STRENGTHS AND LIMITATIONS OF THIS STUDY

We used large, population-based administrative healthcare datasets and risk-adjusted models; therefore, increasing the generalisability of the study.

Medication dispenses were time varying, which allowed for accurate classifications of treatment patterns over time.

Administrative data are collected for hospital administration and not for research, which impacts data availability, particularly on patient clinical characteristics.

The definitions used in the study may have underestimated the number of true oral corticosteroid burst dispenses and severe exacerbations.

Introduction

Chronic obstructive pulmonary disease (COPD) is currently the third leading cause of death globally.1 In Canada between 1950 and 2011, there was an increase in the percentage of deaths attributed to COPD (age 40+) from 0.5% to 4.4%, though for older men, the percentage declined after 1998.2 In Europe between 2005–2007 and 2015–2017, COPD deaths increased among women.3 Research has shown that patients hospitalised with severe exacerbations have a higher risk of mortality in the year after discharge, and the risks of all-cause and COPD-related mortality increase with each rehospitalisation.4

The 2023 Global Initiative for Chronic Obstructive Lung Disease report recommends avoiding regular use of short-acting beta agonists (SABA) as it can lead to resting sinus tachycardia, hypokalaemia and increased oxygen consumption.5 Additionally, as the effect of SABA is short lived and inferior to long-acting beta agonists (LABAs), it is only recommended for immediate symptom relief.5 A study by Fan et al demonstrated the negative effects of SABA overuse on metrics of COPD morbidity, including dyspnoea, use of oxygen and use of maximal inhaled therapy.6 Furthermore, a recent global study of asthma showed SABA overuse to be associated with higher risks of exacerbation and mortality.7 However, similar studies of COPD, examining SABA and other medications such as antibiotics and oral corticosteroids (OCS) in relation to mortality, are lacking.

Many patients with COPD have comorbidities, with cardiovascular disease (CVD) being one of the most common.8 Patients with COPD have an increased cardiopulmonary risk and potential for all-cause mortality,9 possibly due to the systemic inflammatory response associated with exacerbations, hyperinflation, hypoxaemia and shared cardiopulmonary risk factors, such as smoking.10 Given previous research into SABA overuse6 7 and the known link between the pulmonary and cardiovascular systems,11–13 further insight into the impact of SABA overuse on cardiopulmonary outcomes is warranted.

Here we sought to understand, at a population level using real-world data, whether the rates of mortality and major adverse cardiac events (MACE) for patients with COPD varied by medication history. The study objectives were, first, to describe the association of all-cause and COPD-related mortality relative to SABA dispensation, antibiotic dispensation and OCS burst-days in a given year; and second, to describe the association of MACE and postexacerbation MACE relative to SABA dispensation in a given year.

MethodsData sources

Administrative data from the National Ambulatory Care Reporting System, Discharge Abstract Database (DAD), Pharmaceutical Information Network (PIN), Population Registry, Practitioner Claims, Vital Statistics and Alberta Precision Laboratories were obtained from Alberta Health. Alberta Health also provided a COPD cohort based on the Gershon algorithm14 applied to the period between 1 April 1985 and 31 March 2021. This algorithm specified: (1) an International Classification of Diseases, 10th Revision, Canada (ICD-10-CA) diagnosis code for COPD (J41–J44) in any position of DAD at discharge (diagnosis date); or (2) at least two physician claims on separate days within a 2-year period with a primary International Classification of Diseases, 9th Revision, Clinical Modification diagnosis code for COPD (491, 492, 496). The sensitivity, specificity, positive predictive value and negative predicative value of the algorithm were 65.5 (95% CI 56.0 to 74.2), 91.5 (95% CI 87.9 to 94.3), 72.6 (95% CI 62.8 to 80.9) and 88.5 (95% CI 84.7 to 91.7), respectively. The algorithm is recommended for research purposes as it provides superior specificity over sensitivity to accurately rule out those who do not have COPD.

Case definitions and study design

The present study was based on the analysis cohort described in figure 1. Using a retrospective population-based cohort study design, the initial pool of patients with COPD from Alberta Health was restricted to cases ascertained between 1 April 2011 and 31 March 2019. The data are captured until the end of study cohort period, 31 March 2020, to ensure at least 1 year of potential follow-up. The index date (cohort entry date) was set to the COPD diagnosis date for incident cases during this case ascertainment period and to 1 April 2011 for prevalent cases. The cohort was further restricted to patients aged 35+ at index who resided in Alberta for at least 1 year preindex.

