Beta-Blocker Use and Morbidity from Chronic Lung Disease in Patients Undertaking Pulmonary Rehabilitation

Transcription

1 Beta-Blocker Use and Morbidity from Chronic Lung Disease in Patients Undertaking Pulmonary Rehabilitation by Robert Gabriel Varadi A thesis submitted in conformity with the requirements for the degree of Master of Science (Clinical Epidemiology) Graduate Department of Health Policy, Management and Evaluation University of Toronto Copyright by Robert Gabriel Varadi 2014

2 Beta-Blocker Use and Morbidity from Chronic Lung Disease in Patients Undertaking Pulmonary Rehabilitation Abstract Robert Gabriel Varadi Master of Science Graduate Department of Health Policy, Management and Evaluation University of Toronto 2014 Cardiovascular diseases are common in patients with chronic lung diseases. Beta-blockers reduce their morbidity, but are underutilized because of concerns over pulmonary side effects. In this retrospective cohort study, we evaluated the association of beta-blocker use with survival in elderly patients enrolled in pulmonary rehabilitation between Patient characteristics were abstracted from hospital charts and linked to administrative health databases. Primary outcome was time to death or first hospitalization. Matching on propensity score was used to account for potential confounding. No significant increase was seen in the hazard of death or hospitalization in beta-blocker users. In patients with obstructive lung disease, survival was non-significantly longer among beta-blocker users. Residual imbalance in important confounders remained despite repeated refinement of the propensity-score. Survival to death or hospitalization was not significantly associated with beta-blocker use. Beta-blockers should not be withheld from patients with lung disease who have clinical indications for them. ii

3 Acknowledgments Firstly, to my supervisor, Dr. Matthew Stanbrook, and to the members of my thesis committee, Dr. Roger Goldstein and Dr. Don Redelmeier, for their unfailing support, dedication, mentorship and education through all stages of the Master s program. To Dr. Rachael Evans, my one-time office mate, for her invaluable assistance in data collection and verification, and her appreciation for the wit of Eddie Izzard. To the clinical and administrative staff at West Park Healthcare Centre, for their encouragement, good cheer, and their help in locating lost charts and data scattered throughout the hospital. To Brandon Zagorski and the IT group at ICES, for guidance in SAS and troubleshooting in data analysis. To Amber Gertzbein and Errin Barker, for helping me navigate the SGS maze. To my family and friends, for putting up with my erratic hours and absentmindedness, and for always being in my corner. And most of all, to my wife Mariana and my sons, for always being the best possible reasons to close the computer. iii

4 Table of Contents Acknowledgments... iii Table of Contents...iv List of Tables...vi List of Figures...vii Chapter 1 : Background...1 Chapter 2 : Methods Design Population Administrative data sources Exposure Outcome Baseline characteristics and potential confounders Data verification Analysis Sample size / Power...12 Chapter 3 : Results Cohort creation Data verification Validity Baseline characteristics Unadjusted analysis Propensity score matching Adjusted analysis...17 iv

5 8 Obstructive lung disease subgroup...17 Chapter 4 : Discussion Comparison to existing literature Analysis Issues in study design Future directions...26 Chapter 5 : Conclusion...27 References...28 Tables...36 Figures...59 v

6 List of Tables 1. Definition of medication classes. 2. Comorbidities. 3. Cardiorespiratory diagnostic codes. 4. Standardised differences. 5. Missing data. 6. Beta-blocker use by year of entry to study. 7. Baseline characteristics, entire cohort. 8. Baseline characteristics, matched sample, using primary propensity score model. 9. Baseline characteristics, matched sample, using secondary propensity score model. 10. Baseline characteristics, moderate obstructive lung disease subgroup. 11. Baseline characteristics, moderate obstructive lung disease subgroup, matched sample. 12. Survival analysis a. Whole cohort, unadjusted. b. Matched cohort, primary propensity score model. c. Matched cohort, secondary propensity score model. d. Moderate obstructive lung disease subgroup, unadjusted. e. Moderate obstructive lung disease subgroup, matched. vi

