Original Research

Annals of Internal Medicine
Original Research
Comparison of Inhaled Long-Acting -Agonist and Anticholinergic Effectiveness in Older Patients With Chronic Obstructive Pulmonary Disease
A Cohort Study
Andrea Gershon, MD, MSc; Ruth Croxford, MSc, PStat; Teresa To, PhD; Matthew B. Stanbrook, MD, PhD; Ross Upshur, MD, MSc; Paula Sanchez-Romeu, BMath, MMath; and The re ` se Stukel, PhD
Background: Chronic obstructive pulmonary disease (COPD), a largely preventable and manageable respiratory condition, affects an estimated 12% to 20% of adults. Long-acting inhaled -agonists and anticholinergics have both been shown to improve COPD outcomes and are recommended for moderate to severe disease; however, little is known about their comparative effectiveness. Objective: To compare survival in older patients with COPD who initially receive inhaled long-acting -agonists with that of patients who receive anticholinergics. Design: Population-based, retrospective cohort study. Setting: Ontario, Canada. Patients: Patients aged 66 years or older (who carry the largest burden of COPD and for whom data were available) who met a validated case definition of COPD on the basis of health administrative data and were newly prescribed an inhaled long-acting -agonist or a long-acting anticholinergic (but not both) between 2003 and 2007. Patients were followed for up to 5.5 years. Measurements: The primary outcome was all-cause mortality.
Results: A total of 46 403 patients with COPD (mean age, 77 years; 49% women) were included. Overall mortality was 38.2%. Mortality was higher in patients initially prescribed a long-acting anticholinergic than in those initially prescribed a long-acting inhaled -agonist (adjusted hazard ratio, 1.14 [95% CI, 1.09 to 1.19]). Rates of hospitalizations and emergency department visits were also higher in those initially prescribed a long-acting anticholinergic. Limitation: Patients were classified as having COPD on the basis of health administrative records, which did not contain information about lung function. Conclusion: Older adults initially prescribed long-acting inhaled -agonists for the management of moderate COPD seem to have lower mortality than those initially prescribed long-acting anticholinergics. Further research is needed to confirm these findings in younger patients and in a randomized, controlled trial. Primary Funding Source: Government of Ontario, Canada.
Ann Intern Med. 2011;154:583-592. For author affiliations, see end of text. www.annals.org
hronic obstructive pulmonary disease (COPD) is a largely preventable and manageable respiratory condition that affects an estimated 12% to 20% of adults older than 40 years (1, 2). It is also the fth leading cause of death in the world and is projected to be the fourth by 2030 (3, 4). Thus, effective, evidence-based strategies to improve COPD outcomes are crucial. Medications are a mainstay of COPD management. Long-acting -agonists, such as salmeterol and formoterol, and long-acting anticholinergics, also referred to as muscarinic antagonists (of which only tiotropium bromide is currently available for public use), are recommended for the management of moderate to severe COPD (5). Randomized, controlled trials (RCTs) (6, 7) have shown that both types of drug can decrease exacerbations and hospitalizations and improve symptoms, lung function, quality of life, and possibly mortality compared with placebo. However, little is known about their comparative effectiveness in actual practice. This information would allow patients to initiate therapy with the long-acting medication most likely to provide the greatest benet, and possibly avoid use of additional medications that would increase risk, inconvenience, and cost (5).
Downloaded From: http://annals.org/ by Rizka Mulya on 02/22/2013
The purpose of our population-based, retrospective, observational cohort study of older patients with COPD was to compare the effectiveness of long-acting anticholinergics and long-acting -agonists in terms of survival. Older patients were studied because they carry the highest burden of COPD and because population-wide medication records were available for patients aged 65 years or older.
METHODS
We conducted our study by using health administrative data from Ontario, a province of Canada with a diSee also: Print Editors Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584 Web-Only Appendix Appendix Tables Appendix Figure Conversion of graphics into slides
2011 American College of Physicians 583
Original Research
Context
Long-Acting -Agonists and Long-Acting Anticholinergics in COPD
Long-acting inhaled -agonists and anticholinergic agents are both used to treat chronic obstructive pulmonary disease (COPD). Whether one or the other is better for initial therapy is not known.
the 2001, 2003, and 2005 surveys were linked to the health administrative databases on an individual level to obtain additional patient information, including smoking history.
Study Population
Contribution
Patients with physician-diagnosed COPD were identified in a public health administrative database by using a computer algorithm. Patients initially prescribed a long-acting anticholinergic agent had more hospital visits and higher mortality rates than those initially prescribed a long-acting -agonist.
Caution
Patients were identified as having COPD according to administrative data, not clinical guidelines. Lung function testing was not available.
Implication
Further study regarding the relative benefits of long-acting anticholinergics and -agonists for treatment of COPD is warranted. The Editors
verse, multicultural population of approximately 12 million, including about 1.8 million persons aged 65 years or older. The ethics committee of Sunnybrook Health Sciences Centre, Toronto, Ontario, approved our study.
Data Sources
Residents of Ontario have universal public health insurance under the Ontario Health Insurance Plan, the single payer for all medically necessary services across virtually all residents, providers, and hospitals. Service details are captured in health administrative databases, which can be linked on an individual level to provide a complete health services prole for each resident. The Registered Persons Database contains basic demographic information and dates of death. The Ontario Drug Benet Program database contains prescription claim records for all residents aged 65 years or older. Publicly funded prescriptions are subject to a small, means-tested copayment, which does not affect the rate at which they are obtained (8). The Discharge Abstract Database and National Ambulatory Care Reporting System Database of the Canadian Institute of Health Information contain information on all discharges from acute care hospitals, same-day surgical procedures, and emergency department visits. The Ontario Health Insurance Plan Physician Claims database contains information about all services provided by fee-for-service physicians and shadow billings for physicians paid under alternate payment plans. We also used the Canadian Community Health Survey, a national, cross-sectional, population-based survey conducted by Statistics Canada every 2 years. Responses to
584 3 May 2011 Annals of Internal Medicine Volume 154 Number 9
We included patients who met a previously validated case denition of physician-diagnosed COPD on the basis of health administrative data (9), were aged 66 years or older, and lled a rst prescription for either a long-acting anticholinergic or long-acting -agonist between 1 July 2003 (when long-acting anticholinergics were rst commonly used in the community) and 31 March 2007. Compared with real-world clinical evaluation by a physician (which may or may not have included spirometry [10]), the case denition of 1 hospitalization or 3 ambulatory care claims for COPD had a positive predictive value of 81.3% in adults aged 35 years or older (9) and 86.3% in adults aged 65 years or older (unpublished data). This value was probably even higher in patients who met the case denition and were prescribed a long-acting anticholinergic or long-acting -agonist, such as those studied here. The date of lling a rst prescription was used as the study index date. Only new users, dened as those who had not lled a prescription for a long-acting anticholinergic or long-acting -agonist in the previous 12 months (including a combination medication that contained a long-acting -agonist), were included to avoid bias due to either better outcomes among patients who had been receiving their index medication for a longer time or worse outcomes among patients who had been receiving therapy with other long-acting medications that had failed. The study was restricted to patients aged 66 years or older to allow us to look at medication use in the past year.
Baseline Characteristics
We obtained demographic, COPD-related, and general carerelated characteristics from the health administrative databases (Table 1 and Appendix Table 1, available at www.annals.org). Severity of COPD was captured by using information from hospitalizations, emergency department visits, ambulatory care visits, prescription drug use, and interventions before the index date. Socioeconomic status was inferred from neighborhood income, which was derived from postal code and census data (11). Comorbidity was based on diagnostic information in health services records from the 2 years before the index date and was characterized by using the Johns Hopkins Adjusted Clinical Group Case-Mix System (12, 13).
Outcome Measures
The primary outcome was all-cause mortality. Secondary outcomes were hospitalizations and emergency department visits for COPD, related respiratory diseases (inuenza, acute bronchitis, or pneumonia), and cardiovascular disease (acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmias, or cerebrovascular disease). Hospitalizations and emergency dewww.annals.org
Original Research
Table 1. Selected Baseline Characteristics, by Initially Prescribed Medication, Before and After Propensity Score Matching*
Characteristic Before Propensity Score Matching Initially Prescribed Medication Standardized Difference, % After Propensity Score Matching
P Value
Initially Prescribed Medication
Standardized Difference, %
P Value
Long-Acting Long-Acting Anticholinergic -Agonist Patients, n Demographic characteristics Mean age (SD), y Women, % COPD and general care COPD duration, % 1 y 15 y 5 y Previous spirometry, % Median primary care physician visits in previous year (interquartile range), n Specialist visit in previous year, % Median prescriptions for COPD medications filled in previous year (interquartile range), n Medication use in previous year, % Respiratory antibiotic Short-acting -agonist Short-acting anticholinergic Inhaled corticosteroid Oral corticosteroid Long-term oral corticosteroid Methylxanthine Statin ACE inhibitor -Blocker Previous or coexisting medical conditions, % Asthma ever diagnosed Lung cancer Pulmonary embolism Arrhythmias Acute myocardial infarction Other ischemic heart disease Congestive heart failure Cerebrovascular disease Acute events before index date, % Most recent hospitalization for COPD In the past 6 mo 6 mo before index date Never Most recent hospitalization for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent hospitalization for cardiovascular cause In the past 6 mo 6 mo before index date Never 28 563 17 840
Long-Acting Long-Acting Anticholinergic -Agonist 15 532 15 532
77.2 (6.8) 46.2
76.8 (7.0) 53.1
6.0 13.7
0.001 0.001
76.9 (6.9) 50.8
76.9 (6.9) 50.8
0.0 0.0
0.98 1.00
10.3 22.5 67.3 69.7 9 (514) 44.7 6 (114)
9.3 22.0 68.8 74.3 9 (514) 50.5 5 (113)
3.4 1.1 3.2 10.2 1.5 11.6 7.2
0.001 0.23 0.001 0.001 0.001 0.001 0.001
9.8 21.8 68.3 72.4 9 (514) 49.0 5 (113)
9.6 22.4 68.0 73.0 9 (514) 49.1 5 (114)
0.8 1.3 0.7 1.3 0.2 0.2 0.7
0.50 0.25 0.55 0.24 0.89 0.87 0.40
64.6 72.5 54.3 48.3 22.9 3.3 4.1 37.3 43.2 25.9
70.2 71.6 45.7 52.1 27.1 3.4 4.7 38.0 40.4 24.1
11.9 1.9 17.4 7.7 9.8 0.6 3.3 2.1 5.6 4.0
0.001 0.048 0.001 0.001 0.001 0.54 0.001 0.030 0.001 0.001
68.9 72.2 49.6 51.1 25.0 3.3 4.7 38.3 41.5 24.7
68.9 71.9 49.3 51.3 26.0 3.3 4.7 38.3 41.7 24.9
0.0 0.7 0.6 0.3 0.2 0.1 0.4 0.1 0.4 0.4
1.00 0.52 0.59 0.78 0.87 0.92 0.73 0.96 0.70 0.74
37.6 8.6 4.5 39.6 53.3 48.9 40.2 11.7
55.4 8.0 4.6 39.6 53.5 50.9 38.2 9.6
36.3 2.1 0.9 0.1 0.5 4.0 4.1 6.6
0.001 0.031 0.37 0.95 0.59 0.001 0.001 0.001
49.7 8.3 4.6 39.4 53.8 50.5 39.0 10.2
49.7 8.4 4.7 39.9 53.7 50.5 39.2 10.2
0.0 0.5 0.4 1.0 0.3 0.0 0.4 0.0
1.00 0.65 0.75 0.39 0.80 1.00 0.75 1.00
8.0 16.5 75.5
7.3 15.5 77.2
2.5 2.9 4.0
0.010 0.003 0.001
7.8 16.1 76.1
7.8 15.9 76.3
0.1 0.5 0.5
0.90 0.64 0.63
1.8 15.6 82.6
1.9 14.7 83.4
0.4 2.4 2.1
0.65 0.013 0.027
1.8 15.4 82.8
1.9 15.2 82.9
0.7 0.6 0.3
0.56 0.63 0.80
6.0 30.3 63.7
5.2 29.3 65.5
3.5 2.3 3.9
0.001 0.017 0.001
5.6 29.9 64.5
5.5 29.7 64.7
0.2 0.4 0.4
0.88 0.74 0.70
Continued on following page
www.annals.org
3 May 2011 Annals of Internal Medicine Volume 154 Number 9 585
Original Research
Table 1Continued
Characteristic
Before Propensity Score Matching Initially Prescribed Medication Standardized Difference, %
After Propensity Score Matching
P Value
P Value
