There are two common methods for targeting a conditional hazard ratio in analyses of censored time-to-event data include 1) covariate adjustment through traditional outcome modeling and 2) propensity score matching with conditional Cox regression analyses, conditioned on the matched sets. As it is unclear which approach is theoretically correct, simulated data will allow us to compare the results of each approach to a known truth and will enable us to vary the degree of censoring in each exposure group and determine whether the censoring is informative or uninformative. Furthermore, we hypothesize that differences in results between the two approaches may be particularly pronounced in settings with effect modification by time, such as when there is a larger effect later in follow-up as compared to early follow-up. Using simulated data will also allow us to evaluate this phenomenon.
This goal of this project is to compare the two weighting approaches in a simulated setting of longitudinal follow-up with censoring.
Richard Wyss, PhD; Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA
Bruce Fireman, MA; Kaiser Permanente Northern California, Oakland, CA
Rishi J. Desai, PhD, MS; Joshua J. Gagne, PharmD, ScD; Shirley Wang, PhD, ScM; Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA
Margaret Johnson, Darren Toh, ScD; Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, MA
Michael Nguyen, MD; Esther Zhou, MD, PhD; Office of Surveillance and Epidemiology, Center for Drug Evaluation and Research, FDA, Silver Spring, MD
Yueqin Zhao, PhD; Office of Biostatistics, Center for Drug Evaluation and Research, FDA, Silver Spring, MD