7 List of Figures 1. Power plots. vii

8 1 Chapter 1 : Background Heart disease frequently coexists or develops in patients with chronic lung diseases. Investigators have attributed this association to common risk factors such as advanced age and cigarette smoking, to common pathophysiology such as systemic inflammation, and to medication toxicity [1, 2]. Coronary artery disease (CAD) has been described in 10-20% of patients with chronic obstructive pulmonary disease (COPD) [3, 4], while angiographic evidence of CAD has been found in nearly 30% of transplant-listed patients with idiopathic pulmonary fibrosis (IPF) [5-7]. Similarly, up to a third of patients with congestive heart failure (CHF) carry a diagnosis of COPD [8]. The coincidence of both lung and heart disease portends a poorer prognosis than the presence of either in isolation [9-11]. In large cohorts, decreased lung function has been shown to predict increased cardiovascular mortality [12-15]. Cardiovascular complications account for up to a quarter of deaths in COPD [16] and for 10-30% in interstitial lung diseases [17-19]. As such, management of chronic lung diseases must include concurrent strategies to manage coexistent cardiac disease. Beta-adrenergic receptor inhibiting agents, or beta-blockers, reduce mortality and morbidity associated with a wide variety of cardiac diseases, including acute coronary syndromes and myocardial infarction (MI) [20, 21]; stable angina pectoris [22, 23]; congestive heart failure (CHF) [24]; hypertension [25]; and arrhythmias [26-28]. They have also been shown to reduce cardiac morbidity in the perioperative period [29, 30]. There has long been concern that non-selective beta-blockade in patients with obstructive lung diseases would inhibit beta-2- receptor-mediated bronchodilatation, resulting in bronchoconstriction and clinical deterioration [31, 32]. Because of this concern, patients with lung diseases were excluded from many clinical trials of beta-blockers [33-37], while practice guidelines and review articles have listed asthma and COPD as contraindications to beta-blocker use [38-40]. The first generation of beta-blockers were non-selective agents that inhibited both major betaadrenergic receptor subtypes. The subclass of beta-1 receptor blockers are considered cardioselective, as they have a 20-fold greater affinity for the beta-1 adrenergic receptor that predominate in cardiac tissue. With a lower affinity for beta-2 receptors, they carry a lower

9 2 theoretical risk of bronchoconstriction [41]. Meta-analyses of trials of short-term administration of cardioselective beta-blockers in patients with obstructive lung diseases have showed no significant effect on lung function compared to placebo, nor any loss of sensitivity to the bronchodilating effects of short-acting beta-2 agonists (SABA) [42-44]. Though the results are reassuring, the pooled studies were small and focused on surrogate physiologic, not clinical, endpoints. Moreover, deleterious effects on exercise performance, such as increased dynamic hyperinflation in COPD, and on symptoms may be seen without deterioration in static lung function [45]. No clinical trial has yet prospectively tested the effects of long-term beta-blocker use in chronic lung disease. A number of investigators have shown good clinical outcomes in observational studies of patients with chronic lung diseases treated with BB. Mortality was reduced in patients with COPD or asthma treated with beta-blockers following an acute MI, compared to those in whom beta-blockers were withheld [46, 47]. Though not statistically significant, lower hospital readmission rates for COPD or asthma were also seen in the beta-blocker users [46]. Since respiratory disease parameters such as pulmonary function were not available, these studies were unable to adequately adjust for confounding by severity of lung disease. It is possible that those treated with beta-blockers had, on average, less severe lung disease and thus better survival. Moreover, cardiac disease is the most common causes of death in the immediate period after acute MI, and accounts for the majority of deaths over the 1-2 years of follow-up. This may have biased the studies in favour of beta-blockers and masked any deleterious respiratory effects. Chronic use of beta-blockers has been shown to reduce mortality in patients with COPD and hypertension, compared to other antihypertensives [48], and to blunt the apparent cardiac toxicity of SABA [49]. COPD in this study was defined by self-report, which is susceptible to misclassification. Frequency of exacerbations was used to adjust for disease severity, but pulmonary function and other respiratory parameters were not measured. Patients with coexistent COPD and CHF treated with carvedilol, a combined alpha- and nonselective beta-blocker, have shown improvements in cardiovascular function without notable pulmonary complications [50].

10 3 Among patients admitted for acute exacerbations of COPD, in which heart disease frequently contributes to mortality, those treated with beta-blockers were found to have lower in-hospital mortality than those not so treated, irrespective of their indication [3]. This study was unique in focusing on a population whose risk of morbidity was most strongly related to pulmonary as opposed to cardiac disease. However, as only short-term outcomes were considered and follow-up not extended beyond hospital discharge, conclusions about long-term safety and efficacy cannot be drawn. Similar to previous studies, pulmonary function data was not included in the analysis. Owing in part to these studies weaknesses, clinical practice has not been significantly impacted, and studies continue to find beta-blockers are underutilised in chronic lung diseases [51-55]. For instance, in one retrospective study of patients hospitalised for an acute coronary syndrome, only 16% of those diagnosed with lung disease were discharged on beta-blockers [54]. Pulmonary rehabilitation may provide a unique setting in which to investigate the role of betablockers in treating the pulmonary patient. Pulmonary rehabilitation is recognized as a central component in the management of chronic lung diseases [56, 57]. In COPD, strong evidence supports the benefits of pulmonary rehabilitation in reducing dyspnea, improving exercise capacity and quality of life, and reducing hospital and ER admissions and possibly all-cause mortality. An increasing body of literature also supports the value of pulmonary rehabilitation in other lung diseases [58, 59]. Patients enrolling in pulmonary rehabilitation are well characterised at baseline, providing a wealth of variables that may be important potential confounders, including pulmonary function and exercise capacity. Since the patients entering pulmonary rehabilitation generally have more severe disease, heart disease is more prevalent in this population [60, 61]. This should result in a higher rate of cardiac events over time and provide greater power to detect a protective effect of beta-blockers. Access to pulmonary rehabilitation in Canada is limited: it is estimated that rehabilitation program capacity exists for only 1.2% of the eligible population with COPD [62]. Appropriate referral to pulmonary rehabilitation may thus be considered a marker of better quality of respiratory care. As pulmonary rehabilitation demands a major investment of time and effort, patients enrolling are likely more highly motivated and more health-conscious than the general population with lung

