better understand if treatment guidelines for secondary prevention, which recommend all three drugs, are being followed over time. Around 40% of AMI patients started this combination of drugs, but many of those experienced prolonged gaps early in treatment, with only 65% of users persisting at 6 months, dropping quickly to 38% at 2 years and 21% at 5 years. These patients also appeared to experience more short gaps in therapy, with only 85% adherent (PDC ≥0.80) from first prescription to first gap over 90 days, and 45% adherent in the first year. Drug use is dynamic among these patients, with many patients appearing to use all 3 classes immediately following their AMI, but persisting with only one or two, suggesting that true adherence and persistence is somewhere between the limits described above. An important next step in understanding adherence and persistence with drug combinations is to investigate the clinical relevance of these rates, to better understand the true effectiveness of treatment in the population. Clinical trial data indicate that all three drug classes independently decrease the risk of death and secondary cardiac events [18]. If these findings also apply to the general population of AMI patients using any of the three drug classes with PDC ≥0.80, as studies of  91  statins and beta blockers have suggested, then a large proportion of BC patients are potentially benefiting from this treatment [51, 55]. As mentioned in my review of the literature, there appears to be no clinical trial data explicitly supporting the use of all three drugs in combination following AMI. Subgroup analyses in clinical trials for these drug classes suggest that they maintain their efficacy when used in combination, but the marginal benefits and risks of using all three of these classes together have not been measured [18]. Supposing that the benefits of these drug classes are cumulative when used in combination with PDC ≥0.80, then few BC patients are receiving the maximum potential benefit from these treatments. Understanding the difference in outcome between persisting with any or all of these drugs, especially with the lack of clinical trial evidence to support their use in combination, might inform future treatment guidelines and help to define optimal secondary prevention practices.  5.1.2  Measurement techniques A goal of my study was to compare methods to calculate persistence and adherence with  administrative data, because measurement techniques vary widely in the literature. I found that the results for persistence and adherence measured as 1-year PDC were quite similar, while adherence measured as PDC from start to discontinuation (first gap exceeding 90 days) was distinct. The proportion of the population persistent at 1 year was very close to the proportion of the population adherent, with PDC ≥0.80, for the first year. For example, among users of ACE inhibitors 69% were persistent at 1 year, and 67% were adherent over the course of the first year. Both measures also behaved nearly identically in multivariate analysis. The value of adherence, when measured as 1-year PDC, seems to be driven largely by extended gaps or discontinuation in therapy, giving results nearly identical to persistence (when expressed as the 92  proportion of patients who have not discontinued therapy) over the same period. Short gaps in therapy appear to have little influence on population values for 1-year PDC. Of the two measures, persistence allows for the accommodation of censored subjects, does not require a fixed period of observation, and is more easily interpreted, by focusing only on long gaps in drug exposure rather than any gap. In my opinion these features make it better suited than 1-year PDC for characterizing drug use over time. The first definition I chose to use for adherence, the PDC from a patient‟s first prescription to their first extended gap, is somewhat tautological, because I limited the length of gap patients could experience to less than 90 days. It was therefore unlikely that patients would have especially low values for PDC, because only short gaps were included. Similarly, by excluding patients who filled only one prescription I likely biased this value towards adherent patients, because patients who do not persist with therapy could also be more likely to be nonadherent, even when (according to an inventory model) they appear to have drug in-hand. My goal, however, was to identify and attempt to describe these short gaps in therapy, despite these limitations to the measure, and to distinguish them from long gaps and discontinuation with treatment. My results indicate that while patients are persistent with therapy they appear to experience very few gap days, relative to the days they have drug available. This suggests that patients who use medicines generally do so consistently over time; for example, almost 97% of users of any ACE inhibitor, beta blocker or statin, have drug available to them at least 8 days out of 10.  