Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Effectiveness of inhaled corticosteroids in preventing morbidity and mortality in individuals with chronic… Goring, Sarah 2008

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

Item Metadata

Download

Media
24-ubc_2008_fall_goring_sarah.pdf [ 1.19MB ]
Metadata
JSON: 24-1.0066519.json
JSON-LD: 24-1.0066519-ld.json
RDF/XML (Pretty): 24-1.0066519-rdf.xml
RDF/JSON: 24-1.0066519-rdf.json
Turtle: 24-1.0066519-turtle.txt
N-Triples: 24-1.0066519-rdf-ntriples.txt
Original Record: 24-1.0066519-source.json
Full Text
24-1.0066519-fulltext.txt
Citation
24-1.0066519.ris

Full Text

EFFECTIVENESS OF INHALED CORTICOSTEROIDS IN PREVENTING MORBIDITY AND MORTALITY IN INDIVIDUALS WITH CHRONIC OBSTRUCTIVE PULMONARY DISEASE AND THE IMPACT OF COEXISTING ASTHMA  by Sarah Goring  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) August 2008  © Sarah Goring, 2008  Abstract Background: Chronic obstructive pulmonary disease (COPD) is a devastating illness that affects 4.3% of the population of British Columbia over the age of 45 years. Asthma is known to coexist in 10-20% of individuals with obstructive lung disease, and adds to the substantial burden of illness posed by COPD alone. Inhaled corticosteroids (ICS) are currently recommended for the management of COPD among individuals with frequent exacerbations; however, the ability of inhaled corticosteroids to reduce death and hospitalizations among individuals with COPD is controversial. Less is known about the effectiveness of ICS among individuals who are afflicted with both COPD and asthma.  Methods: We used a retrospective cohort study design and administrative data to estimate the relative effectiveness of ICS in reducing hospitalizations or death among individuals with concomitant asthma and COPD, compared with individuals with COPD alone. We used an extended Cox model to estimate this association, with a time-varying measure of exposure to ICS.  Results: We did not find any association between ICS and hazard of death or hospitalization among individuals with COPD alone (HR = 0.99; 95% CI: 0.94 – 1.05), however the hazard was 18% lower (HR = 0.82; 95% CI: 0.69-0.99) among individuals with concomitant disease.  Conclusions: Individuals with combined COPD and asthma show significant benefit from the use of ICS and are more responsive to the effects of ICS than individuals with COPD alone.  ii  Table of Contents Abstract ......................................................................................................................................ii Table of Contents......................................................................................................................iii List of Tables.............................................................................................................................vi List of Figures ..........................................................................................................................vii List of Abbreviations ..............................................................................................................viii Acknowledgements ...................................................................................................................x 1  Introduction.........................................................................................................................1 1.1 1.2 1.3 1.4  2  OVERVIEW......................................................................................................................................1 OVERARCHING OBJECTIVES .............................................................................................................1 RESPONSIBILITIES...........................................................................................................................2 STRUCTURE OF THESIS ...................................................................................................................2  Background on COPD ........................................................................................................4 2.1 2.2  DEFINITION OF COPD.....................................................................................................................4 NATURAL HISTORY OF UNTREATED COPD.......................................................................................5 2.2.1 Lung Function...................................................................................................................5 2.2.2 Exacerbations...................................................................................................................5 2.3 PULMONARY AND SYSTEMIC INFLAMMATION IN COPD.......................................................................6 2.4 COPD AND ASTHMA .......................................................................................................................7 2.4.1 Co-existing COPD and Asthma........................................................................................9 2.5 EPIDEMIOLOGY OF COPD .............................................................................................................10 2.5.1 Incidence and Prevalence ..............................................................................................10 2.5.2 Challenges in estimating true prevalence and incidence of COPD ...............................10 2.5.3 Risk factors.....................................................................................................................12 2.6 BURDEN OF ILLNESS .....................................................................................................................12 2.6.1 Individual burden of illness .............................................................................................12 2.6.2 Societal burden of illness ...............................................................................................13 2.6.3 Economic Burden of illness ............................................................................................13 2.6.4 COPD Hospitalizations ...................................................................................................14 2.7 DISEASE MANAGEMENT ................................................................................................................14 2.7.1 Stable COPD ..................................................................................................................14 2.7.2 Acute Exacerbations.......................................................................................................15 2.8 INHALED CORTICOSTEROIDS .........................................................................................................16 2.8.1 Mechanism of action.......................................................................................................16 2.8.2 Historical and current recommended use of ICS ...........................................................17 2.8.3 Trends in ICS Utilization.................................................................................................18 2.8.4 ICS Efficacy in COPD.....................................................................................................18 2.8.5 ICS Effectiveness in COPD............................................................................................20 2.8.6 Hetereogeneity of response to ICS ................................................................................23 2.9 SUMMARY ....................................................................................................................................25 2.10 FIGURES AND TABLES ...................................................................................................................26  3  Pharmacoepidemiological Context.................................................................................34 3.1  METHODOLOGIC CHALLENGES IN PHARMACOEPIDEMIOLOGICAL STUDIES..........................................34 3.1.1 Selection Bias.................................................................................................................34 3.1.2 Information Bias..............................................................................................................35 3.1.3 Confounding ...................................................................................................................39 3.1.4 Effect modification ..........................................................................................................41 3.1.5 Summary of methodological challenges.........................................................................41  iii  3.2  3.3  PHARMACOEPIDEMIOLOGICAL MODELING APPROACHES...................................................................41 3.2.1 Multiple regression models.............................................................................................42 3.2.2 Extended Cox models with time-dependent covariates .................................................42 3.2.3 Propensity scores ...........................................................................................................44 3.2.4 Instrumental Variables....................................................................................................45 3.2.5 Marginal Structural Models.............................................................................................47 3.2.6 Competing risks framework............................................................................................48 3.2.7 Summary of pharmacoepidemiological modeling approaches ......................................48 FIGURES AND TABLES ...................................................................................................................50  4  Objectives .........................................................................................................................52  5  Methods .............................................................................................................................53 5.1 5.2 5.3 5.4  5.5  5.6  6  Results...............................................................................................................................67 6.1 6.2 6.3 6.4 6.5  6.6  7  OVERVIEW....................................................................................................................................53 DATA QUALITY ..............................................................................................................................53 5.2.1 BC Linked Health Database ...........................................................................................53 ETHICS .........................................................................................................................................54 DATA PREPARATION .....................................................................................................................54 5.4.1 Data verification and transformation...............................................................................54 5.4.2 Cohort Definition.............................................................................................................56 5.4.3 Exposure ........................................................................................................................56 5.4.4 Asthma ...........................................................................................................................58 5.4.5 Comorbidities..................................................................................................................58 5.4.6 Outcomes .......................................................................................................................59 5.4.7 Censoring .......................................................................................................................59 DATA ANALYSIS ............................................................................................................................60 5.5.1 Descriptive statistics .......................................................................................................60 5.5.2 Characterization of ICS use ...........................................................................................60 5.5.3 Predictors of ICS use......................................................................................................60 5.5.4 Association between ICS and time to re-hospitalization or death..................................61 5.5.5 Sensitivity Analyses........................................................................................................63 FIGURES AND TABLES ...................................................................................................................65 DESCRIPTIVE STATISTICS ..............................................................................................................67 CHARACTERIZATION OF ICS USE ...................................................................................................68 PREDICTORS OF ICS USE ..............................................................................................................69 ASSOCIATION BETWEEN ICS AND TIME TO RE-HOSPITALIZATION OR DEATH ......................................69 SENSITIVITY ANALYSES .................................................................................................................71 6.5.1 Cohort definition .............................................................................................................71 6.5.2 Exposure ........................................................................................................................72 6.5.3 Outcomes .......................................................................................................................73 FIGURES AND TABLES ...................................................................................................................74  Discussion ........................................................................................................................85 7.1 7.2 7.3  7.4 7.5 7.6  CONTEXTUALIZATION OF RESULTS .................................................................................................85 CHOICE OF MODELING APPROACH ..................................................................................................86 INTERNAL VALIDITY .......................................................................................................................88 7.3.1 Selection Biases .............................................................................................................88 7.3.2 Information Biases..........................................................................................................90 7.3.3 Confounding ...................................................................................................................95 EXTERNAL VALIDITY ......................................................................................................................97 FUTURE WORK ..............................................................................................................................98 CONCLUDING STATEMENTS ...........................................................................................................99  References .............................................................................................................................100 Appendix 1: Additional Tables .............................................................................................114 iv  Appendix 2: Ethics approval.................................................................................................130  v  List of Tables Table 2.1 COPD disease severity based on CTS classification by symptoms, disability and impairment of lung function (adapted from 2007 CTS guidelines).11 .............................30 Table 2.2 Differences between asthma and COPD (adapted from several sources11,33,37). ......31 Table 2.3 International Classification of Disease (ICD) codes corresponding to COPD............32 Table 2.4 Summary of effect estimates from observational studies of ICS in COPD, according to presence of immortal time bias, reported in peer-reviewed literature. .......................33 Table 3.1 Variables that potentially confound the association between exposure to inhaled corticosteroids and risk of death or hospitalization (see Figure 3.1) ..............................51 Table 5.1 Example of counting process format of data..............................................................66 Table 6.1 Characteristics of individuals discharged alive after a first hospitalization for COPD in BC between 1996 and 2001...........................................................................................76 Table 6.2 Characteristics of drug dispensation to individuals discharged alive after a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001.................77 Table 6.3 Life table describing time to inhaled corticosteroid (ICS) initiation prior to and following discharge hospitalization for COPD in British Columbia (BC) between 1996 and 2001. .......................................................................................................................78 Table 6.4 Compliance with prescribed ICS dose (using cumulative multiple availability measure)128 during follow-up from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001. .......................................................................................79 Table 6.5 Odds ratios and 95% confidence intervals (CIs) of multiple logistic regression model of ICS initiation within ninety days of an index hospitalization for COPD in British Columbia (BC) between 1996 and 2001. .......................................................................80 Table 6.6 Frequency of outcomes and censored observations during follow-up from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001.................81 Table 6.7 Rates and crude rate ratios (RRs) of death or hospitalization for exposure to inhaled corticosteroids (ICS) among individuals discharged alive from a first hospitalization in British Columbia (BC) between 1996 and 2001. ............................................................82 Table 6.8 Adjusted estimates of ICS effectiveness among individuals discharged alive from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001, using an extended Cox model with time-varying exposure. CI: Confidence Interval...............83 Table 6.9 Sensitivity analyses of adjusteda ICS effect estimates among individuals discharged alive from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001, using an extended Cox model with time-varying exposure..................................84 vi  List of Figures Figure 2.1: Representation of decline in lung function over time associated with smoking, adapted from Fletcher and Peto.14 .................................................................................26 Figure 2.2: Theoretical representation of the decline in lung function over time. The upper line represents exacerbation-free decline and the lower line is punctuated by exacerbations and incomplete recovery. Modified from Jones79 ..........................................................27 Figure 2.3: Stepwise approach to therapy for COPD (copied from 2007 CTS guidelines with permission).11 .................................................................................................................28 Figure 2.4 Recommended pharmacotherapy for stable COPD according to disease severity and exacerbation frequency (copied from 2007 CTS guidelines with permission)11 ......29 Figure 3.1 Directed acyclic graph of variables that potentially confound or modify the relationship between inhaled corticosteroids and death or hospitalization (see Table 3.1). ................................................................................................................................50 Figure 5.1 Diagram describing study design including cohort entry, diagnostic classification, exposure classification, and identification of outcomes based on prior hospitalizations and health service utilization recorded in the medical service plan (MSP) database, hospitalization discharge records and drug dispensations.............................................65 Figure 6.1 Flowchart of cohort inclusion among individuals discharged from a first hospitalization for COPD in BC between 1996 and 2001. .............................................74 Figure 6.2 Estimated (smoothed) hazard function for death or hospitalization following discharge from first hospitalization for COPD in BC between 1996 and 2001...............75  vii  List of Abbreviations ACG®  Adjusted Clinical Groups  ATS  American Thoracic Society  BC  British Columbia  BC GPAC  British Columbia Guidelines and Protocols Advisory Committee  BCLHD  British Columbia Linked Health Database  CMA  Continuous multiple availability  COPD  Chronic Obstructive Pulmonary Disease  CTS  Canadian Thoracic Society  DAG  Directed Acyclic Graph  DNA  Deoxyribonucleic Acid  FEV1  Forced Expiratory Volume in one second  FVC  Forced Vital Capacity  GOLD  Global Initiative for Chronic Obstructive Lung Disease  HDAC2  Histone deacetylase 2  HRQL  Health related quality of life  ICD-9  International Classification of Diseases – version 9  ICD-10  International Classification of Diseases – version 10  ICS  Inhaled Corticosteroids  IQR  Interquartile range  ISOLDE  Inhaled Steroids in Obstructive Lung Disease in Europe  IV  Instrumental variables  LABA  Long acting beta agonist  MSM  Marginal structural model  MSP  Medical Services Plan  RCT  Randomized Controlled Trial viii  SD  Standard Deviation  TORCH  TOwards a Revolution in COPD Health  TRISTAN  Trial of Inhaled Steroids And long-acting β2 agonists  ix  Acknowledgements I would like to acknowledge several people who assisted in the completion of this thesis. Adrian Levy, my supervisor, helped greatly in this study by providing support, direction, and assistance and more importantly, he shared epidemiological frameworks and strategies that will be invaluable to my future research endeavours. I would like to thank Dr. Bruce Carleton and Anne Smith for their help in allowing me to gain access to the data used in this study and Dr. Robert Prosser and Rita Sobolev for their contributions to preparing the data for analysis. My supervisory committee, made up of Drs. Bruce Carleton, Robert Levy and Andrew Briggs, provided valuable feedback in many aspects of the study. My partner and close friends have been extremely patient and supportive of me over the past few years, and my family has been supportive of me my whole life, for which I am extremely grateful.  x  1 Introduction 1.1  Overview Chronic obstructive pulmonary disease (COPD) is a devastating illness characterized by  progressive airway obstruction. It affects 4.3 percent of British Columbia (BC) residents over the age of 45 years,1 and is projected to be the third leading cause of death by the year 2020.2 The disease has no cure and places an enormous burden on the health care system due to a lifetime of treatment required to alleviate individuals’ symptoms. COPD and asthma are known to coexist in 10-20% of individuals with obstructive lung disease.3 Although the two diseases share similar clinical characteristics, their combined presence is associated with a much higher burden of disease than the combined burden of both individually.4 Inhaled corticosteroids (ICS) have been demonstrably effective in reducing morbidity and mortality in persons affected by asthma;5 however, there is no consensus as to the effectiveness of the drug in reducing these outcomes among individuals with COPD.6,7 Despite the additional burden posed by individuals with concomitant COPD and asthma, little is known about the effectiveness of ICS in reducing morbidity or mortality in this population with combined disease.  1.2  Overarching objectives The thesis describes much of the background research that was undertaken in order to  design and conduct the study of ICS effectiveness among individuals afflicted with COPD, with and without coexisting asthma. The broad objectives of the thesis were: to review the literature regarding the disease area of COPD and coexisting asthma; to review evidence of ICS treatment effectiveness; to identify sources of bias that could affect observational studies of ICS effectiveness; and to identify analytical techniques to reduce these sources of bias. Ultimately, the goal of the thesis was to use this information to design and conduct a study that compares 1  estimates of ICS effectiveness among COPD-afflicted individuals with and without coexisting asthma, and to interpret the findings of the study in the context of what is known.  1.3  Responsibilities The candidate was responsible for conceiving the specific study question, designing the  methodology study described in this thesis, coding and analyzing all data, and for the authorship of the entire document. The candidate and members of the thesis committee jointly conceived of conducting a study in the area of COPD and identified appropriate sources of data to inform the study. Prescription claims, physician billing and hospitalization data were made available through an existing dataset provided by the BC Linked Health Database to Dr. Carleton who also provided some study variables partially coded for the purpose of a different study by Dr. Robert Prosser. Members of the thesis committee provided methodological and statistical expertise in the conduct of the research and the interpretation of the results (Drs. Adrian Levy and Andrew Briggs) and clinical expertise and guidance and the interpretation of the results (Drs. Robert Levy and Bruce Carleton). Students in the Department of Healthcare and Epidemiology and the Thesis Screening Panel provided feedback and guidance on some aspects of the study design. For literary convention, and to reflect the inputs of all involved in the study, the first person plural is used throughout the thesis, although it is recognized that the candidate is responsible for the contents of the thesis.  1.4  Structure of thesis The remainder of the thesis will be structured as follows. Chapter two introduces the  disease area of COPD – its current definition, epidemiology, burden of illness, management, and the role of ICS in disease management. We also present what is known about coexisting COPD and asthma. In chapter three, we discuss major methodological challenges in pharmacoepidemiological studies and analytic approaches to minimizing the impact of these 2  challenges. The specific objectives of the study are stated in chapter four. Chapter five contains a detailed description of the methods applied in the study with the results of the study described in chapter six. Chapter seven is the discussion section of the thesis, in which the results of the study are interpreted in the context of what is currently known, and what is still not well understood. Additionally, we critically evaluate the internal and external validity of the study and discuss the extent to which the study results represent the true effectiveness of ICS in a population of individuals with coexisting asthma and COPD, compared with individuals with COPD alone. We conclude the thesis with potential directions for future research in the field.  3  2 Background on COPD 2.1  Definition of COPD COPD includes three pathologic subtypes: chronic bronchitis, characterized by excessive  mucous secretion,8 chronic bronchiolitis, characterized by inflammation of the peripheral airways,9 and emphysema, characterized by the destruction of the lung alveoli.10 These pathological subtypes are difficult to diagnose, and most patients have all three pathologic conditions in varying degrees.9 For this reason, the formal definition of COPD, according to the 2007 CTS guidelines is defined functionally: “COPD is a respiratory disorder largely caused by smoking, which is characterized by progressive, partially reversible airway obstruction, systemic manifestations, and increasing frequency and severity of exacerbations.”11 Physicians in BC are currently referred to the 2007 CTS guidelines11 as a guide for diagnosing and treating COPD, with additional recommendations published by the BC Guidelines and Protocols Advisory Committee (BC GPAC).12 Both guidelines emphasize the need for diagnosing COPD at an early stage, using targeted spirometric testing in high-risk patients. The BC GPAC recommends using the criteria of 1) FEV11 less than 80% of the predicted value (based on age, height, sex and race) and 2) FEV1/FVC2 <0.7 postbronchodilator, indicative of a positive diagnosis of COPD.12 Additional disease staging guidelines incorporating clinical symptoms are shown in Table 2.1.  1  Forced Expiratory Volume in 1 second: The volume of air that can be forced out in one second after taking a deep  breath 2  Forced Vital Capacity: The total volume of air that can be forced out after taking a deep breath  4  2.2  Natural History of untreated COPD  2.2.1  Lung Function  The hallmark of COPD is the progressive decline in lung function over time. Studies by Burrows and Earle,13 and Fletcher and Peto14 were among the first to characterize the temporal decline in lung function among individuals with COPD. Their studies showed that the rate of lung function decline changed over time, with more rapid decline associated with more severe disease. Furthermore, they found that smoking cessation retarded the rate of decline such that, eventually, it slowed to the age-related rate of decline observed among non-smokers without COPD (Figure 2.1). The Lung Health Study, which enrolled individuals at high risk for developing COPD, validated this finding in an experimental setting, finding a difference in the rate of lung function decline between two arms of its study: lung function declined an average of 62.3mL/year among patients receiving usual care, compared with an average decline of 47.0mL/year among patients who received a smoking intervention along with usual care.15 Untreated, the rate of lung function decline in COPD-afflicted patients ranges from 47mL/year to 69mL/year16-19 (values observed in the placebo arms of clinical trials of treatment for COPD, measured using FEV1), with larger rates of decline occurring among patients of greater disease severity. 