Clinical & Trials
Propensity Score Methods
In one line
Statistical techniques (matching, weighting, stratification) to balance baseline characteristics in observational comparisons.
Definition
Propensity score methods estimate the probability of treatment assignment given covariates, then balance treated and untreated groups. Critical for using real-world data in regulatory and HTA submissions.
Common techniques include matching, stratification, inverse probability weighting and covariate adjustment. FDA device staff have published methods for propensity-score designs that use registry or historical control data, where the outcome-free design stage is done before outcomes are examined. Propensity scores only balance measured covariates, so unmeasured confounding remains a limitation to address in sensitivity analyses.
Why it matters
FDA's RWE program and EU JCA both expect formal balance diagnostics, sensitivity analyses, and transparent reporting of propensity-score-based comparisons.
Common pitfalls
- •Misinterpreting propensity scores as directly adjusting for confounding rather than balancing covariates is a common mistake.
- •Failing to conduct adequate balance diagnostics can lead to biased results.
- •Omitting sensitivity analyses overlooks potential unmeasured confounding and can undermine the credibility of findings.
- •Improper selection of covariates can introduce bias or reduce the efficiency of the analysis.
- •Applying propensity score methods without understanding their underlying assumptions can lead to invalid conclusions.
Sources
- FDA RWE Framework · FDA
- ClinicalTrials.gov · NIH
- ICH Guidelines · ICH