11 4 disease [63-65]. Finally, a major goal of pulmonary rehabilitation is to increase patients physical activity in day-to-day life. In one study, over 80% of subjects reported adherence to a home exercise program for 6 months after completing pulmonary rehabilitation, and over 50% maintained adherence through one year s follow-up [66]. These habits are likely to reduce the potential for confounding by behavioural and social factors, and strengthen the conclusions drawn on the effects of beta-blockers. We therefore sought to assess the impact of beta-blocker use on morbidity from chronic lung disease, as reflected by death and hospitalization, in patients undertaking pulmonary rehabilitation. We hypothesized that, in patients enrolling in inpatient pulmonary rehabilitation, use of beta-blockers will be associated with longer time to death or first hospitalization for cardiac or respiratory condition.

12 5 Chapter 2 : Methods 1 Design This was a retrospective cohort study, in which data from primary chart abstraction was supplemented by administrative data sources. 2 Population West Park Healthcare Centre (WPHC) provides a program of pulmonary rehabilitation in both inpatient and outpatient settings. The rehabilitation centre is served by an experienced multidisciplinary team, and all patients are assessed by a respiratory physician prior to enrolment. Potential participants must have quit smoking prior to enrolment. The standard inpatient pulmonary rehabilitation course is of 6-weeks duration. Patients undergo individualized exercise training including aerobic exercise at least 3 days per week, according to established practice guidelines [56, 57]. Activities include treadmill, cycling, interval training (alternating high and low power exercise), upper-extremity weight training, and leisure walking. Patients experiencing exercise-induced oxygen desaturation are trained with supplemental oxygen according to routine practice. Other components of the program include education and self-management, nutritional counselling, and psychosocial support. Adverse events arising during rehabilitation are assessed by the attending respirologist. Since WPHC does not have acute care facilities, inpatients suffering any potentially serious complications are discharged from WPHC and transferred to nearby acute care hospitals. Eligible patients were adults with any chronic lung disease, aged at least 66 years on the date of admission to inpatient pulmonary rehabilitation at WPHC. Patients must have undertaken at least one exercise test or exercise training session following admission to be included in the study. A baseline value for the forced expiratory volume in 1 second (FEV1) was required for inclusion. The inception cohort consisted of all eligible patients enrolled between January 1, 1996 and December 31, Patients with COPD were identified as those meeting Canadian Thoracic Society criteria [67], including: 1) compatible chronic respiratory symptoms; 2) FEV1 < 80% of predicted normal; 3) FEV1/FVC 0.7; 4) incomplete bronchodilator

13 6 reversibility of FEV1 and FEV1/FVC (if results available). A subgroup of patients with moderate-to-severe obstructive lung disease was identified as having FEV1 < 80% of predicted normal and FEV1/FVC 0.7, regardless of the underlying diagnosis. The date of admission served as the index date for analysis. Inpatient admission records were used to identify potential study patients, and individual charts were manually reviewed for eligibility. In the event of multiple pulmonary rehabilitation admissions for the same subject, only the first eligible admission was included. 3 Administrative data sources Administrative data sources were used to define the patients exposure to medications, supplement baseline characteristics, and assess outcomes. The Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD) contains demographic and clinical data on all hospital admissions in Ontario from 1988 through the last update of March Coverage is exhaustive, with fewer than 0.01% of values missing for variables containing demographic data, dates of admission, disposition, and main diagnosis [68]. The Registered Persons Database (RPDB) contains the vital status of all persons issued a health insurance number in Ontario since Date of death is gleaned from multiple sources and considered accurate through the date of last CIHI-DAD update. The Ontario Drug Benefits (ODB) program provides publicly-funded coverage for prescription medications to all insured Ontario residents over the age of 65. The ODB database provides records of all drug claims (dispensed prescriptions) paid through the ODB since April 1990, including name, date and quantity of medication dispensed. Coding of ODB records has been validated against original prescriptions, with an error rate of less than 1% [69]. All physician services billed to the provincial public health insurance plan are captured in the Ontario Health Insurance Plan (OHIP) database. Records include the date, setting, and type of service provided. Each individual Ontario resident s records are linked across all these databases using a unique identification number based on the encrypted health insurance number. 4 Exposure The exposure of interest was use of beta-blockers prior to enrolment in pulmonary rehabilitation. Beta-blocker use was determined based on records of beta-blockers