Long-Acting Long-Acting Anticholinergic -Agonist Most recent emergency department visit for COPD In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a cardiovascular cause In the past 6 mo 6 mo before index date Never
Long-Acting Long-Acting Anticholinergic -Agonist
5.6 14.9 79.5
5.2 13.0 81.8
1.5 5.4 5.6
0.111 0.001 0.001
5.5 13.8 80.7
5.5 13.8 80.7
0.1 0.0 0.1
0.94 0.99 0.95
3.8 21.2 75.1
4.4 21.6 74.0
3.0 1.1 2.4
0.002 0.27 0.014
4.2 21.5 74.3
4.3 21.4 74.2
0.8 0.2 0.2
0.46 0.85 0.88
3.0 18.1 78.9
2.8 18.5 78.7
1.5 1.2 0.5
0.118 0.23 0.62
3.1 18.2 78.7
2.9 18.5 78.6
1.1 0.6 0.1
0.33 0.60 0.92
ACE angiotensin-converting enzyme; COPD chronic obstructive pulmonary disease. * See Appendix Table 2 (available at www.annals.org) for complete baseline characteristics. Testing the hypothesis of no difference between the 2 medication cohorts. In Ontario, primary care is provided by family and general physicians, and specialist COPD care is usually provided by pulmonologists, general internists, or geriatricians. Number of prescriptions for long- or short-acting -agonists, long- or short-acting anticholinergics, inhaled corticosteroids, oral corticosteroids, methylxanthine, or respiratory antibiotics. Pneumonia, inuenza, or acute bronchitis. Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmia, or cerebrovascular disease.
partment visits were analyzed as a combined end point with mortality because mortality was common and the factors that caused it were probably exacerbations of those that caused a hospitalization or emergency department visit (making censoring for mortality inappropriate). Patients were followed beginning 7 days after their index prescription, to allow time for patients to start taking their medication and for it to start taking effect (14). Patients who did not die were censored on 15 March 2009. All-cause mortality was used because cause of death was not available in the health administrative data.
Propensity Score Matching and Analysis
Appendix (available at www.annals.org) contains additional information on propensity score methodology. To further ensure that variables not available in the health administrative data, including smoking status, were balanced between treatment groups, we performed a sensitivity analysis by using propensity score calibration and additional information from the Canadian Community Health Survey (17) (Appendix).
Additional Sensitivity Analyses
We used propensity score matching to compare patients with similar observed characteristics, all of whom were potential candidates for both treatments (15). Patients initially prescribed a long-acting anticholinergic were matched 1:1 with those initially prescribed a long-acting -agonist on the basis of age (1 year), sex, asthma status, number of COPD medication prescriptions lled in the previous year, and propensity score. An absolute standardized difference between variables of less than 10% was accepted as adequate balance (16). Outcomes were compared by using Cox proportional hazards regression analysis, adjusted for matched pairs. All statistical tests were 2-sided, with statistical signicance dened as a P value less than 0.05. Analyses were performed by using SAS, version 9.1 (SAS Institute, Cary, North Carolina). The
We conducted several sensitivity analyses to evaluate the robustness of our results. The main analyses were repeated by using a standard proportional hazards regression that adjusted for all covariates, rather than matching on the propensity score. A complementary analysis, which compared the on-treatment effects of long-acting -agonists and long-acting anticholinergics, was also conducted on the basis of current patient use of these medications rather than initial choice of treatment. This was done with a time-on-treatment analysis of the effect of medication exposure by using a standard time-varying proportional hazards analysis with adjustments for all of the variables used to calculate the propensity score. The propensity scorematched analyses were also repeated on subsets of the cohort to look for consistency in the results. First, we examined the subsets of the cohort who had or had not previously received spirometry, to determine whether more or less certainty about the COPD
www.annals.org
Original Research
diagnosis inuenced the results. The original analysis was also stratied by factors of a priori interest, including sex, asthma status, number of prescriptions for COPD medication lled in the previous year, inhaled corticosteroid use, and diagnosis of congestive heart failure. We conducted 2 sensitivity analyses to check assumptions made in the main analyses. One was that follow-up should not begin until 7 days after lling the index prescription, to allow time for patients to start taking their medication and for the medication to start taking effect. To check this, we repeated the analysis, including outcomes that occurred within this period. Another assumption was that hospitalizations and emergency department visits were along the causal pathway to the primary outcome of death and therefore should be combined with it as a composite outcome. To check this, we examined time to rst COPD hospitalization and time to rst COPD emergency department visit as separate outcomes. Finally, we determined whether it was plausible that an unmeasured confounder or the inclusion of persons without COPD, due to misclassication error, was responsible for the observed results by using an array approach. The rate ratio of a theoretical, unmeasured confounder and the imbalance of this confounder between the study cohorts were both varied to see at what point the observed hazard ratio was reduced to 1.0 (18). The same approach was used to estimate the effects of misclassication due to the case denition (Appendix).
Role of the Funding Source
cholinergic and long-acting -agonist groups did not signicantly differ (Table 1 and Appendix Table 2, available at www.annals.org). Within the matched cohorts, median follow-up times for patients initially prescribed longacting anticholinergics or long-acting -agonists were 3.2 and 3.3 years, respectively (maximum, 5.5 years). Overall, 39.9% of patients initially prescribed a long-acting anticholinergic and 36.5% of patients initially prescribed a long-acting -agonist died during the follow-up period (Table 2). Appendix Table 3 (available at www.annals.org) shows absolute survival by year. The Figure shows survival, by initially prescribed medication. Overall, patients who were initially prescribed a long-acting anticholinergic had a modest but signicantly higher adjusted rate of death (adjusted hazard ratio, 1.14 [95% CI, 1.09 to 1.19]) and all other outcomes than those prescribed a long-acting -agonist (Table 2). No evidence indicated that the hazard ratios changed over time. The results of the calibrated propensity score analysis were consistent with those of the main analysis (Appendix).
Additional Sensitivity Analyses
Our study was funded by the Government of Ontario and the Institute for Clinical Evaluative Sciences, which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care. The funding sources had no role in study design, collection, analysis, interpretation of data, writing of the report, or the decision to submit the report for publication.
RESULTS
We found 28 563 new users who were initially prescribed a long-acting anticholinergic and 17 840 who were initially prescribed a long-acting -agonist and met the validated case denition of COPD, on the basis of the health administrative data (Appendix Figure, available at www.annals.org). Compared with users of long-acting -agonists, users of long-acting anticholinergics were less likely to be female or have asthma and less likely to have lled a prescription for a respiratory antibiotic or oral corticosteroid in the previous year, seen a specialist, or received spirometry (Table 1).
Propensity Score Matching
The propensity score provided fair discrimination between treatment groups (c statistic 0.66). Matching produced 15 532 pairs. After matching, the long-acting antiwww.annals.org
The results from the standard proportional hazards regression models that included all of the covariates were similar to the results from the propensity score models (Table 2). The time-on-treatment analysis similarly found that patients currently receiving a long-acting anticholinergic were signicantly more likely to die than those currently receiving a long-acting -agonist, which conrms our main results (Table 3). Separate analyses of patients who had or had not received previous spirometry found no signicant difference in results between these subsets (Table 4). An initial prescription for long-acting -agonists was also consistently associated with improved survival when results were stratied by sex, asthma status, inhaled corticosteroid use, congestive heart failure, and number of prescriptions for COPD medication lled in the previous year (Appendix Table 4, available at www.annals.org). Of interest, stratication by number of prescriptions for COPD medication suggested that the relative survival of those initially treated with a long-acting -agonist, compared with a long-acting anticholinergic, may have been higher in patients with fewer prescriptions (hazard ratio for those with 0 to 2 prescriptions, 1.23 [CI, 1.13 to 1.33]; 3 to 10 prescriptions, 1.14 [CI, 1.06 to 1.23]; 10 prescriptions, 1.07 [CI, 1.00 to 1.15]). An initial prescription for a long-acting -agonist continued to be associated with better outcomes than one for a long-acting anticholinergic when the analysis was repeated, rst with the inclusion of outcomes that occurred within 7 days of the index prescription being lled and then with COPD hospitalizations and emergency department visits for COPD analyzed as separate outcomes
Original Research
Table 2. Adjusted Hazard Ratios for Death and Other Outcomes Associated With Initially Prescribed Inhaled Long-Acting Anticholinergics Compared With Long-Acting -Agonists
Outcome Long-Acting Anticholinergic Long-Acting -Agonist Absolute Difference in Outcomes at 3 Years (95% CI), % Propensity ScoreMatched Regression Standard Covariate-Adjusted Regression
Median Time to Had Outcome, Outcome (95% % CI), d* Primary and secondary outcomes Death Death or hospitalization for COPD Death or hospitalization for COPD or a related respiratory disease Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease** Death, hospitalization, or emergency department visit for COPD Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease** Sensitivity analysis Hospitalization for COPD Emergency department visit for COPD
Had Median Time to Outcome, Outcome (95% % CI), d*
Adjusted Hazard Ratio (95% CI)
Adjusted Hazard P P Value Ratio (95% CI) Value
39.9 55.0 56.8
834 (802862) 36.5 1224 (11901261) 50.5 1144 (11131176) 52.8
920 (891948) 6.3 (5.47.2) 1400 (13561446) 7.6 (6.68.5) 1283 (12421321) 6.9 (6.07.8)
1.14 (1.091.19) 0.001 1.13 (1.091.18) 0.001 1.11 (1.071.16) 0.001 1.13 (1.101.17) 0.001 1.09 (1.051.13) 0.001 1.11 (1.081.14) 0.001
62.1
946 (916976)
59.0
1034 (9981062)
5.9 (5.06.8)
1.07 (1.031.11) 0.001 1.08 (1.051.11) 0.001
61.0
991 (9611020)
56.4
1124 (10931155) 7.6 (6.78.5)
1.11 (1.071.16) 0.001 1.12 (1.091.16) 0.001
64.8
860 (835884)
61.5
925 (903953)
6.0 (5.06.9)
1.07 (1.031.10) 0.001 1.08 (1.051.11) 0.001
70.3
676 (656696)
67.8
726 (702752)
4.9 (4.05.8)
1.05 (1.021.09)
0.003 1.06 (1.031.09) 0.001
33.3 12.2
762 (728801)
30.7
867 (826895)
5.5 (4.56.4)
1.13 (1.051.16) 0.001 1.13 (1.091.17) 0.001 1.09 (1.00,1.17) 0.042 1.09 (1.021.17) 0.007
673 (621741) 11.7
752 (689817) 2.1 (1.42.8)
COPD chronic obstructive pulmonary disease. * Time until 50% of patients experienced the event. Estimated rates were derived from paired KaplanMeier curves. Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for membership in a matched pair. Members of the long-acting anticholinergic and long-acting -agonist cohorts were matched by using propensity score, age, sex, asthma status, and number of COPD medications. Testing the hypothesis that the hazard ratio is 1.00. Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for all variables. Time until 25% of patients experienced the event (25th percentile). Because fewer than one half of patients died or were hospitalized, the median time to these outcomes could not be estimated. Pneumonia, inuenza, or acute bronchitis. ** Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmia, or cerebrovascular disease. Time until 10% of patients experienced the event (10th percentile). Because so few patients visited the emergency department, the median time to this outcome could not be estimated.
(Table 2 and Appendix Table 5, available at www.annals .org). The rate ratio of a theoretical, unmeasured potential confounder would have to be at least 2.0, or the imbalance due to this confounder more than 15%, to reduce the
observed hazard ratio to 1.0. Likewise, if all patients who were misclassied as having COPD were in the long-acting -agonist group and were 30% less likely to die than those with true COPD, or if they were all in the long-acting anticholinergic group and were at least 30% more likely to
www.annals.org
Original Research
die than those with true COPD, the observed hazard ratio would be reduced to 1.0 (Appendix).
Figure. Observed survival, by initial medication prescribed.