2.2.2  Exacerbations  The symptoms of stable COPD are punctuated by periods of intense symptom severity, called exacerbations. Acute exacerbations are characterized by sustained (48 hours or more) worsening of shortness of breath and coughing leading to an increase in the use of maintenance medication and/or supplementation with additional medications.20 They are the most frequent cause of medical visits, hospital visits and death among patients with COPD,20 and the most common cause is a viral or bacterial infection.12 An increased rate of decline in lung function is associated with frequent exacerbations history of respiratory tract infections.21,22 Recovery from an exacerbation may take weeks or 5  months22,23 and a significant proportion of individuals may never show complete resolution of lung function or symptoms.24 This incomplete recovery may partially explain the accelerated rate of decline in lung function among individuals experiencing frequent exacerbations. The proposed mechanism is illustrated in Figure 2.2. Individuals with severe COPD tend to have more exacerbations than those with moderate or mild COPD,25 implying that the frequency of exacerbations increases over time. Although this increase in frequency has yet to be confirmed in longitudinal studies,25 it has been demonstrated that the severity of exacerbation (characterized by increase in symptoms, recovery time and likelihood of hospitalization) increases over time.25  2.3  Pulmonary and systemic inflammation in COPD Airway inflammation is a major component of COPD that causes dyspnea, wheezing, and  hyperinflation of the lungs. The inflammation evident in COPD arises from an abnormal inflammatory response in the central and peripheral airways and the lung parenchyma.9 The prolonged and chronic nature of the inflammation causes the bronchial tissue to become scarred and unable to fully expand, and produces symptoms such as bronchoconstriction, hypersecretion of mucus, and pain.9 The principal cells involved in the inflammatory process of COPD are T lymphocytes, mononuclear cells, neutrophils, macrophages, and scanty eosinophils.9,26 The specific roles of these cells in the inflammatory process, and their effect on lung function is described elsewhere.27,28 Of importance to the current study is that the cell types evident in the lungs of individuals with COPD differ from those that exist in normal inflammatory response, as well as those presenting in other respiratory diseases. The inflammatory process affecting individuals with COPD is not limited to that occurring in the airways. The CTS and Global Initiative for Chronic Obstructive Lung Disease (GOLD) definitions of COPD describe “systemic manifestations”20 and “extrapulmonary effects”8 associated with COPD. Elevated levels of systemic inflammatory markers – C-reactive protein, 6  fibrinogen, leucocytes, tumor necrosis factor-α, and interleukins 6 and 8 – have been found in patients with COPD,29 supporting the notion that individuals with COPD experience systemic effects of the disease. This is further corroborated by the observation that COPD-afflicted individuals often have multiple comorbid medical conditions, including skeletal muscle abnormalities, hypertension, diabetes, coronary artery disease, heart failure, pulmonary infections, cancer, autoimmune disorders, weight loss and nutritional abnormalities and pulmonary vascular disease.30,31 Although this association between systemic inflammation and COPD has been established, it remains to be determined whether the systemic effects that coexist with COPD are caused in part by the abnormal inflammatory processes associated with COPD or whether they arise independently due to the same insult that caused the pulmonary inflammation.30,31  2.4  COPD and asthma Asthma is another respiratory disease that is characterized by restricted airflow. According  to the Canadian guidelines on adult asthma, “Asthma is characterized by paroxysmal or persistent symptoms such as dyspnea, chest tightness, wheezing, sputum production and cough, associated with variable airflow limitation and airway hyper-responsiveness to endogenous or exogenous stimuli. Inflammation and its resultant effects on airway structure are considered to be the main mechanisms leading to the development and maintenance of ashtma”5 The “Dutch Hypothesis,” suggested in 1961 by a group of researchers in the Netherlands, proposes that the two diseases share genetic origins and should be considered as a single disease.32 Although some facets of the argument are sound,33 the weight of evidence lines up against the hypothesis.34 In contemporary clinical practice, the diseases are considered distinct and are defined and managed differently. Although features of the two diseases appear similar, we will highlight some of the principal differences. Functionally, COPD and asthma can be differentiated from one another based on spirometric tests of lung function. The test is administered twice: once before and once after administering a bronchodilator. Compared with individuals who have COPD, those with asthma 7  typically have a greater improvement in FEV1 from the baseline 15 minutes after use of an inhaled short-acting β2-agonist.12 According to the 2007 CTS guidelines, an improvement in FEV1 of more than 0.4L can be suggestive of asthma.11 An oral corticosteroid may also be administered to differentiate between the two diseases:12 symptoms tend to improve with this treatment in individuals with asthma whereas those with COPD may not improve. Neither test is diagnostically definitive because individuals with COPD can have a partially reversible component to their disease, and individuals with asthma can have a partially irreversible component.5,11 Furthermore, the improvement in FEV1 after the use of a bronchodilator is highly variable within individuals35 – in a study of individuals with moderate to severe COPD, approximately 50% changed responder status (defined as an improvement in FEV1 of more than 0.2L) when tested repeatedly over time.35 Therefore, even when using spirometry testing to guide diagnosis, the distinction between asthma and COPD in cases with partial airflow reversibility can be difficult. A major difference between asthma and COPD lies in their inflammatory profiles. As described in section 2.3, the cell types present in the inflammatory process of COPD differ from those in asthma. The inflammatory cells in asthma are primarily CD4+ cells and eosinophils, whereas those in COPD are mainly CD8+ cells and neutrophils. The relevance of this lies in the differential response of these cells to pharmacological treatments36 such as ICS, which has been shown to be more effective in reducing eosinophilic inflammation than neutrophilic inflammation (see section 2.8).36 The different inflammatory profiles in COPD and asthma result in different lung pathologies that manifest into different symptom presentation.36 In asthma, airway smooth muscle contracts as a result of multiple bronchoconstrictor mediators being released from inflammatory cells. This causes episodic wheezing, breathlessness, and coughing.37 In COPD, the airflow limitation arises from structural changes of small airways and closure of small airways as a  8  result of disrupted alveolar attachments, resulting in air trapping and dyspnea. This gives rise to the more constant feeling of breathlessness that is characteristic of COPD.37 A summary of the key differences between COPD and asthma is presented in Table 2.2. 2.4.1  Co-existing COPD and Asthma  Some individuals are described by clinicians as having “COPD-like” asthma and others as having “asthma-like” COPD. However, the two diseases can coexist.8,11 Between 10% and 20% of people with obstructive lung disease have are believed to have both asthma and COPD,3 although given the difficulties in diagnostic definitions, this value is not well defined. A common scenario resulting in the coexistence of both diseases is when asthmatic individuals develop COPD later in life, typically after many years of smoking. Some evidence exists suggesting that asthma is a risk factor for COPD, with higher rates of COPD occurring in asthmatics who smoke than in non-asthmatic smokers.38 The coexistence of asthma and COPD creates a challenge in disease management because of differences in guidelines for the two respiratory diseases.5,11 There is a paucity of information regarding the effectiveness of treatment among individuals with concomitant asthma and COPD, as this sub-population is typically excluded from randomized controlled trials (RCT) in both disease areas. One RCT was recently conducted exclusively in patients with coexisting disease, measuring the efficacy of an anticholinergic in improving lung function. The authors reported a beneficial effect of tiotropium on lung function and symptomatic relief compared with placebo.39 With respect to effectiveness of therapies in reducing outcomes such as death or hospitalizations in this population, more research is needed. Little else is known about co-existing COPD and asthma. The remainder of the introduction focuses on what is known about COPD because the study objectives focus on this population – specifically the comparative effectiveness of ICS across subgroups of this disease.  9  2.5  Epidemiology of COPD  2.5.1  Incidence and Prevalence  In BC, COPD is reported to affect 4.3% of the population aged over 45 years.1 In 2003/2004 there were 8,000 new cases of COPD in BC (0.5% of the BC population over the age of 45 years).1 Worldwide, reported prevalence and incidence estimates vary substantially. Some reasons for this variability will be discussed in section 2.5.2. 2.5.2  Challenges in estimating true prevalence and incidence of COPD  Choosing a method to identify individuals afflicted with COPD that is both sensitive and specific poses a considerable challenge.40 The estimates reported in section 2.5.1 were generated based on the criteria of two physician visits or one hospitalization per year, as recorded in administrative records. Administrative databases contain numeric codes that uniquely identify diseases (International Classification of Diseases – ICD). The study investigators used the ICD codes corresponding to “chronic bronchitis”, “emphysema”, and “chronic airway obstruction, not elsewhere classified” to identify cases (Table 2.3). A fourth ICD-9 code – 490: “bronchitis, not specified as acute or chronic” includes both COPD patients (chronic bronchitis) and non-COPD patients (acute bronchitis); however, it is impossible to determine the patient diagnosis from this code alone. In an analysis of the RAMQ database in Quebec, this code represented 63% of the four ICD codes potentially representing COPD.41 The use of this ambiguous code greatly reduces the specificity of the diagnoses of COPD in administrative databases. Inclusion of this diagnostic code inflates the estimate of COPD incidence and prevalence by including cases of acute bronchitis; however exclusion of the code may provide an underestimate. Additional discussion regarding the validity of recording data into administrative databases is discussed further in section 3.1.2.3. The use of health services utilization to determine incidence and prevalence – including through analysis of administrative databases and surveys of patient-reported diagnosis – may underestimate the true values. Because airflow restriction in COPD occurs gradually over time, 10  rather than with sudden onset, individuals may progressively restrict their activity levels than to seek medical attention, particularly when the disease is in a mild state. Analysis of the Third National Health and Nutrition Examination Survey indicated that 63.3% of subjects with documented low lung function had no prior or current reported diagnosis of any obstructive lung disease,42 suggesting that COPD is a highly under-diagnosed disease, and that records of physician visits underestimate disease prevalence. Conversely, spirometric screening is thought to yield a high number of false positive results, and produces an overestimate of prevalence.43 The burden of obstructive lung disease initiative (BOLD) is presently undertaking a study to estimate the prevalence of COPD in countries around the world in a standardized fashion, incorporating spirometry as well as clinical symptoms and exposure history.44 Another major challenge in comparing prevalence and incidence of COPD across jurisdications and over time lies in the definition of the disease. Historically, there was no standard definition of COPD – the major guidelines published different definitions in different jurisdictions. Applying the spirometry guidelines of the ATS and the ERS on a study population in Northern Italy resulted in airways obstruction prevalence of 57% using the ATS definitions compared with 12% using the ERS guidelines.45 Until recently, there was no international consensus on the definition of COPD. In 1997, the Global Initiative for Chronic Obstructive Lung Disease (GOLD) was formed and comprised an international panel of experts to raise awareness of COPD and to improve prevention and treatment. GOLD developed a consensus definition, severity grading system and disease management guidelines. Several national societies have adopted the GOLD disease definition;46 the CTS has recently done so.11 This common definition will facilitate global comparisons of COPD prevalence, but until it has been adopted internationally, inter-country comparisons will remain unreliable.47 For estimating disease incidence, disease registries are a superior resource because incident cases can be captured at diagnosis; however, such registries do not exist for COPD48 and few estimates of COPD incidence exist.49,50  11  2.5.3  Risk factors  There are many component causes51 that may contribute to the development of COPD. Smoking is the most commonly cited attributable cause, but is neither necessary nor sufficient for developing the disease: 10 to 20% of individuals diagnosed with COPD have never been smokers,52 and 80 to 90% of smokers never develop COPD.53 Exposure to noxious gases and air pollution has been linked to the development of COPD, as well as occupational exposures to grain, isocyanates, cadmium, coal, other mineral dust, and welding fumes.53 There is evidence that a genetic component contributes to COPD susceptibility,48,53 with the α-1-antitrypsin gene commonly cited.8,20,53,54 Several regions of the genome have been identified that contain genes associated with increased risk of COPD,8 but results of these genetic association studies have been inconsistent, and no genes other than α-1-antitrypsin have been definitively identified as influencing the development of COPD.8 Women are thought to be more susceptible to COPD than men: upon diagnosis and in ageand severity-matched studies, females tend to have smoked fewer pack-years than their male counterparts.55-60 Associations between nutrition (low levels of vitamins C and E), socioeconomic status, and lower respiratory infections during childhood have also been linked with development of COPD later in life.53  2.6  Burden of illness  2.6.1  Individual burden of illness  The health-related quality of life (HRQL) of COPD-afflicted individuals is severely compromised, even in the early stages of disease.61 As the disease progresses, the health status of individuals deteriorates.62 Individuals develop increasingly severe feelings of breathlessness and are frequently unable to participate in daily activities or attend work; as a result, they often feel isolated and depressed.49 In the Confronting COPD study, one quarter of  12  the subjects described themselves as having to “stop for breath every few minutes when walking even on level ground” or were “too breathless to leave the house”.63 The intensified symptoms experienced during acute exacerbations affect the HRQL of individuals throughout the period of the exacerbation, and continue to reduce HRQL throughout recovery and between exacerbations,64 due to incomplete recovery and anxiety surrounding future exacerbations.49 2.6.2  Societal burden of illness  The morbidity and mortality associated with COPD has been steadily increasing over recent decades.2,8,40,48,65 Projections of the global burden of illness suggest that by 2020, COPD will become the third leading cause of death worldwide, and the fifth leading cause of morbidity.2 In Canada, the number of deaths from COPD increased by 95% between 1980 and 1995.65 In 1998, it accounted for 4% of all deaths in Canada, the 4th highest among men and 7th highest among women.66 The cause-specific mortality rate for COPD is likely underestimated.49,50 Statistics Canada records only the underlying cause of death in the National Mortality Database,65 omitting deaths where COPD was an intermediate or contributing cause. Inaccuracies in the diagnosis, in chart completion, or chart transcription would also contribute to underestimates, as well omissions due to un-diagnosed cases of COPD.42 2.6.3  Economic Burden of illness  Although COPD poses a considerable societal burden, there is little information regarding the total economic burden of disease in Canada.67 Estimates of direct annual costs from other countries have ranged from USD$200 to $10,800 (year unknown) per patient per year depending on country, disease severity, and study methodology.48 The Confronting COPD study, based on a telephone interview of 401 self-reported COPD patients, used micro-costing to estimate an annual direct medical cost of CDN$1998 (2003) per patient.67 A 1998 report by Health Canada estimated CDN$467 million as the total cost of hospital care and  13  pharmacotherapy for treating persons with COPD.68 This figure excluded the cost of physician visits and of ancillary services. The economic burden of COPD is estimated to be 50% greater than the burden of asthma.4 Among individuals with concomitant COPD and asthma, the direct medical costs are estimated to be greater than that of either disease alone (5.25 times the burden of asthma, and 3.5 times the burden of COPD), as well as greater than the additive combination of the individual costs.4 2.6.4  COPD Hospitalizations  COPD-related hospitalizations are the largest contributor to the direct medical costs of the disease.69 A study of the American National Medical Expenditure Survey estimated the cost of inpatient hospitalization and emergency department care in the United States to account for 73% of total COPD-related expenditures.70 The Confronting COPD study in Canada attributed 57% of the mean annual direct medical cost per patient to inpatient hospitalizations and emergency room visits.67 Individuals with COPD are typically admitted to hospital between one and four times per year, depending on severity, with between 46% and 86% of patients making at least two visits in a year.71 In BC, 9% of patients discharged from a hospital following an acute exacerbation are re-admitted within two weeks of discharge.1  2.7  Disease Management  2.7.1  Stable COPD  COPD cannot be cured. The goals of disease management focus on slowing the progression of disease, alleviating the symptoms, preventing exacerbations, improving exercise tolerance, HRQL, and survival.11 Smoking cessation is the only intervention that is able to slow this accelerated rate of decline.15 Other treatments – pharmacologic therapy, pulmonary rehabilitation, oxygen therapy, and surgical interventions – are used as disease severity increases, illustrated in Figure 2.3.  14  Pharmacologic treatment – bronchodilators and ICS – alleviate symptoms, improve HRQL and have been shown to reduce the occurrence of exacerbations72, however have not been shown to modify the rate of decline in lung function. Bronchodilators act on the airway smooth muscle, widening the airways, improving airflow, and reducing dynamic hyperinflation at rest and during exercise. The action of ICS will be discussed in greater detail in section 2.8. Non-pharmacologic treatment is also important in COPD management. Supplemental oxygen therapy is provided for individuals who have low levels of blood oxygen at rest. It has been shown to improve exercise tolerance and HRQL, and is the only therapy that has been shown to improve survival among COPD patients in an experimental setting.72 Pulmonary rehabilitation – which includes exercise training, nutritional counseling, and education – has been shown to have beneficial effects on exercise capacity, HRQL, duration and frequency of hospitalizations, and feelings of anxiety and depression.8 Surgical interventions have yielded positive results in some COPD patients, however, these interventions are expensive, show unattractive cost-effectiveness ratios in most subgroups,73 and have a high risk of postoperative complications, and are not recommended for the routine management of COPD.8 2.7.2  Acute Exacerbations  Management of exacerbations of COPD depends on their cause and severity. Most exacerbations are caused by viral or bacterial infections;12 however, for approximately one-third of severe exacerbations, the cause is unknown.8 For less severe exacerbations that can be managed at home, recommended treatment involves bronchodilator therapy, a possible course of antibiotics if necessary, and oral corticosteroids if the exacerbation is not resolved within hours.8 For more severe exacerbations that must be managed in hospital, standard treatment may involve controlled oxygen therapy, intravenous methylxanthines, intravenous glucocorticosteroids, and/or noninvasive mechanical ventilation.8  15  2.8  Inhaled Corticosteroids ICS are currently recommended in combination with long acting beta agonists for  individuals with moderate to severe COPD.11 Among individuals with infrequent exacerbations, ICS/LABA is only recommended for individuals who do not respond to bronchodilators alone, whereas this combination is recommended for all individuals who experience frequent exacerbations.11 Glucocortioids (e.g. cortisol, cortisone) are types of corticosteroids naturally produced by the adrenal glands that regulate the metabolism of carbohydrates and proteins and provide anti-inflammatory effects.74 Synthetic glucocorticoids are prescribed to counteract the body’s inflammatory process; however, their effects also include suppression of the immune response, altered metabolism, reduced serum calcium levels, and regulation of mood and behaviour.75 Therefore, prolonged exposure is avoided. When administered topically, the systemic effects of the glucocorticoids are substantially reduced, improving the safety and side-effect profile of the drugs, and allowing for continued use.74 Inhaled corticosteroids are topically administered glucocorticoids used for treatment of respiratory disease. They can be inhaled using a dry powder inhaler, pressurized metered-dose inhaler, or nebules.76 2.8.1  Mechanism of action  Glucocorticoids bind with steroid receptors in the cell cytoplasm and then enter the nucleus and bind to deoxyribonucleic acid (DNA). A cascade of events leads to a reversal of the inflammatory response.75 One mechanism of ICS action of particular relevance is through histone acetylation.77 COPD patients have been observed to have impaired function of histone deacetylase 2 (HDAC2) due to cigarette smoking and oxidative stress.78 The impaired function is thought to reduce the responsiveness to the beneficial effects of ICS,78 which provides some rationale behind why ICS has been shown to provide less symptom relief in individuals with COPD than in asthmatics. 16  ICS is associated with significant reductions in airway inflammation, allowing for improved air exchange.79 Beneficial effects of ICS on lung function have been seen within one week of treatment initiation.80,81 Evidence of long term benefits of ICS on lung function decline, death, symptom relief, and exacerbation frequency, will be reviewed in detail in sections 2.8.4 and 2.8.5. 2.8.2  Historical and current recommended use of ICS  Historically, there is little consensus on the role of ICS in the management of COPD. Only recently is consensus emerging.6,7 The 1992 CTS guidelines state that that “[a]lthough there is clear evidence of airway inflammation in COPD it is less clear that COPD patients, in general, benefit from steroids either taken systemically or inhaled…. If the patient is deemed steroid responsive … one could also consider the use of inhaled steroids. However, data on the response to inhaled steroids in COPD are conflicting82,83 and more research is required.”84 Between 1992 and 2003, no updated guidelines were provided by the CTS regarding use of ICS in management of COPD. The 1995 updates of the American Thoracic Society (ATS) and European Respiratory Society (ERS) noted that the beneficial effects of ICS on lung function and symptoms had not been established, and therefore until recently, no consensus existed as to their effectiveness.85,86 Following the publication of several large-scale clinical trials measuring the efficacy of ICS in COPD, the CTS recommendations released in 2003 suggested high-dose ICS for moderateto-severe patients and patients with high frequencies of exacerbation (e.g. 3 acute exacerbations in 3 years),20 and recommended that ICS be initiated according to the step-wise approach (Figure 2.3). Informed by additional studies of the effectiveness of ICS as well as combined ICS with long-acting β-agonists (ICS/LABA) combination therapy, the CTS published updated guidelines in 2007, recommending ICS/LABA use for individuals with moderate to severe COPD, and depending on frequency of exacerbations (Figure 2.4).11  17  2.8.3  Trends in ICS Utilization  Despite a lack of guidance from the governing bodies in BC, Canada, or internationally, the number of ICS dispensations for COPD increased through the 1990s. In England, the prescribed daily dose of ICS nearly doubled, from 69.9 million per quarter in 1992 to 124.7 million per quarter in 1998.87 Scottish data between 1999 and 2002 showed an increase in dispensations of 10%.88 This increase in dispensation can be partially attributed to the increasing prevalence of the disease; however, the proportion of COPD patients using ICS also increased during the 1990s. An analysis of baseline clinical trial data suggested that the proportion of COPD patients taking ICS in 1987 was 13.2% and rose to 41.4% in 1995.89 A population-based study found a similar increase in ICS dispensations among the elderly, with a reported increase of ICS prescription fills from 42.4% to 53.1% of elderly patients with COPD.20 2.8.4  ICS Efficacy in COPD  The goal of pharmacologic therapy in COPD management is to slow the rate of decline in lung function, ultimately reducing COPD-related mortality. Although no individual clinical trial has shown a reduction in mortality,16-19,26,80,81,90-94 a recent pooled analysis95 of 5085 subjects enrolled in seven randomized controlled trials16-19,80,81,92 (the Inhaled Steroid Effects Evaluation in COPD (ISEEC) study) found that participants randomized to ICS had a lower risk of mortality (HR 0.75; 95% Confidence Interval (CI): 0.57 to 0.99) compared with placebo. The ISEEC study investigators also analyzed the decline in FEV1 among these same seven RCTs96 and concluded that in the first six months of treatment, ICS use was associated with a significant increase in FEV1, compared with placebo. This was quantified as an average increase in postbronchodilator FEV1 of 42mL among men, and a 29mL increase among women. Although the finding at 6 months was statistically significant, the magnitude of the change in FEV1 was acknowledged to be clinically insignificant – 112mL has previously been identified as the minimum change in FEV1 that results in an appreciable change in perceived dyspnea.97  18  Furthermore, after six months, the investigators found no difference – statistical or clinically meaningful – in the rate of FEV1 decline between the ICS and placebo arms. Investigators have examined the effect of ICS on other outcomes and observed that subjects randomized to ICS (either alone, or in combination with long-acting β-agonists80,81,92,94) experience fewer exacerbations17,18,92, less severe exacerbations81,90 and improved HRQL17,92. A systematic review of nine randomized trials reported that the overall rate of exacerbations was 30% lower in subjects randomized to ICS than to placebo (RR=0.70; 95% CI: 0.58 to 0.84).98 The severity of recruited patients varied between studies, and there was evidence to suggest that the choice of study population affected the measured effect sizes. The Copenhagen City Heart Study investigators enrolled patients between 30 and 70 years of age with a baseline post-bronchodilator FEV1 in the ICS group of 86.2% predicted (SD: 20.6%).18 (Table 2.1 contains COPD staging definitions based on values of FEV1.) The EUROSCOP study investigators also recruited younger (30-65 years) patients with mild-to-moderate COPD (100% > FEV1 > 50%). The baseline post-broncholator FEV1 of the ICS group was 76.8% predicted (SD:12.4%).19 Neither of these studies was able to detect a treatment effect on respiratory symptoms or exacerbation frequency.20 The baseline post-bronchodilator FEV1 of the ICS group enrolled in the Lung Health Study was 68.5% predicted (SD: 12.8%). This study detected improvements in the ICS arm in terms of reduced number of outpatient visits for respiratory conditions and reduced symptoms. The largest effect sizes were observed in the studies which enrolled the most severe subjects. The ISOLDE study17 enrolled patients with a baseline FEV1 in the ICS group of 50.3% predicted and reported a 25% reduction in median exacerbation rate. Szafranski et al81, Calverley et al92, the TRISTAN study,80 and the TORCH study94 enrolled subjects with mean baseline FEV1 values that were all below 45% predicted.  