The use of prior event rate ratio adjustment method for controlling unmeasured confounding in pharmacoepidemiologic studies: A cautionary note

Abstract

Background: Unmeasured confounding is one of the principal problems in observational pharmacoepidemiologic studies. Prior event rate ratio (PERR) adjustment method has been proposed to control for unmeasured confounding. Objectives: To assess the performance of the PERR method in realistic pharmacoepidemiologic settings. Methods: Simulation studies were performed in several scenarios with varying effects of prior events on the probability of subsequent exposure, incidence rates, strength of confounders in prior and post periods, and rate of mortality/dropout. Exposure effects were estimated using conventional rate ratio (RR) and PERR adjustmentmethods. For the PERR method, the exposure effect is a ratio of two RRs: RR post exposure initiation and RR prior to initiation of exposure. In each simulation, the sample size was 100000 and each scenario was replicated 10000 times. 95% confidence intervals were estimated in a non-parametric way using the 2.5 and 97.5 percentiles of the 10000 estimates. Results: The exposure effects from the PERR adjustment method are highly biased when “prior” events influence the probability of subsequent exposure or when confounding differs considerably between prior and post periods. For example, the RR ranged from 1.52 to 1.10 (true RR= 2.00) when the effect of prior events on the exposure was RR 1.25 to 1.70, respectively. With a strong effect of prior events on the exposure (e.g. RR= 1.70), the bias of the estimates were more pronounced for PERR method than for the conventional method. In such case, even with a null exposure effect (RR = 1.00), the estimates shifted away from the null. In all settings, the confidence intervals of the estimates were wider for the PERR method than for the conventional method. Conclusions: The PERR adjustment method has significant limitations; in particular situations, e.g. when prior events strongly influence the probability of subsequent exposure, it can be more biased than conventional methods. Hence, caution should be exercised when applying this method and theoretical justification should be provided for underlying assumptions of the PERR.