14 7 prescriptions in the ODB database. Since direct evidence of medication use was not possible in this study design, drug exposure is defined as two or more prescriptions for beta-blockers dispensed within the year prior to admission to pulmonary rehabilitation; at least one prescription must have been filled within 100 days prior to admission, the maximum supply a pharmacy will dispense at a single visit. The practice in the pulmonary rehabilitation program at WPHC is to continue prescribing all non-respiratory medications that an incoming patient had been receiving, unless there is a strong clinical indication to do otherwise. Therefore, patient receiving beta-blockers prior to enrolment would routinely have continued to receive them upon completion of the program. The following beta-blockers are listed for coverage on the ODB formulary: beta-1-selective agents, including atenolol, bisoprolol, metoprolol, acebutolol; nonselective agents, including nadolol, oxprenolol, pindolol, propranolol, timolol; and the combined alpha-/beta-blockers carvedilol and labetalol. Because of its distinct indication as an antidysrrhythmic, prescriptions for sotalol were not considered beta-blocker use. 5 Outcome All outcome data were derived from administrative databases. The primary outcome was the combined endpoint of death or first acute care hospital admission for any cardiac or respiratory diagnosis. Since all patients transferred to an acute care centre from the inpatient pulmonary rehabilitation unit are considered discharged from WPHC, this definition includes those patients experiencing a serious adverse event during the rehabilitation course. Secondary endpoints include all-cause hospitalisation, and cardiac and respiratory admissions considered separately, and death alone. Admission for COPD exacerbation were also considered as a distinct endpoint. 5.1 Death The final study date was March 31, 2011, which was the date of the final update of the CIHI- DAD database for the year Vital status for all patients was determined by linkage to the RPDB. A subject was considered to have died if there was a recorded date of death falling on or before March 31, 2011.

15 8 5.2 Hospital admission Acute-care hospitalisations were identified by linkage to CIHI-DAD. The database contains validated fields for dates of admission and discharge, most responsible diagnosis (MRD), and procedures. An admission episode may include multiple consecutive hospital admissions, each of which is assigned a separate MRD. In such cases, the listed MRD for the first admission within an episode was retained. Diagnoses in CIHI-DAD are classified according to the ICD-9 and -10 systems, depending on the era. Table 3 lists the ICD-9 and -10 diagnosis codes we used to identify admissions as cardiac or respiratory. 5.3 Lung transplantation Patients who undergo lung transplantation have a risk of morbidity and mortality that is unique and substantially different from the general population of patients with chronic lung disease. As such, study patients who received a lung transplant had their outcome data censored at the time of admission for transplant. The following lung transplant procedure codes were identified: 4550 or 4560 under the Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures [CCP]; 1GR85, 1GT85, or 1HY85 under the Canadian Classification of Health Interventions [CCI]. If, in the course of a single admission episode, a subject admitted to a first acute-care institution had subsequently been transferred to a second for the purpose of lung transplantation, the initial non-transplant admission was included in the outcome assessment, while the second was excluded and further data censored. 6 Baseline characteristics and potential confounders Demographic and anthropomorphic data; measures of static pulmonary function, exercise capacity, and gas exchange; and comorbidities were obtained by primary chart abstraction. Primary and secondary respiratory diagnoses were determined from the attending respirologist s diagnosis, supplemented by the available clinical materials. Cardiovascular and metabolic comorbidities were determined based on prespecified criteria (table 2). The Charlson index was used to summarize the total burden of comorbidities [70]. Health care utilization was reflected by the quantity of acute care visits made within the two years prior to the index date. The number of acute care hospitalizations and cumulative length

16 9 of stay in hospital was calculated from CIHI-DAD records. OHIP billing records were used to compute the number of Emergency Department visits. Use of non-beta-blocker medications was determined by linkage to ODB records (table 1). Patients were considered users of a particular class of medication if they had filled at least one prescription in that class within the year prior to the index date. Respiratory medications included inhaled short- and long-acting anticholenergics, inhaled short- and long-acting betaagonists, inhaled corticosteroids, leukotriene antagonists, and methylxanthines. Cardiovascular medications included angiotensin pathway inhibitors, HMG-CoA-reductase inhibitors ( statins ), antiplatelet agents, diuretics, nitrates, digoxin, calcium-channel blockers, and vasodilators. Other potentially important medications included alpha-adrenergic blockers and ophthalmic beta-blockers preparations. Since aspirin is also available for over-the-counter purchase without a prescription, ODB records may not accurately reflect aspirin use in the population. 7 Data verification Average (mean and median), maximum and minimum values for all abstracted data were reviewed; records whose values appeared improbably extreme were re-abstracted. A second investigator, a practicing respirologist, independently abstracted the respiratory diagnoses from 48 patients charts selected at random. The primary and, if applicable, secondary diagnoses were assigned to one of 7 lung disease groupings: COPD; other obstructive diseases; interstitial lung diseases; extraparenchymal restrictive diseases; pulmonary hypertension; lung resection; and other conditions. Concordance between the two investigators was assessed using the unweighted Kappa score [71, 72]. 8 Analysis Event-free survival was computed between the index date and date of endpoint. Survival for the patients who were alive and did not have a recorded hospitalisation by the final study date was censored either at the final study date or at the date of last contact recorded in RPDB, whichever was earliest.