19  In each of these studies, significant reductions in exacerbation rates were observed for ICS alone3 as well as ICS/LABA combination treatment. Improvements in health status have been shown to become progressively wider between ICS and placebo groups after study enrollment,17 suggesting that there was a cumulative benefit of ICS. This has been attributed to a reduction in exacerbations, leading to improved health over the long term.79 2.8.5  ICS Effectiveness in COPD  While randomized controlled trials are designed to measure drug efficacy – whether a medication has the ability to bring about the intended effect – observational studies measure a drug’s effectiveness – whether, in a ‘real world’ setting, a drug in fact achieves its desired effect.99 RCTs are generally viewed as the reference standard for the evaluation of drug efficacy100,101 due to the randomly allocation of individuals to treatment. This ensures that exposed and unexposed groups are mostly balanced in terms of known and unknown confounders, and that any residual differences are due to chance. Observational studies are prone to imbalances in the exposed and unexposed groups, particularly in studies of intended effects of drugs. This is because in the ‘real world’ setting, patients and physicians select treatment based on clinical need, therefore causing exposed and unexposed groups to differ in clinical symptoms.102 Despite these limitations, many observational studies of the intended effects of ICS among COPD-afflicted individuals have been conducted because they are able to measure exposures and outcomes in a large number of subjects over an extended period of time – studies that would be prohibitively expensive to conduct as RCTs. These studies applied strategies to reduce the amount of bias and confounding incurred from the non-randomized treatment  3  Exception: Szafranski et al did not find a significant reduction in exacerbation rates for ICS alone  20  allocation using large administrative databases to estimate the effectiveness of ICS in reducing death, hospitalizations, or both, among individuals with COPD. The earliest of these observational studies to emerge found a 24% reduction in repeat hospitalizations (95% CI: 20-29%), and a 29% reduction in risk of mortality (95% CI: 22-35%) among subjects over 65 years, following discharge from a first hospitalization for COPD.103 This apparent reduction in mortality among ICS users was supported by subsequent observational studies,104,105 suggesting that ICS use caused a significant reduction in mortality; however, it was later realized that these results may have been subject to “immortal time” bias (discussed in detail in section 3.1.2). In two separate studies,106,107 Suissa demonstrated the effect of “immortal time” bias on the effect estimate using an administrative database of Saskatchewan Health. He analysed the data in two ways: first using the methods of previous studies,103,104 and then using methods free of immortal time bias. The re-analysis produced substantial increases in the point estimates of ICS effectiveness, resulting in non-significant effect sizes.106,107 Numerous studies of ICS effectiveness have since been conducted, some free of immortal time bias,108-113 and others still susceptible to the problem.111 Outcomes in these observational studies were all global-clinical rather than physiologic or functional114 – hospitalizations, death, and combined hospitalization or death. Unfortunately, the effect sizes associated with ICS use in these studies remain incongruent, even among those free of immortal time bias. Table 2.4 summarizes the results of these observational studies. The studies with published acknowledgement of immortal time bias are presented separately from those without. Some results listed in the table refer to the same study, with different methods applied to produce the effect estimate. Using hospitalizations as an outcome, no observational studies free of immortal time bias demonstrated a reduction in risk.108-110 The nested case-control study of de Melo et al108 found an increase in risk of hospitalizations in all categorizations of exposure: any use in the year 21  prior to hospitalization, any use in the 15-60 days prior, and any use in the 61-365 days prior. The authors also investigated the effectiveness of ICS according to dose, and found counterintuitive (and likely confounded) results of higher risk among higher dose. Individuals taking ICS in the highest dose category were reported to have a risk of hospitalization that was 2.9 times that of unexposed individuals (95% CI: 1.5-5.9). Although it is possible that ICS is associated with increased risk of hospitalizations, the elevated risk of hospitalization among the exposed group is likely a result of confounding by severity (described in section 3.1), resulting in an artificially elevated estimate of risk in the exposed group. Among the studies using death as a primary endpoint, all studies except for that of Mapel et al112 similarly reported no benefit of ICS. Mapel et al112 used three study designs – Cox proportional hazards, propensity score-matched cohort, and a nested case-control study – to investigate ICS effectiveness. The authors used permutations of exposure and non-exposure to ICS and LABA to produce five different exposure categories (combined ICS/LABA in one inhaler was treated separately from ICS and LABA being used concurrently) during a baseline period of 1 year prior to cohort entry. All three study designs identified statistically significant or borderline significant reductions in the risk of death among any exposure category. Other studies employing similar techniques yielded null results. A main difference between the study conducted by Mapel et al and other studies was that Mapel et al used a time-fixed exposure to ICS, based on exposure at baseline, whereas other studies used time-dependent exposures. Mapel et al argued that the use of a time-fixed exposure misclassifies unexposed individuals as exposed, biasing the results upward, and producing an inflated measure of ICS effectiveness. However, due to the hierarchical nature of treatment and the authors’ definition of cohort entry, the study design excluded the person years of event-free exposure to SABA preceding the use of ICS among those classified as ICS users. This has been described in detail by Suissa115 as excluded immortal time (section 3.1.2), which leads to an overestimate of the event rate in the unexposed (because of the excluded event-free person years in the unexposed group) and 22  thus underestimates the rate ratio. The magnitude of the bias has not been estimated, and it is therefore unclear whether the study results analyzed free from immortal bias would still yield a beneficial effect of ICS. Among the studies with a combined endpoint of death and hospitalization, there was only one result (from a nested case-control study) that indicated a favourable effect of ICS.111 The effect measure was an odds ratio which was meant to estimate the relative risk of death or hospitalization. While the odds ratio approximates the risk ratio when the incidence of outcome is low, it tends to underestimate the risk ratio when outcomes are more frequent.116 The incidence of death or hospitalization during this time period was 53.5%107 – a value at which the odds ratio substantially underestimates the relative risk. However, the point estimate estimated by the authors was 0.71 (95% CI: 0.56-0.90), and even with the underestimate produced by the odds ratio, this still may have suggested a protective effect of ICS. Odds ratios were also used in studies using hospitalization or death alone as an end-point, but because the incidence of hospitalization alone and death alone was not as high as the combined end-point, the odds ratio did not underestimate the relative risk to the extent that it did when using the combined outcome. Although we noted the flaws in the studies that reported a protective effect, the studies detecting a null, or even harmful effect, were also subject to bias. These studies are all subject to some of the common challenges of pharmacoepidemiology that are discussed in greater detail in section 3.1. The most prominent challenge in these studies is controlling for the severity of disease. Uncontrolled studies can result in overestimates of event rates in the exposed, thus overestimating the rate ratio, and at times suggesting harmful effects of drugs. 2.8.6  Hetereogeneity of response to ICS  The pathological heterogeneity of COPD poses a significant challenge in assessing the effectiveness of ICS.117 Historically, COPD was defined as an umbrella term for three distinct disease pathologies: chronic bronchitis, chronic bronchiolitis and emphysema, each of which 23  has distinct pathologies and different inflammatory processes, suggesting that the response to treatment might differ between these groups. Coexisting asthma also contributes to the heterogeneity of the disease because of the additional differences in lung pathology and inflammation profile of asthma compared with COPD (section 2.4). The different systemic manifestations that may arise concurrently also affect the symptoms and outcomes associated with COPD, and may affect an individual’s response to therapy. Another source of heterogeneity within the population of COPD is the exposure to cigarette smoke. The pathology of COPD among current and former smokers differs from that among nonsmokers,118 and the rate of lung function decline is known to differ between current and former smokers.15,60 In order to minimize the heterogeneity of the population in studies of persons with COPD, RCTs generally have strict entry and exclusion criteria. Among RCTs estimating the efficacy of ICS among persons with COPD, frequent exclusion criteria include: individuals with a history of asthma,17,80,81,92-94,119 individuals with bronchodilator reversibility of more than 10-15% of baseline FEV1,17,19,26,80,94 individuals with conditions that could have lead to death within the timeframe of the study,80,81,92,94,119 and individuals with no history of smoking.16,17,19,26,90,91,80,81,92,94,119 A benefit of observational studies is that the sample size can be large, providing the ability to analyze effectiveness within subgroups, while maintaining a large enough sample size to detect differences (providing that internal validity is adequate). Published observational studies have typically included sensitivity analyses, investigating the effect of changing certain operationalized definitions, or looking at effectiveness with or without certain subgroups. Vollmer et al113 measured the effectiveness of ICS among subgroups according to smoking history and co-existing asthma. Although differences between these groups were not tested statistically, the point estimates of ICS effectiveness were generally lower among individuals with coexisting asthma than among individuals without, and the point estimates were also lower in never smokers than in current or former smokers. Fan et al110 also investigated the 24  difference in ICS effectiveness with and without individuals with prior history of asthma. The estimates were also not compared statistically, but the point estimates of ICS effectiveness in reducing mortality were more favourable when asthmatics were included in the cohort. It has been suggested that observational studies should entirely omit individuals with history of asthma because they are considered to be asthmatics misdiagnosed with COPD.117 Several studies have attempted to exclude asthma either using hospital and physician visit records to identify individuals with prior history of asthma, or using the assumption that asthma is rare in an older age-group120 and the majority of obstructive lung disease is due to COPD.103 If patients who are truly asthmatic and do not have COPD are included in a study cohort, evidence suggests that the effect size will be underestimated; however, given that there exists a subgroup of COPD patients with coexisting asthma, we argue that these individuals should be included and the effectiveness in this subgroup should be estimated separately.  2.9  Summary It is known that COPD is a heterogeneous disease with varying pathophysiological  characteristics, symptoms, and responsiveness to medication; however, little is known about the comparative effect sizes of ICS within well-defined sub-groups of the disease. Concomitant COPD and asthma is a subgroup of COPD that places a significantly higher burden on the health care system. We know that ICS is effective in asthmatics, but little is known about its effectiveness in individuals with COPD as well as asthma. In this study, we estimate the effectiveness of ICS among individuals with concomitant COPD and asthma, compared with COPD-afflicted individuals with no history of asthma.  25  2.10 Figures and Tables  Figure 2.1: Representation of decline in lung function over time associated with smoking, adapted from Fletcher and Peto.14  26  Figure 2.2: Theoretical representation of the decline in lung function over time. The upper line represents exacerbation-free decline and the lower line is punctuated by exacerbations and incomplete recovery. Modified from Jones79  27  AECOPD: acute exacerbation of COPD; MRC: Medical Research Council; PRN: as needed; Rx: treatment.  Figure 2.3: Stepwise approach to therapy for COPD (copied from 2007 CTS guidelines with permission).11  28  Figure 2.4 Recommended pharmacotherapy for stable COPD according to disease severity and exacerbation frequency (copied from 2007 CTS guidelines with permission)11  29  Table 2.1 COPD disease severity based on CTS classification by symptoms, disability and impairment of lung function (copied from 2007 CTS guidelines).11 Stage  Symptoms  Lung Function  Mild COPD  Shortness of breath from COPD when hurrying on the level or walking up a slight hill (MRC 2)  FEV1 ≥ 80%, FEV1/FVC< 0.7  Moderate COPD  Shortness of breath from COPD causing the patient to stop after walking approximately 100m (or after a few mintes) on the level (MRC 3 to 4)  Severe COPD  Shortness of breath from COPD resulting in the patient too breathless to leave the house, breathless when dressing or undressing (MRC 5), or the presence of chronic respiratory failure or clinical signs of right heart failure n/a  Very severe COPD  50% ≥ FEV1 < 80% predicted, FEV1/FVC< 0.7  30% ≥ FEV1 < 50% predicted, FEV1/FVC< 0.7  FEV1 < 30% predicted, FEV1/FVC< 0.7  Abbreviations: FEV1: forced expiratory volume in one second; FVC: forced vital capacity; MRC: Medical Research Council dyspnea scale  30  Table 2.2 Differences between asthma and COPD (adapted from several sources11,33,37). Asthma Inflammatory profile Predominant inflammatory cells Clinical Symptoms Frequency Age of onset Smoking history Sputum production Allergies Disease course  Physiology Spirometry Airflow obstruction  COPD  Mast cells Eosinophils CD4+ cells  Neutrophils CD8+ cells  Intermittent and variable Usually < 40 years Not casual Infrequent Often Stable (with exacerbations)  Persistent Usually > 40 years Usually > 10 pack years Often Infrequent Progressive worsening (with exacerbations)  Often normalizes Usually intermittent airflow obstruction but often has a less reversible obstruction Improvement in airway obstruction with bronchodilators and corticosteroids Airway remodeling (epithelial injury and fibrosis)  Never normalizes Progressive airflow obstruction Smaller bronchodilator and corticosteroid response Emphysema (lung destruction) frequent.  31  Table 2.3 International Classification of Disease (ICD) codes corresponding to COPD ICD-9 code  Description  491  Chronic bronchitis  ICD-10 code J41  Description  J42  Simple and mucopurulent chronic bronchitis Unspecified chronic bronchitis  492  Emphysema  J43  Emphysema  496  Chronic airway obstruction, not elsewhere classified  J44  Other chronic obstructive pulmonary disease  32  Table 2.4 Summary of effect estimates from observational studies of ICS in COPD, according to presence of immortal time bias, reported in peer-reviewed literature. End Point Death  Effect Estimate of ICS compared with no ICS (Point estimate, 95% CI) Immortal time bias Free of immortal time bias HR = 0.71, 0.65-0.78103 HRa = 1.10, 0.77-1.59110 HRb = 0.85, 0.64-1.13110 104 HR = 0.62, 0.45-0.85 HRc = 0.94, 0.81-1.09107 105 HR = 0.75, 0.68-0.82 HR = 0.76, 0.61-0.95112 c 107 HR = 0.66, 0.57-0.76 HR = 0.75, 0.58-0.97112 c 107 HR = 0.75, 0.62-0.90 OR = 0.66, 0.46-0.96112 HRd = 1.09, 0.65-1.83113 HRe = 0.76, 0.56-1.02113  Hospitalization  HR = 0.76, 0.71-0.80103  OR = 1.07, 0.91-1.27109 HRa = 1.05, 0.81-1.37110 HRb = 1.13, 0.94-1.36110 OR = 1.27, 1.08-1.48108  Death + Hospitalization  HRc = 0.69, 0.55-0.86106  HRc = 1.00, 0.79-1.26106  HR = 0.69, 0.52-0.93111  OR = 0.71, 0.56-0.90111  HRd = 1.09, 0.79-1.49113 HRe = 1.02, 0.83-1.25113 Abbreviations: HR = hazard ratio; OR = odds ratio; CI = confidence interval; ICS = inhaled corticosteroids a Low-dose ICS b High-dose ICS c A study demonstrating the impact of immortal time bias d Including only current smokers, excluding asthma e Including only ex-smokers, excluding asthma  33  3 Pharmacoepidemiological Context 3.1  Methodologic challenges in pharmacoepidemiological studies Due to the non-experimental nature of pharmacoepidemiological studies, several  challenges arise in measuring effect sizes with precision and without bias. There are three broad categories of bias that distort the true estimate of effect between exposure and outcome in epidemiological and pharmacoepidemiologic studies: selection bias, information bias, and confounding.121,122 3.1.1  Selection Bias  Selection bias exists in studies where there is a systematic difference in the characteristics of the exposed and unexposed samples, often due to the method used to identify study participants.99,121-123 If the likelihood of being included in the study sample is not equal across all factors in the target population, the resulting effect estimate may not accurately represent the true effect size. There are many types of selection bias. We will discuss one form that is prone to affecting pharmacoepidemiological studies conducted using administrative data, and briefly describe two types that are generally avoided when using administrative databases. Individuals who are censored during the follow up period can contribute to a biased results if exposed and unexposed subjects exit the study in a non-differential manner.124 In longitudinal studies, this is also called informative right censoring. In many of the observational studies of ICS effectiveness discussed in section 2.8.5, individuals were censored when they dis-enrolled from the drug plan used to collect the data. In BC, where health coverage is universal, this is unlikely to be related to either exposure or outcome because dis-enrollment would likely be due to a move outside of the province for a reason unrelated to exposure or outcome, and thus would not impose a bias on the results. The administrative drug database in BC, however, does not record drugs dispensed in hospital. Therefore, drug exposure among individuals admitted to long term care facilities cannot be determined, and individuals should be 34  censored upon admission. Because these individuals are typically sicker than non-hospitalized individuals, this does impose a bias. If the main study outcome is time to hospitalization, censorship due to death, may also impose bias. In studies of COPD patients, risk of death is positively associated with the risk of hospitalization, and represents a worse outcome. If the treatment of interest is effective in reducing mortality, the relative risk of hospitalization may appear larger in the exposed group than in the unexposed group because the exposed group lives long enough to be hospitalized, whereas the un-exposed individuals who die will be censored, ‘event-free’. When informative censoring exists, a competing risks framework can be used to obtain unbiased effect estimates.125 This is discussed in further detail in section 3.2.6. Population-based administrative databases are excellent tools for reducing other forms of selection bias related to patient recruitment. In Canada, universal health coverage ensures that every citizen is covered by provincial health coverage and therefore is captured by the database. For this reason, administrative database studies are not prone to referral122 or selfselection bias.122 3.1.2  Information Bias  Information bias refers to any error that arises from when a measurement does not accurately reflect what it is supposed to be measuring, and does so differentially between exposure or outcome categories.99,121-123 Scenarios resulting in misclassification of exposure, diagnosis, and outcomes, and the implications of these misclassifications are discussed in this section. 3.1.2.1 Exposure Misclassification In pharmacoepidemiological studies, drug exposure is difficult to track, especially when using administrative drug databases to measure exposure.126 When a prescription has been filled, it is typically assumed that the individual is subsequently exposed; however, this is not necessarily true. Non-consumption of medication following dispensation occurs in many ways: the medication is not taken at all, the medication is taken intermittently (often due to 35  forgetfulness or delay in re-filling a prescription), or the medication is taken at a dose different from that prescribed (e.g. pill splitting, change in frequency of daily pill-taking).127 Because there is currently no way to track consumption with administrative data, this results in noncompliant individuals being misclassified as ‘exposed’ when they may be un-exposed, or exposed to a lower dose. By measuring compliance in addition to exposure, we can improve the characterization of drug consumption. While still imperfect in its ability to quantify when and how much medication is truly consumed, this approach accommodates for varying levels of compliance with prescribed dose across individuals. The continuous multiple availability (CMA) metric is useful in administrative database studies when there is no evidence of a clinically meaningful compliance cut-point.128 The metric allows for multiple interval measures which are useful when longer-term duration of use or cumulative drug dosage are important aspects of exposure, and the “availability measures” rather than “gap measures” are desirable when studying cumulative drug dosage or dose-response relationships.128 3.1.2.2 Immortal time bias Immortal time bias is an issue relating to exposure misclassification that arises frequently in pharmacoepidemiological literature.115 There are two main types of immortal time bias: misclassified and excluded.129 Misclassified immortal time bias is more common, and arises when individuals are assigned to exposure categories based on prescription fills that follow the cohort entry date. For example, in the study design of Sin and Tu103, individuals were classified as ‘exposed’ if they filled a prescription for ICS in the ninety days following hospital discharge; however, follow-up time started upon discharge. The time spent unexposed was therefore misclassified as ‘exposed’ from the time of discharge to first fill, and was necessarily immortal since individuals must have lived long enough to fill the prescription. Excluded immortal time, on the other hand, occurs in scenarios when the exposed population enters the cohort at the time of (first) drug fill and the unexposed population enters the cohort at another point in disease history, such as at the time of diagnosis. If a step-wise approach to treatment exists, 36  then the exposed population will enter the cohort at a later stage in disease history, and will have immortal time between time of diagnosis and drug fill that was excluded from analysis. By excluding this immortal unexposed time, there will be fewer person-moments in the unexposed, and therefore the event rate in the unexposed will be artificially inflated, making the exposed group look relatively better compared to the unexposed group.129 Appropriate cohort definition can avoid excluded immortal time, and use of time-varying exposure methodology will avoid misclassified immortal time. 3.1.2.3 Diagnostic Misclassification Differential disease misclassification also leads to information bias. The study cohorts used in pharmacoepidemiological studies often include only one disease category and track time to event between exposure categories. Although there is no formal disease-free category, misclassification bias occurs when non-diseased individuals are included in the study cohort and are not balanced between exposure categories. In addition to the issues regarding individuals not being captured in administrative databases due to underutilization of the health system, or having an ambiguous ICS code associated with a visit (discussed in section 2.5.2), administrative records may also yield incorrect ICD codes, arising from infidelity of the transcription of the ICD code on the patient chart to the database, incorrect billing diagnostic codes despite correct clinical diagnoses, or incorrect diagnoses. The transcription from patient chart to database in hospital discharge databases is reportedly highly accurate;130,131 however, the frequency of incorrect billing codes and incorrect diagnoses are more difficult to ascertain, and has not been done for COPD. In addition to inaccuracies, there are also potential omissions from administrative databases due to the limitations of coding. In MSP billing databases, there is room for only one diagnostic code, meaning that patients suffering from two or more comorbid conditions can continue to visit a physician without ever being traced by the MSP database as having that disease. Medication use has been shown to provide good discrimination between COPD and asthma, as well as between COPD and other diseases.132 37  Therefore, we expect that in the event that a disease-free individual is included in the study cohort, they would be more likely to be included in the non-exposed category than the exposed. 3.1.2.4 Outcome Misclassification Outcomes suffer from the same potential for misclassification as previously discussed. Using administrative databases, health service utilization often serves as a proxy measure for a clinical outcome. In database studies, acute exacerbations are typically operationalized as one of: hospitalizations or emergency department visits with COPD as the first or second most responsible diagnostic code; or dispensation of an oral corticosteroid and antibiotic on the same day.108 Non-exacerbations can be misclassified when individuals are admitted to hospital for a reason unrelated to COPD (e.g. diabetes) but COPD, a co-existing condition, is recorded as a responsible diagnosis. Conversely, an exacerbation may be missed (misclassified as noevent) if the individual did not fill an oral corticosteroid and antibiotic on the same day, or if they were admitted to hospital for an acute exacerbation but COPD was not recorded in the operationalized responsible diagnosis field. COPD is often mistaken for congestive heart failure upon admission to hospital and could be missed.133,134 Hospital over-crowding can also lead to missed acute exacerbations, as individuals may be less likely to be admitted. When misclassification is nondifferential across levels of exposure, which is likely in the scenarios described previously, it attenuates the strength of association between exposure and outcome (bias towards the null);122 however, differential misclassification of outcome that arises when exposure to a COPD drug increases the likelihood of being recorded as having COPD can change the size and even direction of the effect. Death is also frequently used as an outcome. Although mortality databases are highly accurate in capturing deaths, the cause of death is typically reduced to a single field, which can lead to poor sensitivity in capturing cause-specific mortality in COPD.49,135 As with misclassified hospitalizations, this can lead to a biased effect estimate if the misclassification of death is differential across levels of exposure. 