Figure 1

Study flow chart for eligible chronic obstructive pulmonary disease (COPD) cases.

Patients were followed from their index date to the earlier of out-of-province migration, death or study end (31 March 2020). A 1-year preindex period was used to identify patient baseline characteristics. Starting at index follow-up was divided into consecutive periods up to 90 days for each patient. Mortality and MACE outcomes were assessed in these follow-up periods (online supplemental figure 1), whereas medication exposure and prior severe exacerbations were ascertained in a 1-year look-back window preceding each period throughout the entire study. This was a modest approach to associate outcomes to exposure that was time varying but not instantaneous. A period can be <90 days in the event of discontinued registration in Alberta, patient death or at study end.

Variables

Demographic variables (age, sex, geographical zone) were determined at index. Comorbid conditions from hospitalisation abstracts 1 year preindex were identified using the Charlson Comorbidity Index (CCI) outlined by Quan et al.15

Medications of interest were selected based on Anatomical Therapeutic Chemical classification codes in the PIN dataset (online supplemental table 1 provides categorisations). In the 1-year history prior to each 90-day follow-up interval, zero SABA, antibiotic and OCS were the most common dispensations. Thus, to facilitate clinical interpretability and to create categories with sufficient number of cases, exposure levels were pragmatically grouped into fewer categories based on regression models from a prior study16 with severe exacerbations as the outcome. There, an iterative process was applied to group dispenses together with similar relative effects (incidence rate ratios, IRR) and statistical associations. As the main research interest was medication dispensing, the smallest usage (non-zero) category was used as the comparison group (ie, the reference). For SABA, levels of exposures were 0, 1 (reference), 2–5 and 6+ canisters or equivalents (see online supplemental table 2 for methods). For antibiotics, exposure categories were 0, 1–2 (reference), 3–5 and 6+ dispenses. For OCS, exposure categories were 0, 1–5 (reference) and 6+ burst-days. OCS bursts were identified as dispensations of 3–28 days with an average daily dose of ≥20 mg prednisone equivalents, plus a COPD primary diagnosis during an emergency department (ED) or physician visit within 7 days before or after dispensation of prednisone. Severe COPD exacerbations were defined as hospitalisations or ED visits with COPD as a primary diagnosis. Severe exacerbations within 14 days of each other were grouped as a single event.

Outcome variables included all-cause and COPD-related mortality. Vital Statistics is considered the gold standard for mortality data. Availability, however, can be lagged by at least 2 years. Thus, the death data for mortality outcomes were supplemented with the Alberta Registry and inpatient hospitalisations (with a discharge disposition relating to death). COPD-related mortality was a subset of all-cause mortality which encompassed COPD, CVD, lung cancer and respiratory failure, identified through codes (online supplemental table 3) for each patient’s main diagnosis in Vital Statistics or inpatient hospitalisation records.

Outcome variables also included five MACE subtypes (CVD death, acute myocardial infarction (AMI), stable/unstable angina, ischaemic stroke and coronary revascularisation (percutaneous coronary intervention or coronary artery bypass graft surgery)) and one MACE composite outcome (of the five subtypes). The composite outcome and CVD death were of primary interest. MACE was identified by the presence of any ED visit or hospitalisation with an ICD-10-CA or CCI code in online supplemental table 3, excluding preadmission comorbidities (codes with an accompanying diagnosis type 1). Postexacerbation MACE was also examined and defined as MACE following a severe COPD exacerbation within 90 days.

Statistical analysis

Demographic and clinical characteristics at index date were summarised descriptively. Time-varying Cox proportional hazards regression models related 90-day mortality (all-cause and COPD-related) to 1-year COPD-related medication history (SABA canisters dispensed, antibiotic dispenses and OCS burst-days) for the analysis cohort (figure 1). Mortality was modelled as a censored time to event for each follow-up period. Each model was adjusted for sex, age (years), calendar year, CCI at index, and, when assessing SABA and antibiotics, 1-year prior history of severe COPD exacerbations. One-year prior exacerbations were not included in the OCS risk-adjusted model due to their collinearity as both the definitions of OCS burst-days and 1-year prior exacerbations included ED visits.