93  5.2  KEY FACTORS INFLUENCING PERSISTENCE AND ADHERENCE Many variables I investigated in multivariate analyses influenced adherence and  persistence as expected, based on findings from past studies. With the exception of the very elderly (aged 90-99 years), patients older than the reference category of 40-49 years, were significantly more likely to be persistent with medication. The effect was particularly strong and consistent among patients between 60 and 69 years. The very elderly were significantly less likely to be persistent with all three drug classes combined, and were no different than the reference group for most other models. These findings support the suggestion in section 2.4.1 that the effect of age on persistence and adherence follows an inverted U-shape [8, 50, 51]. This observed relationship likely arises due to fundamental differences between age groups, not adjusted for in this analysis. The very elderly would be expected to have lower functional status than younger patients, even after adjusting for health status and comorbidities, which is likely to negatively impact their adherence and persistence. By contrast, the youngest patients in the cohort are healthier, are likely to be part of the workforce and may have dependants at home, shaping their perceptions of their own health and the energy they devote to managing their disease. It appears that patients in the middle age categories, perhaps those who are conscious of their disease and functionally fit to manage it, are the most adherent and persistent. The socioeconomic status variables included in my analyses, private insurance, income and social assistance, also showed effects consistent with enabling factors in my conceptual framework (Figure 2.5) and the findings in the literature. Patients with private insurance and patients in the top income category (as compared to the bottom category) were significantly more likely to be persistent and adherent with medication, across analyses by class and drug combination. Similar findings had been previously reported in studies by Kramer and 94  Rasmussen [9, 47]. Receiving drug coverage through social assistance, a program under which beneficiaries, although poor, receive all prescribed medicines without deductibles or copayments, was also positively associated with adherence and persistence but only for users of all three classes in combination. Apart from age and the socioeconomic variables, which tend to have a clear and consistent effect in the literature, many factors I included in my study require further discussion to interpret their results. These include the differences in the effect of sex between drug classes, the effect of health status variables, other pharmaceutical use patterns, and regional variation.  5.2.1  Drug class and sex In multivariate models of both persistence and adherence, measured as 1-year PDC, the  effect of sex changed direction between drug classes. Among users of ACE inhibitors, women were 10% more likely to experience a gap in therapy exceeding 90 days, and were 16% less likely to be adherent, with 1-year PDC ≥0.80. However, the direction of this effect changed among users of beta blockers, with women being 15% less likely to experience a 90-day gap, and 33% more likely to be adherent. Sex did not influence persistence and adherence for statin users, or users of all three drugs in combination. In past studies of adherence and persistence with ACE inhibitors and beta blockers, the effect of sex is unclear. Wherever a statistically significant sex effect is reported in studies of ACE inhibitors, men are more likely to be adherent or persistent [44, 49]. For beta blockers on the other hand, the effect is mixed. A recent review of sex-specific issues in cardiovascular disease reported that differences exist in the presentation, management and outcomes of AMI and other cardiovascular events between men and women, but the reasons for these differences  95  remain unknown [69]. One conclusion was that the underrepresentation of women in clinical trials and the lack of good evidence on sex-specific treatment outcomes has hindered disease management. It may be the case that ACE inhibitors are not as well-tolerated in women as in men, leading to earlier discontinuation in treatment. A large meta-analysis of clinical trials of long-term ACE inhibitor use reported that there were no differences between women and men in the observed improvements in mortality and reinfarction rates [22]. However, the incidence of adverse events (hypotension and renal dysfunction) was not evaluated by sex. In order to interpret and further investigate sex-based differences in adherence and persistence – and ultimately outcomes – with ACE inhibitors, beta blockers and statins, it will be important to understand the potential sex-specific differences in drug safety, efficacy and tolerability.  