38  3.1.3  Confounding  In observational studies, individuals are not randomized to treatment arms. The observed and unobserved characteristics for the treatment groups are unlikely to be balanced. When characteristics associated with the outcome of interest are not distributed equally across treatment groups, this imposes a distortion and when not properly accounted for, yields effect sizes that are inaccurate. To be considered a confounder, a covariate must 1) be a predictor of risk among the unexposed; 2) be a correlate of exposure in the population in the population serving as the source of subjects; and 3) not be an intermediate variable in the causal pathway under study.136 It is important to ensure that non-confounders – particularly variables that lie on the causal pathway between exposure and outcome – are not adjusted for in analysis. Adjustment for non-counfounders will actually impose additional bias to the results rather than correct for existing bias.137 Directed acyclic graphs (DAG) are excellent tools for researchers to visually assess potentially confounding variables137,138 and prevent unnecessary adjustment. Figure 3.1 presents a DAG identifying factors that have been shown to confound the relationship between ICS and COPD. Table 3.1 presents evidence of the association between these factors and exposure to ICS, as well as to risk of death or hospitalization. Confounding by severity is a specific type of confounding that relates to differential disease severities in the exposed and unexposed groups. This generally occurs when a drug is indicated for use at an advanced stage of disease – often in chronic diseases. Confounding by severity differs from confounding by indication: with confounding by severity, all study subjects have the disease, but in varying levels of severity, whereas confounding by indication refers to scenarios where individuals may or may not have a disease that is a risk factor for the outcome.139 When the outcome is associated with severity of disease, confounding by severity causes an increased event rate in the exposed category, compared with the unexposed. Presence of this confounding will increase the effect estimate, and may even suggest a harmful 39  effect of the drug. This does not pose a problem when the outcome is not related to severity of disease, for example, in studies of unintended harmful drug effects (e.g. the relative risk of development of glaucoma based on exposure to ICS).122 When it is necessary to adjust for confounders (through restriction, matching, stratification, or statistical modeling), proxy measures for severity, although imperfect, are used in order to balance exposed and unexposed groups according to severity. In studies of COPD, lung function is an excellent measure of severity with which to balance groups; however, these data are not available in administrative health databases. Protopathic bias can be thought of as an extreme form of confounding by severity. Protopathic bias is not actually a confounder, but is discussed here because of its similarity to confounding by severity. It occurs when individuals become exposed immediately prior to an event because the medication relieves its early symptoms,139 and is therefore problematic in studies of relief medication.139 The causality in this association is reversed: the individuals are taking the medication because they are at high risk; they are not at high risk because they are taking the medication. By including this subset of high-risk individuals in the exposed population, it produces an elevated risk of death among the exposed and may result in a finding that suggests harmful effects of the drug. Conversely, ‘reverse’ protopathic bias can occur when an individual discontinues medication when they begin to feel prodromal symptoms. This yields an opposite consequence on the effect estimate: individuals are categorized as unexposed immediately before the outcome occurs, causing the event rate in the unexposed to be elevated. This yields an artificially low effect size. Epidemiological studies are also subject to unmeasured and unknown confounders. These are underlying differences between exposure groups that are not captured by the data source, or that cannot be quantified by a measurable variable. The use of instrumental variables is a strategy that can account for unmeasured confounders (discussed in section 3.2.4), but when these are unavailable, the best that can be done is to acknowledge their existence and attempt to control for confounders that are measurable. 40  We will discuss analytic approaches that can reduce confounding in pharmacoepidemiological studies in section 3.2. 3.1.4  Effect modification  Effect modifiers are variables such that the effect of one variable (e.g. exposure) on another (e.g. outcome) varies across strata of the effect modifier (e.g. sex).140 This can occur in situations where an exposure is necessary but not sufficient for an outcome: a third variable may need to be present in order for the exposure to be fully effective.141 In extreme examples, presence of a third variable may reverse the effect of the exposure. There is evidence to suggest that the presence of coexisting asthma modifies the effect of ICS in reducing hospitalizations and death among individuals with COPD.113 It is included as an effect modifier in the directed acyclic graph presented in Figure 3.1. 3.1.5  Summary of methodological challenges  Despite their limitations, observational studies are essential for understanding drug effectiveness in the general population. The aforementioned challenges of bias and confounding simply highlight the extent of methodological rigor that must be applied in analyzing drug effectiveness. In conjunction with the results of drug efficacy, a well-conducted observational study where the magnitude and direction of bias is well-documented, is critical in informing policy- and decision-makers whether a drug is being used effectively in the jurisdiction under study.  3.2  Pharmacoepidemiological modeling approaches Many of the challenges discussed previously can be avoided by using an appropriate study  design and careful analysis.142 We will discuss some of the strategies discussed in the literature that account for covariate imbalances,143-145 assuming a retrospective cohort study design using administrative data.  41  3.2.1  Multiple regression models  Simple regression models express an outcome variable as a function of an explanatory variable. In cases where the exposed and unexposed populations differ with respect to risk factors associated with the outcome, additional terms may be added to the model.146 This is called multiple regression (where there is more than one explanatory variable), and can be applied to all types of regression models, including linear, logistic, Poisson, and Cox proportional hazards models.147 By including the confounding variable in the regression model, a portion of the effect size is attributed to that variable, and the effect size associated with the exposure of interest is adjusted. There are several limitations to this approach.148,149 First, this method does not correct of imbalances in unmeasured or immeasurable confounders, as they are, by definition, not measured, and therefore can not be entered into the model. When confounders are unmeasured or immeasurable, and there are no proxy measures for these confounders available, the results will be biased. A second limitation is that if the joint distributions of the covariates do not overlap, this will not be detected by the model, and the model will be calculated anyway, yielding ungeneralizable results. This is not a limitation if the joint distributions are checked and do in fact overlap – thus it is merely a caution. Finally, a fundamental assumption of any model is that it has been properly specified: if the model is misspecified, the estimated effect sizes will be inaccurate. 3.2.2  Extended Cox models with time-dependent covariates  The Cox proportional hazards model147 is the most widely used model for time-to-event analysis with multiple predictors and censored observations. The hazard of event occurrence is defined as the conditional probability of event occurring at time t, given survival to time t or later. Mathematically, it is expressed as  lim P(t ≤ T < t + ∆t | T ≥ t )  λ (t ) = ∆t →0  (1)  ∆t 42  The Cox proportional hazards model provides an estimate of the comparative hazard rates between groups of individuals. The model assumes an underlying (baseline) hazard function, λ0 (t ) , and assumes that for all groups of individuals, the hazard of outcome over time is a constant multiple of this hazard function. The hazard function of an individual, i, with covariates Xi=(X1,X2,X3,..,Xp) is  λi (t ) = λ0 (t )e[X β ]  (2)  i  where β is a vector of coefficients. Assuming that the baseline hazard is identical for all groups, the ratio of the hazard of death over time for individuals i and j with differing covariate values is computed as  λi (t ) λ0 (t )e[X β ] = λ j (t ) λ0 (t )e ⎡⎣X β ⎤⎦ i  (3)  j  Thus, for a one unit increase in a covariate, Xp with all other covariates equal, the hazard ratio can be reduced to e  βp  . The method has been extended to accommodate for variables  that change with time and variables that interact with time. While the Cox proportional hazards model assumes that the baseline hazard for all individuals is proportional over time, the extended Cox model147 is not limited by this restriction and allows covariate values to change over time. The general form of the extended Cox model is  λi (t ) = λ0 (t )e[ X β + X ( t )δ ] i  (4)  i  43  where  λ0 (t ) is the baseline hazard and is a non-negative function of time, Xi is a vector of  time-constant covariates, Xi(t) is a vector of time-varying covariates, and β and δ are vectors of coefficients. In RCTs of drug efficacy, an intention-to-treat approach is often taken, where an individual is assigned as being exposed or unexposed to a drug at baseline, regardless of changes to exposure status following baseline. This is done to preserve the randomization in RCTs and prevent the introduction of bias. In analyzing drug effectiveness, which is already subject to bias due to non-randomization of individuals to treatment, time-varying covariates better represent what occurs in a ‘real world’ setting. Extended Cox models (with time-dependent exposure) have been used in several of the observational studies of ICS effectiveness in COPD.106,107,110,113 The model allows individuals to change their exposure status during followup, and importantly, avoids immortal time bias (section 3.1.2). The underlying assumption of the extended Cox model with time-dependent exposure is that the change of exposure status is unrelated to the risk of subsequent event.150 Violation of this assumption, for example becoming exposed to a drug because one is at higher risk of the outcome, will bias the effect size upward. Because of this assumption, the results of several studies of ICS effectiveness have been criticized in their ability to produce an unbiased effect size of ICS on COPD morbidity and mortality because of the association between ICS initiation and risk of outcome.151,152 When using this model, care must be taken to ensure that this assumption is not violated. 3.2.3  Propensity scores  Propensity scores have been used frequently to adjust for indicated use of medications across treatment groups.153-155 The propensity score was developed by Rosenbaum and Rubin156 as “the conditional probability of assignment to a particular treatment given a vector of observed covariates.” The propensity score is most often calculated using a logistic regression model, where the outcome is the probability of receiving a particular treatment, and covariates 44  associated with exposure to treatment are used as independent predictors; however, other methods such as neural networks, discriminant function analysis, and classification trees may also be used.155 With a given model, a propensity score can be calculated for each individual, based on his or her associated covariates. The propensity score is then incorporated as a covariate in a regression model to model the outcome of interest (e.g. death) to adjust for confounding by indication. The propensity score may be either continuous or categorized into an ordinal categorical variable, which may be used to stratify the regression model or used as a matching variable. Although propensity scores allow for adjustment according to likelihood of receipt of treatment, unmeasured confounders (section 3.1.3), cannot be accounted for by propensity scores. However, because the groups may seem balanced by likelihood of medication receipt, they can give a false impression that exposure groups are balanced.155 The use of propensity scores in research has increased exponentially over the past eight years, however, it is questionable whether its use has improved effect estimates substantially.155 A comparison of effect estimates in 78 analyses using propensity scores and traditional multiple regression found only eight studies in which the propensity score analysis yielded a significantly different score than the multiple regression model.157 3.2.4  Instrumental Variables  Instrumental variables (IV) are used extensively in econometrics, and have been used increasingly in the health care research literature as a method to adjust for unbalanced exposure groups.158,159 This method, unlike the propensity score method, has the ability to adjust for both measured and unmeasured confounders.158 IVs are observable factors associated with the treatment but not with the outcome. For example, in a RCT, the randomization assignment (0 for placebo, 1 for drug) would be a perfect instrumental variable – it is perfectly correlated with the treatment (in this case exposure or non-exposure to the drug)  45  but is not associated with the outcome (e.g. time to death). The technique is performed using two equations:  X i = α 0 + α 1Z i + ν i  (5)  Yi = β 0 + β1 X i + ε i  (6)  where Yi is the outcome, Xi is the treatment received, Zi is the instrumental variable and α1 ≠ 0. Equation 6 can then be substituted into equation 5 to yield:  Yi = ( β 0 + β1α 0 ) + β1α1Z i + ( β1ν i + ε i )  (7)  The instrument must be related to the treatment, but not related to factors explaining the treatment, or to factors explaining the outcome. Mathematically, these criteria are summarized as:158  1) ρX,Z ≠ 0, non-zero correlation between the treatment and the instrument 2) ρZ,ν = 0, zero correlation between the instrument and other factors explaining the exposure. 3) ρZ,ε = 0, zero correlation between the instrument and other factors explaining the outcome (main assumption).  When a strong instrument exists (|ρX,Z| is close to 1) and assumptions have not been violated (ρZ,ν = 0 and ρZ,ε = 0), IV modeling is an excellent way to adjust for known and unknown confounders in unbalanced exposure groups; however, when assumptions are violated, the effect estimates will be biased and imprecise (i.e. large standard error).158 In pharmacoepidemiological studies using administrative databases, it is not always possible to find a variable that is associated with the exposure but not the outcome. Brookhart et al160 used prescription claims database to estimate prescribing physician’s preference for one drug relative to another. They used this as an instrument in their observational study of the 46  effectiveness of COX-2 inhibitors compared with nonselective nonsteroidal anti-inflammatory medications on gastrointestinal complications. While this is theoretically a good instrument, the proxy measure used to define physicians’ prescribing preferences violates the assumptions, leading to a biased effect estimate.161 Not all types of observational studies are well-suited for IVs. In studies with time-varying exposure, individuals may change treatments throughout the study. The example of the COX-2 inhibitors assumes an intent-to-treat type analysis rather than a time-varying analysis of treatment. Identifying a time-varying instrument for time-varying exposure is not often possible.161 Another type of study that is ill-suited to IV models are those that study heterogeneous treatment effects.161  Additional assumptions are required to estimate the  average causal effect of a treatment in the population.161 For example, there should be no additive effect modification by the instrument among the treated (and untreated) patients.162 Hernan et al161 warn against using IV analysis under these circumstances. 3.2.5  Marginal Structural Models  Marginal structural models (MSM) are able to adjust for time-dependent confounding of effect estimates in the presence of a time-dependent exposure.163 First introduced by Robins,163 this technique was used extensively in HIV/AIDS research, but infrequently outside that disease area. MSMs can be used instead of standard approaches to modeling timedependent exposure (e.g. Cox regression with time-varying covariate) when the following conditions apply: 1) there exists a time-dependent covariate that is a risk factor for, or a predictor of, the event of interest and also predicts subsequent exposure, and 2) past exposure history predicts subsequent level of the covariate.163 This tends to hold true when confounding by severity or indication exists, particularly when the treatment regimen changes over time according to a person’s response to the treatment. While this is a promising method for controlling for bias in studies of time-dependent exposure, this approach is limited to data-rich studies where values of intermediate variables 47  (such as CD4 lymphocyte count) exist. In studies using administrative data, without clinical variables, the ability to implement this model is limited. 3.2.6  Competing risks framework  Regardless of the modeling approach used, pharmacoepidemiologic studies may be limited in their ability to characterize the risk of experiencing non-fatal outcomes. Standard survival techniques can only be meaningfully interpreted when the outcome of interest is independent of the censoring mechanism. In studies where the outcome is time to hospitalization, censoring of deaths is typically not independent of the risk of hospitalization. The competing risks framework allows for unbiased analysis of competing events by fitting cause-specific models separately for each type of failure.125 Although the most frequently used method does not allow for contolling of confounders,164 Lunn and McNeil165 recently developed a method for dealing with competing risks using an adaptation of the Cox proportional hazards model. This method has yet to be incorporated into standard statistical software, in such a way as to accommodate for time-varying covariates and interaction across strata of a second variable. 3.2.7  Summary of pharmacoepidemiological modeling approaches  Several methods exist to account for confounding and bias in pharmacoepidemiological studies. Multiple regression models, with or without time-varying covariates, are the most commonly used methods for modeling the association between drug exposure and outcome. Propensity scores, while quantifying the likelihood of receipt of medication, may not improve significantly upon multiple regression mostly because they do not offer any advantage over multiple regression approaches in accounting for unmeasured confounders. In theory, the IV approach is an excellent tool for accounting for unmeasured confounding and therefore providing a method for balancing exposure categories. Unfortunately, the ability to identify valid instruments in pharmacoepidemiological research is highly limited, and furthermore, IV methods are not well-suited to address time-varying exposure or heterogeneity in treatment 48  response. MSMs can handle time-varying exposure as well as time-varying confounders; however, the use of these models is limited to data-rich environments with respect to clinical outcome measures; their use with administrative data sets is limited. It seems that, while many advances are being made in pharmacoepidemiological modeling approaches, advanced models are only applicable in certain scenarios. In a study of timevarying exposure with administrative data, multiple regression with time-varying exposure is the most feasible approach. As it is a widely used approach, this allows for comparison of estimates with other studies employing the same technique; however, it does not avoid the problems of unmeasured confounders, model misspecification, or violation of assumptions. Care must be taken in minimizing these problems as much as possible and understanding their impact, in order to generate the least biased effect estimates.  49  3.3  Figures and Tables  LABA: long-acting beta-agonists; ICS: inhaled corticosteroids; ACG®: adjusted clinical group; LOS: length of stay in hospital; ICU: intensive care unit. Arrow heads indicate the direction of the association. The arrow head from “Asthma” points to the association between “Exposure to ICS” and “Hospitalization or Death”, indicating that it modifies the effect.  Figure 3.1 Directed acyclic graph of variables that potentially confound or modify the relationship between inhaled corticosteroids and death or hospitalization (see Table 3.1).  50  Table 3.1 Variables that potentially confound the association between exposure to inhaled corticosteroids and risk of death or hospitalization (see Figure 3.1) Characteristic Demographic information Age Sex  Hospitalization information Admission to intensive care unit (ICU)  Measurement  Association with exposure to ICS  Association with risk of death or hospitalization  10-year intervals  Younger individuals more likely to receive treatment166 Women associated with asthma diagnosis168 who are more likely to receive ICS  Older individuals at advanced risk of death and hospitalization167 Mortality and re-hospitalization for COPD higher in men than women167  Admission to ICU implies more severe disease; associated with ICS initiation Longer LOS implies more severe disease; associated with ICS initiation Lower Charlson score associated with higher likelihood of ICS166 Assumed to be similar to association with Charlson score  Admission to ICU implies more severe disease; higher risk of outcome  Binary: M/F  Binary: Y/N  Length of stay  Days (continuous)  Charlson comorbidity index  Number  Adjusted Clinical Group (ACG®)  Binary: Y/N  MSP information History of asthma  Drug information Prior use of long-acting beta agonist (LABA)  Prior use of anticholinergic  Longer LOS implies more severe disease; higher risk of outcome Higher Charlson comorbidity index associated with higher risk of death169 Belonging to ACG® group 4940 and 5070 are associated with higher risk of mortality and hospitalization170  Binary: Y/N  ICS is the mainstay of asthma treatment and is associated with higher likelihood of use5  History of asthma implies better response to ICS, although unknown if this is ‘cancelled out’ by existing COPD  Binary: Y/N  Because of stepwise approach to COPD management, expect individuals already taking LABA to be more likely to take ICS11 Because of stepwise approach to COPD management, expect individuals already taking anticholinergics to be more likely to take ICS11  Expect greater ICS effectiveness when used in conjunction with LABA  Binary: Y/N  Expect greater ICS effectiveness when used in conjunction with LABA  51  4 Objectives The overarching aim of this pharmacoepidemiologic study was to determine whether the effectiveness of ICS in persons with COPD differs in the presence of concomitant asthma. The specific objective was to compare the reduction in the hazard of hospitalizations or death conferred by ICS among COPD-afflicted individuals without coexisting asthma to the hazard reduction among individuals afflicted with both diseases.  52  5 Methods 5.1  Overview We used a retrospective cohort study to estimate the effect of ICS on death or  hospitalization among two subgroups of individuals afflicted with COPD: individuals afflicted with both COPD and asthma; and individuals afflicted with COPD but not with asthma. Data obtained from the BC Linked Health Database (BCLHD) between April 1, 1991 and March 31, 2001. We measured the intended effect of ICS treatment on the risk of re-hospitalization or death among individuals diagnosed with COPD, and individuals diagnosed with both asthma and COPD. We applied design features and statistical methods to account for common challenges in using observational data for pharmacoepidemiological studies discussed in section 3.1.  5.2  Data Quality  5.2.1  BC Linked Health Database  The BCLHD is a deterministically linked set of administrative databases that include information on health care resources used by all registered individuals in BC (approximately 97% of residents). The linked database integrates health service records, prescription data, population health data and census statistics anonymously at the individual level.171 General issues pertaining to the reliability and validity of administrative databases were discussed in section 3.1.2. There are no validation studies that validate the BCLHD against survey data for COPD. Validity of physician billing databases for identification of COPD in Québec administrative databases compared with survey data is reportedly low;65 however, there is no gold standard against which to compare COPD prevalence, and survey data may overestimate COPD.42 Given the accuracy of Canadian hospital discharge databases compared with patient charts in other disease areas,130,131 as well as in respiratory disease,172  53  the use of hospitalization data to identify patients with COPD is likely a reasonable representation of reality. The validity of PharmaNet (College of Pharmacists) prescription refill data for reflecting adherence among individuals with heart failure taking beta-blockers was reportedly high.173 No validation studies have been performed assessing the validity of the PharmaNet database in COPD patients; however, coding accuracy is reportedly high in other Canadian drug databases.174 Therefore, although there is no certainty that the drugs were consumed, the PharmaNet database is likely accurate with respect to the dispensation of drugs.  5.3  Ethics Ethical approval for this study was provided by the University of British Columbia  Behavioural Research Ethics Board (Appendix 2).  5.4  Data Preparation Data were received from the BCLHD in six files: data from the province’s Medical Services  Plan (MSP) (1991-1995; 1996-2001), hospital discharge data (1991-1995; 1996-2001), PharmaNet data (1996-2001) and Adjusted Clinical Group (ACG®) codes (1996-2001). Demographic and mortality statistics were included in these files. We queried the database using Structured Query Language (SQL) on an Oracle® database through the Toad® interface. We prepared the data, created new variables, and conducted all statistical analysis using R version 2.7.0 (© The R Foundation for Statistical Computing). 5.4.1  Data verification and transformation  In order to prepare the data for analysis, a series of checks of data integrity were undertaken. We first checked the records of ICS fills contained in the PharmaNet dispensing data. ICS are dispensed in containers (inhalers, nebules, or blisters) that contain anywhere between 8 to 240 units (e.g. inhalations) of medication, depending on the brand and packaging of the drug. We prepared a table containing, for each brand, possible packaging quantities, and minimum and maximum daily doses, and the corresponding days of supply that could be 54  delivered from one package. These data were obtained from the Compendium of Pharmaceuticals and Specialties in years concurrent with our study period.76,175 We crosschecked the ‘quantity’ and ‘days’ fields of the PharmaNet database with the values in the table. We reviewed the records for which the number of days supplied was outside the possible days calculated in our table, and records for which the quantity supplied was not a multiple of the possible packaging units. We used these records to prepare a standard set of rules to apply in order to detect and correct errors in data entry systematically. We identified records where the quantity dispensed was equal to 1, 2, 3 or 4 and the minimum possible supply was greater than 1, 2, 3, or 4 respectively. We changed these entries to be multiples of the number of units per canister (e.g. if one inhaler provides 200 doses, we changed the value of “1” to “200” and the value of “2” to “400”). We also identified records where the number of days provided was greater than the maximum possible days, calculated from the minimum daily dose, and the quantity supplied. We changed these values to the quantity dispensed divided by the minimum daily dose. We repeated this process for values that were below the minimum possible days supplied. We systematically investigated all records where the ‘days’ field was greater than 200. We corrected entries that were clearly typographical errors (e.g. ‘days’ supply’ field was 2800 but all other fills for the same medication for that person were for 200 days). In total, approximately 1% of the dispensation 141,714 records were modified. We adjusted the hospitalization data in two ways. First, we removed any hospitalizations for which the individual was not currently registered with MSP, as it was unclear whether drug data would be captured for these individuals. Second, we merged records for patients who were transferred between hospitals (these are entered as separate records). We applied a rule such that individuals who were discharged and re-admitted within one calendar day were considered as the same hospitalization. We did not use the ‘transfer to’ and ‘transfer from’ fields, because an initial inspection of these fields showed that they were not entered consistently (e.g. ‘transfer to’ field complete in one record but ‘transfer from’ field in subsequent record with same calendar date was empty). Because some individuals were transferred to 55  extended-care hospitals directly from acute-care hospitals, we tracked the level-of-care fields for each of the hospital admissions. For privacy reasons, the month of death is provided, but not the day of death. We set the death dates to the last day of the month, to ensure that events that occurred in the same month as death could not appear to occur after the death date. 5.4.2  Cohort Definition  The source population was all residents in BC who were registered with MSP between the April 1, 1991 and May 2001. The study sample included a dynamic population of individuals who were continuously enrolled in MSP between April 1, 1996 and March 31, 2001 (Figure 2.1). Individuals entered the cohort upon discharge (to the community) from a first hospitalization for COPD between October 1, 1996 and March 31, 2001. We excluded individuals who were hospitalized for more than ninety days, because this was unlikely to be for an acute hospital episode. We cross-referenced hospitalization records between April 1, 1991 and September 30, 1996 to verify there were no prior hospitalizations for COPD. To be considered a hospitalization for COPD, the first or second most responsible diagnosis had to correspond to the ICD-9 codes used to define COPD (Table 2.3). We excluded the ICD-9 code ‘490’ (bronchitis, not specified as acute or chronic) due its low specificity for COPD (discussed in section 2.5.2).41 Individuals were included in the cohort if they were aged 45 years or older at the time of discharge in order to exclude asthmatics who did not have COPD.120 5.4.3  Exposure  Exposure to ICS was assessed using the PharmaNet database, which captures prescriptions dispensed from community pharmacies to all individuals enrolled in BC’s MSP. Exposure to ICS was defined as prescriptions containing belomethasone dipropionate, budesonide, fluticasone propionate, or triamcinolone as any of the active ingredients, in which the administration route was inhalation. Nasal sprays and oral corticosteroids were excluded. Combination products, such as ICS/LABA inhalers, were included. 