100 95 90
DISCUSSION
We conducted an observational, longitudinal, population-based study of new use of long-acting anticholinergics and long-acting -agonists by older patients who met a validated case denition of COPD on the basis of health administrative data. Patients who were initially prescribed a long-acting anticholinergic seemed to have a 14% higher adjusted mortality rate than those initially prescribed a long-acting -agonist and were also more likely to be hospitalized or to visit an emergency department for COPD or a related condition. The incremental risk associated with long-acting anticholinergics seemed to be independent of patient sex or coexisting medical conditions or whether the diagnosis of COPD was conrmed with spirometry. Thus, our ndings suggest that long-acting -agonists may be more effective than long-acting anticholinergics at improving survival in older patients with COPD. To our knowledge, our observational study is the rst to directly compare mortality with long-acting -agonists and mortality with long-acting anticholinergics in a large population of older patients with COPD. A few small RCTs and meta-analyses that compared such outcomes as lung function or health status between patients receiving long-acting -agonists and those receiving long-acting anticholinergics have yielded conicting results. However, most found that both drug classes were equally effective in preventing exacerbations and hospitalizations (19 22). A larger RCT designed to examine the relative efcacy of long-acting -agonists versus longacting anticholinergics on exacerbations (but not mortality, as in our study) has been completed (23), but the results have not yet been released. Although RCTs remain the gold standard for the study of the efcacy of COPD medication, observational
Survival (95% CI), %
Long-acting -agonist Long-acting anticholinergic
85 80 75 70 65 60 55 50 45 0 1 2 3 4 5
Log-rank P < 0.001
Time From Index Date, y At risk, n 11 891 8401
Long-acting 15 532 anticholinergic Long-acting -agonist 15 532
13 603
5311
2114
13 780
12 245
8731
5577
2095
studies, such as ours, provide complementary information on medication effectiveness (which might differ from RCT results) for several reasons. First, RCTs often have limited generalizability because they exclude patients who are part of the population to which the therapy will be applied, such as those with severe disease or comorbidity (24). For example, our study examined a large, older, frailer populationmany of whom would have been excluded from previous RCTs. Second, RCTs may recruit patients who are not new users of medications (more than one half of the patients enrolled in the TORCH [TOwards a Revolution in COPD Health] trial [7], which studied the effectiveness of inhaled cortico-
Table 3. Adjusted Hazard Ratios for Death and Other Outcomes Associated With the On-Treatment Effect of Inhaled Long-Acting Anticholinergics Compared With Long-Acting -Agonists
Outcome Death Death or hospitalization for COPD Death or hospitalization for COPD or a related respiratory disease Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease Death, hospitalization, or emergency department visit for COPD Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease Adjusted Hazard Ratio (95% CI)* 1.10 (1.041.17) 1.05 (1.011.10) 1.02 (0.981.07) 1.02 (0.971.06) 1.05 (1.011.10) 1.00 (0.961.04) 1.00 (0.961.04)
P Value
0.002 0.030 0.35 0.46 0.026 0.96 0.87
COPD chronic obstructive pulmonary disease. * We treated time spent receiving long-acting anticholinergics or long-acting -agonists as time-dependent covariates. The hazard ratio compares the risk of patients currently receiving a long-acting anticholinergic with that of patients currently receiving a long-acting -agonist. Hazard ratios are adjusted for the covariates used to calculate the propensity score. Testing the hypothesis that the hazard ratio is 1.00. Pneumonia, inuenza, or acute bronchitis. Congestive heart failure, ischemic heart disease (including acute myocardial infarction), cardiac arrhythmia, or cerebrovascular disease.
www.annals.org 3 May 2011 Annals of Internal Medicine Volume 154 Number 9 589
Original Research
Table 4. Adjusted Hazard Ratios for Death and Other Outcomes Associated With Initially Prescribed Inhaled Long-Acting Anticholinergics Compared With Long-Acting -Agonists
Outcome Previously Received Spirometry (11 193 Matched Pairs) Did Not Previously Receive Spirometry (3987 Matched Pairs) Adjusted Hazard Ratio (95% CI)* 1.15 (1.061.25) 1.09 (1.011.17) 1.07 (1.001.15) 1.03 (0.961.11) 1.09 (1.021.17) 1.04 (0.971.11) Relative Difference
Adjusted Hazard Ratio (95% CI)* Death Death or hospitalization for COPD Death or hospitalization for COPD or a related respiratory disease Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease Death, hospitalization, or emergency department visit for COPD Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease 1.09 (1.041.15) 1.11 (1.061.16) 1.09 (1.041.14) 1.05 (1.011.10) 1.10 (1.051.15) 1.05 (1.001.10)
P Value
0.001 0.001 0.001 0.017 0.001 0.039
P Value
0.001 0.025 0.054 0.84 0.017 0.27
Ratio of Hazard Ratios (95% CI) 0.95 (0.861.05) 1.02 (0.941.11) 1.01 (0.931.10) 1.02 (0.941.11) 1.01 (0.931.10) 1.01 (0.931.09)
P Value
0.31 0.60 0.75 0.63 0.82 0.88
1.03 (0.991.07)
0.171
1.01 (0.941.08)
0.80
1.02 (0.941.10)
0.62
COPD chronic obstructive pulmonary disease. * Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for membership in a matched pair. Members of the long-acting anticholinergic and long-acting -agonist cohorts were matched by using propensity score, age, sex, asthma status, and number of COPD medications. Testing the hypothesis that the hazard ratio is 1.00. Testing the hypothesis that the ratio of the hazard ratios is 1.00. Pneumonia, inuenza, or acute bronchitis. Congestive heart failure, ischemic heart disease (including acute myocardial infarction), cardiac arrhythmia, or cerebrovascular disease.
steroids and long-acting -agonists, had used at least 1 of these medications in the year before study entry) and who, because they tolerated the regimen and volunteered to receive it in a study, may be more likely to have favorable outcomes. To avoid this, we examined only new users of medication. Finally, RCTs are often not large enough or long enough to look at relatively rare events, such as mortality. Our study followed tens of thousands of patients for up to 5.5 years. Our study suggests a relative difference in mortality between patients initially prescribed long-acting anticholinergics and those prescribed long-acting -agonists, but it was not designed to determine the reasons why. One possibility is that both types of drug effectively reduce mortality, but long-acting -agonists do so more than long-acting anticholinergics. A second possibility is that long-acting -agonists do not or only marginally reduce mortality, whereas long-acting anticholinergics increase mortality. The latter hypothesis is supported by several studies and meta-analyses (2530), which found an increase in allcause and cardiovascular mortality associated with inhaled anticholinergics but not with inhaled -agonists. However, few of these studies focused specically on long-acting (as opposed to short-acting) medications. In contrast, a recent meta-analysis (6, 31), which included the large UPLIFT (Understanding Potential Long-term Impacts on Function with Tiotropium) study, found no evidence of increased mortality associated with long-acting anticholinergics; however, this work has the same limitations as the RCTs it
was based on, most notably limited generalizability and the inclusion of patients who had previously received the study medications. Our study has limitations. First, a drawback of all observational studies is potential confounding by indication or by disease severity. Because our data did not contain a precise measure of disease severity (such as lung function), we cannot be sure that such confounding did not occur. However, we consider confounding unlikely because we controlled for many prognostically important variables. As further assurance that an unmeasured confounder, such as smoking, was accounted for, a propensity score calibration analysis was used to conrm the results. In addition, a potential unmeasured confounder would have to have either a much greater association with death than that of smoking or a greater imbalance between the cohorts than that seen for any of the variables except asthma to negate our results (32). Such a confounder also could not be strongly correlated with any of the other variables (such as asthma) already adjusted for in the analysis. Second, cases of other respiratory diseases, most likely asthma, could have been misclassied as COPD. In an extreme scenario in which all such misclassication occurred in the long-acting -agonist group, patients incorrectly classied as having COPD would have had to be about 30% less likely to die than those with true COPD to explain our results. Because all-cause mortality has been estimated to be 27% lower in patients aged 65 years or older with asthma than in patients of the same age with
www.annals.org
Original Research
COPD (2, 33), misclassication could explain our observed results. In a second extreme scenario, in which misclassication only occurred in the long-acting anticholinergic group, patients misclassied as having COPD would have had to be about 30% more likely to die than those with true COPD to explain our results. If the misclassied patients had asthma, mismanagement of their disease could have increased their risk for death. However, this is unlikely to have caused a 30% increase in mortality in practice, especially because a large proportion of patients receiving a long-acting anticholinergic were also receiving asthma medications, and long-acting anticholinergics seem to benet patients with asthma (34). Because the medication groups were matched on a codiagnosis of asthma, neither extreme scenario is likely. Finally, patients were not completely adherent to their index medication regimen and often received the other study medication, a situation that is common in the real world. However, the results of a time-varying analysis, which looked only at current use of medication, were consistent with our main results and, on average, patients spent more time taking the initially prescribed drug than taking the alternative (Appendix). In summary, we conducted a population-based, retrospective, observational study of new use of long-acting anticholinergics and long-acting -agonists in older patients who met a validated case denition of COPD on the basis of health administrative data. We found that patients initially prescribed a long-acting anticholinergic seemed to have a 14% higher adjusted mortality rate than those initially prescribed a long-acting -agonist. This suggests that long-acting -agonists might be a better initial therapy for patients with moderate to severe COPD. Future research is needed to conrm these ndings in RCTs and in younger patients. Research is also needed to examine the relative safety proles of these medications so that their risk benet ratios can be compared.
From the Institute for Clinical Evaluative Sciences, Sunnybrook Health Sciences Centre, The Hospital for Sick Children, and University of Toronto, Toronto, Ontario, Canada.
Disclaimer: The opinions, results, and conclusions reported in this paper are those of the authors and are independent of the funding sources. No endorsement by the Institute for Clinical Evaluative Sciences or the Ontario Ministry of Health and Long-Term Care is intended or should be inferred. Acknowledgment: The authors thank Brogan, Ottawa, Ontario, Canada, for the use of their Drug Product and Therapeutic Class Database. Grant Support: By the Government of Ontario; a Career Scientist
by an annual grant from the Ontario Ministry of Health and Long-Term Care.
Potential Conflicts of Interest: Disclosures can be viewed at www.acponline
.org/authors/icmje/ConictOfInterestForms.do?msNumM10-1643.
Reproducible Research Statement: Study protocol and statistical code:
Available from Ms. Croxford (e-mail, ruth.croxford@ices.on.ca). Data set: Not available.
Requests for Single Reprints: Andrea Gershon, MD, MSc, Institute for Clinical Evaluative Sciences, G106, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada; e-mail, andrea.gershon@ices.on.ca.
Current author addresses and author contributions are available at www .annals.org.
References
1. Buist AS, McBurnie MA, Vollmer WM, Gillespie S, Burney P, Mannino DM, et al; BOLD Collaborative Research Group. International variation in the prevalence of COPD (the BOLD Study): a population-based prevalence study. Lancet. 2007;370:741-50. [PMID: 17765523] 2. Gershon AS, Wang C, Wilton AS, Raut R, To T. Trends in chronic obstructive pulmonary disease prevalence, incidence, and mortality in ontario, Canada, 1996 to 2007: a population-based study. Arch Intern Med. 2010;170:560-5. [PMID: 20308643] 3. Jemal A, Ward E, Hao Y, Thun M. Trends in the leading causes of death in the United States, 1970-2002. JAMA. 2005;294:1255-9. [PMID: 16160134] 4. Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med. 2006;3:e442. [PMID: 17132052] 5. Global Initiative for Chronic Obstructive Lung Disease. Global Strategy for the Diagnosis, Management, and Prevention of COPD. Global Initiative for Chronic Obstructive Lung Disease; 2008. 6. Tashkin DP, Celli B, Senn S, Burkhart D, Kesten S, Menjoge S, et al; UPLIFT Study Investigators. A 4-year trial of tiotropium in chronic obstructive pulmonary disease. N Engl J Med. 2008;359:1543-54. [PMID: 18836213] 7. Calverley PM, Anderson JA, Celli B, Ferguson GT, Jenkins C, Jones PW, et al; TORCH investigators. Salmeterol and uticasone propionate and survival in chronic obstructive pulmonary disease. N Engl J Med. 2007;356:775-89. [PMID: 17314337] 8. Hux JE, Naylor CD, Fielding DA. The Ontario Drug Benet Program Copayment: Its Impact on Access for Ontario Seniors and Charges to the Program. Toronto: Institute for Clinical Evaluative Sciences; 1997. 9. Gershon AS, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. Identifying individuals with physcian diagnosed COPD in health administrative databases. COPD. 2009;6:388-94. [PMID: 19863368] 10. Joo MJ, Au DH, Lee TA. Use of spirometry in the diagnosis of chronic obstructive pulmonary disease and efforts to improve quality of care. Transl Res. 2009;154:103-10. [PMID: 19665686] 11. Wilkins, R. PCCF Version 4D Users Guide (Geocodes/PCCF). Automated Geographic Coding Based on the Statistics Canada Postal Code Conversion Files, Including Postal Codes to December 2003. Ottawa: Statistics Canada; 2004. 12. Reid RJ, MacWilliam L, Verhulst L, Roos N, Atkinson M. Performance of the ACG case-mix system in two Canadian provinces. Med Care. 2001;39:86-99. [PMID: 11176546] 13. The Johns Hopkins University ACG Case-Mix System, version 8.0. Baltimore: Johns Hopkins Univ; 2006. 14. Durham MC. Tiotropium (Spiriva): a once-daily inhaled anticholinergic medication for chronic obstructive pulmonary disease. Proc (Bayl Univ Med Cent). 2004;17:366-73. [PMID: 16200123] 15. DAgostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17: 2265-81. [PMID: 9802183] 16. Austin PC, Grootendorst P, Anderson GM. A comparison of the ability of different propensity score models to balance measured variables between treated
Award from the Ontario Ministry of Health and Long-Term Care and a Research Fellowship from the Canadian Institutes of Health Research, Institute of Population and Public Health, and The Public Health Agency of Canada (Dr. Gershon); and the Dales Award in Medical Research from the University of Toronto, (Dr. To). Support was also provide by the Institute for Clinical Evaluative Sciences, which is funded
www.annals.org
Original Research