17 10 Baseline characteristics were summarized with descriptive statistics. Crude unadjusted estimates of survival among beta-blocker users and non-users were generated by Kaplan-Meier method accounting for censoring. Multivariable analysis was performed to account for possible confounding. Statistical analysis was performed using SAS (SAS 9.2 for Windows, SAS Institute Inc., Cary, NC, USA). The level of significance for all tests was Propensity score matching In circumstances in which the number of outcome events is small relative to the number of variables analysed, survival analysis by multivariable regression modeling may produce results that are biased and unreliable [73, 74]. Adjustment based on the propensity score does not suffer from this limitation [75]. The propensity score is defined as the conditional probability of having received an intervention, given a set of covariates. This probability provides a single number as summary of a set of measured covariates [76, 77]. However, it does not necessarily balance important confounders that are unmeasured or not included in the propensity score model [78]. Propensity score can be estimated using a logistic regression model: eˆ logit ˆ ei i ( PS ) = log = ˆ β βˆ j jx where e ) i is the estimated propensity score for an individual patient, representing in this study the probability of beta-blocker use; and β j X j is the vector of covariates and their regression coefficients. All potentially important covariates were included in the initial propensity score model as main effects. Based on residual imbalances between beta-blocker users and nonusers, the model was then refined using an iterative approach, allowing for interaction terms, and quadratic and cubic terms for continuous variables [79]. Three techniques are commonly employed to reduce confounding using the propensity score: matching, stratification, and covariate adjustment; more recently a new technique, inverse probability of treatment weighting, has been introduced. Covariate adjustment methods may be less reliable, and are sensitive to inaccurate modeling of the propensity score in the final regression model [76, 80]. Matching and stratification techniques are not dependent on correct model specification, and are therefore preferred. From studies on stratification by a single

18 11 continuous variable, it has been demonstrated that stratifying on quintiles of propensity score should similarly remove 90% of the bias due to imbalance in all measured covariates used in the construction of propensity score [79] [81]. The stratified analysis may use data from a greater proportion of eligible patients, while the matched analysis discards untreated patients who are dissimilar to treated patients. However, stratification may result in a greater imbalance in covariates than is seen with matching [80]. As such, matching was chosen as the primary analytic method. A 1:1 matching ratio was employed in a greedy matching strategy, using a standard technique. A random beta-blocker user was selected and paired with the non-beta-blocker user who had the closest match on the logit of the propensity score, within a caliper width of 0.2 times the standard deviation of the logit of the propensity score [82]. Matching was performed without replacement; once matched, the non-beta-blocker user was removed from the sample and could not serve as match for any of the following beta-blocker users. Matching continued until all beta-blocker users had been matched, or until no satisfactory match was identified. An optimal matching algorithm, by contrast, would select the best non-user matches so as to minimize the total within-pair differences in propensity score across the entire matched cohort. Such an algorithm is computationally very demanding, and may not provide a significant benefit to balance [102]. The success of balancing measured covariates was assessed numerically by computing standardised differences for each measured covariate (table 4). It has been suggested that a standardised difference of more than 10% represents meaningful imbalance in a covariate [83]. It was expected that many patients would be missing values for some important covariates abstracted by chart review. A propensity score model based solely on patients with complete data would exclude patients with any missing data and thus restrict the sample size. Missing values were imputed using the SAS Multiple Imputation procedure, with 20 imputed data sets created. Their estimates were pooled in order to develop the propensity score model. At this point, the imputed data were discarded; assessment of covariate balance and analysis of treatment effect was performed on the original data set containing actual, not imputed, data.

19 12 Cox proportional-hazard modelling was used to estimate the effect of beta-blocker use on event-free survival, stratified on matched-pair. Analysis was performed for the primary and secondary endpoints. A planned subgroup analysis was performed in patients with moderate obstructive lung disease. Matching was done using the same primary propensity score model as for the primary cohort. 9 Sample size / Power An estimated 1200 eligible elderly patients (120 patients admitted each year, 75% over age 65) have enrolled in pulmonary rehabilitation at the study centre over the 13 year accrual period; 10-20% of these were expected to be beta-blocker users. Patients have a high morbidity following completion of pulmonary rehabilitation. The literature suggests that 35-40% of patients with severe disease are admitted to hospital within one year after discharge from a program [84, 85], and approximately 5% die within one year [61, 86]. In experimental and observational studies of patients with acute MI, chronic CHF or hypertension, treatment with beta-blockers has been associated with relative reductions of 15-40% in the risk of death and hospitalization, even in patients with COPD [46-48, 87, 88]. Given the lower baseline cardiac risk in the population in pulmonary rehabilitation, a more conservative estimate of risk reduction of 10-20% was expected. In estimating the required sample size, a median event-free survival time of 24 months was assumed for the non-beta-blocker user group. With 900 patients accrued over 13 years, 20% of whom were beta-blocker users, the study would have at least 80% power to detect a hazard ratio of 0.78, or a reduction in hazard of death or hospitalisation of at least 22% (two-sided alpha=0.05) (figure 1, upper panel). With the same sample size of 900 patients, the smallest detectable HR would range from 0.83 to 0.74, if median survival fell in the range of months (figure 1, lower panel).

