Calendar time-specific propensity score analysis for observational data: a case study estimating the effectiveness of inhaled long-acting beta-agonist on asthma exacerbations
Propensity scores (PS) are frequently used in observational studies. PS are usually estimated over the entire study period without consideration of the effect of changing patterns of the included variables over time. This study sought to compare PS estimated using the entire study period (conventional PS) and PS estimated for specific periods (calendar time-specific PS (CTS-PS)), and to determine whether there are differences in estimated treatment effects using these approaches.
We conducted a claims data analysis. Asthmatic patients who received an asthma controller during 1997–2008 were included. Exposed patients were those who received an inhaled long-acting beta-2 agonist. Conventional PS used the entire period to estimate a PS for individuals. CTS-PS approach divided the study period into 1-year periods and estimated PS separately for each period. Each individual had two PS. Both PS approaches were used to estimate adjusted hazard ratio (HR) for asthma exacerbations using Cox proportional hazard models.
A total of 288,518 patients with an average age of 11.9 ± 5.8 years were included. The difference between conventional PS and CTS-PS in each period ranged from −0.213 to 0.098. The adjusted HR of conventional PS-matched cohort was 1.20 (95%CI: 1.18–1.22), whereas the estimate for the CTS-PS-matched cohort was 1.24 (95%CI: 1.23–1.37).
Focusing on a specific year, there was a difference between conventional PS estimated versus CTS-PS for that year. However, there was minimal effect of CTS-PS on the observed treatment effects compared with conventional PS approach. Copyright © 2013 John Wiley & Sons, Ltd.