Poisson regression models related the six MACE outcomes (and their respective postexacerbation subsets), identified in each 90-day follow-up period, to patients’ 1-year prior history of SABA dispensation (figure 1). Models accounted for repeat observations for the same individuals over time using generalised estimation equation. Risk-adjusted models included the same covariates as in mortality. Statistical analyses were completed using SAS V.9.4 (SAS Institute). When appropriate, statistical significance was identified from results with associated p values <0.05 or when point estimates of rates (HR or IRR) did not contain the value of 1.

Due to the population level and deidentified nature of the data, a waiver of consent was requested and approved by the ethics board prior to study commencement. The authors used the Strengthening the Reporting of Observational Studies in Epidemiology guidelines in writing this report.17

Patient and public involvement

No patients were involved in the conduct of the study; however, pulmonologists, who provided their expertise in COPD and insights into a patient’s journey in clinical practice, were involved in the design and execution of the study.

Results

The study cohort included 188 969 patients with COPD, among which 54% were newly diagnosed at the time of cohort entry. The mean (SD) and median (IQR) follow-up times were 4.9 (3.2) and 4.7 (2.0–8.6) years, respectively. Additional baseline characteristics are presented in table 1 and online supplemental table 4.

Table 1

Demographic and clinical characteristics of patients with COPD in Alberta, Canada, at index date, April 2011–March 2019

The 90-day rates of all-cause mortality by SABA dispensation in a given year were similar in the unadjusted and risk-adjusted models. Both showed dose–response effects, with increasing hazards for greater numbers of canister dispenses. In the risk-adjusted model, all-cause mortality rate was 6% higher for patients dispensed 2–5 (HR: 1.06, 95% CI 1.03 to 1.09) and 20% higher for patients receiving 6+ (vs 1) SABA canisters (HR: 1.20, 95% CI 1.16 to 1.24; figure 2). A similar observation was made for 90-day rate of COPD-related mortality, it was 13% higher for the 2–5 canisters group (HR: 1.13, 95% CI 1.09 to 1.18) and 40% higher for the 6+ SABA canisters group (HR: 1.40, 95% CI 1.34 to 1.46; figure 2).

Figure 2

Risk-adjusted associations between 90-day mortality (A: all-cause, B: COPD-related) and 1-year prior history of medication dispensation for patients in Alberta, Canada, fiscal years 2011–2019. HRs (dots) are shown with 95% CIs (lines). Associations are adjusted for patient sex, age at index, year of index, Charlson Comorbidity Index (CCI) and severe COPD exacerbation count (1-year prior history). The horizontal line represents no association. Symbols above the line represent increased rates. Symbols below the line represent decreased rates. COPD, chronic obstructive pulmonary disease; OCS, oral corticosteroid; SABA, short-acting beta agonist.

Similarly, there was a dose–response effect in the association between 90-day rate of mortality (all-cause and COPD-related) and antibiotic dispensation in a given year. Both the unadjusted and risk-adjusted models showed increasing hazards for greater numbers of antibiotic dispenses. All-cause and COPD-related mortality rates were both over 20% higher with 3–5 antibiotic dispenses (vs 1–2) and were the highest at 62% (all-cause) and 43% (COPD-related) for patients with 6+ (vs 1–2) antibiotic dispenses (risk-adjusted all-cause HR: 1.62, 95% CI 1.57 to 1.66; COPD-related HR: 1.43, 95% CI 1.38 to 1.49; figure 2).

Furthermore, for OCS, there was generally a dose–response effect in the unadjusted and adjusted models for both all-cause and COPD-related mortality, where hazards increased with greater OCS burst-days. In the adjusted models, the rate for 6+ (vs 1–5) OCS burst-days was 27% (HR: 1.27, 95% CI 1.18 to 1.36; figure 2) and 29% (HR: 1.29, 95% CI 1.19 to 1.40) higher for all-cause and COPD-related mortality, respectively.