Ann Intern Med. 2008;149:380-90. [PMID: 18794557] 26. Boehringer Ingelheim. Tiotropium (Spiriva) Respimat: Evaluation of Fatal EventsFebuary 2010. Accessed at http://trials.boehringer-ingelheim.com/res /trial/data/pdf/Pooled_Analysis_U10-3255-01.pdf on 11 March 2011. 27. Ogale SS, Lee TA, Au DH, Boudreau DM, Sullivan SD. Cardiovascular events associated with ipratropium bromide in COPD. Chest. 2010;137:13-9. [PMID: 19363211] 28. Anthonisen NR, Connett JE, Enright PL, Manfreda J; Lung Health Study Research Group. Hospitalizations and mortality in the Lung Health Study. Am J Respir Crit Care Med. 2002;166:333-9. [PMID: 12153966] 29. Rodrigo GJ, Nannini LJ, Rodr guez-Roisin R. Safety of long-acting betaagonists in stable COPD: a systematic review. Chest. 2008;133:1079-87. [PMID: 18460518] 30. Singh S, Loke YK, Furberg CD. Inhaled anticholinergics and risk of major adverse cardiovascular events in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis. JAMA. 2008;300:1439-50. [PMID: 18812535] 31. Celli B, Decramer M, Leimer I, Vogel U, Kesten S, Tashkin DP. Cardiovascular safety of tiotropium in patients with COPD. Chest. 2010;137:20-30. [PMID: 19592475] 32. Critchley JA, Capewell S. Mortality risk reduction associated with smoking cessation in patients with coronary heart disease: a systematic review. JAMA. 2003;290:86-97. [PMID: 12837716] 33. Gershon AS, Guan J, Wang C, To T. Trends in asthma prevalence and incidence in ontario, Canada, 1996 2005: a population study. Am J Epidemiol. 2010;172:728-36. [PMID: 20716702] 34. Peters SP, Kunselman SJ, Icitovic N, Moore WC, Pascual R, Ameredes BT, et al; National Heart, Lung, and Blood Institute Asthma Clinical Research Network. Tiotropium bromide step-up therapy for adults with uncontrolled asthma. N Engl J Med. 2010;363:1715-26. [PMID: 20979471]
and untreated subjects: a Monte Carlo study. Stat Med. 2007;26:734-53. [PMID: 16708349] 17. Stu rmer T, Schneeweiss S, Avorn J, Glynn RJ. Adjusting effect estimates for unmeasured confounding with validation data using propensity score calibration. Am J Epidemiol. 2005;162:279-89. [PMID: 15987725] 18. Schneeweiss S. Sensitivity analysis and external adjustment for unmeasured confounders in epidemiologic database studies of therapeutics. Pharmacoepidemiol Drug Saf. 2006;15:291-303. [PMID: 16447304] 19. Barr RG, Bourbeau J, Camargo CA, Ram FS. Inhaled tiotropium for stable chronic obstructive pulmonary disease. Cochrane Database Syst Rev. 2005: CD002876. [PMID: 15846642] 20. Puhan MA, Bachmann LM, Kleijnen J, Ter Riet G, Kessels AG. Inhaled drugs to reduce exacerbations in patients with chronic obstructive pulmonary disease: a network meta-analysis. BMC Med. 2009;7:2. [PMID: 19144173] 21. Baker WL, Baker EL, Coleman CI. Pharmacologic treatments for chronic obstructive pulmonary disease: a mixed-treatment comparison meta-analysis. Pharmacotherapy. 2009;29:891-905. [PMID: 19637942] 22. Salpeter SR, Buckley NS. Systematic review of clinical outcomes in chronic obstructive pulmonary disease: beta-agonist use compared with anticholinergics and inhaled corticosteroids. Clin Rev Allergy Immunol. 2006;31:219-30. [PMID: 17085795] 23. Boehringer Ingelheim. Tiotropium Once Daily 18 Mcg Versus Salmeterol Twice Daily 50 Mcg on Time to First Exacerbation in COPD Patients [Clinical trial]. ClinicalTrials.gov registration number: NCT00563381. Accessed at www .clinicaltrials.gov/ct2/show/NCT00563381?termNCT00563381&rank1 on 11 March 2011. 24. Travers J, Marsh S, Caldwell B, Williams M, Aldington S, Weatherall M, et al. External validity of randomized controlled trials in COPD. Respir Med. 2007;101:1313-20. [PMID: 17113277] 25. Lee TA, Pickard AS, Au DH, Bartle B, Weiss KB. Risk for death associated with medications for recently diagnosed chronic obstructive pulmonary disease.
VISIT
THE
ANNALS BOOTH
AT
SUBSPECIALTY MEETINGS
Annals staff will be at these upcoming meetings. Digestive Disease Week, Chicago, 8 10 May 2011 American Thoracic Society, Denver, 1318 May 2011 American Society of Clinical Oncology, Chicago, 3 6 June 2011 American Diabetes Association, San Diego, 24 27 June 2011 Stop by the ACP/Annals booth and register to be a peer reviewer or discuss your thoughts for submissions or topic coverage with Annals staff.
www.annals.org
Annals of Internal Medicine

