[Indication bias in retrospective observational studies: applications of propensity scores]

Analytical observational studies are useful for evaluating the effects of interventions when randomized controlled trials are not feasible. However, the absence of randomization introduces confounding by indication, as external factors known as confounders may be associated both with the probability of receiving a treatment and with the occurrence of an outcome (without being part of the causal pathway), potentially distorting observed results. Among the available strategies to adjust for potential confounders and emulate randomization, the Propensity Score (PS) is one of the most widely used methods. It is defined as the probability of receiving an intervention given a specific set of covariates, which should be selected based on causal considerations and followed by rigorous assessment of covariate balance between groups using standardized differences and graphical tools.The PS can be implemented through several approaches, including direct adjustment, stratification, matching, or inverse probability of treatment weighting (IPTW). Each technique yields different types of estimands (conditional or marginal) depending on the analytic objective. Although these methods do not guarantee causal inference on their own, they represent robust approaches for reducing confounding in observational studies, underscoring the importance of their proper understanding and application.