There was an observable dose–response effect for the composite MACE and CVD deaths that was even stronger when examining postexacerbation MACE. In both the unadjusted and risk-adjusted models, the incidence of these outcomes was generally higher among patients with more SABA dispenses. In the risk-adjusted model for postexacerbation MACE, patients with 2–5 (vs 1) SABA dispenses were 26% (IRR: 1.26, 95% CI 1.16 to 1.36) and 27% (IRR: 1.27, 95% CI 1.16 to 1.40) more likely to experience a postexacerbation MACE and CVD death within 90-day follow-up period, respectively (figure 3). Additional MACE results are presented in online supplemental table 5.

Figure 3

Risk-adjusted associations between 90-day major adverse cardiac event (MACE) and 1-year prior history of short-acting beta agonist (SABA) dispensation (A: overall, B: within 90 days after severe exacerbation) for patients with chronic obstructive pulmonary disease (COPD) in Alberta, Canada, fiscal years 2011–2019. Incidence rate ratios (dots) are shown with 95% CIs (lines). Associations are adjusted for 1-year prior history of COPD medication (oral corticosteroid (OCS) burst-days and number of antibiotics), sex, age in years at index date, and calendar year of index date, CCI at index and 1-year prior history of severe exacerbations. The horizontal line represents no association. Symbols above the line represent increased rates. Symbols below the line represent decreased rates. CCI, Charlson Comorbidity Index; CVD, cardiovascular disease; MACE, major adverse cardiac event; SABA, short-acting beta agonist.

Two post hoc sensitivity analyses were performed. The first restricted to COPD-related antibiotics. The second accounted for dispensation LABA and/or long-acting muscarinic antagonist in the year prior to each follow-up period. Both yielded results similar to original analyses (online supplemental table 6).

Discussion

Among 188 969 patients with COPD in Alberta, Canada, we found that higher SABA, antibiotic and OCS dispensation in a given year was associated with significantly higher rates of subsequent 90-day all-cause and COPD-related mortality. Higher SABA dispensation was also associated with higher rates of subsequent 90-day postexacerbation MACE and CVD death. This study shows increased SABA, antibiotic and OCS dispensation to be potential indicators of greater mortality rate. These indicators can be used by healthcare providers to identify patients with uncontrolled or suboptimal management of disease, which may warrant a review of their current COPD and cardiac medications.

A novel finding from our study was the dose–response trend between 1-year history of SABA dispensation and 90-day rate of mortality (all-cause and COPD-related) and MACE, which persisted after adjusting for risk factors such as prior exacerbations. To our knowledge, this is the first study to examine all-cause and COPD-related mortality relative to SABA dispense history among patients with COPD. Studies examining cardiopulmonary risks relative to SABA medications are also sparse.18 19 One nested case–control study examined this association using administrative data from Canada from 1980 to 1997 and found no association between SABA dispensation and risk of AMI.18 In contrast, a population-based nested case–control study of patients with COPD in the UK from 1998 to 2018 showed that new prescriptions of SABA were associated with a twofold higher risk of MACE compared with new prescriptions of short-acting muscarinic antagonists.19

It is plausible that the association between high SABA dispensation and COPD-related mortality, and particularly CVD death, is mediated by severe exacerbations in these patients. High SABA use has been significantly associated with increased risk of exacerbations in several studies,20 21 and the risk of MACE has been found to increase markedly following the occurrence of exacerbations.11 12 In a population-based case-crossover study, the risk of MACE increased nearly four times following the onset of acute exacerbations in patients with COPD, and the association was strongest for CVD death,12 as in our study. A meta-analysis found similar results, suggesting a 1.7 times higher risk of stroke and a 2.4 times higher risk of AMI in the 1–3 months following exacerbation, respectively.11 While the biological mechanisms underlying COPD exacerbation and risk of acute cardiovascular outcomes are beyond the scope of this paper, some mechanisms include systematic inflammation, hypoxaemia and hyperinflation.11 22 The fact that the association was also observed for all-cause mortality in our study suggests that high SABA dispensing could be a general indicator to help healthcare providers identify patients with uncontrolled or suboptimal management of disease.