56  We tracked exposure using a time-dependent measure, in an effort to avoid immortal time bias.107 We used the CMA128 metric to measure compliance using the days’ worth of medication dispensed, as recorded in the PharmaNet database. We calculated the CMA for each individual from first recorded ICS fill until last recorded fill. The CMA following the last recorded fill was assumed to be the same as between the penultimate fill and the last. We used the CMA following the last recorded fill to define an ‘expected next fill’. Individuals who did not fill a prescription on or before this day were then considered unexposed. Changing between brands of ICS was not considered to be discontinuation from ICS. We added a 30 day grace period to the end of the ‘expected next fill’ because of the imprecision of the death date in our data. At the time of discharge from hospital, the CMA was defined as 0 for those who had not yet filled an ICS prescription and for those who had filled prescriptions prior to hospitalization, the current level of CMA was used. We used a counting process147 to identify transition days when an individual’s level of compliance changed. Table 5.1 shows an example of the counting process form of records for an individual who is 100% compliant for the first 90 days, 80% compliant for the next 100 days, and 50% for the following 200 days. We categorized CMA according to the step function:  ⎧0 if CMA = 0 ⎪1 if 0<CMA < 0.25 ⎪⎪ ICS = ⎨ 2 if 0.25 ≤ CMA < 0.5 ⎪3 if 0.5 ≤ CMA < 0.75 ⎪ ⎪⎩ 4 if 0.75 ≤ CMA  (8)  For the main analysis, we categorized ICS exposure according to the CMA:  ⎧0 if CMA = 0 ICS = ⎨ ⎩1 if CMA > 0  (9)  57  5.4.4  Asthma  We assumed that individuals with co-occurring COPD and asthma were diagnosed with asthma prior to COPD. Therefore, we assigned individuals to the coexisting asthma stratum if they had any history of a hospitalization or specialist visit coded for asthma (ICD-9 493.xx). We intentionally excluded general practitioner (GP) visits, because these records have poor discriminatory ability to distinguish between asthma and COPD;132 however, the definition of asthma was broadened to include GP visits in a sensitivity analysis (section 5.5.5). 5.4.5  Comorbidities  We used two measures of comorbidities in our analysis: the ACG® system176 and the Charlson comorbidity index.169 The ACG® system was developed by investigators at Johns Hopkins University. ACGs® are a set of 106 mutually exclusive health states defined by morbidity, age, and sex. Using administrative data, individuals are first assigned one or more Aggregated Diagnosis Groups (ADGs®) according to the ICD codes present in their resource history use. Based on the combination of ADGs®, individuals are then assigned to an ACG®. In our study, the ACGs® were provided by the BC Ministry of Health, and incorporated all health service utilization data from the fiscal year prior to the index hospitalization. Two ACGs® (4940 – 6-9 Other ADG® Combinations, Age 35+, 4+ Major ADGs®; 5070 – 10+ Other ADG® Combinations, Age 18+, 4+ Major ADGs®) were identified in the literature as being significant predictors of death and future hospitalization.170 We used this categorization to produce a single binary variable that coded for belonging to one of these two groups. To calculate the Charlson comorbidity index, we used an algorithm adapted for administrative databases.177 We calculated the index using the diagnostic codes in the other 15 responsible diagnosis codes of the index hospitalization, as we did not have access to non-respiratory-related health services utilization of our study population, and, by definition, this was the first respiratory-related hospitalization of this study population. This likely underestimated the true value of the index. The Charlson index was developed specifically to predict mortality (within one year), and 58  weights presence of comorbid disease according to the associated risk of death.169 We considered using both of these indices because they were calculated in very different ways, and captured different sets of health utilization data – the ACG® capturing data from the fiscal year prior to index hospitalizations, and the Charlson index capturing listed conditions at the time of discharge from the index hospitalization. 5.4.6  Outcomes  The primary outcome of interest was a second hospitalization for COPD or all-cause mortality. We chose to use a combined end-point because it is an effective method of handling competing risks (discussed in section 3.2.6), and reduces selection bias due to informative censoring arising from death (section 3.1.1). All-cause mortality, rather than cause-specific death, was used because of the poor sensitivity of death codes could underestimate the causespecific mortality rate and there is potential for this misclassification to be differential across levels of exposure (discussed in section 3.1.2). Furthermore, given the systemic effects of COPD, some non-pulmonary deaths may be associated with COPD (e.g. heart failure), and thus a reduction in the frequency of these deaths among individuals exposed to ICS may reflect systemic anti-inflammatory properties of ICS. 5.4.7  Censoring  We administratively censored individuals on March 31, 2001 if they had not yet experienced an event. We also censored individuals if they de-registered from MSP, because after that date, there were no data on health service utilization or prescription fill data. Finally, we censored individuals upon entry to long-term care facilities. These events can occur on a prescheduled date, and do not reflect an exacerbation or death, for which the hospitalizations are meant to serve as a proxy measure, and furthermore, dispensations of medications by hospitals and long-term care facilities are not captured by PharmaNet and therefore, exposure status cannot be determined following admission.  59  5.5  Data Analysis The data analysis was designed to: 1) describe the hospitalization and prescription fill data  of the study cohort; 2) characterize ICS use temporally and to identify patient characteristics associated with ICS initiation; and 3) estimate factors associated with time to death or rehospitalization, using an extended Cox model. 5.5.1  Descriptive statistics  We calculated descriptive statistics of the study cohort, subdivided by presence of coexisting asthma, including counts, percents, and parametric and non-parametric descriptive statistics of study covariates at baseline. We also calculated mean and median follow-up times for each stratum, and tested for differences in proportions of outcomes using standard statistical methods.178 5.5.2  Characterization of ICS use  We calculated the number of ICS dispensations over the study period, as well as the most frequently dispensed brands and types of drug, stratified by presence of asthma. To characterize the timing of ICS initiation with respect to first hospitalization, we calculated the time to ICS initiation among the asthma and non-asthma sub-groups. We categorized individuals who were previously taking ICS upon hospitalization as initiating ICS at day 0, and treated ICS initiation as an event. We generated a life table to summarize days to event, taking into account censored observations. 5.5.3  Predictors of ICS use  We used a logistic model179 to identify predictors of ICS within the three months (90 days) following a hospitalization so that we could compare these results with other observational studies (section 2.8.5). Three months is a time frame that is short enough to capture dispensations for which the hospitalization served as the impetus for ICS initiation, yet long enough that individuals who were given ICS in hospital (which would not be captured by PharmaNet) would re-fill within this time frame. For this analysis, we excluded individuals who 60  died within the three month period. The anticipated association between characteristics and likelihood of ICS initiation is included in the table of potential confounders, in Table 3.1. We fit a multiple logistic regression model to estimate the relationship between covariates and the likelihood of initiating ICS following hospitalization. We obtained a regression coefficient for each variable which represents the change in the odds of ICS initiation for 1-unit increase in the covariate value, or the change in odds for a change in category, compared with a reference. We left all predictors in the final model to display all information. We recognize that this leads to model over-fitting; however, the purpose of the model was descriptive rather than predictive – the model is intended only for describing the data. We tested the significance of individual covariates to identify potentially predictive variables, and the significance of the association fitted by the model using the log likelihood ratio. 5.5.4  Association between ICS and time to re-hospitalization or death  We estimated and compared the unadjusted and adjusted association between exposure to ICS and the composite outcome of COPD hospitalizations or all-cause mortality in the two strata defined by presence of asthma. 5.5.4.1 Person Time We first calculated the counts and proportion of events – censorship, all-cause mortality, and hospitalization – across exposure categories and sub-grouped according to presence of co-existing asthma. We then used person-time analysis180 to calculate the crude event rate among exposed and unexposed person-years. The event rate was calculated as the number of outcomes divided by the person-years at risk. We used the ratio of these two event rates to determine the unadjusted measure of relative risk of re-hospitalization or death among exposed and unexposed populations, stratified by presence of co-existing asthma. 5.5.4.2 Cox Model147 The choice of modeling approach to measure the association between exposure to ICS and death or hospitalization is discussed in more detail in sections 3.2.7 and 7.2. Briefly, we chose 61  to use the extended Cox model over other modeling approaches because it was the most suitable model given that we were using administrative data without any clinical data and based on its ability to model varying exposure over time, adjust for potentially confounding variables, and allow for stratification of the cohort based on presence of co-existing asthma. In order to produce an adjusted estimate of ICS effectiveness, we compared the hazard of death or hospitalization among individuals exposed to ICS against individuals unexposed to ICS, at a given time, t. 5.5.4.3 Preliminary analysis We calculated the smoothed hazard function of death or hospitalization over time for individuals with and without coexisting asthma in order to provide insight about the conditional failure rates in these two groups and to support the decision to use a non-parametric form of the Cox model. 5.5.4.4 Adjusted analysis of ICS effectiveness The covariates considered in the model are listed in Table 3.1. We chose these variables based on the directed acyclic graph (Figure 3.1). Variables which, according to the directed acyclic graph, did not appear to be confounders, were not included in the model. In the interest of developing a parsimonious model with high explanatory power, we removed variables from the model by comparing log likelihood ratios of nested models with and without the variables, and by testing the effect of removing that variable on changes to the estimates of other coefficients. We did not anticipate there being any interaction terms, other than the interaction between asthma and ICS. This was based on there being no evidence in the observational literature and no a priori reasoning for interaction in the selected variables. To model the interaction between ICS exposure and presence of co-existing asthma, we used strata by covariate interaction terms.147 By fitting the extended Cox model, we obtained the log hazard ratios associated with each variable. Exponentiating the estimates produced the hazard ratios. For continuous variables, 62  the hazard ratio represents the change in hazard associated with a one-unit increase in the covariate at each point in time. Categorical variables were converted to dummy variables, giving each category its own hazard ratio, compared with the referent category. The hazard ratio associated with the ICS by asthma interaction was multiplied by the ICS hazard ratio to produce the hazard ratio associated with ICS in the stratum of individuals with coexisting asthma.147 The significance of the individual regression estimates were tested using the Wald statistic,147 which tests the null hypothesis that the regression estimate is equal to zero. We used the likelihood ratio test147 to assess the significance of the association represented by the model. 5.5.5  Sensitivity Analyses  In order to understand the impact of changing operationalized definitions in the study design, and to quantify the potential effect of potential sources of bias (discussed in section 3.1), we conducted several sensitivity analyses. We investigated the impact of changing our definitions of cohort inclusion, exposures, and outcomes. Analyzing the sensitivity of our results to various parameters was essential in being able to draw robust conclusions regarding the relationship between exposure and outcomes.181 5.5.5.1 Cohort definition The original analysis included individuals at the time of first hospitalization where the first or second responsible hospital discharge diagnostic code was COPD. We assessed the impact of restricting the cohort to individuals whose first diagnostic code was COPD, excluding those with a second responsible diagnostic code of COPD. In a second analysis, we investigated the effect of broadening the definition of the asthma. Initially, individuals were categorized into the coexisting asthma and COPD stratum based on a prior specialist visit or hospitalization coded for asthma, intentionally excluding GP visits. This definition was broadened to include any prior physician visit coded for asthma.  63  5.5.5.2 Exposure In the main analysis, individuals were considered to be exposed when they filled consecutive prescription fills, according to the methods of CMA calculation. In the literature, 75% compliance is frequently considered a cut-point of compliance. We investigated several compliance cut-points and estimated the effect sizes when creating cut-points of 10%, 25%, 50%, and 75% compliance. We investigated the extent to which the use of time-varying exposure in the main analysis avoided exposure misclassification in this study. First, we eliminated the time-varying exposure between the time of hospital discharge and first ICS fill. This yields misclassified immortal time129 and has been demonstrated to underestimate the risk of outcome in the exposed.106 We changed the exposure status so that individuals who filled a prescription for ICS during the follow-up period were categorized as ‘exposed’ from the time of hospital discharge until first ICS fill, rather than changing from ‘unexposed’ between hospital discharge to first ICS fill to ‘exposed’ at the time of first ICS fill. Second, we additionally eliminated time-varying exposure following an ICS fill. In the main analysis, individuals become ‘unexposed’ if they stopped filling ICS prescriptions (section 5.4.3) in order to reflect discontinued use medication; however, in this sensitivity analysis, individuals filling any prescription for ICS were considered ‘exposed’ from first ICS fill to the end of the follow-up period. Therefore, an individual’s exposure was considered time-fixed over the entire follow-up period – ‘exposed’ if they filled any prescription for ICS, and ‘unexposed’ if they did not fill any prescriptions for ICS. 5.5.5.3 Outcomes Using the Cox-proportional hazards model, we limited the deaths to respiratory-specific deaths, using the ICD-10 equivalents of the death codes used in the analysis of Sin and Man.105 We then measured the effectiveness of ICS in reducing the hazard of rehospitalization alone, all-cause mortality alone, and cause-specific mortality alone, compared between strata of concomitant asthma. 64  5.6  Figures and Tables  Figure 5.1 Diagram describing study design including cohort entry, diagnostic classification, exposure classification, and identification of outcomes based on prior hospitalizations and health service utilization recorded in the medical service plan (MSP) database, hospitalization discharge records and drug dispensations.  65  Table 5.1 Example of counting process format of data  Subject identification 1 1 1 …  Interval (days)  Compliance  Covariate 2  Covariate 3  (0,90] [90,190) [190, 390) …  100% 80% 50% …  x1 x1 x1 …  y1 y1 y1 …  …  66  6 Results 6.1  Descriptive statistics The study cohort consisted of 14,153 individuals who were first hospitalized for COPD  between October 1, 1996 and March 31, 2001 (Figure 6.1), 14.4% (2041) of whom had a history of asthma (Table 6.1). Overall, there were more men (57.8%) than women, and the mean age was 73.7 years (SD 10.0 years). Within the subset of subjects with coexisting asthma and COPD, there were proportionally fewer men (47.1%) but the mean age was similar (73.1 years; SD 10.2 years). More than half of the study cohort (59.7%) had a Charlson comorbidity index of one of more. 10.7% of the cohort had an index of two or three, which indicates that at the time of hospitalization, these individuals had three or more of the comorbid conditions that are associated with death within one year,169 or had one or more of the serious comorbid conditions that are associated with a higher risk of death. These numbers were similar in the subset of individuals with concomitant COPD and asthma – 60.5% of this sample had a Charlson index of one or more, and 9.8% had an index greater than one. Based on health services utilization in the year prior to the index hospitalization, 14.9% of all subjects were assigned to either of the ACG® groups 5070 or 4940, meaning that they had six or more comorbid conditions, four of which were considered to be serious (e.g. malignancy).170 An ICD code corresponding to COPD (Table 2.3) was listed as the first most responsible reason for hospitalization in 42.9% of the index hospitalizations, and as the second most responsible reason in 58.0% of the hospitalizations (0.9% had a code for COPD in both the first and second most responsible diagnosis fields). ‘Chronic airway obstruction, not elsewhere classified’ (ICD-9 code 496) was the most frequently recorded diagnostic code (23.8% of first most responsible diagnoses, and 45.4% of second most responsible diagnoses). Although pneumonia was not used as a criterion for cohort inclusion, it was recorded as the primary 67  diagnostic code in 11.3% of the index hospitalizations. Additional data describing diagnostic codes in the first and second most responsible diagnosis fields is presented in Table 1a, b of Appendix 1. Subjects were hospitalized for a median of five days (IQR: 3-9 days). 5.7% were admitted to ICU for a median of three days (IQR 1-5 days). The median follow-up time was 14.4 months (mean of 17.8 months), until the rehospitalization, death, or censoring, whichever came first. During this period, a total of 8540 study subjects filled a prescription for ICS. This represented 55.0% of the COPD-alone study sample and 92.2% of the COPD/asthma sample (Table 6.2). Of 59,135 prescriptions filled, Flovent® inhalers (250 µg/AEM) were most frequently dispensed, and the majority of dispensations were delivered using the MDI/dry powder formulation (90.4%). A summary of all drug dispensations is provided in Table 2 of Appendix 1.  6.2  Characterization of ICS use Temporal characteristics of ICS initiation are presented as a life table in Table 6.3. Only the  number of ICS-naïve subjects who were alive and uncensored at the end of each time interval is presented in the first column, although the cumulative proportion of individuals remaining ICS-naïve (presented in the third results column) was calculated using all person-time during that interval. 50.1% of all subjects (43.9% of COPD-only and 87.3% of COPD plus asthma) had filled an ICS prescription prior to their first hospitalization. The cumulative proportion of individuals who remained ICS naïve ninety days following hospital discharge was 49.8% among individuals with COPD only, and 10.0% among individuals with COPD and asthma. After two years, this proportion was 42.4% and 7.0%, respectively. Among all individuals who were dispensed at least one prescription of ICS, the median level of compliance was 70.3% (IQR: 44.1%-96.1%), based on individuals’ cumulative prescription refill compliance at the time of last fill (Table 6.4). The distribution of compliance was roughly Gaussian, with the most frequent compliance levels between 70% and 80%. Individuals who refilled their prescription prior to depleting their available supply had a cumulative refill 68  compliance that was calculated to be greater than 100%; however, because this represented a period of oversupply due to refill timing, these individuals were re-assigned a 100% level of refill compliance rather than a value greater than 100%. This resulted in a second peak in the distribution at 100%.  6.3  Predictors of ICS use There were 5711 individuals who were ICS-naïve at the time of discharge from  hospitalization and who did not die in the subsequent 90 days, 13.4% (764) of whom filled a first prescription for ICS in the ninety days following discharge. The estimates of the multiple logistic regression model show that there was a significant association between the likelihood of ICS initiation and sex, presence of asthma, and comorbid conditions (Table 6.5). The odds of ICS initiation during this 90 day period among individuals with concomitant asthma were approximately two times the odds among individuals without: 21.7% of ICS-naïve asthmatics filled within the ninety days following hospitalization, compared with 11.6% of ICS-naïve nonasthmatics. Men showed 25% lower odds of ICS initiation as compared with women (OR: 0.75; 95% CI: 0.64 - 0.88), and the presence of major comorbidities in the year prior to hospitalization (ACG® group) was associated with a 29% reduction in the odds of receiving ICS (OR: 0.71; 95% CI: 0.55 - 0.92). Women and asthma were correlated (χ2= 18.6; p <0.001); however, we retained both of these variables so as to present all information and because we do not intend to use these results for predictive purposes. There was a significant association between the likelihood of ICS initiation and the variables in the model; the log-likelihood ratio was 68.2, p <0.001 for 12 degrees of freedom, suggesting that at least one regression coefficient was different from zero.  6.4  Association between ICS and time to re-hospitalization or death Overall, there were 8385 (59.2%) censored observations due to MSP deregistration, entry  to a long-term care facility, or reaching the end of the study period (Table 6.6). All other outcomes in the follow-up period were either death or second hospitalization, which were 69  combined into a single outcome in the main analysis, as they represent competing risks. Among individuals with COPD alone, 14.8% of outcomes following first hospitalization for COPD were deaths; 9.6% of outcomes were death among individuals with concomitant COPD and asthma. The lower proportion of deaths in the stratum of individuals with concomitant COPD and asthma may not represent fewer deaths during follow-up, but rather that individuals were re-hospitalized prior to death more frequently in this group. Re-hospitalizations represented 34.5% of outcomes in the COPD plus asthma stratum, and 24.5% of outcomes in the stratum of individuals with COPD. Table 6.7 presents the crude rate ratios of death or hospitalization across ICS exposure categories. In the stratum containing individuals with COPD-only, the unadjusted event rate in the exposed group was 0.29 events per person year, and was 0.25 events per person year in the unexposed group, representing a rate ratio of 1.14 (95% CI: 1.07 – 1.21). Among individuals afflicted with both asthma and COPD, the unadjusted event rates were 0.30 and 0.32 in the exposed and unexposed categories, respectively, resulting in a rate ratio of 0.94 (95% CI: 0.79 – 1.12). The overall rate ratio (RR = 1.13; 95% CI: 1.08 – 1.19) was similar to that among the COPD-alone stratum. Table 6.8 shows the hazard ratios and 95% CIs of the extended Cox model, estimating the adjusted association between ICS exposure and hazard of hospitalization or death. The adjusted hazard ratio was not significantly associated with ICS exposure at time t, in the stratum of individuals with COPD alone (HR=0.99; 95% CI: 0.94-1.05). In the stratum containing individuals afflicted with both COPD and asthma, the hazard ratio of exposure to ICS at time t compared with non-exposure was 18% lower than in the stratum of individuals with COPD alone (HR = 0.82; 95% CI: 0.69 – 0.99). This represents a significant difference in hazard ratios between strata, and a beneficial association between ICS exposure and hazard of death or hospitalization.  70  Except for prior use of LABA, and time spent in ICU, all of the anticipated confounding variables were found to be significantly associated with the hazard of death or hospitalization. Increased age was associated with higher hazard ratios, as was increased Charlson comorbidity index and increased length of stay. Men had a higher hazard ratio compared with women (1.16; 95% CI: 1.10 – 1.22), and individuals assigned to high-risk ACG® groups were associated with increased hazard of death or hospitalization (HR = 1.55; 95% CI: 1.45 – 1.65). Although prior use of LABA was not found to be significantly associated with the hazard ratio, because some evidence that suggests that ICS and LABA together are more effective than ICS alone, we chose to keep prior use of LABA in the model. The log likelihood ratio of the model was 1079 with p < 0.001 on 15 degrees of freedom, suggesting that at least one regression coefficient is different from zero. The hazard functions for each stratum are displayed presented in Figure 2.1. For most of the study period, the hazard of death or hospitalization was higher among individuals with coexisting asthma and COPD than among individuals with COPD alone. This figure does not take into account differences in exposure status. It simply indicates that individuals with coexisting disease have a different underlying hazard function than individuals with COPD alone.  6.5  Sensitivity Analyses The sensitivity analyses are summarized in Table 6.9 according to changes in cohort  definition, exposure, and outcomes. The hazard ratios and Wald statistics for the variables included in each analysis are provided in Appendix 1, Table 3 a-k. The log-likelihood ratios for each model had p-values less than 0.001, suggesting that in each model, at least one hazard ratio was different from one. 6.5.1  Cohort definition  In limiting the analysis to include only individuals for whom the index hospitalization had a first responsible diagnosis was COPD (rather than a first or second), we found a greater 71  effectiveness of ICS in the COPD-only stratum. The hazard ratio in this stratum was 0.92 (95% CI: 0.84-1.00). The hazard ratio associated with the COPD/asthma times ICS interaction term indicated that there was greater effectiveness in this stratum than in the COPD-only stratum (HR = 0.92; 95% CI: 0.70-1.21). The point estimate of ICS effectiveness in the COPD/asthma stratum, obtained by multiplying the ICS hazard ratio with the COPD/asthma-ICS interaction hazard ratio, was 0.846, which is similar to the estimate in the main analysis. Changing the definition of co-existing asthma to include individuals with any history of physician visit or hospitalization with a responsible diagnostic code corresponding to asthma resulted in an increase in the number of individuals with co-existing asthma from 2041 to 5552 (39.2% of the cohort). Using this definition, the population defined as having coexisting COPD and asthma had a hazard ratio associated with ICS exposure that was lower than that among individuals with COPD alone (HR = 0.77; 95% CI: 0.68 – 0.86), consistent with the main analysis. 