Confounding adjustment with propensity scores for overlap weighting in observational studies: a concise primer.

Overlap weighting has emerged as a valuable method for addressing confounding in observational studies, particularly in real-world data settings characterized by imbalanced covariates and limited overlap between treatment groups. Its ability to produce stable, interpretable estimates makes it an attractive alternative to inverse probability of treatment weighting (IPTW), which can suffer from extreme weights and instability. This report outlines the methodological basis of overlap weighting and contrasts it with IPTW. The limitations of IPTW are illustrated through a clinical example comparing clopidogrel and prasugrel, where substantial baseline differences lead to poor propensity score (PS) overlap. Overlap weighting is discussed as a solution that emphasizes individuals in clinical equipoise (i.e. PS near 0.5), minimizes the influence of outliers, and achieves exact covariate balance. Additional sections address stabilized weights, recent literature applications, and considerations for target estimands. Overlap weighting is well-suited for observational studies with moderate to poor overlap and should be considered a preferred approach in many real-world contexts. Presenting results from multiple PS methods, including standardized mortality ratio (SMR) weighting, IPTW, PS adjustment, and overlap weighting, can help assess robustness and enhance the credibility of causal inferences.