5.2.2  Health status and related variables In my analysis, only the patients with the poorest health, with 5 or more major ADGs,  were significantly less likely to be adherent and persistent. The effect of worse health status may have been attenuated by the effect of using more drug classes following AMI, as the variables were somewhat correlated and are conceptually similar. The other variables included in the analysis that were potentially related to health status appeared to show the opposite effect: patients living in long-term care, with potentially worse health status, had improved adherence and persistence, while patients with a short hospitalization for initial AMI, who were potentially healthier, had worse adherence and persistence. These effects are unlikely to be related to health status alone, and are more likely due to interaction with the health care system and patient perception of health status. Patients in long-term residential care were more likely to be persistent with ACE inhibitors and beta blockers, and were also more likely to be adherent with those drug classes, 96  when using the definition of PDC from first prescription to first gap exceeding 90 days. Although these patients have worse health status, necessitating residential care, they have much more contact with caregivers and health services providers who can help to ensure that they use their medicines appropriately over time. Residents in long-term care also receive prescription drug benefits through PharmaCare Plan B, and do not face the same economic barriers to accessing medicines. There are many interrelated aspects to the variable for long-term care, and health status is likely a minor component; patients living in long-term residential care use pharmaceutical therapy under a very different set of circumstances than the average outpatient. Referring back to my conceptual framework, residence in long-term care appears to be an enabling factor for consistent drug use, as opposed to health status, which is a predisposing factor. Patients with a short index hospitalization, on the other hand, were more likely to experience a gap in therapy with all three drug classes. These patients may have had a less severe index AMI and better health status, but this may also mean that they perceive their cardiovascular disease differently than other patients. Previous studies have also shown that having a more severe index event (AMI vs. unstable angina, for example) or undergoing a revascularization procedure increase the likelihood of persistence with cardiovascular drugs [44, 49]. Patient perception is a recurring theme in the Andersen-Newman framework for health services use and other frameworks related to adherence: if patients do not believe their disease is severe, or do not perceive any benefits to treatment, they will be less likely to seek care and use treatments consistently [2, 59]. This is likely the effect being observed among patients with an index hospitalization under 3 days, rather than a true health status-related effect.  97  5.2.3  Pharmaceutical use patterns The variables most consistently associated with adherence and persistence in my study  were those related to use of other pharmaceuticals, but variables like these have not been wellexplored in the literature. The use of the different drug classes in this study appeared highly interrelated. The use of ACE inhibitors, beta blockers or statins in the 120 days prior to first AMI increased the likelihood of subsequent adherence and persistence with each of those classes respectively. The prior use of ACE inhibitors in particular also influenced beta blocker, statin and combination use in models of persistence and 1-year PDC. Using more prescription drugs after AMI was also found to increase adherence and persistence. The use of all three drugs in combination following AMI significantly increased the likelihood of adhering to and persisting with each of the drug classes individually. Looking at general pharmaceutical use, patients using at least four different drug classes, for any indication, in the 120 days following discharge for AMI were significantly more likely than patients using only 1-3 classes to be persistent and adherent with any drug class or combination. Beta blocker use prior to AMI was shown to be positively associated with beta blocker persistence in two of the studies I reviewed, supporting the findings of this study [51, 53]. There may be many reasons why past drug use appears to predict future drug use. In the AndersenNewman framework, for example, an individual‟s propensity to use health services, as evidenced by their past use, is a key predisposing factor predicting their future health services use [59]. In the case of cardiovascular drugs, patients who used them consistently before their AMI may continue to do so because it is already a habitual behaviour. They may also have an increased awareness of their disease and a positive perception of treatment, factors known to be positively associated with adherence.  