Current Author Addresses: Drs. Gershon and Stukel, Ms. Croxford, and Ms. Sanchez-Romeu: Institute for Clinical Evaluative Sciences, G106, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada. Dr. To: Child Health Evaluative Sciences, Research Institute, The Hospital for Sick Children, 555 University Avenue, Toronto, Ontario M5G 1X8, Canada. Dr. Stanbrook: Asthma & Airway Centre, Toronto Western Hospital, 7th Floor, East Wing, 399 Bathurst Street, Toronto, Ontario M5T 2S8, Canada. Dr. Upshur: Sunnybrook Health Sciences Centre, 2075 Bayview Avenue, Room A100, Toronto, Ontario M4N 3M5, Canada. Author Contributions: Conception and design: A. Gershon, R. Croxford, M.B. Stanbrook, R. Upshur, T. Stukel. Analysis and interpretation of the data: A. Gershon, R. Croxford, M.B. Stanbrook, R. Upshur, P. Sanchez-Romeu, T. Stukel. Drafting of the article: A. Gershon, R. Croxford, R. Upshur, P. SanchezRomeu. Critical revision of the article for important intellectual content: A. Gershon, R. Croxford, T. To, M.B. Stanbrook, R. Upshur, T. Stukel. Final approval of the article: A. Gershon, R. Croxford, T. To, M.B. Stanbrook, R. Upshur, P. Sanchez-Romeu, T. Stukel. Statistical expertise: R. Croxford, T. To, P. Sanchez-Romeu, T. Stukel. Obtaining of funding: A. Gershon. Administrative, technical, or logistic support: R. Croxford. Collection and assembly of data: A. Gershon, R. Croxford, P. SanchezRomeu.
quacy of the model used to estimate the propensity score. A properly constructed propensity score ensures that the 2 matched treatment groups are comparable with respect to all of the measured covariates. Appendix Table 2 shows the standardized between-group differences and P values before and after matching. After matching, all standardized differences were less than 10%, and all P values were far removed from statistical signicance, which indicates good balance. Cox proportional hazards survival analysis was used to compare outcomes for patients in the long-acting anticholinergic treatment cohort with those in the long-acting -agonist cohort, taking membership in a matched pair into account. Our primary analysis found improved outcomes among patients initially treated with a long-acting -agonist compared with those initially treated with a long-acting anticholinergic (Table 2). To supply context for the reported hazard ratios, Table 2 also presents the absolute difference in survival at 3 years. Appendix Table 3 contains more complete information, showing the percentage of patients in the group initially treated with long-acting -agonists who survived to the end of each of the 5 years, and the survival difference between the treatment groups. The percentage of patients who survived in the longacting anticholinergic cohort can be calculated as the percentage who survived in the long-acting -agonist cohort minus the difference in survival. Calibrated Propensity Score Analysis As a check that variables not available in the health administrative data were balanced between the treatment groups, we performed a sensitivity analysis by using propensity score calibration (17). A representative subcohort of 694 patients (250 from the long-acting -agonist cohort and 444 from the long-acting anticholinergic cohort) participated in the Canadian Community Health Survey and had data on smoking status, including exposure to secondhand smoke, body mass index, immigration status, and self-reported health, to augment the data from their original health administrative variables. A gold standard propensity score was calculated for these patients, on the basis of their original data plus the additional data. This gold standard score was compared with their previous propensity score, using ordinary leastsquares regression, to obtain the relationship between the 2 scores. The resulting equation was then used to transform the propensity scores of all study patients (not just those with additional data) into gold standard scores. Finally, the matching and outcome analyses were repeated, as with the primary analysis. Results of the calibrated propensity score analysis were consistent with the main analysis (Appendix Table 6). Additional Sensitivity Analyses We performed several additional analyses to check the robustness of the results. Appendix Tables 4 to 8 provide supporting information and results from these analyses. The analyses were repeated by using a standard proportional hazards regression that adjusted for all covariates rather than the matched propensity score analysis (Table 2). Appendix Table 5 presents the hazard ratios for each covariate in this analysis.
3 May 2011 Annals of Internal Medicine Volume 154 Number 9 W-203
APPENDIX: EFFECTIVENESS OF LONG-ACTING INHALED -AGONISTS AND ANTICHOLINERGICS IN OLDER PATIENTS WITH COPD
Cohort Assembly The Appendix Figure shows how patients in the administrative health databases were selected for inclusion in our study. Propensity Score Analysis We used propensity score methodology to control for confounding by baseline cohort characteristics. The propensity score reects the probability that a given patient was initially prescribed a long-acting anticholinergic, given that patients particular pattern of baseline covariates, and was calculated by using logistic regression. A set of 57 covariates, which covered patient demographic characteristics, COPD severity, and overall health, were used to calculate the propensity score. Appendix Table 1 shows the variables used and the associated odds ratios. One-to-one matching between patients in the long-acting anticholinergic and long-acting -agonist cohorts was based on the propensity score and was made only if both scores agreed to within a caliper of 0.2 times the SD of the scores. Matching on the propensity score allows unbiased estimation of the treatment effect, to the extent that unmeasured confounders are correlated with the measured covariates. Matching also restricts the analysis to those who are eligible to receive either treatment. Patients were also matched on the basis of age (1 year), sex, asthma status, and number of COPD medication prescriptions lled in the previous year. Each matched pair was unique; data from unmatched patients were excluded. After matching, the 2 treatment groups were compared with respect to each of the measured covariates to determine the adewww.annals.org
Appendix Table 6 presents the results of the propensity scorematched analysis, stratied by sex, asthma status, number of COPD prescriptions lled in the previous year, inhaled corticosteroid use, and a diagnosis of congestive heart failure. In each stratied analysis, long-acting -agonists continued to be associated with improved survival. Appendix Table 7 provides the results obtained when all analyses (calculation of the propensity score, matching, and the matched proportional hazards regression) were repeated with the inclusion of outcomes that occurred within 7 days of the index prescription being lled. Including these outcomes did not signicantly change the study results. A time-on-treatment analysis of the effect of medication exposure conrmed the primary ndings of better survival in patients who were initially prescribed a long-acting -agonist (Table 4). Appendix Table 8 provides information on the number of person-years and percentage of total person-years spent receiving a long-acting -agonist or long-acting anticholinergic in each of the treatment groups in each year of follow-up. It also provides information about medication crossover and inhaled corticosteroid use during the follow-up period. Adherence to the index medication was not optimal in either medication group, and patients commonly received the nonindex medication. Patients who received a long-acting anticholinergic were more adherent and less likely to have received a long-acting -agonist during the follow-up period than vice versa. They were also less likely to have received an inhaled corticosteroid. Accounting for Potential Misclassification The validated case denition of COPD based on health administrative data had a positive predictive value of 0.86 when applied to patients aged 65 years or older (9). This means that an estimated 14% of those included in our analysis may not have had COPD. The inclusion of misclassied patients in the cohort could have led to incorrect results in 2 ways. First, the misclassied patients could have been more (or less) healthy than those
correctly identied and could have been concentrated in 1 of the 2 treatment groups, thus contributing to better (or worse) outcomes in that group. Second, misclassication could have resulted in an apparent increase in the precision of results, thus decreasing the P value and narrowing the CI for the estimated hazard ratios. To determine how each of these might have inuenced the results, we estimated the true hazard ratio and its CI under a range of assumptions about the health of patients who were misclassied relative to those who truly had COPD and the effect of misclassication on the precision of the SE. Appendix Table 9 shows that our results could be explained if misclassication resulted in sufciently healthier patients in the long-acting -agonist group (for example, because they had asthma, which commonly has lower mortality than COPD) or sufciently sicker patients in the long-acting anticholinergic group (for example, because their true disease was not being recognized or treated properly). Under the assumptions that all patients misclassied as having COPD were in the long-acting -agonist group and that they were 30% less likely to die (1.0 0.70 0.3, or 30%) than those with true COPD, the true hazard ratio would be 1.04 (CI, 1.00 to 1.09). Although the point estimate for the hazard ratio is still greater than 1.0, the lower bound of the CI is compatible with no drug difference (the P value for the comparison would no longer be statistically signicant). Likewise, if all misclassied patients were in the long-acting anticholinergic group and were 30% more likely to die, the lower bound of the CI for the true hazard ratio would be 1.00. The assumption that all misclassied patients are in the same treatment group is conservative because increased inequality in the distribution of the misclassied patients leads to lower true hazard ratios. Finally, placing all patients who were misclassied in the long-acting -agonist group and assuming a 20% reduction in the risk for death meant that misclassication would have had to decrease the observed SE of the estimate by more than 50% to widen the true CI to include a value of 1.0.
W-204 3 May 2011 Annals of Internal Medicine Volume 154 Number 9
www.annals.org
Appendix Table 1. Variables Used to Calculate the Propensity Score*

Variable Odds Ratio of Being Prescribed a LongActing Anticholinergic Instead of a Long-Acting -Agonist (95% CI) 2.65 (1.813.89)
P Value
Intercept Demographic characteristics Age Male sex Resident in a long-term care facility Neighborhood income quintile 1 (low) 2 3 4 5 (high) Urban or rural residence COPD and general care COPD duration 1 y 15 y 5 y Previous spirometry Number of primary care physician visits in the previous year Specialist visit in previous year Influenza vaccination in previous year Previous lung volume reduction surgery Number of prescriptions for COPD medications filled in the previous year Medication use in previous year Respiratory antibiotic Short-acting -agonist Short-acting anticholinergic Inhaled corticosteroid Oral corticosteroid Long-term oral corticosteroid use Methylxanthine Statin ACE inhibitor -Blocker Number of classes of medications prescribed Comorbidity** Acute minor Acute major Likely to recur Chronic medical unstable Chronic medical stable Chronic specialty unstable Chronic specialty stable Eye or dental Psychosocial Prevention, administration Previous or coexisting medical conditions Asthma ever diagnosed Active asthma Pneumonia, influenza, or acute bronchitis Cor pulmonale Lung cancer Pulmonary embolism Arrhythmias Acute myocardial infarction Other ischemic heart disease Congestive heart failure Cerebrovascular disease Hypertension Fragility fracture due to osteoporosis Diabetes Any cancer (other than lung cancer) Mental health visit (in previous year)
0.0001 0.045 0.001 0.001 0.32 0.27 0.90 0.86 0.21
1.00 (1.001.01) 1.20 (1.161.25) 1.16 (1.071.26) 1.00 (reference) 1.03 (0.971.09) 1.03 (0.971.10) 1.00 (0.941.07) 0.99 (0.931.06) 0.96 (0.911.02)
1.00 (reference) 0.94 (0.871.01) 0.97 (0.901.04) 0.96 (0.921.01) 1.00 (1.001.01) 0.90 (0.860.93) 1.02 (0.981.06) 1.13 (0.941.53) 1.01 (1.011.10)
0.094 0.42 0.086 0.003 0.001 0.28 0.43 0.001
0.89 (0.840.95) 1.01 (0.961.06) 1.37 (1.311.44) 0.98 (0.941.03) 0.86 (0.810.90) 1.00 (0.941.06) 0.94 (0.851.04) 1.04 (0.991.08) 1.11 (1.061.16) 1.07 (1.011.12) 0.98 (0.970.98) 0.87 (0.800.94) 1.02 (0.961.09) 1.00 (0.961.05) 1.56 (1.451.68) 1.01 (0.951.07) 0.96 (0.911.01) 0.95 (0.891.02) 1.02 (0.981.07) 1.06 (1.021.11) 1.00 (0.961.04) 0.64 (0.610.67) 0.64 (0.610.67) 0.97 (0.931.01) 1.21 (0.901.63) 1.06 (0.991.14) 1.00 (0.911.10) 0.99 (0.951.04) 0.98 (0.941.03) 0.93 (0.890.98) 1.03 (0.981.08) 1.12 (1.041.20) 0.96 (0.901.03) 1.07 (0.981.17) 0.96 (0.921.01) 0.95 (0.910.99) 0.95 (0.881.02)
0.001 0.80 0.001 0.44 0.001 0.92 0.24 0.15 0.001 0.014 0.001 0.001 0.56 0.91 0.001 0.83 0.083 0.15 0.33 0.007 0.94 0.001 0.001 0.16 0.20 0.10 0.97 0.72 0.45 0.005 0.20 0.002 0.24 0.16 0.10 0.022 0.17
www.annals.org
Appendix Table 1Continued