6.5.2  Exposure  We observed a trend in the point estimates of ICS effectiveness when we adjusted our definition of exposure as a function of compliance. When we required compliance to be greater than 75% in order to be considered exposed, and considered all compliance levels below 75% to be unexposed, ICS exposure was associated with an increased hazard of death or hospitalization in the COPD-only stratum (HR = 1.13; 95% CI: 1.05 – 1.22). Lowering this compliance threshold resulted in a lowered association between ICS and hazard of death or hospitalization in the COPD-only stratum: at the 50% compliance threshold, the hazard ratio associated with exposure was 1.09 (95% CI: 1.02 – 1.17); at 25% the hazard ratio was 1.08 (95% CI: 1.02 – 1.15) and at 10% it was 1.04 (95% CI: 0.98 – 1.10). In the asthma plus COPD stratum, the multiplicative combination of the stratum specific hazard ratio with the hazard ratio in the COPD-alone stratum yielded a similar decrease in effect estimates with decreasing cutpoints. 72  Changing the time-varying exposure variable to a time-fixed exposure was done in two steps (section 5.5.5.2). In the first step, where we demonstrated the effect of including immortal time between the time of hospital discharge and first ICS, 1444 individuals who had been categorized as ‘unexposed’ in the main analysis were re-categorized as ‘exposed’ during this period, resulting in 918.2 person years of misclassified immortal time. The inclusion of this immortal time among the exposed resulted in a lower point estimate of ICS effectiveness in the COPD-only stratum (HR = 0.80; 95% CI: 0.75-0.85) and in the co-existing COPD and asthma stratum (HR = 0.85 * 0.80 = 0.68) (Table 6.9), compared with the main analysis (Table 6.8). In the second step, where individuals who filled a prescription for ICS was categorized as ‘exposed’ for all time periods following first fill, 2144.4 person years were re-assigned as ‘exposed’ and 598 outcomes (deaths or hospitalizations) were re-assigned as ‘exposed’. The combination of these two steps resulted in a time-fixed exposure variable rather than timevarying. The overall effect estimate suggested a reduction in the risk of hospitalization among individuals with COPD alone (HR = 0.88; 95% CI: 0.83 - 0.93) and no significant difference in effectiveness between individuals with COPD alone and individuals with combined COPD and asthma (HR = 0.97; 95% CI: 0.74 - 1.25) (Table 6.9). 6.5.3  Outcomes  The choice of outcome measure impacted the point estimates of effectiveness of ICS in the COPD-alone stratum, but the point estimate associated with ICS effectiveness was consistently lower in the stratum of coexisting COPD and asthma, regardless of outcome measure. ICS was associated with a 58% reduction in hazard of all-cause mortality (HR = 0.42; 95% CI: 0.39 – 0.46) in the COPD-alone stratum, and a 41% reduction in hazard of cause-specific mortality (HR = 0.59; 95% CI: 0.51 – 0.69). In the COPD plus asthma stratum, the point estimates of these hazard ratios were 0.25 and 0.26, respectively. The combined outcome of cause-specific mortality or hospitalization yielded a hazard ratio of 1.47 in the COPD-only stratum (95% CI: 1.37 – 1.58), and hospitalization alone resulted in a hazard ratio of 1.64 (95% CI: 1.52 – 1.77). 73  6.6  Figures and Tables  Abbreviations: MSP = Medical Services Plan; LOS = length of stay.  Figure 6.1 Flowchart of cohort inclusion among individuals discharged from a first hospitalization for COPD in BC between 1996 and 2001.  74  Figure 6.2 Estimated (smoothed) hazard function for death or hospitalization following discharge from first hospitalization for COPD in BC between 1996 and 2001.  75  Table 6.1 Characteristics of individuals discharged alive after a first hospitalization for COPD in BC between 1996 and 2001.  n, % Demographics Age - mean, SD Sex - n male, %male Comorbid Conditions Charlson Index - n, % 0 - Total comorbid score = 0 1 - Total comorbid score = 1-2 2 - Total comorbid score = 3-4 3 - Total comorbid score = 5-6 ACG® Group - n, % 5070 - 4+ Major ADG®, 10+ conditions 4940 - 4 Major ADG®, 10 conditions Index Hospitalization Data Most frequently coded primary diagnosis - n, % 496 - Chronic airway obstruction, not elsewhere classified 491 - Chronic bronchitis 486 - Pneumonia Most frequently coded secondary diagnosis - n, % 496 - Chronic airway obstruction, not elsewhere classified 491 - Chronic bronchitis 492 - Emphysema Admission to intensive care unit - n, % Length of Stay (days) - median, IQR In ICU (if admitted) In hospital (total stay)  COPD without Asthma 12112 100  COPD with Asthma 2041 100  All Subjects 14153 100  73.8 7222  10.0 59.6  73.1 961  10.2 47.1  73.7 8183  10.0 57.8  4890 5911 1110 201  40.4 48.8 9.2 1.7  807 1034 176 24  39.5 50.7 8.6 1.2  5697 6945 1286 225  40.3 49.1 9.1 1.6  1396 435  11.5 3.6  250 32  12.2 1.6  1646 467  11.6 3.3  2821 1780 1386  23.3 14.7 11.4  552 366 218  27.0 17.9 10.7  3373 2146 1604  23.8 15.2 11.3  5596 888 649 695  46.2 7.3 5.4 5.7  825 177 68 118  40.4 8.7 3.3 5.8  6421 1065 717 813  45.4 7.5 5.1 5.7  3 5  1-5 3-9  2 6  1-5 3-9  3 5  1-5 3-9  Abbreviations: ACG®: Adjusted Clinical Group; ADG®: Aggregated Diagnosis Group; ICU: intensive care unit; SD: standard deviation; IQR: inter-quartile range.  76  Table 6.2 Characteristics of drug dispensation to individuals discharged alive after a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001.  Dispensations - n, % Use of ICS Never dispensed ICS At least one dispensation of ICS  COPD without Asthma 44851  COPD with Asthma 14284  All Subjects 59135  5454  45.0  159  7.8  5613  39.7  6658  55.0  1882  92.2  8540  60.3  41008 3282 561  91.4 7.3 1.3  12442 1544 298  87.1 10.8 2.1  53450 4826 859  90.4 8.2 1.5  Most frequently prescribed drugs - n fills, % fills Flovent Inhaler 7163 16.0 Pulmicort Turbuhaler 6238 13.9 Becloforte Inhaler 6132 13.7  2307 2246 2013  16.2 15.7 14.1  9470 8484 8145  16.0 14.3 13.8  Drug use prior to 1st hospitalization 638 LABA - n, % 4052 Anticholinergics - n, %  433 1276  21.2 62.5  1071 5328  7.6 37.6  ICS Drug Type - n fills, % fills ICS: MDI/dry powder ICS: nebulized Salmeterol + ICS  5.3 33.5  Abbreviations: ICS: Inhaled corticosteroids; MDI: metered-dose inhaler; LABA: long-acting beta-agonist.  77  Table 6.3 Life table describing time to inhaled corticosteroid (ICS) initiation prior to and following discharge hospitalization for COPD in British Columbia (BC) between 1996 and 2001. COPD without asthma  Days Pre-hospitalization 1 2 - 30 31 - 60 61 - 90 91 - 180 181 - 365 365 - 730 731 - 1600  n at risk 12112 6798 5751 5224 4861 4123 3152 1710 31  n initiating ICS 5314 120 386 135 71 126 183 193 130  % unexposed 56.1 55.1 51.8 50.6 49.8 48.5 46.1 42.4 35.5  COPD with asthma n at risk 2041 259 213 183 170 140 103 51 0  n initiating ICS 1782 3 32 13 4 14 12 12 0  % unexposed 12.7 12.5 11.0 10.3 10.0 9.2 8.3 7.0 0  All Subjects n at risk 14153 7057 5964 5407 5031 4263 3255 1761 31  n initiating ICS 7096 123 418 148 75 140 195 205 140  % unexposed 49.9 49.0 45.9 44.7 44.1 42.8 40.6 37.3 31.0  Abbreviations: ICS: Inhaled corticosteroids.  78  Table 6.4 Compliance with prescribed ICS dose (using cumulative multiple availability measure)128 during follow-up from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001. Median IQR  COPD without asthma 0.700 (0.430 - 0.962)  COPD with asthma 0.700 (0.474 - 0.954)  All subjects 0.700 (0.441 - 0.961)  Abbreviations: IQR: Inter-quartile range.  79  Table 6.5 Odds ratios and 95% confidence intervals (CIs) of multiple logistic regression model of ICS initiation within ninety days of an index hospitalization for COPD in British Columbia (BC) between 1996 and 2001. Intercept Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Coexisting Asthma (vs. no coexisting asthma) Prior use of LABA (vs. no prior use) Prior use of anticholinergics (vs. no prior use) Days spent in intensive care unit ACG® = 5070 or 4940 Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3  Abbreviations: ACG®: Adjusted Clinical Group  Odds Ratio 0.16 0.75  95% CI (0.11 - 0.23) (0.64 - 0.88)  1.25 1.14 0.92 0.81 1.13 2.06 0.65 1.08 1.02 0.71  (0.86 - 1.84) (0.79 - 1.62) (0.64 - 1.31) (0.54 - 1.23) (0.37 - 3.41) (1.47 - 2.89) (0.34 - 1.23) (0.88 - 1.33) (0.97 - 1.08) (0.55 - 0.92)  1.00 0.76 0.56  (0.84 - 1.18) (0.56 - 1.05) (0.24 - 1.30)  80  Table 6.6 Frequency of outcomes and censored observations during follow-up from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001. COPD without asthma Deaths nd  2 Hospitalizations Censored observations Total  COPD with asthma  n  %  n  1794  14.8  195  3075 7243 12112  25.4 59.8 100.0  704 1142 2041  %  All subjects n  %  9.6  1989  14.1  34.5 56.0 100.0  3779 8385 14153  26.7 59.2 100.0  81  Table 6.7 Rates and crude rate ratios (RRs) of death or hospitalization for exposure to inhaled corticosteroids (ICS) among individuals discharged alive from a first hospitalization in British Columbia (BC) between 1996 and 2001. Exposed  COPD without Asthma COPD with Asthma All subjects  Events  Person years  2253 736 2989  7763 2455 10218  Unexposed Event Rate (event·py-1) 0.29 0.30 0.29  Events 2616 163 2779  Person years 10270 511 10781  Event Rate (event·py-1) 0.25 0.32 0.26  RR  95% CI  1.14 0.94 1.13  1.07 - 1.21 0.79 - 1.12 1.08 - 1.19  Abbreviations: CI: Confidence Interval.  82  Table 6.8 Adjusted estimates of ICS effectiveness among individuals discharged alive from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001, using an extended Cox model with time-varying exposure. CI: Confidence Interval. Hazard Ratio Stratum-specific effectiveness of ICS Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Adjustment Variables Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG® = 5070 or 4940 (vs other) Prior use of long acting beta agonists Prior use of anticholinergics  Abbreviations: ICS: Inhaled Corticosteroids; ACG®: Adjusted Clinical Group  95% CI  0.99 0.82  (0.94 - 1.05) (0.69 - 0.99)  1.16  (1.1 - 1.22)  1.39 1.84 2.26 2.73 3.19 1.01  (1.17 - 1.65) (1.57 - 2.16) (1.93 - 2.65) (2.31 - 3.23) (2.23 - 4.56) (1.01 - 1.01)  1.17 1.38 2.06 1.55 1.04 1.63  (1.11 - 1.24) (1.26 - 1.51) (1.73 - 2.46) (1.45 - 1.65) (0.94 - 1.16) (1.54 - 1.72)  83  Table 6.9 Sensitivity analyses of adjusteda ICS effect estimates among individuals discharged alive from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001, using an extended Cox model with time-varying exposure. Cohort Definition Asthma as any prior HSU coded with 493 Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum First responsible diagnostic code of COPD only Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Exposure Compliance >75% Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Compliance >50% Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Compliance >25% Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Compliance >10% Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum No time varying exposure to first fill Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Time-fixed exposure Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Outcomes Cause-specific death or hospitalization Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Hospitalization (death censored) Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum All cause mortality Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum Cause-specific mortality Exposure to ICS - COPD only stratum Exposure to ICS * COPD plus asthma stratum  Hazard Ratio  95% CI  1.06 0.77  (0.98 - 1.14) (0.68 - 0.86)  0.92 0.92 Hazard Ratio  (0.84 – 1.00) (0.70 - 1.21) 95% CI  1.13 1.07  (1.05 - 1.22) (0.91 - 1.26)  1.09 1.17  (1.02 - 1.17) (1.01 - 1.36)  1.08 1.05  (1.02 - 1.15) (0.90 - 1.23)  1.04 0.93  (0.98 - 1.10) (0.78 - 1.10)  0.80 0.85  (0.75 - 0.85) (0.71 - 1.02)  0.88 0.97 Hazard Ratio  (0.83 - 0.93) (0.75 - 1.25) 95% CI  1.47 0.86  (1.37 - 1.58) (0.68 - 1.09)  1.64 0.92  (1.52 - 1.77) (0.71 - 1.19)  0.42 0.59  (0.39 - 0.46) (0.48 - 0.74)  0.59 0.44  (0.51 - 0.69) (0.30 - 0.64)  a Adjusted for same variables as main model: age, sex, length of stay, comorbidity index, ACG group, prior use of long acting beta agonists, prior use of anticholinergics Abbreviations: CI: Confidence Interval; ICS: Inhaled Corticosteroids  84  7 Discussion In this study of individuals with COPD discharged from a first hospitalization in BC between 1996 and 2001, we observed a difference in the effectiveness of ICS among individuals with COPD alone and individuals with concomitant COPD and asthma. The hazard ratio of death or hospitalization was 18% lower in the subset of individuals with concomitant COPD and asthma compared with the subset of individuals with COPD alone (HR = 0.82, 95% CI: 0.69 – 0.99). We did not detect a reduction in hazard of death or hospitalization associated with exposure to ICS in the COPD-only stratum (HR = 0.99; 95% CI: 0.94-1.05); however, the hazard ratio in the stratum of individuals with COPD and asthma suggested that there was a beneficial effect of ICS in this subgroup. Because co-existing asthma and COPD is characterized by partially reversible airflow, this may suggest that post-bronchodilator reversibility may be an important factor in identifying individuals with COPD who respond well to ICS.  7.1  Contextualization of results Prior to this study, no other published studies have estimated the comparative effectiveness  of ICS in a population of individuals with COPD alone and individuals with co-existing asthma and COPD. Two other observational studies explored the impact of a prior history of asthma;110,113 however, neither explicitly measured the difference in effectiveness between the two groups, and both studies used a broad definition of co-existing asthma, likely including individuals with “pure” asthma, rather than individuals with co-existing asthma and COPD. Although not explicitly quantified, the results of both studies suggested greater ICS effectiveness among individuals afflicted with COPD who had a prior history of asthma compared those with no history, which is consistent with the findings of the current study. There are no experimental studies to which the results can be compared, as there are no studies of ICS efficacy among individuals with concomitant asthma and COPD, nor are there subgroup analyses addressing this issue, because individuals with history of asthma and 85  individuals with bronchodilator reversibility of more than 10% of baseline FEV1 are typically excluded from RCTs (discussed in section 2.8.6). Patients with COPD are thought to have impaired HDAC2 function, consequently reducing the effectiveness of ICS (section 2.8.1).78 Therefore, in theory, the comparatively greater effectiveness of ICS on the asthmatic component of the disease could be overridden by the impaired function of HDAC2. However, the results of this study suggest that this may not be the case, since ICS maintains its improved effectiveness in asthma compared with COPD, even when both diseases coexist. Given the macro-level of our findings, it is injudicious to draw conclusions regarding the biological mechanisms that may or may not explain the effect modification we observed in our data.182 The previously suggested mechanism should be investigated further. Causal relationships become more credible with the support of additional evidence and logical reasoning.183 Additional studies should be conducted to strengthen the body of evidence supporting the causal relationship between ICS and reduced rates of death and hospitalization in asthmatics with COPD.  7.2  Choice of modeling approach We implemented an extended Cox model with a time-varying covariate representing use of  ICS. Cox models are a superior method for representing risk using censored longitudinal data,147 and their extension to allow for time-varying covariates provides an analytic tool for reducing immortal time bias in pharmacoepidemiological studies.115 Because exposure changes over time, it means that the hazard ratio between two groups of individuals changes over follow-up time. As a result, the hazard ratio associated with ICS exposure applies only to the hazard ratio at a specified time, and the hazard ratio of an exposed group of individuals can only be assessed at specified times, compared with an unexposed group of individuals with equal baseline covariates. Therefore, these results cannot be used to predict the estimated time to event among individuals with known covariates.184 86  This implies that the results from this study must be interpreted carefully: in the subpopulation of asthmatics with COPD, the hazard of death at a specified time is 19% lower among those exposed to ICS at that time compared with those who are not, assuming all baseline characteristics are equal; and in the subpopulation of individuals with COPD alone, the hazard of death is not different between exposure categories of ICS, at a given time, t (HR = 0.99; 95% CI: 0.94-1.05). The major assumption in using a model with time-varying exposure is that the change in exposure occurs in a random fashion. This has been a criticism of other studies using time varying covariates in the estimation of ICS effectiveness in COPD.185 Because frequent exacerbations are currently an indication for ICS initiation,11 we anticipated that by restricting cohort entry to individuals at the time of discharge from their first hospitalization for COPD, ensuing changes in exposure status would be related to this common event, and not to a difference in underlying risk of death or re-hospitalization. Only 10.2% (1444) of subjects initiated ICS following hospitalization, with just 4.8% (680) initiating more than 90 days postdischarge (Table 6.3). Thus, the impetus for ICS initiation in these cases was likely the index hospitalization. Therefore, the use of a time-varying exposure is unlikely to have biased our results any more than a time-fixed exposure, and the time-varying exposure had the additional advantage of reducing immortal time bias, and incorporating discontinued use of ICS. The choice to use this model over other modeling approaches discussed in section 3.2 was in part because of the limitations of administrative databases in pharmacoepidemiological research. The instrumental variable approach may have been more effective in reducing any unknown confounding; however, within the available data, we were unable to identify an instrument that was related to the exposure but not to the outcome. We were also unable to implement a marginal structural model due to the unavailability of a data source that captures one or more intermediate variables. Had an intermediate variable been available, such as  87  FEV1, it could have been incorporated into the model to predict subsequent exposure, and may have reduced bias. The choice not to use propensity scores approach was based on the fact that it does not significantly improve the accuracy of effect estimates when data is limited nor does it control for unmeasured confounders.155 We did identify predictors of ICS use using a logistic regression model; however, due to the large number of individuals who filled their first prescription prior to the index hospitalization, the model only predicted use for 8.0% of individuals who filled one or more ICS prescriptions (Table 6.3).  7.3  Internal validity Bias and confounding exist in all epidemiological studies. Confounding is particularly a  concern in observational studies of effectiveness because individuals are not randomized to treatment groups and thus predictors of outcome may not be equally distributed across categories of exposure.123 While some types of bias are avoidable through careful study design and analysis, many types are not. The direction and magnitude of the biases introduced in section 3.1 are discussed in this section in order to inform inferences regarding the results of the study.123 7.3.1  Selection Biases  As administrative databases were used to form the study cohort, this study is mainly free of types of selection bias relating to subject recruitment; however, informative right-censoring of individuals may have contributed some amount of bias. Because death and hospitalization are both events that are associated with COPD disease severity and are competing risks, we used a composite end point of death and hospitalization. This allowed for the measurement of hospitalizations as an outcome, while avoiding informative right censoring of deaths. Censoring deaths only induces bias if the exposed and unexposed groups have different underlying mortality rates, yielding an over-estimate of event-free survival in the group with the 88  higher death rate. In the sensitivity analysis, the association between ICS and hazard of hospitalization alone in the COPD-only stratum was 1.47 (95% CI: 1.37-1.58) (Table 6.9), suggestive of a harmful effect of ICS on mortality. Further analysis of the differential censoring between exposure groups confirmed that there were proportionally more deaths among the unexposed (18.3%) than among the exposed (8.2%). This resulted in more ‘hospitalizationfree’ person years in the un-exposed category, because individuals died before being hospitalized, and artificially inflated the hazard ratio. We also censored individuals who entered long-term care facilities. This was a reasonable decision because these hospital admissions are typically pre-scheduled, do not represent an acute exacerbation, and because medication is supplied by the long-term care facility and data for these medications cease to be recorded in the PharmaNet database. However, the individuals who required long-term care and surgery likely represented a more advanced stage of disease and higher risk of death. As with the censorship of death, bias would be incurred if censorship occurred differentially between the exposed and unexposed groups. In this case, there were more censored events due to level-of-care in the exposed compared with the unexposed (3.9% in the exposed, and 1.5% in the unexposed). Assuming that individuals who were censored were at higher risk for events, our effect size would be biased downward, in favour of ICS. This difference was more marked in the asthma subpopulation (1.2% unexposed censored due to long-term care, versus 4.2% exposed censored). Therefore, the effect of differential censorship may account in some part for the lower effect size measured among those individuals with asthma; however because of the small number of censored observations due to surgery and long-term care, the change in effect size was small. We investigated the effect of changing these censored observations to ‘events’ and the point estimates of the hazard ratios changed by 0.01 units. Therefore, although theoretically the hazard ratios are biased downwards due to differential censoring of long-term care, the magnitude of the bias is very small, and has little effect on the estimates or conclusions that be drawn from the results. 89  7.3.2  Information Biases  Several potential sources of information bias were discussed in section 3.1.2: exposure misclassification (with special consideration of immortal time bias), diagnostic misclassification, and outcome misclassification. This section addresses the extent to which these were avoided in this study, and the impact of the biases on the effect estimates. 7.3.2.1 Immortal time bias Given the concern about immortal time bias in studies of ICS effectiveness in COPD,129 we were careful to design this study in such a way as to avoid it. By defining cohort entry at the time of discharge from hospital, rather than according to a hierarchical drug utilization benchmark, we avoided excluded immortal time.107 To avoid misclassified immortal time, we used a time-varying exposure, with ‘days’ as the unit of time.106 We conducted an analysis to investigate the impact of using a time-fixed exposure variable rather than time-varying exposure. Suissa106 originally demonstrated this difference, using the work of Sin and Tu103 as an exemplar. Suissa found that eliminating the misclassified immortal time changed the point estimate of ICS effectiveness from 0.69 in the analysis subject to immortal time bias to 1.00 in the analysis free of immortal time. Using the data from the current study, addition of immortal time (because the current study was designed to exclude it) changed the estimate of ICS effectiveness in the COPD-only stratum from 0.99 in the analysis free of immortal time to 0.80 in the analysis including it. We suspect that the magnitude of the change was not as large as that demonstrated by Suissa because his study enrolled individuals who were ICS-naïve upon discharge from hospital, and thus all individuals filling a prescription for ICS accumulated immortal time, whereas in the current study, approximately 50% of individuals who were categorized as exposed at any time had filled an ICS prescription prior to hospitalization, and thus did not accumulate immortal time. The second step in investigating the effect of using a time-fixed exposure variable involved categorizing individuals as ‘exposed’ between the time of first ICS fill and the end of follow-up, despite ICS discontinuation. Although 90  this does not represent immortal time bias, we will discuss it in this section as it relates to the use of time-fixed exposure. After changing exposure from ‘unexposed’ to ‘exposed’ for all time periods following an ICS fill, the point estimate of ICS effectiveness in the COPD-only stratum increased from 0.80 to 0.88, which was lower than the initial analysis, yet higher than in the first step. The anticipated direction of change of estimate depends upon the association between ICS discontinuation and outcome. Because there was a decrease in the number of unexposed person years, if the number of events were to remain the same, the event rate in the unexposed would increase, suggesting a larger benefit of ICS. However, the proportion of events among those who discontinued was higher than among those who did not, leading to an increase in the effect estimate when individuals were analyzed according to an intent to treat analysis. 7.3.2.2 Exposure Misclassification Proper exposure classification using administrative data is challenging. In this study, exposure was assumed based on filled prescriptions for ICS; however, the actual consumption of the medication was never verified due to the administrative nature of the data. The use of the CMA metric accounted for hoarding because unused doses carry forward from one refill interval to the next,128 preventing misclassification of individuals as unexposed during a perceived lapse in prescription fill. It is more likely that individuals were misclassified as exposed when they were truly unexposed than vice versa. Twenty-five percent of ICS users were less than 44.1% compliant (cumulative compliance) at the time of their last fill (Table 6.4), meaning that they were either taking less than half of the recommended daily dose throughout, or used the full recommended daily dose during periods of use, but on more than half of the days, they took no medication at all. The effect of this misclassification on the effect estimate is complicated, as it depends on when and how the misclassification occurred. If the level of compliance increases over time, with individuals becoming more compliant near the end of the study period, then these event-free periods of truly un-exposed person-time would inflate the 91  person-time denominator of the exposed category, leading to an underestimate of the hazard ratio (suggesting a protective effect of ICS). However, it is more likely that misclassified time occurred immediately prior to an event, causing events to be mistakenly assigned to the exposed category, thus inflating the event rate in the exposed category and overestimating the hazard ratio (towards a harmful effect of ICS). Without further analysis of the temporal trends in compliance, it is difficult to conclude the effect of exposure misclassification on the results. The sensitivity analyses elucidated some additional aspects regarding the impact of exposure misclassification. When classification to the ‘exposed’ category was restricted to individuals with a compliance level greater than or equal to 75%, the hazard ratio was higher in both strata than in the original analysis (where exposure was defined as compliance greater than 0%) (Table 6.8). The hazard ratios became smaller as the compliance threshold used to define exposure was lowered. A similar trend was identified previously using increasing levels of dose to define exposure.108 If we assume that compliance and dose are associated with risk of outcome (i.e. higher compliance among individuals with worse health), then these results are rational: periods of lower compliance were event-free, and the period of higher compliance were more strongly associated with events, thus inflating the hazard ratio. 