98  Other drug effects, such as number of classes used and number of prescribers, need to investigated further to more fully understand their influence on adherence and persistence. Past studies have reported that patients who use more than 1 or more than 3 drugs are more likely to be persistent with ACE inhibitors, beta blockers or statins, but none have investigated higher categories [8, 49, 50]. There may be a point where a patient‟s use of more drugs no longer improves persistence and adherence, as regimen complexity or adverse events arising from polypharmacy become an issue. For example, in some adherence and persistence models, the use of ≥10 classes had no effect, while the use of 4 to 9 classes did. The studies by Perreault et al that found a positive association between persistence and use of ≥3 drug classes, also found that the more daily doses (of any drug) a patient had, the less likely they were to persist, suggesting there might be a tradeoff between the two factors [8, 50]. Studies investigating the additional use of specific drug classes or of known contraindicated drugs might also help to elucidate the effect of increased drug use on adherence and persistence. Finally, the effect of the number of unique prescribers, for any drug in the 120 days after discharge, was not consistent across models. I had considered it an enabling variable in my conceptual framework, with patients who had a regular prescriber being more likely to be adherent and persistent, but this was not the case in my analysis. Patients with 2 or 3 prescribers had better persistence and adherence with statins or any class, and patients with 4 or more prescribers had worse adherence and persistence with all three drugs in combination. The vast majority of drug users in the cohort (around 85%) had two or more unique prescribers in the 120-days after AMI, likely due to the fact that their first prescriptions would be from their attending physician upon discharge, and subsequent prescriptions would be from their regular doctor. Therefore, having more than one prescriber may simply be an indication that a patient filled more than one prescription. Changing the reference category to 1 or 2 prescribers, instead 99  of only 1 prescriber, to account for both discharge prescriptions and prescriptions from a family doctor might have improved the quality of this variable. Among users of all three drugs, having more than four prescribers was negatively associated with adherence and persistence. This may suggest that in order to consistently use multiple classes of drugs over time, some degree of prescriber continuity is important. It would be valuable in future studies to better quantify prescriber continuity using pharmacy claims data, accommodating both number of unique prescribers and number of physician visits or prescriptions, to understand its relationship with adherence and persistence.  5.2.4  Regional variation The results of my analysis suggest that some regional variation exists in adherence and  persistence, although no HSDA had consistent effects across all classes or combinations. In general, adherence and persistence rates were lowest in the three Island HSDAs and Fraser East. Adherence and persistence rates were also lower in Kootenay Boundary, especially for statins, and higher in East Kootenay for ACE inhibitors. My choice of Vancouver as the reference HSDA likely influenced these results; null findings could be due to both similarities between HSDAs, for example Richmond and Fraser South are both urban areas similar to Vancouver, or to wide confidence intervals arising from small samples sizes, as in the Northwest and Northeast. The next step for future studies of regional variation in medication adherence and persistence would be to incorporate regional-level variables, including characteristics of the external environment or health system, as shown in my conceptual framework, to determine what factors in each region might be influencing drug use patterns. Studies by Perreault et al in Quebec, for example, incorporated a simple variable for rural vs. urban residence and found that 100  rural drug users had significantly higher persistence with statins and antihypertensives [8, 50]. Among rural patients, one study found that those who reported that they were less satisfied with the care they received and who faced difficulties traveling to their appointments were significantly less likely to fill their prescriptions [70]. There are many variables at play within regions, such as access to and availability of health care providers, and understanding their impact on long-term drug use would be a valuable next step. Using a smaller level of analysis would also be ideal, because HSDAs are quite coarse geographic divisions, with significant heterogeneity within regions.  