20 13 Chapter 3 : Results 1 Cohort creation Administrative records at West Park Healthcare Centre identified 3800 admissions to the inpatient respiratory unit over the inception period from January 1, 1996, through December 31, admissions involved patients aged at least 66 years. Excluding 735 repeat admissions, 1171 individual patient admission records were identified for full review. Data were abstracted from several discrete sources, including the primary hospital chart, the outpatient Respiratory Medicine clinic chart, the pulmonary function laboratory, and the departments of Respiratory Therapy and of Physiotherapy. 132 patients were excluded after chart review: 33 were missing value for the FEV1; 32 had a permanent tracheostomy; 34 were admitted for mechanical ventilation assessment only; 22 were admitted for but did not participate in pulmonary rehabilitation; and 11 charts could not be located patients were included in the final cohort. The cohort was then linked to provincial administrative databases held at the Institute for Clinical Evaluative Sciences. For each subject, linkage was performed using a unique identification number based on the encrypted health insurance number. All 1039 (100%) patients in the cohort were successfully linked to the administrative databases. 2 Data verification Data were reviewed to identify outliers with improbable values; these variables were then reabstracted from the original source. Only 4 errors required correction. No errors were identified in variables considered important potential confounders. Variables for which data were incomplete are listed in table 5. Complete data were available for all demographic, comorbidity, medication, and health care utilization variables. Fewer than 10% of values for gas exchange and exercise capacity variables were missing. MRC dyspnea score was not provided or could not be accurately computed for 17% of patients. The Chronic

21 14 Respiratory Questionnaire health status score was not available for over 50% of patients; as such, this parameter was not included in the analysis. Event dates recorded in the Registered Persons Database were compared against similar dates recorded in Discharge Abstracts Database. In no instance did a hospital admission date fall after the recorded date of death or date of last contact. 3 Validity There was substantial agreement between the independent data abstractors on the respiratory diagnosis considered primary (Kappa=0.74). There was very strong agreement on the presence of a respiratory diagnosis, irrespective of its being considered primary or secondary (Kappa=0.89). Agreement was similarly very strong when the primary diagnosis was dichotomized as COPD or other (Kappa=0.82). 4 Baseline characteristics 53 (5.1%) patients met the criteria for beta-blocker use, having filled a total of 365 prescriptions for beta-blockers in the year prior to the index date. The majority of patients received beta-1-selective agents including metoprolol (20 patients [34%]), atenolol (18 [31%]), and bisoprolol (13 [22%]). Other beta-blockers agents included carvedilol (4 [7%]), acebutolol (3 [5%]), and propranolol (1 [2%]). 6 patients had filled prescriptions for two distinct betablockers. One subject prescribed bisoprolol had also filled 4 prescriptions for sotalol in the year prior to index date. Beta-blocker use varied with era of study, with a greater proportion of beta-blocker use seen in patients entering the cohort in later years (table 6). Beta-blocker users accounted for only 1.6% of patients entering the cohort between , but 13.4% of patients entering between Among beta-blocker users, the median year of entry to the cohort was 2006, as compared to 2001 among non-beta-blocker users. Overall, patients in the more recent era were more likely to have a non-copd diagnosis, more frequently had cardiac comorbidities, and had greater use of cardiac and long-acting respiratory medications.

22 15 Baseline characteristics of beta-blocker and non-beta-blocker users are presented in table 7. Beta-blocker users were on average older, less likely to be married, and residing in more affluent neighbourhoods. There was no significant difference in use of acute medical care in the 2 years prior to index date. A greater proportion of beta-blocker users did not have a diagnosis of COPD, and had reported never smoking. Pulmonary function was on average severely impaired in the cohort. Spirometric values and total lung capacity in beta-blocker users were less severely perturbed, though this may also reflect the greater proportion of nonobstructive lung diseases. The diffusing capacity was severely reduced to a similar degree in both groups. Self-reported dyspnea, as measured by both the MRC score and the CRQ dyspnea subscale, was more severe in beta-blocker users, and exercise capacity and resting oxygen levels were poorer. Overall, 653 (62.8%) patients in the cohort had a diagnosis of at least one cardiovascular or metabolic disease, while 597 (57.5%) patients had at least one diagnosis considered an indication for beta-blockers. Beta-blocker users had a greater burden of comorbidities, as measured by the Charlson Comorbidity Index, and individual cardiovascular and metabolic comorbidities were more common in beta-blocker users. The mean number of distinct medications used in the past year was significantly greater in beta-blocker users. A greater proportion of beta-blocker users had filled prescriptions for important cardiac medications. The pattern of use of respiratory medications between the groups was mixed: significantly more beta-blocker users were prescribed inhaled long-acting anticholenergic and beta-agonist agents, while prescriptions for inhaled corticosteroids, short-acting bronchodilators, and methylxanthines were significantly less common in this group. 5 Unadjusted analysis There were 4067 hospital admission episodes among all 1039 patients during the follow-up period. Among the 53 beta-blocker users, there were 129 admissions: 64 (50%) for a respiratory diagnosis, 50 (39%) for COPD, and 16 (12%) for a cardiac diagnosis. Among the 986 non-beta-blocker users, there were 3938 admissions: 2565 (65%) for a respiratory diagnosis, 2149 (55%) for COPD, and 225 (6%) for a cardiac diagnosis. 97 patients, including

