Propensity score methods in medical research: a concise practical guide with applications to emergency medicine.

Abstract

Propensity score (PS) methods can reduce measured confounding in observational emergency medicine research, but validity depends more on design choices and diagnostics than on the treatment-prediction model. This clinical review presents a practical workflow for defining the target estimand, selecting pretreatment covariates, choosing matching or weighting, checking covariate balance and treatment overlap, and analyzing a binary outcome in the population created by the PS design. The average treatment effect (ATE), average treatment effect in the treated (ATT), and average treatment effect in the overlap population (ATO) answer different clinical questions. A worked example of 1,200 simulated emergency patients uses a common, explicit time zero to show how confounding by indication can make immediate treatment appear harmful before adjustment and how PS matching, stabilized inverse probability of treatment weighting, and overlap weighting change both the comparison population and estimated effect. Standardized mean differences, PS distributions, weight summaries, and effective sample size are presented before treatment-specific risks, risk differences, and risk ratios. A standalone supplement provides package-based R code, expected output, and analysis-derived figures. Survival analysis is outside the scope. PS methods are most credible when used as design tools: define the question, construct a supported comparison, verify diagnostics, and only then analyze outcomes.