5.3  STRENGTHS AND LIMITATIONS This study provides insight into the state of ACE inhibitor, beta blocker and statin drug  use following AMI in BC, using a rich, population-wide administrative dataset. By using pharmaceutical claims data, which included some prescriber and payer information, linked to hospital records, and MSP registration and billing records, I was able to construct a conceptually-driven model of adherence and persistence with a wide range of analysis variables. Few studies have been able to incorporate individual-level demographic, socioeconomic, health status, and drug use variables into analyses of adherence and persistence. This study also has the advantage of a long observation period, from the start of 2001 to the end of 2006, and by using multiple definitions of adherence and persistence I was able to describe different, but related, aspects of drug use over time. The limitations faced by this study are much the same as the limitations faced by many other studies of adherence and persistence using administrative data. The inventory model I constructed to determine daily drug availability is susceptible to errors in the data (despite  101  excluding the outlying values for calculated daily dose) and assumes that patients use their medicines as prescribed, with one days‟ supply per day. Another limitation of my study is that I chose not to investigate factors or events associated with gaps in therapy, which means this study cannot distinguish between appropriate and potentially inappropriate gaps in therapy. Patients who experience adverse events from the drugs used, or patients who are hospitalized for unrelated events, may discontinue use of these drugs on the advice of their physicians. These patients may appear to not adhere to therapy, but ideally they would not be classified as non-adherent because gaps in their therapy are entirely appropriate. In a similar vein, this study cannot distinguish between intentional and unintentional gaps in therapy. As I alluded to in my discussion of health status and pharmaceutical use variables, further information is needed to address more qualitative factors related to patient perceptions and attitudes. Consistent use of medication over time is a complex patient behaviour, and in order to more fully understand the reasons behind a patient‟s adherence and persistence with medication, information is needed that cannot be derived solely from administrative data. All of the conceptual frameworks I reviewed focus heavily on patient beliefs, which shape patients‟ attitudes towards illness and medical care. A study of patient decision-making and treatment adherence hypothesized that intentional and unintentional are separate entities, influenced by different factors [71]. Unintentional non-adherence, for example missing doses or experiencing a lag between prescription fills, may be influenced by demographic variables, health status or not having a regular source of care. Intentional non-adherence on the other hand, is closely related to a patient‟s perception of their illness, and the balance of pros and cons they associate with treatment [71]. It is important to recognize these limitations to my study, and to the rates of adherence and persistence I report. However, this study still contributes important information to the 102  understanding of drug use, by presenting a high-level overview of adherence and persistence rates. While it is important in future studies to delve deeper into reasons for discontinuing treatment, understanding the scale and scope of the issue is an important step which can only be achieved through large studies such as this one, making use of comprehensive administrative datasets.  