The missing puff

Abstract

Background: Severity of chronic pulmonary obstructive disease (COPD), with key component percent predicted forced expiratory volume in 1 second (ppFEV1), is an important predictor of COPD outcomes and mortality. Objectives: To evaluate the effect of missingness in ppFEV1 for the assessment of COPD severity using data on spirometry and symptoms from the Clinical Practice Research Datalink and estimate mortality rates and rate ratios (RRs) adjusting by COPD severity when complete case analysis (CCA) and multiple imputation (MI) are used. Methods: Cohort study of new users of aclidinium and LABA medications between 2012 and 2015, aged ≥40 years with COPD. Severity was classified as GOLD A, B, C, or D according to (1) ppFEV1 (recorded or estimated using expected FEV1 and FEV1, or the Global Lungs Initiative European Respiratory Society Task Force formulas); (2) symptoms from the mMRC dyspnoea grade, the COPD Assessment Test (CAT), or physician‐recorded breathlessness; and (3) exacerbation history. The mortality rates and RRs (aclidinium vs LABA) were obtained using a negative binomial regression model adjusting for COPD severity, sex, age, smoking status, and BMI. CCA and MI were employed to account for missing data on ppFEV1. MI was performed using the fully conditional specification method with 10 imputations. The imputation model included the outcome, exposure, and many a priori relevant variables. Results: The study included 3555 new users of aclidinium and 4797 new users of LABA. Spirometry results on ppFEV1 (recorded or estimated) were missing for 12.4% of the patients. Approximately 72% of the patients with available data had a ppFEV1 >50%. The distribution of COPD severity among users of aclidinium using CCA vs MI was 24.3% vs 24.7% in category A, 17.2% vs 17.0% in B, 21.7% vs 22.5% in C, and 36.8% vs 35.8% in D. For users of LABA, it was 38.1% vs 38.0% in category A, 19.4% vs 19.2% in B, 22.5% vs 23.0% in C, and 20.0% vs 19.8% in D. The adjusted mortality rates per 1000 person‐years when using CCA vs MI were 23.08 vs 25.63 among users of aclidinium and 29.77 vs 33.90 among users of LABA. The adjusted mortality RRs for aclidinium vs LABA were 0.78 (95% CI, 0.58‐1.03) with CCA and 0.76 (95% CI, 0.57‐1.00) with MI. Conclusions: Minor differences were observed in the distribution of COPD severity and in the mortality RRs employing CCA vs MI to handle missing data on ppFEV1. Use of MI allowed the inclusion of all the study population in the analyses.