Using Propensity Score Matching for Sequential Recruitment in Multigroup Cohort Studies

Multigroup cohort studies are used to compare exposed and unexposed groups on important outcomes. Recruitment for these studies can be done simultaneously or sequentially. When recruitment is simultaneous, there are many advanced statistical techniques, including propensity score matching, that can be used to select balanced groups to account for confounders of the effect of the exposure on the outcome. This paper presents a novel method for using propensity score matching for sequential recruitment using a five-step modular approach: 1) estimate propensity scores, 2) create propensity clusters, 3) match participants, 4) create holdout groups, and 5) enroll. These steps are repeated each recruitment period. We explain each step, providing R code to assist researchers in implementation, then demonstrate the performance of the method in a multigroup cohort study examining the effects of posttraumatic stress disorder on cardiovascular risk markers among trauma-exposed individuals. The method performed well and improved balance compared to using no selection technique, but it also performed comparably to propensity score matching with simultaneous recruitment. This method provides a novel approach to sequential recruitment in multigroup cohort studies. Future research should incorporate best practices from machine learning and evaluation of different parameter settings based on research context.