5.4  CONCLUSIONS My study found that ACE inhibitors, beta blockers, statins and their combination, are  frequently used immediately following AMI, and generally used consistently thereafter. Most patients use as least one drug for several years after AMI, but the use of all three classes in combination, as recommended by treatment guidelines, diminishes quickly over time. Understanding the clinical relevance of these persistence and adherence rates, especially for drug combinations, is an important next step in understanding patient outcomes and determining optimal treatment levels. Older patients, with the exception of the very elderly, and patients with higher socioeconomic status are more likely to be persistent and adherent, as has been repeatedly shown in the literature. Very poor health status negatively influences adherence and persistence, while use of more drug classes, both before and after AMI, positively influences adherence and persistence. In my conceptual model, age, health status and other pharmaceutical use were classified as predisposing factors, influencing patients‟ propensity to be adherent and persistent with medication. Socioeconomic status variables were classified as enabling factors, facilitating patients‟ drug use over time.  103  This study investigated some factors, such as residence in long-term care, short hospitalization and prescriber continuity, which may capture important aspects of adherence and persistence that could not described with administrative data. These factors may be related to patient perception of illness and perceived benefits of treatment, which are predisposing factors for drug use, and interaction with the health care system, an enabling factor. These areas would benefit from further investigation to better understand patient decision-making processes and behaviours. After adjusting for individual-level demographic, socioeconomic, health status and drug use variables, there remained some unexplained variation in adherence and persistence rates across HSDAs. These differences may be accounted for by incorporating regional-level variables into future analysis.  104  References 1.  Morgan, S., et al., The Canadian Rx Atlas, 2nd Edition. 2008, Centre for Health Services and Policy Research: Vancouver (BC).  2.  Osterberg, L. and T. Blaschke, Adherence to medication. N Engl J Med, 2005. 353(5): p. 487-97.  3.  Cramer, J.A., et al., Medication compliance and persistence: terminology and definitions. Value Health, 2008. 11(1): p. 44-7.  4.  Andrade, S.E., et al., Methods for evaluation of medication adherence and persistence using automated databases. Pharmacoepidemiology and Drug Safety, 2006. 15(8): p. 565-574.  5.  Antman, E., et al., ACC/AHA guidelines for the management of patients with ST-elevation myocardial infarction; A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (Committee to Revise the 1999 Guidelines for the Management of patients with acute myocardial infarction). J Am Coll Cardiol, 2004. 44(3): p. E1-E211.  6.  Austin, P.C., et al., Use of evidence-based therapies after discharge among elderly patients with acute myocardial infarction. Canadian Medical Association Journal, 2008. 179(9): p. 895-900.  7.  Akincigil, A., et al., Long-term adherence to evidence based secondary prevention therapies after acute myocardial infarction. JOURNAL OF GENERAL INTERNAL MEDICINE, 2008. 23(2): p. 115-121.  8.  Perreault, S., et al., Persistence and determinants of statin therapy among middle-aged patients for primary and secondary prevention. BRITISH JOURNAL OF CLINICAL PHARMACOLOGY, 2005. 59(5): p. 564-573.  9.  Rasmussen, J.N., et al., Use of statins and beta-blockers after acute myocardial infarction according to income and education. Journal of Epidemiology and Community Health, 2007. 61(12): p. 1091-1097.  10.  Morgan, S., et al., The British Columbia Rx Atlas. 2005, Centre for Health Services and Policy Research: Vancouver (BC).  11.  Manuel, D.G., et al., Burden of cardiovascular disease in Canada. Can J Cardiol, 2003. 19(9): p. 997-1004.  12.  Yusuf, S., et al., Global burden of cardiovascular diseases - Part I: General considerations, the epidemiologic transition, risk factors, and impact of urbanization. Circulation, 2001. 104(22): p. 2746-2753.  13.  Chow, C.M., et al., Regional variation in self-reported heart disease prevalence in Canada. Can J Cardiol, 2005. 21(14): p. 1265-71. 105  14.  Tanuseputro, P., et al., Risk factors for cardiovascular disease in Canada. Can J Cardiol, 2003. 19(11): p. 1249-59.  15.  Tu, J.V., et al., Outcomes of acute myocardial infarction in Canada. CANADIAN JOURNAL OF CARDIOLOGY, 2003. 19(8): p. 893-901.  16.  Heidenreich, P.A. and M. McClellan, Trends in treatment and outcomes for acute myocardial infarction: 1975-1995. AMERICAN JOURNAL OF MEDICINE, 2001. 110(3): p. 165174.  17.  Jackevicius, C.A., et al., Acute treatment of myocardial infarction in Canada 1999-2002. CANADIAN JOURNAL OF CARDIOLOGY, 2005. 21(2): p. 145-152.  18.  van der Elst, M.E., et al., Drug Therapy for Prevention of Recurrent Myocardial Infarction. Ann Pharmacother, 2003. 37(10): p. 1465-1477.  19.  Freemantle, N., et al., beta Blockade after myocardial infarction: systematic review and meta regression analysis. British Medical Journal, 1999. 