This study (NCT04138758) was conducted using data from the HealthCore Integrated Research Database (HIRD; January 2013–March 2019), a large administrative healthcare claims database maintained by HealthCore that has previously been used for the study of numerous diseases, including COPD [28,29,30,31,32]. The study was conducted using administrative claims in the form of a limited data set pursuant to data agreements between HealthCore and participating health plans in compliance with the US Health Insurance Portability and Accountability Act. The study did not require ethics committee approval, nor were subjects required to provide informed consent.
HIRD includes longitudinal medical and pharmacy claims data for health plan members residing in the USA. Member enrollment, medical care (professional and facility claims), outpatient prescription drug use, outpatient laboratory test result data, and healthcare utilization may be tracked for health plan members in the database dating back to January 2006, and with diagnoses recorded in International Classification of Diseases, version 10 (ICD-10) since October 2015. Laboratory data are available for those tests that have been performed using two large, national reference laboratories (Quest and LabCorp).
Study Population and Design
Patients in each cohort were required to have at least one prescription for a fixed-dose combination (FDC) inhaler of tiotropium/olodaterol or LABA/ICS between 1 January 2013 and 31 March 2019, with the first prescription defined as the index date. Patients were also required to have at least one diagnosis of COPD at any time prior to the index date and at least 1 year of continuous health plan eligibility prior to the index date. Patients were excluded for the following reasons: age less than 40 years at the index date; diagnosis of asthma in the year prior to the index date, lung cancer, interstitial lung disease, or lung transplant; or pre-index use of tiotropium/olodaterol, LABA/ICS, or LABA/LAMA/ICS in free or fixed form.
All analyses were conducted for the total population, as well as in subgroups of patients at high or low risk of exacerbation based on (1) previous history of exacerbations in the year preceding cohort entry (low exacerbation history, 0 hospitalizations and 0–1 outpatient exacerbations; high exacerbation history, at least 1 hospitalization or at least 2 outpatient exacerbations); and (2) blood eosinophil count (baseline eosinophils less than 300 cells/µL; baseline eosinophils at least 300 cells/µL) as identified on the basis of the laboratory result value that was closest but prior to the index date (within 6 months).
Exposure measures were based on pharmacy dispensation of tiotropium/olodaterol and LABA/ICS over up to 1 year of follow-up. Exposure was based on the current use of tiotropium/olodaterol or LABA/ICS as defined by the dispensation date plus the day’s supply recorded at the time of dispensation.
A gap of 15 days was allowed between dispensations to allow for delays in obtaining medication refills and for continued use beyond the days supplied where medication has been missed because of imperfect adherence. The 15-day gap was increased in the sensitivity analyses to 30 and 60 days.
Patients were followed from the index date until the earliest of the following: first occurrence of a COPD exacerbation, community-acquired pneumonia, escalation to triple therapy, switch in treatment, discontinuation of COPD treatment, the end of the study period, the end of continuous health plan eligibility, or (for the main analyses) 1 year after cohort entry. The treatment segment ended at the earliest of the following events: (1) 15 days after the end of the observed day’s supply for the medication received on the index date without a subsequent dispensing of COPD medication (i.e., discontinuation); (2) initiation of triple therapy; or (3) any other change in use of study medication, including to a different combination therapy, from a fixed-form to a free-form combination therapy, or from combination therapy to monotherapy.
The primary outcome was to compare the risk of first COPD exacerbation after initiating treatment with tiotropium/olodaterol versus LABA/ICS. COPD exacerbation was defined as a COPD-related hospitalization (severe exacerbation) or an emergency department visit for COPD and/or prescription of an antibiotic for a respiratory condition the same day as an oral corticosteroid (moderate exacerbation). To increase the specificity of our definition of exacerbation, the prescription of either an antibiotic or oral corticosteroid alone was not included. Severe and moderate exacerbations were considered as a composite for the main analyses.
Secondary outcomes included the risk of hospitalization for community-acquired pneumonia, risk of pharmacy dispensation indicating escalation to triple therapy, and risk of any of the following events occurring: COPD exacerbation, hospitalization for pneumonia, or escalation to triple therapy. Pneumonia was defined as hospitalization with one of the following diagnostic codes: 481.x–486.x, 487.0, 507.x, 507.0, 507.1, 507.8, 510.0, 510.9, 511.0, 513.0, 514.x, 517.1, 519.8, 530.84 (as per International Classification of Diseases, version 9, Clinical Modification [ICD-9-CM]; used up to October 2015) and J10.0, J11.0, J12–J18, J22, J69, J85.0, J85.1, and J86 (ICD-10; used from October 2015 onwards) (pneumonia without hospitalization was not captured). These definitions from hospitalization data have been used in several previous studies in patients with COPD  and are detailed in Table S1 in the supplementary material. As a result of the nature of the ICD diagnostic codes (e.g., J10.0: “influenza due to other identified influenza virus with pneumonia” [ICD-10]), it was not possible to verify pneumonia as the primary driver of hospitalization. Escalation to triple therapy was defined as any combination of pharmacy dispensations resulting in the overlapping use of a LAMA, LABA, and ICS through any fixed or free combination for at least 1 day.
Outcome rates are described using incidence rates (IRs) and associated 95% confidence intervals (CIs). Differences in the risk of a first COPD exacerbation, pneumonia, escalation to triple therapy, or any one of these events (COPD exacerbation, or pneumonia, or escalation to triple therapy) for tiotropium/olodaterol use relative to LABA/ICS use were assessed using hazard ratios (HRs) and associated 95% CIs. Cox proportional hazard models were performed as an as-treated analysis to derive HRs; a first treatment carry-on analysis was conducted as a sensitivity analysis. Confounding was controlled via fine stratification and reweighting of an exposure propensity score and by including potential confounding factors in the models for variables that remained imbalanced after reweighting. Data trimming was applied on the basis of the overlap of propensity score distributions .
Data were analyzed separately and for subgroups based on blood eosinophil levels (for those with available results) and exacerbation history using the same approach. Calculation of propensity scores and fine stratification and reweighting were repeated within the patient subgroups to create weighted populations.
Propensity scores including both pre-specified and high-dimensional, data-derived variables were calculated; scores were calculated on the basis of the following covariates: sex, age, calendar year of cohort entry, season of index date, US census region of residence, insurance type, medication history, exacerbation history and pre-specified chronic comorbidities, in addition to the most frequently occurring diagnoses, medications, and procedures observed. Adjustment for the propensity score used the fine stratification method to create a weighted pseudo-population . Ten strata were defined on the basis of the distribution of the propensity score within tiotropium/olodaterol-exposed individuals for the region of overlap with the LABA/ICS group, and stratum-specific weights were applied to create balance between the exposure groups. The balance of patient characteristics between the treatment groups was described before and after propensity score application. The Cox proportional hazard model was further adjusted for patient characteristics found to be imbalanced after application of the propensity score, where imbalance was defined as standardized differences greater than 10%. Several sensitivity analyses and bias analyses were performed, as described in the supplementary methods.