7.3.2.3 Diagnostic Misclassification Cohort entry was defined as the time of discharge from a first hospitalization for COPD, as captured by the first or second responsible diagnosis in the hospital discharge record. This choice was made because hospitalizations for COPD are not always captured in the first diagnostic field. This choice of including individuals with a first or second responsible diagnosis of COPD may have included resulted in the inclusion of individuals who were hospitalized for reasons unrelated to COPD. The top first- and second- responsible diagnoses for the index admissions are listed in Table 1 of Appendix 1. In the primary diagnostic field, most of the primary diagnostic codes were pulmonary in nature, however, cardiac diseases (heart failure, cardiac dysrhythmias, and acute myocardial infarction) ranked among the top ten (accounting 92  for 9.5% of the subjects). A sensitivity analysis including only first hospitalizations for which the primary diagnostic code was for COPD yielded an estimate of 0.92 (95% CI: 0.84-1.00) in the COPD-alone stratum (Table 6.9). The asthmatic subgroup was not found to be statistically different from the COPD group; however, the point estimate of effectiveness in the asthmatic group (0.92*0.92=0.846) is in the same direction and of a similar magnitude as in the original analysis (Table 6.8). The lower hazard ratio found in the COPD-alone stratum suggests that differential diagnostic misclassification may have occurred in the original analysis, leading to an overestimate of the hazard ratio. This is corroborated by the fact that there were proportionally fewer events (39.0%) and more unexposed individuals (63.1%) in the group of individuals with COPD as a second diagnostic code than in those with COPD as the first (43.0% and 50.0% respectively). Diagnostic misclassification may also have occurred with respect to the definition of asthma. In this study, individuals were categorized as having coexisting asthma if there was any record of a specialist visit or hospitalization for asthma prior to the index hospitalization. This definition resulted in 14.4% of the study sample to be considered to have co-existing asthma. This is consistent with the estimates that 10%-20% of the COPD population has coexisting asthma, but differs from the study by Vollmer et al113, where 42.5% of the study sample was categorized as having co-existing asthma, according to any prior health care utilization for with asthma was listed as the diagnostic code. These authors acknowledged that this was likely to include individuals who did not have truly co-existing disease, but rather individuals whose diagnosis of asthma was re-assessed and found to be truly COPD, and vice versa. The more stringent criteria requiring a specialist visit of hospital encounter for asthma categorization likely produced a more representative sample of individuals with who were truly afflicted with both diseases. In a sensitivity analysis, the definition of asthma was broadened to match the definition used by Vollmer et al. Using this definition, 39.2% of the COPD population was categorized as having asthma. The point estimate of hazard of death or re-hospitalization 93  associated with ICS use in the coexisting COPD and asthma stratum was similar to that in the original analysis: (HR = 1.06*0.77 = 0.82) (Table 6.9). The point estimate in the COPD-alone stratum increased when potential asthmatics were re-categorized to the asthma-plus COPD stratum. This finding suggests that any diagnostic misclassification in the study made with respect to categorization of asthma would not change the point estimate of effectiveness in the concomitant asthma plus COPD stratum, nor would it change the conclusion that ICS has greater effectiveness in this stratum, compared with the COPD alone stratum; however, it does suggest that the estimate in the COPD-only stratum would be slightly higher. 7.3.2.4 Outcome misclassification Sensitivity analyses of the choice of outcome measures investigated the impact of changing the end-point from all-cause mortality or re-hospitalization to permutations of all-cause mortality, pulmonary-specific mortality and re-hospitalization (Table 6.9). While all-cause mortality is not subject to misclassification, the cause of death and cause of hospitalization may have been misclassified differentially between exposure categories. Changing the outcome to pulmonary-specific mortality or re-hospitalization resulted in a hazard ratio of 1.47 (95% CI: 1.37-1.58), which suggests a harmful effect of ICS, and is higher than our original estimate of 0.99 in the COPD-only stratum. In the asthma subpopulation, the estimate also suggests a harmful effect (HR = 1.47 * 0.86 = 1.26). A higher hazard ratio associated with cause-specific death compared with all-cause mortality was also detected in the sensitivity analysis using death alone as an outcome (ignoring hospitalizations) (Table 6.9). This inflated risk that resulted from changing all-cause mortality to cause-specific mortality may be explained by a proportionally higher frequency of cause-specific deaths among the exposed compared with the unexposed. 39% of the deaths in the exposed category were respiratory-related, whereas only 24% of the unexposed category died due to respiratory-related causes. This difference in proportional cause-specific deaths may be due to poor sensitivity in recording respiratory-related deaths,186-188 and may 94  occur differentially if individuals who are taking COPD medication at the time of death are more likely to have their cause of death attributed to COPD than non-medicated individuals. If this difference in cause-specific morality is indeed due to misclassification, the choice of using an all-cause mortality end-point rather than cause-specific mortality in measuring effectiveness of ICS in COPD is justified. 7.3.3  Confounding  This study employed techniques to identify confounders, and to reduce the degree of confounding by severity, protopathic bias (and reverse protopathic bias). To be considered a confounder, a variable must be associated with the drug exposure and the outcome of interest, without being in the causal pathway between the drug exposure and the outcome.122 Using variables that we suspected, a priori, to be associated with ICS exposure, we measured the association between variables and exposure to ICS in our logistic regression analysis. The odds ratios presented in Table 6.5 suggest that sex, history of coexisting asthma, and presence of severe comorbid conditions (ACG®) were associated with exposure to ICS in the ninety days following discharge from hospitalization. The direction of these associations were consistent with the anticipated directions based on the literature, presented in Table 3.1. Although we did not detect significant associations in our analysis between ICS initiation and age, prior use of LABA, prior use of anticholinergics, or Charlson comorbidity index, data from the literature suggested that they are associated, and furthermore, these factors have been associated with death and hospitalizations in previous studies (Table 3.1). Therefore, we considered these variables to be potential confounders of the association between ICS exposure and outcome, and adjusted for them in the Cox models. Disease severity was not explicitly captured in the administrative health databases. However, to reduce the amount of bias imposed from unbalanced severity levels across categories, we restricted the definition of cohort entry to individuals who had been admitted to hospital exactly one time with the responsible diagnostic code of COPD, since severity is 95  associated with frequency of acute exacerbations and hospitalizations.25 We assumed that the change in severity over the median 14.4 of months of follow-up from the index hospitalization was similar across individuals, and therefore that the severity would be similar across individuals at any time during follow-up, except for periods of symptom exacerbation experienced by individuals with COPD. Confounding by severity was likely minimized as a result of the study time frame. Between 1992 and 2003 there were no formal guidelines on when ICS should be administered with respect to disease severity.84 The recommendation that ICS be restricted to those with moderate-to-severe COPD who experience frequent exacerbations was first published in the 2003 CTS guidelines;20 prior to this, guidelines were unclear regarding ICS use in COPD management. The dataset used in this study captured data between 1996 and 2001, in the absence of recommendations to prescribe ICS to individuals with advanced disease, suggesting that ICS may not have been prescribed according to disease severity to the extent that it would be following publication of the guidelines. A study of Belgian physicians conducted in 2002 supports the conjecture that ICS was not being prescribed exclusively to individuals with advanced disease: 59.4% of general practitioners initiated ICS in all patients presenting with COPD, rather than initiating according to severity (32.8% and 7.3% initiated ICS “sometimes” or “in specific cases”, respectively, while 0.5% reported that they never initiated ICS).189 As discussed in section 3.1.3, because ICS are not intended for use as relief medication, it is unlikely that protopathic bias affected the results (resulting in an overestimate of risk among the exposed). Of the individuals who were ever exposed to ICS, 89% had filled their first ICS prescription either prior to, or within one month of discharge from hospital (Table 6.3). Therefore, the protopathic bias would likely only affect the remaining 11% of ICS initiators. Other observational studies have not found strong indication of protopathic bias in ICS use.108  96  On the contrary, reverse protopathic bias may have affected our results. The exposure field was coded such that individuals may ‘discontinue’ ICS use. This approach differs from other studies of ICS effectiveness, but is consistent with patterns of ICS use described in the literature, where between 47.6% and 67.0% of COPD patients were found to discontinue ICS use within 1 year.190 It is known that patients experience withdrawal effects from systemic steroids, and some evidence that this is the case with ICS.191 In studies using administrative data, it is impossible to determine the direction of causation in a scenario where an individual discontinued ICS shortly before an event: the individual may have discontinued ICS because of his or her deterioration, or the individuals’ health may have deteriorated because of ICS discontinuation. For this reason, we did not take any measures to correct for potential ‘reverse protopathic bias’ as we may be covering up important effects of ICS discontinuation. It is likely that there are additional unknown factors that are associated with both the exposure and the outcome and that are not in the causal pathway; however, the known confounders were controlled for to the extent possible using administrative data, using methods of restriction in the design phase and adjustment in the analysis phase.  7.4  External validity The sample population was all individuals aged greater than 45 years in BC registered with  MSP who experienced a first hospitalization for COPD between 1996 and 2001, categorized by prior history of asthma, and by use of ICS (beclomethasone dipropionate, budesonide, fluticasone propionate, or triamcinolone). The target population, to whom the results are generalizable, are older individuals who have moderate COPD and have experienced at least one acute exacerbation. The mean age of the study sample was 73.7 years (SD 10.0 years), and there were proportionally more men than women. Because COPD currently tends to be under-diagnosed among women,168 in coming years, the target population may comprise more women. Because  97  the results were adjusted for sex, and there is sufficient representation of women in the data set, this should not affect their generalizability. The operationalized definition of coexisting asthma and COPD used in this study is thought to represent the true population of individuals afflicted with both diseases because of the requirement that the physician visit be for a specialist. Therefore, we expect that these results are generalizable to the population of individuals who truly have both diseases, rather than to a population of individuals with COPD who were mis-diagnosed with asthma, or vice versa.  7.5  Future work The current study estimated effectiveness of ICS using data between 1996 and 2001  among individuals with COPD alone and individuals with concomitant COPD and asthma. A complement to the current study would be to include a third population: individuals with asthma alone. This would allow comparison of effectiveness estimates in the group with concomitant disease with the effectiveness estimates in groups with each disease on its own. This would allow for comparison to see if individuals with concomitant asthma and COPD respond to ICS in a way that is more “asthma-like” or “COPD-like”. Additional studies could be done to identify other sub-populations that may respond well to ICS, based on clinical measures such as FEV1 reversibility and airway hyperresponsiveness, other exposures, such as smoking status, race, or presence of other comorbid conditions. These studies would contribute to a greater understanding of ICS mechanisms in COPD, and would provide a targeted treatment strategy for practitioners. The drugs used during the study period were primarily delivered as monotherapy. Only one brand of drug existed that delivered a combined dose of ICS and LABA together, and it represented only 1.5% (885) of all ICS dispensations. The combined ICS/LABA therapy is currently the recommended delivery of ICS for the treatment of individuals with COPD,11 Future studies could be conducted comparing ICS/LABA combination therapy with ICS monotherapy. 98  Germaine to the study of ICS in COPD is the issue of adverse events. Recent publications suggest that there is an increased risk of pneumonia among individuals using ICS.94 Investigation of this issue was outside the scope of the current study, however, an observational study design could lend itself well to answering the question of whether ICS is associated with a greater risk of pneumonia.  7.6  Concluding statements Prior to this study, the effectiveness of ICS had not been studied in a population of  individuals with concomitant COPD and asthma. Currently, very little is understood regarding the diagnosis, management and treatment effectiveness in this subpopulation. Our findings suggest that ICS is more effective in this population than among individuals with COPD alone, and that use of ICS significantly reduces the hazard of death and hospitalization associated with the combined presence of asthma and COPD. Given the uncertain effectiveness of ICS in COPD, this finding suggests that there is at least one subpopulation of individuals afflicted with COPD that stands to benefit from ICS. Further research is needed to identify additional subpopulations that may benefit from treatment in the highly heterogeneous population of individuals afflicted with COPD. This will allow for targeted treatment of the disease, allowing for improved outcomes and a potential reduction of the considerable burden of illness attributable to death and hospitalization.  99  References 1.  Platt H. Chronic Disease Management: A snapshot of COPD Care in British Columbia 2003/2004. British Columbia Ministry of Health, 2005.  2.  Murray CJ, Lopez AD. Alternative projections of mortality and disability by cause 19902020: Global Burden of Disease Study. Lancet 1997;349(9064):1498-504.  3.  Soriano JB, Davis KJ, Coleman B, Visick G, Mannino D, Pride NB. The proportional Venn diagram of obstructive lung disease: two approximations from the United States and the United Kingdom. Chest 2003;124(2):474-81.  4.  Du D, Akazawa MO, Blanchette CM, Wang J, Mapel DW, Dalal A, Scharf SM. Burden of Concomitant Asthma and COPD in a Medicaid Population. Chest 2008.  5.  Lemiere C, Bai T, Balter M, Bayliff C, Becker A, Boulet LP, Bowie D, Cartier A, Cave A, Chapman K, Cowie R, Coyle S, Cockcroft D, Ducharme FM, Ernst P, Finlayson S, FitzGerald JM, Hargreave FE, Hogg D, Kaplan A, Kim H, Kelm C, O'Byrne P, Sears M, Markham AW. Adult Asthma Consensus Guidelines Update 2003. Can Respir J 2004;11(Suppl A):9A-18A.  6.  Barnes PJ. Inhaled corticosteroids are not beneficial in chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2000;161(2 Pt 1):342-4; discussioin 344.  7.  Calverley PM. Inhaled corticosteroids are beneficial in chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2000;161(2 Pt 1):341-2; discussion 344.  8.  Buist AS, Anzueto A, Calverley PM, deGuia TS, Fukuchi Y, Jenkins C, Khaltaev N, Kiley JP, Kocabas A, Lopez MV, Nizankowska-Mogilnicka E, Rabe KF, Roisin RR, van der Molen T, van Weel C, al. e. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease. Global Initiative for Chronic Obstructive Lung Disease (GOLD), 2006.  9.  Caramori G, Adcock I. Pharmacology of airway inflammation in asthma and COPD. Pulm Pharmacol Ther 2003;16(5):247-77.  10.  McIvor A, Lowry J, Bourbeau J, Borycki E. Chapter 2: Assessment of COPD. In: Bourbeau J, Nault D, Borycki E, eds. Comprehensive Management of Chronic Obstructive Pulmonary Disease. Hamilton, Ontario: BC Decker Inc, 2002;19-31.  11.  O'Donnell DE, Aaron S, Bourbeau J, Hernandez P, Marciniuk DD, Balter M, Ford G, Gervais A, Goldstein R, Hodder R, Kaplan A, Keenan S, Lacasse Y, Maltais F, Road J, Rocker G, Sin D, Sinuff T, Voduc N. Canadian Thoracic Society recommendations for management of chronic obstructive pulmonary disease - 2007 update. Can Respir J 2007;14 Suppl B:5B-32B.  12.  Guidelines and Protocols Advisory Committee: Chronic Obstuctive Pulmonary Disease (COPD). British Columbia Medical Association, 2005.  13.  Burrows B, R.H. E. Course and prognosis of chronic obstructive lung disease. A prospective study of 200 patients. N Engl J Med 1969;280:397-404. 100  14.  Fletcher CM, Peto R, Tinker CM, Speizer FE. The natural history of chronic bronchitis and emphysema: an eight year study of early chronic obstructive lung disease in working men in London. Toronto: Oxford UP, 1976.  15.  Anthonisen NR, Connett JE, Kiley JP, Altose MD, Bailey WC, Buist AS, Conway WA, Jr., Enright PL, Kanner RE, O'Hara P, et al. Effects of smoking intervention and the use of an inhaled anticholinergic bronchodilator on the rate of decline of FEV1. The Lung Health Study. Jama 1994;272(19):1497-505.  16.  Effect of inhaled triamcinolone on the decline in pulmonary function in chronic obstructive pulmonary disease. N Engl J Med 2000;343(26):1902-9.  17.  Burge PS, Calverley PM, Jones PW, Spencer S, Anderson JA, Maslen TK. Randomised, double blind, placebo controlled study of fluticasone propionate in patients with moderate to severe chronic obstructive pulmonary disease: the ISOLDE trial. Bmj 2000;320(7245):1297-303.  18.  Vestbo J, Sorensen T, Lange P, Brix A, Torre P, Viskum K. Long-term effect of inhaled budesonide in mild and moderate chronic obstructive pulmonary disease: a randomised controlled trial. Lancet 1999;353(9167):1819-23.  19.  Pauwels RA, Lofdahl CG, Laitinen LA, Schouten JP, Postma DS, Pride NB, Ohlsson SV. Long-term treatment with inhaled budesonide in persons with mild chronic obstructive pulmonary disease who continue smoking. European Respiratory Society Study on Chronic Obstructive Pulmonary Disease. N Engl J Med 1999;340(25):1948-53.  20.  O'Donnell DE, Aaron S, Bourbeau J, Hernandez P, Marciniuk D, Balter M, Ford G, Gervais A, Goldstein R, Hodder R, Maltais F, Road J. Canadian Thoracic Society recommendations for management of chronic obstructive pulmonary disease--2003. Can Respir J 2003;10 Suppl A:11A-65A.  21.  Kanner RE, Anthonisen NR, Connett JE. Lower respiratory illnesses promote FEV(1) decline in current smokers but not ex-smokers with mild chronic obstructive pulmonary disease: results from the lung health study. Am J Respir Crit Care Med 2001;164(3):358-64.  22.  Donaldson GC, Seemungal TA, Bhowmik A, Wedzicha JA. Relationship between exacerbation frequency and lung function decline in chronic obstructive pulmonary disease. Thorax 2002;57(10):847-52.  23.  Spencer S, Jones PW. Time course of recovery of health status following an infective exacerbation of chronic bronchitis. Thorax 2003;58(7):589-93.  24.  Seemungal TA, Donaldson GC, Bhowmik A, Jeffries DJ, Wedzicha JA. Time course and recovery of exacerbations in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2000;161(5):1608-13.  25.  Donaldson GC, Seemungal TA, Patel IS, Lloyd-Owen SJ, Wilkinson TM, Wedzicha JA. Longitudinal changes in the nature, severity and frequency of COPD exacerbations. Eur Respir J 2003;22(6):931-6.  101  26.  Hattotuwa KL, Gizycki MJ, Ansari TW, Jeffery PK, Barnes NC. The effects of inhaled fluticasone on airway inflammation in chronic obstructive pulmonary disease: a doubleblind, placebo-controlled biopsy study. Am J Respir Crit Care Med 2002;165(12):15926.  27.  Jeffery PK. Comparison of the structural and inflammatory features of COPD and asthma. Giles F. Filley Lecture. Chest 2000;117(5 Suppl 1):251S-60S.  28.  Siafakas N, Tzortzaki E, Bouros D. Why do only some smokers develop COPD? In: Pearson M, Wedzicha W, eds. Chronic Obstructive Pulmonary Disease - critical debates. Malden, Ma: Blackwell, 2003.  29.  Gan WQ, Man SF, Senthilselvan A, Sin DD. Association between chronic obstructive pulmonary disease and systemic inflammation: a systematic review and a metaanalysis. Thorax 2004;59(7):574-80.  30.  Fabbri LM, Rabe KF. From COPD to chronic systemic inflammatory syndrome? Lancet 2007;370(9589):797-9.  31.  Agusti A, Soriano JB. COPD as a Systemic Disease. Copd 2008;5(2):133-8.  32.  Orie N, Sluiter H, De Vries K, Tammeling G, Wiktop J. The host factor in bronchitis. In: Orie N, Sluiter H, eds. Bronchitis. Assen, The Netherlands: Royal Van Gorcum, 1961;43-59.  33.  Bleecker ER. Similarities and differences in asthma and COPD. The Dutch hypothesis. Chest 2004;126(2 Suppl):93S-95S; discussion 159S-161S.  34.  Barnes PJ. Against the Dutch hypothesis: asthma and chronic obstructive pulmonary disease are distinct diseases. Am J Respir Crit Care Med 2006;174(3):240-3; discussion 243-4.  35.  Hansen EF, Vestbo J. Bronchodilator reversibility in COPD: the roguish but harmless little brother of airway hyperresponsiveness? Eur Respir J 2005;26(1):6-7.  36.  Buist AS. Similarities and differences between asthma and chronic obstructive pulmonary disease: treatment and early outcomes. Eur Respir J Suppl 2003;39:30s35s.  37.  Barnes PJ. Mechanisms in COPD: differences from asthma. Chest 2000;117(2 Suppl):10S-4S.  38.  Silva GE, Sherrill DL, Guerra S, Barbee RA. Asthma as a risk factor for COPD in a longitudinal study. Chest 2004;126(1):59-65.  39.  Magnussen H, Bugnas B, van Noord J, Schmidt P, Gerken F, Kesten S. Improvements with tiotropium in COPD patients with concomitant asthma. Respir Med 2008;102(1):506.  40.  Halbert RJ, Natoli JL, Gano A, Badamgarav E, Buist AS, Mannino DM. Global burden of COPD: systematic review and meta-analysis. Eur Respir J 2006;28(3):523-32.  102  41.  Lacasse Y, Montori VM, Lanthier C, Maltis F. The validity of diagnosing chronic obstructive pulmonary disease from a large administrative database. Can Respir J 2005;12(5):251-6.  42.  Mannino DM, Gagnon RC, Petty TL, Lydick E. Obstructive lung disease and low lung function in adults in the United States: data from the National Health and Nutrition Examination Survey, 1988-1994. Arch Intern Med 2000;160(11):1683-9.  43.  Screening for chronic obstructive pulmonary disease using spirometry: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med 2008;148(7):529-34.  44.  Buist AS, Vollmer WM, Sullivan SD, Weiss KB, Lee TA, Menezes AM, Crapo RO, Jensen RL, Burney PG. The Burden of Obstructive Lung Disease Initiative (BOLD): rationale and design. Copd 2005;2(2):277-83.  45.  Viegi G, Pedreschi M, Pistelli F, Di Pede F, Baldacci S, Carrozzi L, Giuntini C. Prevalence of airways obstruction in a general population: European Respiratory Society vs American Thoracic Society definition. Chest 2000;117(5 Suppl 2):339S-45S.  46.  Celli BR, MacNee W. Standards for the diagnosis and treatment of patients with COPD: a summary of the ATS/ERS position paper. Eur Respir J 2004;23(6):932-46.  47.  Lindberg A, Jonsson AC, Ronmark E, Lundgren R, Larsson LG, Lundback B. Prevalence of chronic obstructive pulmonary disease according to BTS, ERS, GOLD and ATS criteria in relation to doctor's diagnosis, symptoms, age, gender, and smoking habits. Respiration 2005;72(5):471-9.  48.  Chapman KR, Mannino DM, Soriano JB, Vermeire PA, Buist AS, Thun MJ, Connell C, Jemal A, Lee TA, Miravitlles M, Aldington S, Beasley R. Epidemiology and costs of chronic obstructive pulmonary disease. Eur Respir J 2006;27(1):188-207.  49.  Halpin DM, Miravitlles M. Chronic obstructive pulmonary disease: the disease and its burden to society. Proc Am Thorac Soc 2006;3(7):619-23.  50.  Bourbeau J, Siok MA, Nault D, Borycki E. Chapter 1: Optimizing Medical Care Delivery. In: Bourbeau J, Nault D, Borycki E, eds. Comprehensive Management of Chronic Obstructive Pulmonary Disease. Hamilton, Ontario: BC Decker Inc, 2002;1-17.  51.  Rothman KJ, Greenland S. Causation and Causal Inference. In: Rothman KJ, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: Lippincott-Raven, 1998.  52.  The Health Benefits of Smoking Cessation: A Report of the Surgeon General. Washington, DC: U.S. Department of Health and Human Services, 1990.  53.  Anto JM, Vermeire P, Vestbo J, Sunyer J. Epidemiology of chronic obstructive pulmonary disease. Eur Respir J 2001;17(5):982-94.  54.  Petty TL. The History of COPD. International Journal of COPD 2006;1(1):3-14.  55.  Watson L, Vestbo J, Postma DS, Decramer M, Rennard S, Kiri VA, Vermeire PA, Soriano JB. Gender differences in the management and experience of Chronic Obstructive Pulmonary Disease. Respir Med 2004;98(12):1207-13. 103  56.  Watson L, Vonk JM, Lofdahl CG, Pride NB, Pauwels RA, Laitinen LA, Schouten JP, Postma DS. Predictors of lung function and its decline in mild to moderate COPD in association with gender: results from the Euroscop study. Respir Med 2006;100(4):74653.  57.  Vestbo J, Soriano JB, Anderson JA, Calverley P, Pauwels R, Jones P. Gender does not influence the response to the combination of salmeterol and fluticasone propionate in COPD. Respir Med 2004;98(11):1045-50.  58.  Langhammer A, Johnsen R, Gulsvik A, Holmen TL, Bjermer L. Sex differences in lung vulnerability to tobacco smoking. Eur Respir J 2003;21(6):1017-23.  59.  de Torres JP, Casanova C, Hernandez C, Abreu J, Aguirre-Jaime A, Celli BR. Gender and COPD in patients attending a pulmonary clinic. Chest 2005;128(4):2012-6.  60.  Anthonisen NR, Skeans MA, Wise RA, Manfreda J, Kanner RE, Connett JE. The effects of a smoking cessation intervention on 14.5-year mortality: a randomized clinical trial. Ann Intern Med 2005;142(4):233-9.  61.  Ferrer M, Alonso J, Morera J, Marrades RM, Khalaf A, Aguar MC, Plaza V, Prieto L, Anto JM. Chronic obstructive pulmonary disease stage and health-related quality of life. The Quality of Life of Chronic Obstructive Pulmonary Disease Study Group. Ann Intern Med 1997;127(12):1072-9.  62.  Oga T, Nishimura K, Tsukino M, Sato S, Hajiro T, Mishima M. Longitudinal deteriorations in patient reported outcomes in patients with COPD. Respir Med 2007;101(1):146-53.  63.  Rennard S, Decramer M, Calverley PM, Pride NB, Soriano JB, Vermeire PA, Vestbo J. Impact of COPD in North America and Europe in 2000: subjects' perspective of Confronting COPD International Survey. Eur Respir J 2002;20(4):799-805.  64.  Miravitlles M, Ferrer M, Pont A, Zalacain R, Alvarez-Sala JL, Masa F, Verea H, Murio C, Ros F, Vidal R. Effect of exacerbations on quality of life in patients with chronic obstructive pulmonary disease: a 2 year follow up study. Thorax 2004;59(5):387-95.  65.  Lacasse Y, Brooks D, Goldstein RS. Trends in the epidemiology of COPD in Canada, 1980 to 1995. COPD and Rehabilitation Committee of the Canadian Thoracic Society. Chest 1999;116(2):306-13.  66.  Respiratory Disease in Canada. In: Bryanton V, Chen Y, Johanson H, Leeb K, McRae L, Michaelson P, Nair C, Stewart P, eds. Ottawa, Canada: Canadian Institute for Health Information, Canadian Lung Association, Health Canada, Statistics Canada, 2001.  67.  Chapman KR, Bourbeau J, Rance L. The burden of COPD in Canada: results from the Confronting COPD survey. Respir Med 2003;97 Suppl C:S23-31.  68.  Economic Burden of Illness in Canada, 1998. Ottawa: Policy Research Division, Strategic Policy Directorate, Population and Public Health Branch, Health Canada, 1998.  104  69.  Sullivan SD, Ramsey SD, Lee TA. The economic burden of COPD. Chest 2000;117(2 Suppl):5S-9S.  70.  Sullivan SD, Strassels SA, Smith DH. Characterization of the costs of chronic obstructive pulmonary disease (COPD) in the US [abstract]. Eur Respir J 1996;9(suppl 23):421S.  71.  Wang Q, Bourbeau J. Outcomes and health-related quality of life following hospitalization for an acute exacerbation of COPD. Respirology 2005;10(3):334-40.  72.  Sin DD, McAlister FA, Man SF, Anthonisen NR. Contemporary management of chronic obstructive pulmonary disease: scientific review. Jama 2003;290(17):2301-12.  73.  Ramsey SD, Sullivan SD, Kaplan RM. Cost-effectiveness of lung volumn reduction surgery. Proc Am Thorac Soc 2008;5:406-411.  74.  Levine SR, McLaughlin AJ. Pharmacology in Respiratory Care. Toronto: The McGrawHill Companies, 2001.  75.  Cadario BJ, Leathem AM, eds. Drug Information Reference. 5th ed. Vancouver, BC: The BC Drug and Poison Information Centre, 2003.  76.  Repchinsky C, ed. Compendium of Pharmaceuticals and Specialties. The Canadian Drug Reference for Health Professionals. 42nd ed. Ottawa: Canadian Pharmacists Association, 1998.  77.  Ito K, Ito M, Elliott WM, Cosio B, Caramori G, Kon OM, Barczyk A, Hayashi S, Adcock IM, Hogg JC, Barnes PJ. Decreased histone deacetylase activity in chronic obstructive pulmonary disease. N Engl J Med 2005;352(19):1967-76.  78.  Barnes PJ, Ito K, Adcock IM. Corticosteroid resistance in chronic obstructive pulmonary disease: inactivation of histone deacetylase. Lancet 2004;363(9410):731-3.  79.  Jones PW. Clinical effects of inhaled corticosteroids in chronic obstructive pulmonary disease. Proc Am Thorac Soc 2004;1(3):167-70.  80.  Calverley P, Pauwels R, Vestbo J, Jones P, Pride N, Gulsvik A, Anderson J, Maden C. Combined salmeterol and fluticasone in the treatment of chronic obstructive pulmonary disease: a randomised controlled trial. Lancet 2003;361(9356):449-56.  81.  