23 16 5 (9.4%) beta-blocker users and 92 (9.3%) non-beta-blocker users, did not have a record of death or hospital admission after the index date. In unadjusted analysis including all 1039 patients, there was no significant difference in survival between all beta-blocker users and non-users to the primary combined endpoint of death or cardiorespiratory hospitalization (table 12a). No significant difference was detected in survival to combined death or all-cause hospitalization. 798 (77%) patients died during the follow-up period, with a median survival from index date of 47.5 months. There were 38 deaths among beta-blocker users, and 760 among non-beta-blocker users. The Kaplan-Meier estimate of median survival to death was significantly shorter in beta-blocker users (33 vs 50 months, HR 1.40, 95% CI ). Beta-blocker users had a significantly increased hazard of death or hospitalisation for cardiac disease only (HR 1.57, 95% CI ), but nonsignificantly longer survival to combined death or hospitalisation for respiratory disease (HR 0.85, 95% CI ). 6 Propensity score matching Despite repeated refinements to the propensity score model, there remained significant residual imbalance in important covariates between matched beta-blocker users and non-users. A primary propensity score model was chosen to achieve balance in the greatest number of important cardiovascular and respiratory covariates. 31 of 53 (58%) beta-blocker users were successfully matched to non-beta-blocker user controls (table 8). Matched cases and controls were well balanced on cardiovascular, respiratory and metabolic comorbidities. Among important medications, matched beta-blocker users were less commonly prescribed angiotensin antagonists, calcium-channel blockers and nitrates, but more commonly prescribed long-acting inhaled anticholinergics. Matched beta-blocker users had lower baseline 6MWD and diffusing capacity, and were more likely to have smoked; they less commonly reported severe dyspnea and were less likely to experience exertional desaturation. Though the number of hospital admissions were similar between the two groups, matched beta-blocker users spent fewer days in hospital over the 2 years prior to index date. A secondary propensity score model was then designed to achieve balance in important respiratory parameters; with this model, 35 (66%) beta-blocker users were successfully

24 17 matched to controls (table 9). As expected, the groups demonstrated balance on baseline 6MWD, MRC dyspnea, oxygenation measures, respiratory medication use, and smoking status; however, mean FEV1 was greater in matched beta-blocker users than in non-users. There was residual imbalance in comorbidities, with matched beta-blocker users having a higher mean Charlson comorbidity score, and a greater proportion of beta-blocker users having a diagnosis of CAD and CHF. The main difference in this model was the removal of a term for era from the secondary propensity score model. 7 Adjusted analysis The primary adjusted analysis was performed with 62 patients matched on the primary propensity score (31 each beta-blocker users and non-users). No significant difference was detected between the matched groups in survival to the combined endpoint of death or cardiorespiratory (table 12b). Similarly, no significant association was seen between betablocker use and survival to death alone or any of the combined endpoints. The point estimates of the hazard ratio were closer to unity in all of the primary matched analyses, as compared to the unadjusted estimates. The secondary adjusted analysis was performed with 70 patients matched on the secondary propensity score (35 each beta-blocker users and non-users). No significant difference in survival was detected for the primary endpoint. A significant decrease in survival to death alone remained among beta-blocker users, with a point estimate of effect that was higher than in the unadjusted analysis (HR 2.22, 95% CI ) (table 12c). 8 Obstructive lung disease subgroup 864 patients met criteria for at-least moderate obstructive lung disease (FEV1/FVC < 0.7, and FEV1 < 80% of predicted normal), of whom 29 (3.4%) were beta-blocker users. COPD was the primary diagnosis in 95% of patients. Baseline characteristics are presented in table 10. While FEV1 was less impaired on average in beta-blocker users, exercise capacity and dyspnea were more severely limited. As expected, beta-blocker users were more likely to have a cardiac comorbidity and to be receiving cardiac medications. Long-acting bronchodilators were more commonly used among beta-blocker users.

25 18 18 (62%) beta-blocker users were successfully matched to non-beta-blocker controls. The matched groups were generally well balanced on important comorbidities and cardiorespiratory medication use. Matched beta-blocker users had a tendency to milder lung disease, with a higher mean baseline 6MWD and less severe dyspnea than non-beta-blocker users (table 11). In unadjusted analysis of all 864 patients, no significant difference was detected between the exposure groups in survival to any endpoint (table 12d). In the matched analysis including 36 patients (18 each beta-blocker users and non-users), no significant differences were found between the groups in survival to any endpoint (table 12e). The point estimates for effect were lower for all endpoints than in the unadjusted analysis, and nealy all favoured a survival advantage among beta-blocker users, though without achieving statistical significance.

