Performance of propensity score methods in observational studies: a systematic review

Observational studies are increasingly used in health research. Although several studies have examined the performance of propensity score (PS) methods, a critical need remains for a comprehensive synthesis of these approaches. This review aimed to summarize existing evidence and offer practical guidance for researchers to estimate treatment effects in observational studies.

This narrative synthesis systematic review was conducted in accordance with PRISMA guidelines and included peer-reviewed studies published between 2000 and 2025. Eligible studies reported methodological performance indicators such as treatment effect estimation, bias, mean squared error (MSE), and confidence interval (CI) coverage. All the search results were managed via Zotero to remove duplicates. Two independent reviewers screened titles and abstracts for relevance. Studies were categorized by design, estimand, and the PS method used for comparison.

Propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) generally improved covariate balance and acceptable CI coverage across multiple evaluation metrics. Caliper-based optimal and full matching reduce bias and improve covariate balance for odds ratios (ORs) and marginal hazard ratios (MHR). IPTW, particularly when associated with doubly robust (DR) methods or stabilized weights, offers low MSE and reliable inference. PS adjustment is also effective in reducing bias and estimating relative risk (RR). In contrast, PS stratification often underperforms due to its sensitivity to effect size, treatment prevalence, and PS distribution, leading to greater bias and less precise estimates.

PSM, DR-IPTW, and full matching are among the most effective methods for reducing bias, achieving covariate balance, minimizing MSE, and enhancing precision in observational studies.