From association to causation: interpreting propensity score-based analyses in real-world evidence

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Propensity score (PS) methods are common in real-world evidence (RWE) studies for approximating randomized comparisons.
• Misinterpretation of statistical adjustment as causality can lead to flawed clinical conclusions.

Purpose of the Study:

• To present a clinically oriented framework for interpreting PS-based analyses.
• To highlight key principles of causal inference in RWE studies.

Main Methods:

• Framework development focusing on estimand definition (ATE vs. ATT).
• Covariate selection based on causal structure.
• Assessment of positivity/overlap and robustness to unmeasured confounders.
• Demonstration using anesthesiology studies.

Main Results:

• Discrepancies between analytical methods and interpretations are frequent in PS analyses.
• PS matching often estimates ATT, limiting generalizability; inverse probability of treatment weighting may yield unstable estimates with limited overlap.
• Unmeasured confounders remain a critical limitation, even with balanced covariates.

Conclusions:

• Causal interpretation of PS-based analyses necessitates alignment of estimand, assumptions, and methods.
• Researchers and clinicians must critically evaluate overlap, confounders, and generalizability beyond statistical adjustment.
• Transitioning from association to causation requires advanced statistical methods and transparent, causal reasoning-grounded interpretations.