Variable Odds Ratio of Being Prescribed a LongActing Anticholinergic Instead of a Long-Acting -Agonist (95% CI)
P Value
Acute events before index date Most recent hospitalization for COPD In the past 6 mo 6 mo before index date Never Most recent hospitalization for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent hospitalization for cardiovascular cause In the past 6 mo 6 mo before index date Never Most recent emergency department visit for COPD In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a cardiovascular cause In the past 6 mo 6 mo before index date Never
1.00 (reference) 0.98 (0.901.08) 0.93 (0.861.00) 1.00 (reference) 1.06 (0.911.24) 1.02 (0.881.18) 1.00 (reference) 0.92 (0.941.01) 0.91 (0.821.00) 1.00 (reference) 1.04 (0.941.15) 0.89 (0.810.97)
0.71 0.055 0.42 0.83 0.098 0.061 0.41 0.010
1.00 (reference) 1.10 (0.991.23) 1.15 (1.031.27) 1.00 (reference) 0.92 (0.811.05) 0.96 (0.851.08)
0.072 0.009 0.21 0.48
ACE angiotensin-converting enzyme; COPD chronic obstructive pulmonary disease. * Patients were also categorized by the area of Ontario in which they lived. Odds ratio from the logistic regression model predicting the probability that a patients prescription was for a long-acting anticholinergic rather than a long-acting -agonist. Testing the hypothesis that the odds ratio is 1.00. Reference category is rural. In Ontario, primary care is provided by family and general physicians, and specialist COPD care is usually provided by pulmonologists, general internists, or geriatricians. Number of prescriptions for long- or short-acting -agonists, long- or short-acting anticholinergics, inhaled corticosteroids, oral corticosteroids, methylxanthine, or respiratory antibiotics. ** As indicated by Johns Hopkins Collapsed Ambulatory Diagnostic Groups (13). Pneumonia, inuenza, or acute bronchitis. Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmia, or cerebrovascular disease.
www.annals.org
Appendix Figure. Study flow diagram.
Residents of Ontario, Canada, aged 66 y between 1 July 2003 and 31 March 2007 (n = 1 834 953)
Met case definition of COPD, on the basis of health administration data, by 31 March 2007 (n = 233 449)
Filled prescription for long-acting anticholinergic or -agonist (n = 99 429)
Excluded (n = 53 026) Medication use: 51 085 Received prescriptions for both longacting anticholinergic and long-acting -agonist on the index date: 10 246 Filled a prescription for long-acting anticholinergic or -agonist in the previous 12 mo: 40 839 Insufficient information on covariates: 821 Did not have 2 y of history in health administration databases before index date: 170 Missing socioeconomic data: 651 Had outcome within 7 d of index prescription: 1120
Included in analysis (n = 46 403)
Newly received a long-acting anticholinergic (n = 28 563)
Newly received a long-acting -agonist (n = 17 840)
COPD chronic obstructive pulmonary disease.
www.annals.org
Appendix Table 2. Baseline Characteristics, by Initially Prescribed Medication, Before and After Propensity Score Matching*
P Value
P Value
Long-Acting Anticholinergic Patients, n Demographic characteristics Mean age (SD), y Women, % Resident in a long-term care facility, % Neighborhood income quintile, % 1 (low) 2 3 4 5 (high) Urban residence, % COPD and general care COPD duration, % 1 y 15 y 5 y Previous spirometry, % Median primary care physician visits in previous year (interquartile range), n Specialist visit in previous year, % Influenza vaccination in previous year, % Previous lung volume reduction surgery, % Median prescriptions for COPD medications filled in the previous year (interquartile range), n Medication use in previous year Respiratory antibiotic, % Short-acting -agonist, % Short-acting anticholinergic, % Inhaled corticosteroid, % Oral corticosteroid, % Long-term oral corticosteroid, % Methylxanthine, % Statin, % ACE inhibitor, % -Blocker, % Mean classes of medications prescribed (SD), n Comorbidity, % Acute minor Acute major Likely to recur Chronic medical unstable Chronic medical stable Chronic specialty unstable Chronic specialty stable Eye or dental Psychosocial Prevention, administration 28 563
Long-Acting -Agonist 17 840
Long-Acting Anticholinergic 15 532
Long-Acting -Agonist 15 532
77.2 (6.8) 46.2 11.3
76.8 (7.0) 53.1 8.3
6.0 13.7 9.6 0.3
0.001 0.001 0.001 0.77
76.9 (6.9) 50.8 9.3
76.9 (6.9) 50.8 9.0
0.0 0.0 0.8 0.9
0.98 1.00 0.50 0.43
25.1 23.0 19.5 17.3 15.0 81.2
25.5 22.8 19.0 17.3 15.5 85.2
10.7
0.001
25.6 23.4 19.0 17.1 15.0 83.7
25.5 23.0 19.1 17.1 15.3 83.9
0.6
0.60
10.3 22.5 67.3 69.7 9 (514)
9.3 22.0 68.8 74.3 9 (514)
3.4 1.1 3.2 10.2 1.5
0.001 0.23 0.001 0.001 0.001 0.001 0.61 0.28 0.001
9.8 21.8 68.3 72.4 9 (514)
9.6 22.4 68.0 73.0 9 (514)
0.8 1.3 0.7 1.3 0.2
0.50 0.25 0.55 0.24 0.89
44.7 42.0 0.5 6 (114)
50.5 42.2 0.4 5 (113)
11.6 0.5 1.0 7.2
49.0 41.9 0.5 5 (113)
49.1 42.5 0.4 5 (114)
0.2 1.2 0.5 0.7
0.87 0.31 0.67 0.40
64.6 72.5 54.3 48.3 22.9 3.3 4.1 37.3 43.2 25.9 11.6 (5.1)
70.2 71.6 45.7 52.1 27.1 3.4 4.7 38.0 40.4 24.1 12.1 (5.2)
11.9 1.9 17.4 7.7 9.8 0.6 3.3 2.1 5.6 4.0 9.9
0.001 0.048 0.001 0.001 0.001 0.54 0.001 0.030 0.001 0.001 0.001
68.9 72.2 49.6 51.1 25.0 3.3 4.7 38.3 41.5 24.7 12.0 (5.2)
68.9 71.9 49.3 51.3 26.0 3.3 4.7 38.3 41.7 24.9 12.0 (5.2)
0.0 0.7 0.6 0.3 0.2 0.1 0.4 0.1 0.4 0.4 0.4
1.00 0.52 0.59 0.78 0.87 0.92 0.73 0.96 0.70 0.74 0.71
91.1 87.7 73.0 93.8 84.9 19.5 8.1 28.1 44.2 46.0
93.3 88.9 75.7 89.9 86.7 21.7 9.5 29.5 44.0 45.7
8.1 3.5 6.2 14.5 5.1 5.4 5.1 3.2 0.5 0.7
0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.59 0.48
92.7 88.8 74.8 92.1 86.2 21.2 8.9 29.2 44.4 46.0
92.9 88.8 75.1 92.0 86.3 20.9 8.9 29.1 44.0 45.7
0.7 0.0 0.1 0.4 0.2 0.6 0.1 0.1 0.7 0.5
0.57 0.94 0.56 0.72 0.83 0.59 0.97 0.95 0.55 0.64
www.annals.org

P Value
P Value
Long-Acting Anticholinergic Previous or coexisting medical conditions, % Asthma ever diagnosed Active asthma Pneumonia, influenza, or acute bronchitis Cor pulmonale Lung cancer Pulmonary embolism Arrhythmias Acute myocardial infarction Other ischemic heart disease Congestive heart failure Cerebrovascular disease Hypertension Fragility fracture due to osteoporosis Diabetes Any cancer (other than lung cancer) Mental health visit in previous year Acute events before index date, % Most recent hospitalization for COPD In the past 6 mo 6 mo before index date Never Most recent hospitalization for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent hospitalization for cardiovascular cause** In the past 6 mo 6 mo before index date Never Most recent emergency department visit for COPD In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a COPD-related condition In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a cardiovascular cause** In the past 6 mo 6 mo before index date Never
Long-Acting -Agonist
Long-Acting Anticholinergic
Long-Acting -Agonist
37.6 23.3 35.0 0.6 8.6 4.5 39.6 53.3 48.9 40.2 11.7 80.1 5.9 24.9 32.7 7.5
55.4 39.4 37.0 0.4 8.0 4.6 39.6 53.5 50.9 38.2 9.6 81.2 5.1 26.5 33.7 8.2
36.3 35.9 4.1 2.3 2.1 0.9 0.1 0.5 4.0 4.1 6.6 2.8 3.5 3.7 2.2 2.7
0.001 0.001 0.001 0.016 0.031 0.37 0.95 0.59 0.001 0.001 0.001 0.003 0.001 0.001 0.021 0.005
49.7 33.3 36.2 0.5 8.3 4.6 39.4 53.8 50.5 39.0 10.2 80.5 5.3 26.4 33.4 8.0
49.7 33.2 36.5 0.4 8.4 4.7 39.9 53.7 50.5 39.2 10.2 80.9 5.3 26.2 33.9 8.0
0.0 0.2 0.7 0.5 0.5 0.4 1.0 0.3 0.0 0.4 0.0 0.9 0.0 0.5 1.0 0.1
1.00 0.85 0.55 0.67 0.65 0.75 0.39 0.80 1.00 0.75 1.00 0.42 1.00 0.67 0.30 0.92
8.0 16.5 75.5 1.8 15.6 82.6 6.0 30.3 63.7
7.3 15.5 77.2 1.9 14.7 83.4 5.2 29.3 65.5
2.5 2.9 4.0 0.4 2.4 2.1 3.5 2.3 3.9
0.010 0.003 0.001 0.65 0.013 0.027 0.001 0.017 0.001
7.8 16.1 76.1 1.8 15.4 82.8 5.6 29.9 64.5
7.8 15.9 76.3 1.9 15.2 82.9 5.5 29.7 64.7
0.1 0.5 0.5 0.7 0.6 0.3 0.2 0.4 0.4
0.90 0.64 0.63 0.56 0.63 0.80 0.88 0.74 0.70
5.6 14.9 79.5
5.2 13.0 81.8
1.5 5.4 5.6
0.111 0.001 0.001
5.5 13.8 80.7
5.5 13.8 80.7
0.1 0.0 0.1
0.94 0.99 0.95
3.8 21.2 75.1
4.4 21.6 74.0
3.0 1.1 2.4
0.002 0.27 0.014
4.2 21.5 74.3
4.3 21.4 74.2
0.8 0.2 0.2
0.46 0.85 0.88
3.0 18.1 78.9
2.8 18.5 78.7
1.5 1.2 0.5
0.118 0.23 0.62
3.1 18.2 78.7
2.9 18.5 78.6
1.1 0.6 0.1
0.33 0.60 0.92
ACE angiotensin-converting enzyme; COPD chronic obstructive pulmonary disease. * Patients were also categorized by the area of Ontario in which they lived. Testing the hypothesis of no difference between the 2 medication cohorts. In Ontario, primary care is provided by family and general practitioners, and specialist COPD care is usually provided by pulmonologists, general internists, or geriatricians. Number of prescriptions for long- or short-acting -agonists, long- or short-acting anticholinergics, inhaled corticosteroids, oral corticosteroids, methylxanthine, or respiratory antibiotics. As indicated by Johns Hopkins Collapsed Ambulatory Diagnostic Groups (13). Pneumonia, inuenza, or acute bronchitis. ** Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmia, or cerebrovascular disease.
www.annals.org
Appendix Table 3. Survival at the End of Each Year of Follow-up