318(7200): p. 1730-1737.  20.  Pedersen, T.R., et al., Randomized Trial of Cholesterol-Lowering in 4444 Patients with CoronaryHeart-Disease - the Scandinavian Simvastatin Survival Study (4s). Lancet, 1994. 344(8934): p. 1383-1389.  21.  Collins, R., et al., MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin in 20536 high-risk individuals: a randomised placebo-controlled trial. Lancet, 2002. 360(9326): p. 722.  22.  Flather, M.D., et al., Long-term ACE-inhibitor therapy in patients with heart failure or leftventricular dysfunction: a systematic overview of data from individual patients. Lancet, 2000. 355(9215): p. 1575-1581.  23.  van der Elst, M.E., et al., Effect of drug combinations on admission for recurrent myocardial infarction. HEART, 2007. 93(10): p. 1226-1230.  24.  Austin, P.C., et al., Factors associated with the use of evidence-based therapies after discharge among elderly patients with myocardial infarction. Canadian Medical Association Journal, 2008. 179(9): p. 901-908.  25.  Verrall, T., N. Hudema, and R. Gezer, Chapter 2. Chronic Diseases: Post-Heart Attack (Acute Myocardial Infarction) Care, in Quality Insight, T. Verrall, et al., Editors. 2008, Health Quality Council of Saskatchewan: Saskatoon, SK.  26.  Sabaté, E., Adherence to Long-Term Therapies: Evidence for Action. 2003, Geneva: World Health Organization.  27.  Hess, L.M., et al., Measurement of adherence in pharmacy administrative databases: a proposal for standard definitions and preferred measures. Ann Pharmacother, 2006. 40(7-8): p. 1280-88.  106  28.  Steiner, J.F. and A.V. Prochazka, The assessment of refill compliance using pharmacy records: Methods, validity, and applications. Journal of Clinical Epidemiology, 1997. 50(1): p. 105-116.  29.  Steiner, J.F. and M.A. Earnest, Lingua medica - The language of medication-taking. Annals of Internal Medicine, 2000. 132(11): p. 926-930.  30.  Caetano, P.A., J.M. Lam, and S.G. Morgan, Toward a standard definition and measurement of persistence with drug therapy: Examples from research on statin and antihypertensive utilization. Clin Ther, 2006. 28(9): p. 1411-24; discussion 1410.  31.  Hawkshead, J. and M.A. Krousel-Wood, Techniques for Measuring Medication Adherence in Hypertensive Patients in Outpatient Settings: Advantages and Limitations. Disease Management & Health Outcomes, 2007. 15: p. 109-118.  32.  Tamblyn, R., et al., The use of prescription claims databases in pharmacoepidemiological research: the accuracy and comprehensiveness of the prescription claims database in Quebec. J Clin Epidemiol, 1995. 48(8): p. 999-1009.  33.  Miller, E., B. Blatman, and T.R. Einarson, A survey of population-based drug databases in Canada. CMAJ, 1996. 154(12): p. 1855-64.  34.  Anderson, G.M., et al., Trends and determinants of prescription drug expenditures in the elderly: data from the British Columbia Pharmacare Program. Inquiry, 1993. 30(2): p. 199-207.  35.  Steiner, J.F., et al., A general method of compliance assessment using centralized pharmacy records. Description and validation. Med Care, 1988. 26(8): p. 814-23.  36.  Choo, P.W., et al., Validation of patient reports, automated pharmacy records, and pill counts with electronic monitoring of adherence to antihypertensive therapy. Med Care, 1999. 37(9): p. 846-57.  37.  Grymonpre, R., et al., Validity of a prescription claims database to estimate medication adherence in older persons. Med Care, 2006. 44(5): p. 471-7.  38.  Simpson, E., et al., Drug prescriptions after acute myocardial infarction: Dosage, compliance, and persistence. AMERICAN HEART JOURNAL, 2003. 145(3): p. 438-444.  39.  Hudson, M., et al., Comparison of measures of medication persistency using a prescription drug database. American Heart Journal, 2007. 153(1): p. 59-65.  40.  Cramer, J.A., et al., The significance of compliance and persistence in the treatment of diabetes, hypertension and dyslipidaemia: a review. Int J Clin Pract, 2008. 62(1): p. 76-87.  41.  Blackburn, D.F., et al., Adherence to statins, beta-blockers and angiotensin-converting enzyme inhibitors following a first cardiovascular event: A retrospective cohort study. Canadian Journal of Cardiology, 2005. 21(6): p. 485-488.  42.  Brookhart, M.A., et al., Physician follow-up and provider continuity are associated with long-term medication adherence: a study of the dynamics of statin use. Arch Intern Med, 2007. 167(8): p. 847-52. 107  43.  Brookhart, M.A., et al., Adherence to lipid-lowering therapy and the use of preventive health services: An investigation of the healthy user effect. AMERICAN JOURNAL OF EPIDEMIOLOGY, 2007. 166(3): p. 348-354.  44.  Eagle, K.A., et al., Adherence to evidence-based therapies after discharge for acute coronary syndromes: An ongoing prospective, observational study. AMERICAN JOURNAL OF MEDICINE, 2004. 