26 19 Chapter 4 : Discussion 1 Comparison to existing literature Historical concerns about the deleterious effects of beta-blockers may no longer be as relevant. Many early reports of BB-induced bronchoconstriction, reduced sensitivity to inhaled betaagonists, and increased respiratory symptoms, involved small series of patients exposed to nonselective beta-blockers such as propranolol [89-91]. These agents have largely been replaced by the cardioselective BB, which have much lower affinity for the beta-2 adrenergic receptor in the lungs. In a meta-analysis of 20 studies including 278 patients with obstructive lung diseases, cardioselective beta-blockers were not found to have significant impact on lung function or symptoms [42]. Similarly, combined alpha/beta-blockers such as carvedilol may not produce an important decline in lung function [50, 92]. 86% of beta-blockers prescriptions in our study were for beta-1 selective agents, and only 1 patient received propranolol. Our study design did not permit identifying patients who had previously suffered adverse effects of beta-blockers. There is a growing body of evidence that beta-blockers are safe for use in patients with chronic lung disease. While our results are consistent with this, we did not demonstrate an improvement in mortality or hospitalization. Our study included patients with any chronic lung disease, while previous research centered on COPD and asthma. 23% of beta-blocker users in our cohort had a primary diagnosis of an interstitial lung disease, often pulmonary fibrosis, for which prognosis is poor and no effective treatment is available [93]. This may have reduced our ability to detect a treatment effect. When the analysis was restricted to those patients with obstructive disease of at least moderate severity, we found estimates of mortality and hospitalization reduction in line with previous research, though not achieving statistical significance. Previous studies on beta-blocker use have selected populations whose cardiovascular risk is highest, such as following MI [46, 47] or vascular surgery [94], in chronic CHF [95, 96] or in chronic hypertension [48]. Our study selected a population with more severe lung disease and therefore greater pulmonary risk, as evidenced by the preponderance of hospitalizations for respiratory diagnoses. Two studies have observed lower

27 20 in-hospital mortality among beta-blocker users who were hospitalized with COPD exacerbation; however, follow-up was not extended after hospital discharge [3, 97]. Moreover, by defining medication use by in-hospital prescriptions, these studies were vulnerable to immortal time bias. Two recent large population-based studies of beta-blocker use in COPD have yielded conflicting results. Short and colleagues retrospectively analysed a patient registry in Scotland, and found that patients using beta-blockers had 22% lower mortality than non-users over a 10-year span, irrespective of respiratory medication use [98]. This cohort was wellcharacterized and accounted for FEV1 in the analysis. In a prospective study, Ekstrom and colleagues evaluated patients with severe COPD upon initiation of home oxygen therapy [99]. They incorporated a time-dependent analysis to account for continuity of medication exposure throughout the study period. Beta-blocker use was associated with a statistically significant 19% increase in mortality over 4 years follow-up, while other cardiac medications such as angiotensin-pathway antagonists and statins were associated with non-significant mortality reduction. In addition to the difference in analytic techniques, the conflicting results may in part be attributable to severity of illness. Ekstrom et al included only severely impaired patients with chronic hypoxemic respiratory failure, whose mean FEV1 was severely reduced at less than 40% of predicted normal. This severity of illness more closely matches that seen in our study. Short et al included a broader range of range of patients, only 30% of whom had spirometrically severe COPD. It is possible that beta-blockers have differential effects based on severity of disease. Our small number of beta-blocker users did not permit additional analysis of this potential interaction. 2 Analysis 2.1 Choice of analytic technique In non-randomized studies, it is important to account for differences in baseline characteristics that may confound the relationship between treatment and outcome. Multivariable regression modeling is commonly used to adjust for potential confounding. However, regression modeling has been shown to be unreliable when the number of confounders is large relative to the sample size. It has been shown that when there are fewer than 10 outcome events in the

28 21 smallest of the primary exposure groups for each variable analysed (generally referred to as events per variable ), the model proves unreliable, with excessively biased estimates of effect and questionable validity of significance testing [73, 100, 101]. This finding has been described in both logistic and Cox proportional hazard regression. The problems associated with few events per variable may be more pronounced when the association between the primary exposure and the outcome is weak, and when the sample size is small [74]. In our study, the number of outcome events among beta-blocker users was small (at most 53); the number of potential confounders large (as many as 50); and the association between betablocker use and morbidity weak (hazard ratios no larger than 2). As such, simple multivariable regression adjustment was not an appropriate method to account for confounding. The propensity score provides a single summary of a set of known confounders, and may be estimated as the probability of having received the treatment of interest based on measured baseline characteristics. The propensity score is a balancing score, such that after conditioning on the propensity score, the distribution of measured characteristics is similar between treated and untreated patients. There are a number of advantages posited to propensity score methods over multivariable regression methods alone [102]. Firstly, propensity score methods have greater flexibility in accounting for confounders when outcomes are few, and may provide correct estimates of treatment effect when sample size is small [103]. For instance, a simulation study with binary outcome showed less bias and smaller standard errors in the estimate of treatment effect using stratification on the propensity score as compared to logistic regression, when the number of outcome events was small [75]. Secondly, it is easier and more explicit to assess the success of propensity score methods in achieving balance of baseline characteristics by direct comparison, than to assess residual confounding and appropriateness of model specification in regression modeling. Thirdly, propensity score methods separate the balancing of treatment groups with respect to potential confounders, from the analysis of treatment effect on outcome. Since these steps are combined in regression modeling, a concern has been raised that adjustment of the regression model may be continued in an effort to further address confounding, until a desired or expected treatment effect on outcome is seen [102].