Outcome Death Event-free, %* Difference (95% CI), percentage points Death or hospitalization for COPD Event-free, %* Difference (95% CI), percentage points Death or hospitalization for COPD or a related respiratory disease Event-free, %* Difference (95% CI), percentage points Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease Event-free, %* Difference (95% CI), percentage points Death, hospitalization, or emergency department visit for COPD Event-free, %* Difference (95% CI), percentage points Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Event-free, %* Difference (95% CI), percentage points Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease Event-free, %* Difference (95% CI), percentage points Sensitivity analysis Hospitalization for COPD Event-free, %* Difference (95% CI), percentage points Emergency department visit for COPD Event-free, %* Difference (95% CI), percentage points Year 1 88.7 2.9 (2.33.5) Year 2 80.0 4.8 (4.15.6) Year 3 71.9 6.3 (5.47.2) Year 4 64.7 7.4 (6.48.4) Year 5 58.0 8.2 (7.09.3)
78.9 3.8 (3.04.6)
67.1 6.0 (5.16.9)
57.9 7.6 (6.68.5)
50.2 8.5 (7.59.5)
43.9 8.8 (7.79.2)
77.0 3.4 (2.64.1)
64.8 5.5 (4.66.3)
55.5 6.9 (6.07.8)
47.8 7.6 (6.68.6)
41.6 8.0 (6.99.1)
72.4 2.9 (2.13.8)
59.0 4.9 (4.05.8)
49.3 5.9 (5.06.8)
41.3 6.3 (5.47.3)
35.1 6.4 (5.37.5)
74.7 4.5 (3.75.3)
61.5 6.4 (5.57.3)
51.9 7.6 (6.78.5)
44.0 8.2 (7.39.2)
38.0 8.7 (7.69.7)
70.6 3.5 (2.74.4)
56.5 5.0 (4.15.9)
46.5 6.0 (5.06.9)
38.4 6.2 (5.37.2)
32.5 6.7 (5.67.7)
65.0 3.0 (2.13.9)
49.9 4.3 (3.45.2)
39.9 4.9 (4.05.8)
32.0 5.0 (4.15.9)
26.1 5.1 (4.16.1)
85.8 2.4 (1.73.1) 94.0 1.1 (0.71.5)
78.0 4.1 (3.34.9) 90.1 1.4 (0.82.1)
71.8 5.5 (4.56.4) 87.2 2.1 (1.42.8)
66.4 7.0 (5.98.0) 84.1 2.5 (1.63.4)
61.4 7.3 (6.18.5) 81.9 3.2 (2.14.3)
COPD chronic obstructive pulmonary disease. * Among patients rst prescribed a long-acting -agonist. Percentage of event-free patients in the long-acting -agonist group minus the percentage of event-free patients in the long-acting anticholinergic group. Pneumonia, inuenza, or acute bronchitis. Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmias, or cerebrovascular disease.
www.annals.org
Appendix Table 4. Adjusted Hazard Ratios for Death Associated With Initially Prescribed Inhaled Long-Acting Anticholinergics Compared With Long-Acting -Agonists, Stratified by Various Factors
Characteristic Absolute Difference in Death at 3 Years (95% CI), % Propensity Score Matching Conventional Proportional Hazards Regression Adjusted Hazard Ratio (95% CI) 1.13 (1.081.19) 1.14 (1.081.20)
Adjusted Hazard Ratio (95% CI)* Sex Male Female Asthma Yes No Number of prescriptions for COPD medications filled in the previous year 02 310 10 Inhaled corticosteroid use Yes No Congestive heart failure Yes No 5.7 (4.46.9) 6.5 (5.37.7) 1.13 (1.071.20) 1.14 (1.071.21)
P Value
P Value
0.001 0.001
0.001 0.001
7.3 (6.18.5) 3.4 (2.44.5)
1.16 (1.091.23) 1.12 (1.061.19)
0.001 0.001
1.11 (1.051.17) 1.14 (1.091.20)
0.001 0.001
7.0 (5.68.4) 6.4 (4.97.8) 4.7 (3.16.3)
1.23 (1.131.33) 1.14 (1.061.23) 1.07 (1.001.15)
0.001 0.001 0.042
1.21 (1.121.29) 1.14 (1.071.22) 1.08 (1.021.14)
0.001 0.001 0.010
7.1 (5.78.5) 6.0 (4.97.1)
1.12 (1.031.22) 1.14 (1.071.22)
0.008 0.001
1.12 (1.061.19) 1.13 (1.081.19)
0.001 0.001
5.1 (4.26.0) 7.2 (5.88.8)
1.18 (1.081.29) 1.13 (1.041.23)
0.001 0.003
1.13 (1.081.19) 1.13 (1.081.20)
0.001 0.001
COPD chronic obstructive pulmonary disease. * Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for membership in a matched pair. Members of the long-acting anticholinergic and long-acting -agonist cohorts were matched by using propensity score and, where appropriate, age, sex, asthma status, and number of COPD medications. Testing the hypothesis that the hazard ratio is 1.00. Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for all variables. Number of prescriptions for long- or short-acting -agonists, long- or short-acting anticholinergics, inhaled corticosteroids, oral corticosteroids, methylxanthine, or respiratory antibiotics. Analysis used 7425 matched pairs not receiving inhaled corticosteroids and 3643 matched pairs receiving inhaled corticosteroids. Analysis used 6013 matched pairs with congestive heart failure and 2618 matched pairs without congestive heart failure.
Appendix Table 5. Repeat of the Primary Analysis Without Excluding Outcomes That Occurred Within 7 Days of the Index
Prescription Being Filled
Outcome Propensity Score Calibration Adjusted Hazard Ratio (95% CI)* Death Death or hospitalization for COPD Death or hospitalization for COPD or a related respiratory disease Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease Death, hospitalization, or emergency department visit for COPD Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease 1.11 (1.061.15) 1.11 (1.071.15) 1.09 (1.051.13) 1.06 (1.021.09) 1.10 (1.061.14) 1.06 (1.021.10) 1.03 (1.001.07)
P Value
0.001 0.001 0.001 0.003 0.001 0.001 0.059
COPD chronic obstructive pulmonary disease. * Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for membership in a matched pair. Members of the long-acting anticholinergic and long-acting -agonist cohorts were matched by using propensity score, age, sex, asthma status, and number of COPD medications. Testing the hypothesis that the hazard ratio is 1.00. Pneumonia, inuenza, or acute bronchitis. Congestive heart failure, ischemic heart disease (including acute myocardial infarction), cardiac arrhythmia, or cerebrovascular disease.
www.annals.org 3 May 2011 Annals of Internal Medicine Volume 154 Number 9 W-211
Appendix Table 6. Results of the Calibrated Propensity Score Analysis: Adjusted Hazard Ratios for Death and Other Outcomes Associated With Initially Prescribed Inhaled Long-Acting Anticholinergics Compared With Long-Acting -Agonists, Using Calibrated Propensity Score for Matching
Outcome Propensity Score Calibration Adjusted Hazard Ratio (95% CI)* Death Death or hospitalization for COPD Death or hospitalization for COPD or a related respiratory disease Death or hospitalization for COPD, a related respiratory disease, or cardiovascular disease Death, hospitalization, or emergency department visit for COPD Death, hospitalization, or emergency department visit for COPD or a related respiratory disease Death, hospitalization, or emergency department visit for COPD, a related respiratory disease, or cardiovascular disease 1.08 (1.041.13) 1.07 (1.031.11) 1.05 (1.011.09) 1.03 (1.001.07) 1.07 (1.031.11) 1.05 (1.011.09) 1.03 (0.991.06)
P Value
0.001 0.001 0.009 0.094 0.001 0.007 0.126
COPD chronic obstructive pulmonary disease. * Reects the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Hazard ratios are adjusted for membership in a matched pair. Members of the long-acting anticholinergic and long-acting -agonist cohorts were matched by using the propensity score, age, sex, asthma status, and number of COPD medications. Testing the hypothesis that the hazard ratio is 1.00. Pneumonia, inuenza, or acute bronchitis. Congestive heart failure, ischemic heart disease (including acute myocardial infarction), cardiac arrhythmia, or cerebrovascular disease.
www.annals.org
Appendix Table 7. Cox Proportional Hazards Regression Model Results for Time to Death, Using Unmatched Data and Adjusting for All Covariates*
Variable Demographic characteristics Age Male sex Resident in a long-term care facility Neighborhood income quintile 1 (low) 2 3 4 5 (high) Urban or rural residence COPD and general care COPD duration 1 y 15 y 5 y Previous spirometry Number of primary care physician visits in the previous year Specialist visit in previous year Influenza vaccination in previous year Previous lung volume reduction surgery Number of prescriptions for COPD medications filled in the previous year Medication use in previous year Respiratory antibiotic Short-acting -agonist Short-acting anticholinergic Inhaled corticosteroid Oral corticosteroid Long-term oral corticosteroid use Methylxanthine Statin ACE inhibitor -Blocker Number of classes of medications prescribed Comorbidity Acute minor Acute major Likely to recur Chronic medical unstable Chronic medical stable Chronic specialty unstable Chronic specialty stable Eye or dental Psychosocial Prevention, administration Previous or coexisting medical conditions Asthma ever diagnosed Active asthma Pneumonia, influenza, or acute bronchitis Cor pulmonale Lung cancer Pulmonary embolism Arrhythmias Acute myocardial infarction Other ischemic heart disease Congestive heart failure Cerebrovascular disease Hypertension Fragility fracture due to osteoporosis Diabetes Any cancer (other than lung cancer) Mental health visit (in previous year) Hazard Ratio (95% CI) 1.05 (1.051.05) 1.33 (1.281.38) 1.66 (1.561.76) 1.00 (0.951.07) 1.02 (0.961.08) 0.94 (0.881.00) 0.97 (0.911.04) 1.00 (reference) 1.00 (0.951.06)
P Value
0.001 0.001 0.001 0.89 0.55 0.041 0.42 0.90
1.05 (0.981.12) 0.99 (0.941.03) 1.00 (reference) 0.90 (0.860.93) 1.00 (1.001.00) 1.18 (1.131.23) 0.91 (0.870.94) 0.94 (0.731.22) 1.01 (1.011.01)
0.190 0.56 0.001 0.26 0.001 0.001 0.66 0.001
0.92 (0.870.98) 1.04 (0.991.10) 1.07 (1.021.11) 0.87 (0.840.91) 1.00 (0.951.04) 0.99 (0.931.04) 1.01 (0.931.10) 0.76 (0.730.79) 1.01 (0.971.05) 1.00 (0.961.05) 1.04 (1.031.04)
0.007 0.102 0.004 0.001 0.89 0.65 0.73 0.001 0.63 0.92 0.001
0.99 (0.911.07) 0.97 (0.901.03) 0.88 (0.840.92) 1.26 (1.151.38) 0.91 (0.860.96) 0.90 (0.860.94) 0.92 (0.860.98) 0.96 (0.921.00) 1.08 (1.041.13) 0.98 (0.951.02)
0.75 0.30 0.001 0.001 0.001 0.001 0.007 0.028 0.001 0.41
0.83 (0.800.87) 0.88 (0.840.92) 1.15 (1.101.19) 1.41 (1.141.74) 1.73 (1.631.83) 1.05 (0.971.14) 1.06 (1.021.11) 1.03 (0.981.07) 0.96 (0.921.00) 1.35 (1.291.41) 1.21 (1.141.28) 0.87 (0.820.92) 1.18 (1.101.26) 1.08 (1.041.13) 1.20 (1.161.25) 1.03 (0.971.10)
0.001 0.001 0.001 0.001 0.001 0.21 0.004 0.28 0.069 0.001 0.001 0.001 0.001 0.001 0.001 0.36