Like SABA, we found a dose–response association between 1-year prior history of antibiotic dispensation and OCS with mortality rate. Antibiotics are often recommended for treatment of exacerbations in patients with moderate and severe COPD.5 Longer courses of OCS are associated with higher risks of all-cause mortality up to 1 year after exposure.23 As for SABA, the associations we observed may not have been linked to medication dispensing itself, but rather to frequent antibiotic and OCS dispensing as an indicator of more severe disease, frequent exacerbations and possibly suboptimal control or management with recommended treatments.6 24 25

The strengths of our study warrant discussion. First, we used large, population-based administrative healthcare datasets to address our research questions. These datasets included detailed medication data, including dosage, which enabled a comprehensive understanding of real-world treatment patterns and enabled us to link medication dispensing with health outcomes. The partitioning of follow-up into 90-day periods allowed medication dispenses to be time varying rather than a single level of exposure averaged throughout the entire follow-up, which allowed for accurate classifications of treatment patterns over time. Capturing the medication history that preceded the outcomes also allowed us to assess the temporal association between exposure and outcome. Given the population-based nature of our study and our use of risk-adjusted models, the associations we observed are likely generalisable to other regions.

Our study also had limitations. Administrative data are collected for hospital administration, which impacts data availability, particularly on patient clinical characteristics. Medication history was based on dispensation data, which may overestimate usage. Furthermore, we identified OCS bursts using a strict definition requiring an average daily dose, duration and an ED or practitioner visit with a COPD diagnosis, which may have underestimated the true number of OCS burst dispenses. Grouping dispenses can potentially obscure variation within each category, but these are likely minimal due to the iterative process to identify these categories. Additionally, we may have underestimated severe exacerbations since we identified them using an administrative data algorithm based on hospitalisations and ED visits with COPD as the main diagnosis (compared with all hospitalisations or removing the COPD main diagnosis criterion). The derived CCI will not capture long-term chronic conditions that did not result in inpatient hospitalisations 1 year prior to index. We only examined the primary diagnosis for COPD-related mortality, which may not have captured all deaths related to COPD. Lastly, we did not adjust for multiple comparisons in the statistical analysis; however, accounting for multiple comparisons would lead to the same estimates but with wider CIs. Therefore, the positive associations observed between medications, mortality and cardiopulmonary outcomes would persist.

Crucially, our study was designed to estimate the association between mortality and COPD-related medication history; a causal relationship cannot be inferred. Further, unmeasured variables, such as lifestyle risk factors, or clinical characteristics such as forced expiratory volume in 1 s or disease severity, may have been linked to SABA, OCS and antibiotic medication dispensation and mortality rates, thereby confounding the associations in our study. This limitation may impact the generalisability of our findings should those factors differ substantially between regions. A potential avenue for future research is identifying the key drivers of COPD-related medication usage, particularly those deviating from clinical guidelines.

Conclusion

Our study provides novel real-world evidence that a history of frequent SABA, antibiotic or OCS dispensing in a given year was associated with significantly higher rates of all-cause and COPD-related mortality. Associations between COPD medications and mortality (all-cause and COPD specific), MACE and CVD death (the latter two within 90 days after severe exacerbation) all followed a dose–response trend. A history of higher SABA (2+ dispensations), antibiotics (3+ dispensations) or OCS (6+ burst-days) in a year was associated with an increased burden and rate of mortality, MACE outcomes and CVD death. The consistency of these findings with the asthma literature, combined with the dose–response relations we observed and a plausible intermediate role for severe exacerbations and MACE, is compelling. Further research could help understand if a causal relationship exists between medication use and mortality and if other risk factors may be driving treatment patterns. Overall, our study suggests COPD reliever or exacerbation management medication history may be a useful proxy for healthcare professionals to identify patients with uncontrolled or suboptimal management of disease and adjust the care strategy if appropriate. Studies building on our findings will be critical for informing treatment guidelines and improving patient care.

Data availability statement

Data may be obtained from a third party and are not publicly available. The dataset supporting the conclusions of this article was derived from Alberta Health administrative data. Deidentified data were released to Medlior Health Outcomes Research by Alberta Health following ethical approval and a data request. Data from this study are not publicly available and cannot be shared for privacy reasons and ethical restrictions as per the research agreement with Alberta Health.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study involves human participants and was approved by the Health Research Ethics Board of Alberta–Community Health Committee (HREBA.CHC-21-0011). A waiver of consent was requested as the study involved analysis of anonymised data from Alberta Health.

Acknowledgments

The authors acknowledge Heather Neilson and Tram Pham, who contributed to the writing of this manuscript, and Kellie Maclean for figure development. This study is based in part on data provided by Alberta Health and Alberta Health Services.