Study design and data source
This retrospective cohort study included data for patients with COPD aged between 40 years and 90 years from the Taiwan National Health Insurance Research Database (NHIRD) in Health and Welfare Data Science Center 1 January 2002 to 31 December 2016. NHIRD was established in 1995 and covers 99% of the 23 million inhabitants in Taiwan under compulsory national health insurance. The database includes personal information, codes for diagnoses and procedures from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and Tenth Revision (ICD-10). The patient’s medication was classified according to the Anatomical Therapeutic Chemical and National Health Insurance codes. These codes are widely accepted drug classification systems coordinated by the World Health Organization Collaborating Centre for Drug Statistics Methodology. The study was approved by the Institutional Review Board of Kaohsiung Medical University.
Identification of the COPD cohort
Using the database, we identified adults aged between 40-90 years with newly diagnosed COPD (ICD-9 codes 490, 491, 492 and 496) who had received more than one inpatient diagnosis or three or more outpatient diagnoses between January 1, 2002, and December 31, 2015. Patients with COPD should receive one of the following medications for more than one month: long-acting β adrenergic agonists (LABAs); long-acting anti-muscarinic agonists (LAMAs); or oral methylxanthines combined with short-acting β2-agonists (SABAs) or short-acting muscarinic antagonists (SAMAs). [30-32] Patients were excluded if they met any of the following conditions: asthma, lung transplantation, lung cancer or death within one month after diagnosis of COPD.
Identification of the BZRA cohort
The BZRA cohort was identified from COPD patients newly exposed to BZRAs and divided into an oral BZD group and an oral non-BZD BZRA group. Patients were required to continue using the BZRA without change for 30 days. Patients were excluded if they received BZRA injections during the study period or died before the index date. Each patient was followed until an outcome occurred: mortality or the end of the study (31 December 2016). The BZRA cohort was divided into three groups based on the initial prescription: those taking BZDs (the BZD cohort), those taking non-BZD BZRAs (the non-BZD cohort), and those prescribed both BZDs and non-BZD drugs simultaneously in the initial prescription (the mix cohort). A cohort of patients with COPD who were not taking BZRAs (the nonuser cohort) was also identified.
Exposure and index date assessment
A new prescription for BZRAs was defined as the patient receiving his or her first prescription for BZRAs during the study period following entry into the cohort. We excluded patients who received their first BZRA prescription prior to the diagnosis of COPD. When patients received more than prescription during the study period, only the first was included in the analysis. The index date was the date of the first BZD, non-BZD or Mixed-use prescription after COPD diagnosis.
Covariates
The baseline characteristics and comorbidities of each patient were collected within one year prior to the index date. These included COPD severity, hypertension, diabetes mellitus, ischemic heart disease, heart failure, atrial fibrillation, arrhythmia, chronic kidney disease, malignancy, depression and anxiety. In addition, the frequency of COPD acute exacerbations was recorded. Severe acute exacerbation was defined as patients requiring admission or visits to the emergency department. Moderate acute exacerbation was defined as a patent prescription for either antibiotics or oral corticosteroids from an outpatient department. Other COPD treatment medications, including LABAs, LAMAs, inhaled corticosteroids, SABAs, SAMAs, systematic beta-2 agonists and methylxanthines, were recorded for analysis.
Propensity score matching to create the final groups
Propensity score matching was applied on a 1:1:1 basis to the BZD, non-BZD and mix cohorts to create three final groups (the BZD, non-BZD and mix groups) with balanced baseline characteristics. The matching factors included age group, sex, comorbidities and COPD medication. [34] Propensity score matching was also used to match the patients in the three BZRA groups and the patients in the nonuser cohort 1:1 to perform sensitivity analyses.
Outcome assessment
For each patient, we collected any of the following outcomes that occurred during the 30 days following the index date: outpatient visits for respiratory exacerbations, hospital admission for acute COPD exacerbation or for respiratory exacerbations, emergency department attendance for COPD or pneumonia and all-cause mortality.
Statistical analysis
Differences in baseline characteristics were evaluated by the chi-square test (for categorical variables). Univariate and multivariate Poisson regression with robust error variance analysis was used to generate crude and adjusted incidence rate ratios (IRRs) with 95% CIs for each type of outcome. Consequently, the IRRs for all the study outcomes and further sensitivity analyses were evaluated by the Poisson regression model with a robust error variance [35] between the BZD, non-BZD and mix groups. Time-to-event analyses were performed using Kaplan–Meier plots and the log-rank test. Various sensitivity analyses were performed. First, we used several different follow-up periods (60, 90, 180 and 365 days) to ensure that the results were consistent with those obtained for the 30-day follow-up. Second, we compared risks between the three BZRA groups and the nonuser group. Third, because BZDs with a longer half-life would be expected to cause more serious side effects than those with shorter half-lives, [15, 36] we stratified the BZD group according to half-life to reduce any pharmacokinetic effects. SAS statistical software (Version 9.4; SAS Institute, Inc., Cary, NC) was used to perform all the analyses. A two-sided p value <0.05 was considered statistically significant.
Patient and public involvement
There was no patient or public involvement in developing the hypothesis, the specific aims or the research questions, nor in developing plans for the design or implementation of the study.