Szafranski W, Cukier A, Ramirez A, Menga G, Sansores R, Nahabedian S, Peterson S, Olsson H. Efficacy and safety of budesonide/formoterol in the management of chronic obstructive pulmonary disease. Eur Respir J 2003;21(1):74-81.  82.  Callaghan CM, Dittus RS, Katz BP. Oral corticosteroid therapy for patients with stable chronic obstructive pulmonary disease. Ann Intern Med 1991;114:216-223.  83.  Shim CS, Williams MH, Jr. Aerosol beclomethasone in patients with steroid-responsive chronic obstructive pulmonary disease. Am J Med 1985;78(4):655-8.  84.  Guidelines for the assessment and management of chronic obstructive pulmonary disease. Canadian Thoracic Society Workshop Group. Cmaj 1992;147(4):420-8. 105  85.  Standards for the diagnosis and care of patients with chronic obstructive pulmonary disease. American Thoracic Society. Am J Respir Crit Care Med 1995;152(5 Pt 2):S77121.  86.  Siafakas NM, Vermeire P, Pride NB, Paoletti P, Gibson J, Howard P, Yernault JC, Decramer M, Higenbottam T, Postma DS, et al. Optimal assessment and management of chronic obstructive pulmonary disease (COPD). The European Respiratory Society Task Force. Eur Respir J 1995;8(8):1398-420.  87.  Majeed A, Ferguson J, Field J. Prescribing of beta-2 agonists and inhaled steroids in England: trends between 1992 and 1998, and association with material deprivation, chronic illness and asthma mortality rates. J Public Health Med 1999;21(4):395-400.  88.  Boyter AC, Steinke DT. Changes in prescribing of inhaled corticosteroids (1999-2002) in Scotland. Pharmacoepidemiol Drug Saf 2005;14(3):203-9.  89.  Van Andel AE, Reisner C, Menjoge SS, Witek TJ. Analysis of inhaled corticosteroid and oral theophylline use among patients with stable COPD from 1987 to 1995. Chest 1999;115(3):703-7.  90.  Paggiaro PL, Dahle R, Bakran I, Frith L, Hollingworth K, Efthimiou J. Multicentre randomised placebo-controlled trial of inhaled fluticasone propionate in patients with chronic obstructive pulmonary disease. International COPD Study Group. Lancet 1998;351(9105):773-80.  91.  Bourbeau J, Rouleau MY, Boucher S. Randomised controlled trial of inhaled corticosteroids in patients with chronic obstructive pulmonary disease. Thorax 1998;53(6):477-82.  92.  Calverley PM, Boonsawat W, Cseke Z, Zhong N, Peterson S, Olsson H. Maintenance therapy with budesonide and formoterol in chronic obstructive pulmonary disease. Eur Respir J 2003;22(6):912-9.  93.  Albers M, Schermer T, van den Boom G, Akkermans R, van Schayck C, van Herwaarden C, van Weel C. Efficacy of inhaled steroids in undiagnosed subjects at high risk for COPD: results of the detection, intervention, and monitoring of COPD and asthma program. Chest 2004;126(6):1815-24.  94.  Calverley PM, Anderson JA, Celli B, Ferguson GT, Jenkins C, Jones PW, Yates JC, Vestbo J. Salmeterol and fluticasone propionate and survival in chronic obstructive pulmonary disease. N Engl J Med 2007;356(8):775-89.  95.  Sin DD, Wu L, Anderson JA, Anthonisen NR, Buist AS, Burge PS, Calverley PM, Connett JE, Lindmark B, Pauwels RA, Postma DS, Soriano JB, Szafranski W, Vestbo J. Inhaled corticosteroids and mortality in chronic obstructive pulmonary disease. Thorax 2005;60(12):992-7.  96.  Soriano JB, Sin DD, Zhang X, Camp PG, Anderson JA, Anthonisen NR, Buist AS, Burge PS, Calverley PM, Connett JE, Petersson S, Postma DS, Szafranski W, Vestbo J. A pooled analysis of FEV1 decline in COPD patients randomized to inhaled corticosteroids or placebo. Chest 2007;131(3):682-9.  106  97.  Redelmeier DA, Goldstein RS, Min ST, Hyland RH. Spirometry and dyspnea in patients with COPD. When small differences mean little. Chest 1996;109(5):1163-8.  98.  Alsaeedi A, Sin DD, McAlister FA. The effects of inhaled corticosteroids in chronic obstructive pulmonary disease: a systematic review of randomized placebo-controlled trials. Am J Med 2002;113(1):59-65.  99.  Gordis L. Epidemiology. 3rd ed. Philidelphia: Elsevier Saunders, 2004.  100.  Sorensen HT, Lash TL, Rothman KJ. Beyond randomized controlled trials: a critical comparison of trials with nonrandomized studies. Hepatology 2006;44(5):1075-82.  101.  Ioannidis JP, Haidich AB, Pappa M, Pantazis N, Kokori SI, Tektonidou MG, Contopoulos-Ioannidis DG, Lau J. Comparison of evidence of treatment effects in randomized and nonrandomized studies. Jama 2001;286(7):821-30.  102.  Laupacis A, Mamdani M. Observational studies of treatment effectiveness: some cautions. Ann Intern Med 2004;140(11):923-4.  103.  Sin DD, Tu JV. Inhaled corticosteroids and the risk of mortality and readmission in elderly patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2001;164(4):580-4.  104.  Soriano JB, Vestbo J, Pride NB, Kiri V, Maden C, Maier WC. Survival in COPD patients after regular use of fluticasone propionate and salmeterol in general practice. Eur Respir J 2002;20(4):819-25.  105.  Sin DD, Man SF. Inhaled corticosteroids and survival in chronic obstructive pulmonary disease: does the dose matter? Eur Respir J 2003;21(2):260-6.  106.  Suissa S. Effectiveness of inhaled corticosteroids in chronic obstructive pulmonary disease: immortal time bias in observational studies. Am J Respir Crit Care Med 2003;168(1):49-53.  107.  Suissa S. Inhaled steroids and mortality in COPD: bias from unaccounted immortal time. Eur Respir J 2004;23(3):391-5.  108.  de Melo MN, Ernst P, Suissa S. Inhaled corticosteroids and the risk of a first exacerbation in COPD patients. Eur Respir J 2004;23(5):692-7.  109.  Bourbeau J, Ernst P, Cockcoft D, Suissa S. Inhaled corticosteroids and hospitalisation due to exacerbation of COPD. Eur Respir J 2003;22(2):286-9.  110.  Fan VS, Bryson CL, Curtis JR, Fihn SD, Bridevaux PO, McDonell MB, Au DH. Inhaled corticosteroids in chronic obstructive pulmonary disease and risk of death and hospitalization: time-dependent analysis. Am J Respir Crit Care Med 2003;168(12):1488-94.  111.  Kiri VA, Pride NB, Soriano JB, Vestbo J. Inhaled corticosteroids in chronic obstructive pulmonary disease: results from two observational designs free of immortal time bias. Am J Respir Crit Care Med 2005;172(4):460-4.  107  112.  Mapel DW, Nelson LS, Lydick E, Soriano J, Yood MU, Davis KJ. Survival among COPD patients using fluticasone/salmeterol in combination versus other inhaled steroids and bronchodilators alone. Copd 2007;4(2):127-34.  113.  Vollmer WM, Peters D, Crane B, Kelleher C, Buist AS. Impact of regular inhaled corticosteroid use on chronic obstructive pulmonary disease outcomes. Copd 2007;4(2):135-42.  114.  Gross NJ. Outcome measures for COPD treatments: a critical evaluation. Copd 2004;1(1):41-57.  115.  Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008;167(4):492-9.  116.  Zhang J, Yu KF. What's the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes. Jama 1998;280(19):1690-1.  117.  Barber P. "Chronic obstructive pulmonary disease": the diagnostic last refuge of the intellectually challenged? Eur Respir J 2004;24(4):710-1; author reply 711.  118.  Rutgers SR, Postma DS, ten Hacken NH, Kauffman HF, van Der Mark TW, Koeter GH, Timens W. Ongoing airway inflammation in patients with COPD who Do not currently smoke. Chest 2000;117(5 Suppl 1):262S.  119.  Aaron SD, Vandemheen KL, Fergusson D, Maltais F, Bourbeau J, Goldstein R, Balter M, O'Donnell D, McIvor A, Sharma S, Bishop G, Anthony J, Cowie R, Field S, Hirsch A, Hernandez P, Rivington R, Road J, Hoffstein V, Hodder R, Marciniuk D, McCormack D, Fox G, Cox G, Prins HB, Ford G, Bleskie D, Doucette S, Mayers I, Chapman K, Zamel N, FitzGerald M. Tiotropium in combination with placebo, salmeterol, or fluticasonesalmeterol for treatment of chronic obstructive pulmonary disease: a randomized trial. Ann Intern Med 2007;146(8):545-55.  120.  Vestbo J. Another piece of the inhaled corticosteroids-in-COPD puzzle. Am J Respir Crit Care Med 2001;164(4):514-5.  121.  Rothman KJ, Greenland S. Precision and validity in epidemiologic studies. In: Rothman KJ, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: LippincottRaven, 1998.  122.  Collet J, Boivin J. Bias and confounding in pharmacoepidemiology. In: Strom B, ed. Pharmacoepidemiology. Chichester: John Wiley & Sons, 2000;765-784.  123.  Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet 2002;359(9302):248-52.  124.  Kristman V, Manno M, Cote P. Loss to follow-up in cohort studies: how much is too much? Eur J Epidemiol 2004;19(8):751-60.  125.  Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks and multi-state models. Stat Med 2007;26(11):2389-430.  108  126.  Crystal S, Akincigil A, Bilder S, Walkup JT. Studying prescription drug use and outcomes with medicaid claims data: strengths, limitations, and strategies. Med Care 2007;45(10 Supl 2):S58-65.  127.  Peterson AM, McGhan WF. Pharmacoeconomic impact of non-compliance with statins. Pharmacoeconomics 2005;23(1):13-25.  128.  Steiner JF, Prochazka AV. The assessment of refill compliance using pharmacy records: methods, validity, and applications. J Clin Epidemiol 1997;50(1):105-16.  129.  Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf 2007;16(3):241-9.  130.  Levy AR, Tamblyn RM, Fitchett D, McLeod PJ, Hanley JA. Coding accuracy of hospital discharge data for elderly survivors of myocardial infarction. Can J Cardiol 1999;15(11):1277-82.  131.  Rawson NS, D'Arcy C. Assessing the validity of diagnostic information in administrative health care utilization data: experience in Saskatchewan. Pharmacoepidemiol Drug Saf 1998;7(6):389-98.  132.  McKnight J, Scott A, Menzies D, Bourbeau J, Blais L, Lemiere C. A cohort study showed that health insurance databases were accurate to distinguish chronic obstructive pulmonary disease from asthma and classify disease severity. J Clin Epidemiol 2005;58(2):206-8.  133.  Morrison LK, Harrison A, Krishnaswamy P, Kazanegra R, Clopton P, Maisel A. Utility of a rapid B-natriuretic peptide assay in differentiating congestive heart failure from lung disease in patients presenting with dyspnea. J Am College Cardiology 2002;39(2):202209.  134.  Iversen KK, Kjaergaard J, Akkan D, Kober L, Torp-Pedersen C, Hassager C, Vestbo J, Kjoller E. Chronic obstructive pulmonary disease in patients admitted with heart failure. J Intern Med 2008.  135.  Mannino DM, Brown C, Giovino GA. Obstructive lung disease deaths in the United States from 1979 through 1993. An analysis using multiple-cause mortality data. Am J Respir Crit Care Med 1997;156(3 Pt 1):814-8.  136.  Greenland S, Robins JM. Confounding and misclassification. Am J Epidemiol 1985;122(3):495-506.  137.  Hernan MA, Hernandez-Diaz S, Werler MM, Mitchell AA. Causal knowledge as a prerequisite for confounding evaluation: an application to birth defects epidemiology. Am J Epidemiol 2002;155(2):176-84.  138.  Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology 1999;10(1):37-48.  139.  Salas M, Hofman A, Stricker BH. Confounding by indication: an example of variation in the use of epidemiologic terminology. Am J Epidemiol 1999;149(11):981-3.  109  140.  Greenland S, Rothman KJ. Concepts of Interaction. In: Rothman KJ, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: Lippincott-Raven, 1998.  141.  Miettinen O. Confounding and effect-modification. 1974. Am J Epidemiol 1995;141(12):1113-6, discussion 1111-2.  142.  Strom B. Study designs available for pharmacoepidemiology studies. In: Strom B, ed. Pharmacoepidemiology. 3rd ed. Toronto: John Wiley and Sons, 2000.  143.  McMahon AD. Approaches to combat with confounding by indication in observational studies of intended drug effects. Pharmacoepidemiol Drug Saf 2003;12(7):551-8.  144.  Psaty BM, Koepsell TD, Lin D, Weiss NS, Siscovick DS, Rosendaal FR, Pahor M, Furberg CD. Assessment and control for confounding by indication in observational studies. J Am Geriatr Soc 1999;47(6):749-54.  145.  Klungel OH, Martens EP, Psaty BM, Grobbee DE, Sullivan SD, Stricker BH, Leufkens HG, de Boer A. Methods to assess intended effects of drug treatment in observational studies are reviewed. J Clin Epidemiol 2004;57(12):1223-31.  146.  Vittinghoff E, Glidden DV, Shiboski SC, McCullock CE. Regression Methods in Biostatistics: Linear, Logistic, Survival and Repeated Measures Models. Statistics for Biology and Health. New York: Springer, 2005.  147.  Therneau TM, Grambsch PM. Modeling Survival Data: extending the Cox Model. Statistics for biology and health. New York: Springer-Verlag, 2000.  148.  Christenfeld NJ, Sloan RP, Carroll D, Greenland S. Risk factors, confounding, and the illusion of statistical control. Psychosom Med 2004;66(6):868-75.  149.  Brenner H, Blettner M. Controlling for continuous confounders in epidemiologic research. Epidemiology 1997;8(4):429-34.  150.  Rothman KJ. A potential bias in safety evaluation during open-label extensions of randomized clinical trials. Pharmacoepidemiol Drug Saf 2004;13(5):295-8.  151.  Sin DD, Man SF, Tu JV. Inhaled glucocorticoids in COPD: immortal time bias. Am J Respir Crit Care Med 2003;168(1):126-7.  152.  Suissa S. Observational studies of inhaled corticosteroids in chronic obstructive pulmonary disease: misconstrued immortal time bias. Am J Respir Crit Care Med 2006;173(4):464; author reply 464-5.  153.  Perkins SM, Tu W, Underhill MG, Zhou X, Murray MD. The use of propensity scores in pharmacoepidemiologic research. Pharmacoepidemiol Drug Saf 2000;9:93-101.  154.  Seeger JD, Kurth T, Walker AM. Use of propensity score technique to account for exposure-related covariates: an example and lesson. Med Care 2007;45(10 Supl 2):S143-8.  155.  Glynn RJ, Schneeweiss S, Sturmer T. Indications for propensity scores and review of their use in pharmacoepidemiology. Basic Clin Pharmacol Toxicol 2006;98(3):253-9. 110  156.  Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70(1):41-55.  157.  Shah BR, Laupacis A, Hux JE, Austin PC. Propensity score methods gave similar results to traditional regression modeling in observational studies: a systematic review. J Clin Epidemiol 2005;58(6):550-9.  158.  Martens EP, Pestman WR, de Boer A, Belitser SV, Klungel OH. Instrumental variables: application and limitations. Epidemiology 2006;17(3):260-7.  159.  Johnston KM, Gustafson P, Levy AR, Grootendorst P. Use of instrumental variables in the analysis of generalized linear models in the presence of unmeasured confounding with applications to epidemiological research. Stat Med 2008;27(9):1539-56.  160.  Brookhart MA, Wang PS, Solomon DH, Schneeweiss S. Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable. Epidemiology 2006;17(3):268-75.  161.  Hernan MA, Robins JM. Instruments for causal inference: an epidemiologist's dream? Epidemiology 2006;17(4):360-72.  162.  Brookhart MA, Wang PS, Solomon DH, Schneeweiss S. Instrumental variable analysis of secondary pharmacoepidemiologic data. Epidemiology 2006;17(4):373-4.  163.  Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11(5):550-60.  164.  Tai BC, Machin D, White I, Gebski V. Competing risks analysis of patients with osteosarcoma: a comparison of four different approaches. Stat Med 2001;20(5):661-84.  165.  Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics 1995;51(2):524-32.  166.  Mapel DW, Hurley JS, Roblin D, Roberts M, Davis KJ, Schreiner R, Frost FJ. Survival of COPD patients using inhaled corticosteroids and long-acting beta agonists. Respir Med 2006;100(4):595-609.  167.  McGhan R, Radcliff T, Fish R, Sutherland ER, Welsh C, Make B. Predictors of rehospitalization and death after a severe exacerbation of COPD. Chest 2007;132(6):1748-55.  168.  Chapman KR, Tashkin DP, Pye DJ. Gender bias in the diagnosis of COPD. Chest 2001;119(6):1691-5.  169.  Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40(5):373-83.  170.  Petersen LA, Pietz K, Woodard LD, Byrne M. Comparison of the predictive validity of diagnosis-based risk adjusters for clinical outcomes. Med Care 2005;43(1):61-7.  111  171.  Chamberlayne R, Green B, Barer ML, Hertzman C, Lawrence WJ, Sheps SB. Creating a population-based linked health database: a new resource for health services research. Can J Public Health 1998;89(4):270-3.  172.  Delfino RJ, Becklake MR, Hanley JA. Reliability of hospital data for population-based studies of air pollution. Arch Environ Health 1993;48(3):140-6.  173.  Dahri K, Shalansky SJ, Jang L, Jung L, Ignaszewski AP, Clark C. Accuracy of a provincial prescription database for assessing medication adherence in heart failure patients. Ann Pharmacother 2008;42(3):361-7.  174.  Levy AR, O'Brien BJ, Sellors C, Grootendorst P, Willison D. Coding accuracy of administrative drug claims in the Ontario Drug Benefit database. Can J Clin Pharmacol 2003;10(2):67-71.  175.  Repchinsky C, ed. Compendium of Pharmaceuticals and Specialties. The Canadian Drug Reference for Health Professionals. 42nd ed. Ottawa: Canadian Pharmacists Association, 2000.  176.  Weiner J. The Johns Hopkins ACG(R) Case-Mix System: Reference Manual v7.0. Baltimore, MD: Johns Hopkins University, 2005.  177.  D'Hoore W, Bouckaert A, Tilquin C. Practical considerations on the use of the Charlson comorbidity index with administrative data bases. J Clin Epidemiol 1996;49(12):142933.  178.  Fleiss J, Levin B, Paik M. Statistical methods for rates and proportions. Wiley Series in Probability and Statistics. 3rd ed. Hoboken, New Jersey: Wiley, 2003.  179.  Kleinbaum D, Klein M. Logistic Regression. A self learning text. Statistics for Biology and Health. 2nd ed. New York: Springer, 2002.  180.  Greenland S, Rothman KJ. Introduction to categorical statistics. In: Rothman KJ, Greenland S, eds. Modern Epidemiology. 2nd ed. Philidelphia, PA: Lippincott - Raven, 1998.  181.  Greenland S. Basic methods for sensitivity analysis of biases. Int J Epidemiol 1996;25(6):1107-16.  182.  Thompson WD. Effect modification and the limits of biological inference from epidemiologic data. J Clin Epidemiol 1991;44(3):221-32.  183.  Hill AB. The Environment and Disease: Association or Causation? Proc R Soc Med 1965;58:295-300.  184.  Fisher LD, Lin DY. Time-dependent covariates in the Cox proportional-hazards regression model. Annu Rev Public Health 1999;20:145-57.  185.  Kiri VA, Vestbo J, Pride NB, Soriano JB. Inhaled steroids and mortality in COPD: bias from unaccounted immortal time. Eur Respir J 2004;24(1):190-1; author reply 191-2.  112  186.  Lloyd-Jones DM, Martin DO, Larson MG, Levy D. Accuracy of death certificates for coding coronary heart disease as the cause of death. Ann Intern Med 1998;129(12):1020-6.  187.  Hunt LW, Jr., Silverstein MD, Reed CE, O'Connell EJ, O'Fallon WM, Yunginger JW. Accuracy of the death certificate in a population-based study of asthmatic patients. Jama 1993;269(15):1947-52.  188.  Smyth ET, Wright SC, Evans AE, Sinnamon DG, MacMahon J. Death from airways obstruction: accuracy of certification in Northern Ireland. Thorax 1996;51(3):293-7.  189.  Decramer M, Bartsch P, Pauwels R, Yernault JC. Management of COPD according to guidelines. A national survey among Belgian physicians. Monaldi Arch Chest Dis 2003;59(1):62-80.  190.  Blais L, Bourbeau J, Sheehy O, LeLorier J. Inhaled corticosteroids in COPD: determinants of use and trends in patient persistence with treatment. Can Respir J 2004;11(1):27-32.  191.  van der Valk P, Monninkhof E, van der Palen J, Zielhuis G, van Herwaarden C. Effect of discontinuation of inhaled corticosteroids in patients with chronic obstructive pulmonary disease: the COPE study. Am J Respir Crit Care Med 2002;166(10):1358-63.  113  Appendix 1: Additional Tables Table 1a First most responsible ICD-9 codes for index hospitalization for COPD in British Columbia (BC) between 1996 and 2001. Diagnosis 1 496 491 486 428 492 466 427 786 482 410  Description Chronic airway obstruction, not elsewhere classified Chronic bronchitis Pneumonia Heart Failure Emphysema Acute bronchitis and bronchiolitis Cardiac dysrhythmias Symptoms involving respiratory system and other chest symptoms Other bacterial pneumonia Acute MI Other NA Total  COPD without Asthma n % 2821 23.3 1780 14.7 1386 11.4 783 6.5 482 4.0 266 2.2 226 1.9  COPD with Asthma n % 552 27.0 366 17.9 218 10.7 89 4.4 67 3.3 44 2.2 39 1.9  All Subjects n % 3373 23.8 2146 15.2 1604 11.3 872 6.2 549 3.9 310 2.2 265 1.9  202 182 179 3653 152 12112  36 19 21 572 18 2041  238 201 200 4225 170 14153  1.7 1.5 1.5 30.2 1.3 100.0  1.8 0.9 1.0 28.0 0.9 100.0  1.7 1.4 1.4 29.9 1.2 100.0  114  Table 1b Second most responsible ICD-9 codes for index hospitalization for COPD in British Columbia (BC) between 1996 and 2001. Diagnosis 2 496 491 492 428 486 799 401 427 250 465  Description Chronic airway obstruction, not elsewhere classified Chronic bronchitis Emphysema Heart Failure Pneumonia Other ill-defined and unknown causes - respiratory arrest? Essential hypertension Cardiac dysrhythmias Diabetes Acute upper respiratory infections of multiple or unspecified sites Other NA Total  COPD without Asthma n % 5596 46.2 888 7.3 649 5.4 402 3.3 311 2.6  COPD with Asthma n % 825 40.4 177 8.7 68 3.3 80 3.9 40 2.0  All Subjects n % 6421 45.4 1065 7.5 717 5.1 482 3.4 351 2.5  203 206 158 143  1.7 1.7 1.3 1.2  34 30 24 34  1.7 1.5 1.2 1.7  237 236 182 177  1.7 1.7 1.3 1.3  122 2088 1346 12112  1.0 17.2 11.1 100.0  33 380 316 2041  1.6 18.6 15.5 100.0  155 2468 1662 14153  1.1 17.4 11.7 100.0  115  Table 2: Dispensations of ICS, by DIN during follow-up from a first hospitalization for COPD in British Columbia (BC) between 1996 and 2001.  DIN 2213613 851752 2215055 2213605 872334 851760 897353 1978926 2174774 1978918 2174766 2237246 374407 2240836 2237247 2240837 852074 2213591 2213729 828548 2215047 2229099 1950002 2174758 790486 2237245 1949993 2216531  NAME FLOVENT INHALERS - AEM INH-ORL 250MCG/AEM PULMICORT TURBUHALER 200 MCG/DOSE BECLOFORTE INHALER - AEM INH 250MCG/AEM FLOVENT INHALERS - AEM INH-ORL 125MCG/AEM ALTI-BECLOMETHASONE DIPROPRIONATE INHALER 50MCG/MD PULMICORT TURBUHALER 400 MCG/DOSE BECLOFORTE 250MCG/AEM PULMICORT NEBUAMP 0.5 MG/ML FLOVENT INHALERS - AEM INH-ORL 250MCG/ACTUATION PULMICORT NEBUAMP 0.25 MG/ML FLOVENT INHALERS - AEM INH-ORL 250MCG/ACTUATION FLOVENT DISKUS VANCERIL AEM 50MCG ADVAIR 250 DISKUS FLOVENT DISKUS ADVAIR 500 DISKUS PULMICORT TURBUHALER 100 MCG/DOSE FLOVENT INHALERS - AEM INH-ORL 50MCG/AEM BECLODISK - PWR INH 200MCG/BLISTER BECLODISK PWR 200MCG/BLISTER BECLOVENT ROTACAPS - INH 200 MCG/CAP PULMICORT NEBUAMP 0.125 MG/ML BECLOVENT ROTACAPS - INH 200 MCG FLOVENT INHALERS - AEM INH-ORL 50MCG/ACTUATION BRONALIDE INHAL 250MCG/AEM FLOVENT DISKUS BECLOVENT ROTACAPS 100 MCG BECLOVENT - AEM 50MCG/AEM  COPD without asthma n % 7483 14.7% 7216 14.2% 6749 13.3% 7528 14.8%  COPD with asthma n % 2402 15.2% 2582 16.3% 2246 14.2% 1438 9.1%  All Subjects n % 9885 14.8% 9798 14.7% 8995 13.5% 8966 13.4%  6484 2519 2644 1890 1872 1868 1603 494 536 395 167 159 243 203 113 94 83 82 68 77 43 56 28 29  1211 1254 838 1064 873 586 337 160 114 143 163 159 25 32 48 13 20 11 18 6 31 6 29 10  7695 3773 3482 2954 2745 2454 1940 654 650 538 330 318 268 235 161 107 103 93 86 83 74 62 57 39  12.7% 5.0% 5.2% 3.7% 3.7% 3.7% 3.2% 1.0% 1.1% 0.8% 0.3% 0.3% 0.5% 0.4% 0.2% 0.2% 0.2% 0.2% 0.1% 0.2% 0.1% 0.1% 0.1% 0.1%  7.6% 7.9% 5.3% 6.7% 5.5% 3.7% 2.1% 1.0% 0.7% 0.9% 1.0% 1.0% 0.2% 0.2% 0.3% 0.1% 0.1% 0.1% 0.1% 0.0% 0.2% 0.0% 0.2% 0.1%  11.5% 5.7% 5.2% 4.4% 4.1% 3.7% 2.9% 1.0% 1.0% 0.8% 0.5% 0.5% 0.4% 0.4% 0.2% 0.2% 0.2% 0.1% 0.1% 0.1% 0.1% 0.1% 0.1% 0.1%  116  DIN 828521 2240835 768707 893633 2213583 634549 2213710 1926314 2174731 2242030 2215039 545333 2242029  NAME BECLODISK - PWR 100MCG/BLISTER ADVAIR 100 DISKUS BECLOFORT INHALER BECLOVENT AEM 50MCG/AEM FLOVENT INHALERS - AEM INH-ORL 25MCG/AEM PULMICORT INHALER 200MCG BECLODISK - PWR INH 100MCG/BLISTER AZMACORT FLOVENT INHALERS - AEM INH-ORL 25MCG/ACTUATION QVAR 100 MCG BECLOVENT ROTACAPS - INH 100 MCG/CAPSULE BECLOVENT ROTACAPS 200MCG QVAR 50 MCG Total  COPD without asthma n % 34 0.1% 29 0.1% 16 0.0% 14 0.0% 18 0.0% 12 0.0% 14 0.0% 6 0.0% 6 0.0% 4 0.0% 0 0.0% 2 0.0% 0 0.0% 50881  COPD with asthma n % 2 0.0% 0 0.0% 4 0.0% 5 0.0% 1 0.0% 3 0.0% 0 0.0% 1 0.0% 1 0.0% 1 0.0% 3 0.0% 0 0.0% 1 0.0% 15841  All Subjects n % 36 0.1% 29 0.0% 20 0.0% 19 0.0% 19 0.0% 15 0.0% 14 0.0% 7 0.0% 7 0.0% 5 0.0% 3 0.0% 2 0.0% 1 0.0% 66722  117  Table 3a Cox proportional hazards model using cause-specific deaths and hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001.  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.49 1.09  lower 95% CI 1.38 1.02  upper 95% CI 1.60 1.16  1.48 1.79 2.08 2.29 2.09 1.01  1.21 1.48 1.73 1.87 1.25 1.01  1.81 2.16 2.51 2.80 3.47 1.01  1.06 0.94 0.87 1.39 1.03 1.82 0.89  0.99 0.84 0.60 1.27 0.92 1.70 0.70  1.13 1.06 1.28 1.51 1.16 1.94 1.13  118  Table 3b Cox proportional hazards model measuring time to re-hospitalization among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 (death is censored)  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.67 1.06  lower 95% CI 1.54 0.99  upper 95% CI 1.80 1.13  1.42 1.72 1.93 1.89 1.28 1.01  1.16 1.42 1.59 1.53 0.67 1.00  1.75 2.08 2.32 2.32 2.45 1.01  1.06 0.93 0.93 1.38 1.02 1.78 0.95  0.99 0.82 0.63 1.26 0.90 1.66 0.72  1.14 1.06 1.39 1.51 1.15 1.91 1.24  119  Table 3c Cox proportional hazards model measuring time to death (all-cause) among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 0.42 1.39  lower 95% CI 0.38 1.29  upper 95% CI 0.45 1.50  1.51 2.46 3.61 5.03 7.83 1.01  1.13 1.90 2.78 3.85 5.09 1.01  2.00 3.20 4.67 6.57 12.05 1.02  1.41 2.07 2.11 1.77 1.10 1.52 0.58  1.30 1.85 1.57 1.62 0.93 1.40 0.47  1.53 2.32 2.84 1.92 1.31 1.64 0.73  120  Table 3d Cox proportional hazards model measuring time to death (respiratory related) among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 0.58 1.40  lower 95% CI 0.50 1.22  upper 95% CI 0.68 1.60  2.03 3.24 5.39 7.87 10.86 1.02  1.07 1.77 2.96 4.27 4.48 1.02  3.87 5.94 9.81 14.51 26.32 1.03  1.13 0.87 0.67 1.28 1.55 2.43 0.41  0.99 0.67 0.28 1.08 1.21 2.11 0.28  1.30 1.14 1.62 1.53 1.98 2.80 0.61  121  Table 3e Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – compliance less than 75% was considered unexposed  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.69 1.17  lower 95% CI 1.57 1.11  upper 95% CI 1.81 1.23  1.40 1.80 2.23 2.74 3.35 1.01  1.18 1.53 1.90 2.31 2.34 1.01  1.67 2.11 2.61 3.25 4.80 1.01  1.17 1.40 1.32 1.61 0.97 1.49 1.05  1.10 1.28 1.02 1.51 0.87 1.41 0.91  1.24 1.53 1.73 1.72 1.08 1.57 1.22  122  Table 3f Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – compliance less than 50% was considered unexposed  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.33 1.17  lower 95% CI 1.25 1.11  upper 95% CI 1.42 1.23  1.40 1.81 2.23 2.73 3.39 1.01  1.18 1.54 1.90 2.30 2.37 1.01  1.67 2.13 2.62 3.23 4.84 1.01  1.17 1.40 1.35 1.62 0.97 1.51 1.20  1.11 1.28 1.03 1.51 0.87 1.42 1.03  1.24 1.53 1.75 1.73 1.08 1.59 1.39  123  Table 3g Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – compliance less than 25% was considered unexposed  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.17 1.16  lower 95% CI 1.10 1.10  upper 95% CI 1.24 1.23  1.41 1.84 2.26 2.75 3.35 1.01  1.19 1.57 1.93 2.32 2.34 1.01  1.68 2.17 2.65 3.26 4.80 1.01  1.17 1.39 1.32 1.61 0.99 1.55 1.09  1.11 1.27 1.01 1.51 0.89 1.46 0.93  1.24 1.52 1.72 1.72 1.10 1.64 1.28  124  Table 3g Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – compliance less than 10% was considered unexposed  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.07 1.17  lower 95% CI 1.01 1.11  upper 95% CI 1.14 1.23  1.42 1.86 2.29 2.77 3.36 1.01  1.19 1.58 1.95 2.33 2.35 1.01  1.69 2.19 2.68 3.28 4.81 1.01  1.17 1.38 1.31 1.60 1.01 1.59 0.94  1.11 1.26 1.01 1.50 0.91 1.50 0.79  1.24 1.51 1.71 1.71 1.13 1.68 1.12  125  Table 3h Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – individuals were categorized as exposed between time of hospital discharge and first ICS fill.  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 0.80 1.15  lower 95% CI 0.75 1.09  upper 95% CI 0.85 1.21  1.40 1.86 2.28 2.70 3.05 1.01  1.17 1.58 1.94 2.28 2.13 1.01  1.66 2.18 2.67 3.20 4.36 1.01  1.17 1.36 2.02 1.52 1.06 1.71 0.85  1.10 1.25 1.69 1.42 0.96 1.62 0.71  1.24 1.49 2.41 1.63 1.18 1.81 1.02  126  Table 3i Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – time-fixed exposure  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 0.88 1.15  lower 95% CI 0.83 1.09  upper 95% CI 0.93 1.21  1.40 1.85 2.27 2.73 3.17 1.01  1.18 1.58 1.94 2.30 2.22 1.01  1.66 2.17 2.66 3.23 4.53 1.01  1.17 1.37 2.04 1.53 1.05 1.67 0.97  1.11 1.26 1.71 1.43 0.94 1.58 0.75  1.24 1.50 2.43 1.64 1.16 1.77 1.25  127  Table 3j Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – – asthma was coded as prior history of any ICD-9 code of 493 – not limited to specialists  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 1.07 1.17  lower 95% CI 0.99 1.11  upper 95% CI 1.15 1.24  1.42 1.87 2.30 2.75 3.28 1.01  1.19 1.59 1.96 2.32 2.29 1.01  1.69 2.20 2.70 3.26 4.69 1.01  1.17 1.38 1.30 1.59 1.02 1.59 0.76  1.11 1.26 1.00 1.49 0.92 1.50 0.68  1.24 1.51 1.69 1.70 1.13 1.69 0.85  128  Table 3k Cox proportional hazards model measuring time to all-cause death or hospitalizations among individuals with COPD following a first hospitalization in British Columbia (BC) between 1996 and 2001 – cohort was restricted to individuals with a first diagnosis of 491, 492, 496.  Exposure to ICS Male (vs. Female) Age (vs 45-54) 55-64 65-74 75-84 85-94 95-104 Length of Stay Comorbidity Index (vs 0) Charlson Index = 1 Charlson Index = 2 Charlson Index = 3 ACG = 5070 or 4940 (vs other) Prior Use of LABA Prior use of Anticholinergics Asthma * ICS interaction term (vs. no Asthma)  HR 0.92 1.21  lower 95% CI 0.84 1.12  upper 95% CI 1.00 1.31  1.57 2.04 2.35 2.65 1.71 1.01  1.23 1.61 1.87 2.07 0.82 1.01  2.02 2.57 2.95 3.39 3.54 1.02  1.04 1.07 0.83 1.60 1.09 1.67 0.92  0.95 0.93 0.53 1.44 0.94 1.54 0.70  1.13 1.24 1.32 1.79 1.26 1.82 1.21  129  Appendix 2: Ethics approval  130  131  

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.24.1-0066519/manifest

Comment

Related Items