117(2): p. 73-81.  45.  Gislason, G.H., et al., Long-term compliance with beta-blockers, angiotensin-converting enzyme inhibitors, and statins after acute myocardial infarction. European Heart Journal, 2006. 27(10): p. 1153-1158.  46.  Ho, P.M., et al., Impact of medication therapy discontinuation on mortality after myocardial infarction. Arch Intern Med, 2006. 166(17): p. 1842-7.  47.  Kramer, J.M., et al., National evaluation of adherence to beta-blocker therapy for 1 year after acute myocardial infarction in patients with commercial health insurance. AMERICAN HEART JOURNAL, 2006. 152(3): p. 454-+.  48.  Lachaine, J., et al., Persistence and adherence to cholesterol lowering agents: evidence from Regie de l'Assurance Maladie du Quebec data. Am Heart J, 2006. 152(1): p. 164-9.  49.  Newby, L.K., et al., Long-term adherence to evidence-based secondary prevention therapies in coronary artery disease. Circulation, 2006. 113(2): p. 203-212.  50.  Perreault, S., et al., Persistence with treatment in newly treated middle-aged patients with essential hypertension. ANNALS OF PHARMACOTHERAPY, 2005. 39(9): p. 1401-1408.  51.  Rasmussen, J.N., A. Chong, and D.A. Alter, Relationship between adherence to evidence-based pharmacotherapy and long-term mortality after acute myocardial infarction. Jama-Journal of the American Medical Association, 2007. 297(2): p. 177-186.  52.  Wei, L., et al., Adherence to statin treatment and readmission of patients after myocardial infarction: a six year follow up study. HEART, 2002. 88(3): p. 229-233.  53.  Wei, L., et al., Use and adherence to beta-blockers for secondary prevention of myocardial infarction: who is not getting the treatment? PHARMACOEPIDEMIOLOGY AND DRUG SAFETY, 2004. 13(11): p. 761-766.  54.  Bouchard, M.H., et al., Impact of adherence to statins on coronary artery disease in primary prevention. Br J Clin Pharmacol, 2007. 63(6): p. 698-708.  55.  Blackburn, D.F., et al., Cardiovascular morbidity associated with nonadherence to statin therapy. Pharmacotherapy, 2005. 25(8): p. 1035-43.  56.  Dormuth, C.R., et al., Statin Adherence and Risk of Accidents A Cautionary Tale. Circulation, 2009. 119(15): p. 2051-2057.  108  57.  Phillips, K.A., et al., Understanding the context of healthcare utilization: assessing environmental and provider-related variables in the behavioral model of utilization. Health Serv Res, 1998. 33(3 Pt 1): p. 571-96.  58.  Andersen, R. and J.F. Newman, Societal and individual determinants of medical care utilization in the United States. Milbank Mem Fund Q Health Soc, 1973. 51(1): p. 95-124.  59.  Andersen, R.M., Revisiting the behavioral model and access to medical care: does it matter? J Health Soc Behav, 1995. 36(1): p. 1-10.  60.  Gordis, L., Chapter 9: Cohort Studies, in Epidemiology, 3rd ed. 2004, Elsevier Saunders: Philadelphia. p. 149-158.  61.  Tu, J.V., et al., Methods Appendix - Chapter 5: Acute myocardial infarction outcomes in Ontario, in Cardiovascular health and services in Ontario: an ICES atlas, C.D. Naylor and P.M. Slaughter, Editors. 1999, Institute for Clinical Evaluative Sciences: Toronto. p. 25-30.  62.  Jackevicius, C.A., P. Li, and J.V. Tu, Prevalence, predictors, and outcomes of primary nonadherence after acute myocardial infarction. Circulation, 2008. 117(8): p. 1028-36.  63.  Law, M.R., et al., A longitudinal study of medication nonadherence and hospitalization risk in schizophrenia. JOURNAL OF CLINICAL PSYCHIATRY, 2008. 69(1): p. 47-53.  64.  PharmaCare Homepage. 2009 [cited 2009 June 16]; Available from: http://www.health.gov.bc.ca/pharmacare/#.  65.  Hanley, G.E., et al., Distributional consequences of the transition from age-based to income-based prescription drug coverage in British Columbia, Canada. Health Economics, 2008. 17(12): p. 1379-1392.  66.  BC Stats: Mapping and Geography: Health Maps. 2008 [cited 2009 Aug. 11]; Available from: http://www.bcstats.gov.bc.ca/DATA/POP/maps/hsda.asp.  67.  Weiner, J.P., et al., Development and application of a population-oriented measure of ambulatory care case-mix. Medical Care, 1991. 29(5): p. 452-472.  68.  About the ATC/DDD System. 2008 [cited 2009 June 16]; Available from: http://www.whocc.no/atcddd/.  69.  Pilote, L., et al., A comprehensive view of sex-specific issues related to cardiovascular disease. Canadian Medical Association Journal, 2007. 176(6): p. S1-S44.  70.  Wroth, T.H. and D.E. Pathman, Primary medication adherence in a rural population: the role of the patient-physician relationship and satisfaction with care. J Am Board Fam Med, 2006. 19(5): p. 478-86.  71.  Wroe, A.L., Intentional and unintentional nonadherence: a study of decision making. J Behav Med, 2002. 25(4): p. 355-72.  109  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0070908/manifest

Comment

Related Items