www.annals.org 3 May 2011 Annals of Internal Medicine Volume 154 Number 9 W-213

Variable Acute events before index date Most recent hospitalization for COPD In the past 6 mo 6 mo before index date Never Most recent hospitalization for a COPD-related condition** In the past 6 mo 6 mo before index date Never Most recent hospitalization for cardiovascular cause In the past 6 mo 6 mo before index date Never Most recent emergency department visit for COPD In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a COPD-related condition** In the past 6 mo 6 mo before index date Never Most recent emergency department visit for a cardiovascular cause In the past 6 mo 6 mo before index date Never Initial medication Long-acting -agonist Long-acting anticholinergic Hazard Ratio (95% CI)
P Value
1.28 (1.201.36) 1.15 (1.091.20) 1.00 (reference) 1.13 (1.011.27) 1.13 (1.071.18) 1.00 (reference) 1.30 (1.201.41) 1.10 (1.041.16) 1.00 (reference) 1.17 (1.081.26) 1.06 (1.011.12) 1.00 (reference) 0.95 (0.871.04) 0.92 (0.880.97) 1.00 (reference) 1.00 (0.911.10) 0.96 (0.921.01) 1.00 (reference)
0.001 0.001 0.031 0.001 0.001 0.001 0.001 0.023 0.27 0.001 0.99 0.110
1.00 (reference) 1.13 (1.091.18)
0.002
ACE angiotensin-converting enzyme; COPD chronic obstructive pulmonary disease. * Patients were also categorized by the area of Ontario in which they lived. Testing the hypothesis that the hazard ratio is 1.00. Reference category is rural. In Ontario, primary care is provided by family and general physicians, and specialist COPD care is usually provided by pulmonologists, general internists, or geriatricians. Number of prescriptions for long- or short-acting -agonists, long- or short-acting anticholinergics, inhaled corticosteroids, oral corticosteroids, methylxanthine, or respiratory antibiotics. As indicated by Johns Hopkins Collapsed Ambulatory Diagnostic Groups (13). ** Pneumonia, inuenza, or acute bronchitis. Acute myocardial infarction, congestive heart failure, ischemic heart disease, cardiac arrhythmia, or cerebrovascular disease.
Appendix Table 8. Person-Years and Percentage of Total Person-Years Receiving Long-Acting Anticholinergics, Long-Acting -Agonists, and Inhaled Corticosteroids in Each Year of Follow-up for Patients Who Were Initially Prescribed Long-Acting Anticholinergics and Long-Acting -Agonists
Group and Medication Year 1 Long-acting anticholinergic group Long acting anticholinergic Long-acting -agonist Inhaled corticosteroids Long-acting -agonist group Long acting anticholinergic Long-acting -agonist Inhaled corticosteroids 8593 (59.2) 1389 (9.6) 3867 (26.6) Year 2 6345 (49.5) 2019 (15.7) 3800 (29.6) Person-Years Receiving Medication, n (%) Year 3 4984 (49.2) 2058 (20.3) 3255 (32.1) Year 4 3478 (50.6) 1662 (24.2) 2376 (34.6) Year 5 1967 (52.3) 1081 (28.6) 1427 (37.9)
1469 (10.1) 6687 (45.9) 6864 (47.2)
2429 (18.6) 4806 (36.8) 5322 (40.8)
2476 (23.7) 3867 (37.0) 4291 (41.1)
2077 (28.8) 2760 (38.3) 3047 (42.3)
1274 (33.2) 1558 (40.6) 1708 (44.5)
www.annals.org
Appendix Table 9. Influence of Potential Misclassification on Study Results

Assumptions and Observed Hazard Ratio for Death (95% CI) Assumed Misclassified Patients in Each Treatment Group, %* Assumed Risk Ratio for Death for Misclassified Patients Relative to Those Correctly Classified True Hazard Ratio (95% CI) Based on Assumptions
Long-Acting Anticholinergic Group All misclassification occurred in the long-acting -agonist group and misclassified patients were healthier 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) All misclassification occurred in the long-acting anticholinergic group and misclassified patients were less healthy 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) Assumed risk ratio constant at 0.60 and distribution of misclassified patients varied between the groups 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) 1.14 (1.091.19) Assumed risk ratio constant at 0.80, all misclassified patients were placed in the long-acting -agonist group, and the SE of the observed hazard ratio increased 1.14 (1.081.19); 10% increase in SE 1.14 (1.081.20); 30% increase in SE 1.14 (1.071.21); 50% increase in SE 1.14 (1.051.21); 75% increase in SE
Long-Acting -Agonist Group
0 0 0 0 0
28 28 28 28 28
0.90 0.80 0.70 0.60 0.55
1.11 (1.061.15) 1.07 (1.031.12) 1.04 (1.001.09) 1.01 (0.971.05) 0.99 (0.951.04)
28 28 28 28 28
0 0 0 0 0
1 0.90 1.11 1 0.80 1.25 1.30 1 0.70 1.43 1 0.60 1.67
1.10 (1.061.15) 1.06 (1.021.11) 1.05 (1.001.10) 1.02 (0.971.06) 0.96 (0.921.00)
1 3 5 10
27 25 23 18
0.60 0.60 0.60 0.60
1.02 (0.981.06) 1.04 (0.991.08) 1.05 (1.011.10) 1.10 (1.051.15)
0 0 0 0
28 28 28 28
0.80 0.80 0.80 0.80
1.07 (1.011.14) 1.07 (1.011.14) 1.07 (1.011.15) 1.07 (0.991.16)
* The overall misclassication rate was 14%. Because the matched analysis included equal numbers of patients in the long-acting anticholinergic and long-acting -agonist cohorts, if every misclassied patient was in the same treatment cohort, 28% of the patients in that cohort would be misclassied, whereas 0% of those in the other cohort would be misclassied. Adjusted hazard ratios reect the risk in the long-acting anticholinergic cohort compared with the long-acting -agonist cohort. Survival analysis produces a parameter estimate and associated SE. A CI is calculated for the parameter estimate, and then the point estimate and the lower and upper bounds of the CI are exponentiated to report a hazard ratio and CI for the hazard ratio. Ination factors were applied to the SE for the parameter estimate.
www.annals.org

Original Research

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Original Research

Uploaded by

Copyright:

Available Formats

Annals of Internal Medicine

Long-Acting -Agonists and Long-Acting Anticholinergics in COPD

Downloaded From: http://annals.org/ by Rizka Mulya on 02/22/2013

Long-Acting -Agonists and Long-Acting Anticholinergics in COPD

Initially Prescribed Medication

Long-Acting Long-Acting Anticholinergic -Agonist 15 532 15 532

77.2 (6.8) 46.2

76.8 (7.0) 53.1

76.9 (6.9) 50.8

76.9 (6.9) 50.8

10.3 22.5 67.3 69.7 9 (514) 44.7 6 (114)

9.3 22.0 68.8 74.3 9 (514) 50.5 5 (113)

3.4 1.1 3.2 10.2 1.5 11.6 7.2

0.001 0.23 0.001 0.001 0.001 0.001 0.001

9.8 21.8 68.3 72.4 9 (514) 49.0 5 (113)

9.6 22.4 68.0 73.0 9 (514) 49.1 5 (114)

0.8 1.3 0.7 1.3 0.2 0.2 0.7

0.50 0.25 0.55 0.24 0.89 0.87 0.40

37.6 8.6 4.5 39.6 53.3 48.9 40.2 11.7

55.4 8.0 4.6 39.6 53.5 50.9 38.2 9.6

36.3 2.1 0.9 0.1 0.5 4.0 4.1 6.6

0.001 0.031 0.37 0.95 0.59 0.001 0.001 0.001

49.7 8.3 4.6 39.4 53.8 50.5 39.0 10.2

49.7 8.4 4.7 39.9 53.7 50.5 39.2 10.2

0.0 0.5 0.4 1.0 0.3 0.0 0.4 0.0

1.00 0.65 0.75 0.39 0.80 1.00 0.75 1.00

8.0 16.5 75.5

7.3 15.5 77.2

2.5 2.9 4.0

0.010 0.003 0.001

7.8 16.1 76.1

7.8 15.9 76.3

0.1 0.5 0.5

0.90 0.64 0.63

1.8 15.6 82.6

1.9 14.7 83.4

0.4 2.4 2.1

0.65 0.013 0.027

1.8 15.4 82.8

1.9 15.2 82.9

0.7 0.6 0.3

0.56 0.63 0.80

6.0 30.3 63.7

5.2 29.3 65.5

3.5 2.3 3.9

0.001 0.017 0.001

5.6 29.9 64.5

5.5 29.7 64.7

0.2 0.4 0.4

0.88 0.74 0.70

Continued on following page

3 May 2011 Annals of Internal Medicine Volume 154 Number 9 585

Downloaded From: http://annals.org/ by Rizka Mulya on 02/22/2013

Long-Acting -Agonists and Long-Acting Anticholinergics in COPD

Before Propensity Score Matching Initially Prescribed Medication Standardized Difference, %

After Propensity Score Matching

Initially Prescribed Medication

Long-Acting Long-Acting Anticholinergic -Agonist

5.6 14.9 79.5

5.2 13.0 81.8

1.5 5.4 5.6

0.111 0.001 0.001

5.5 13.8 80.7