|
Abstract:
|
Arguably, the practice of statistics involves becoming skilled at gleaning deeper insights from a dataset or research study than may be available from simplistic analysis. One technique admirably demonstrating this aspect of the beauty of statistics is the use of propensity scores to enable causal inference from observational data. Propensity scores, defined as the probability of receiving a particular treatment given a set of covariates, are balancing scores, i.e., the distribution of the covariates given the propensity score is balanced. Consequently, we can estimate average treatment effects from observational data, going well beyond the "correlation analysis" that a naïve summary may rest with. In this presentation, we describe the results of extensive simulations comparing three propensity score methods (stratification, inverse probability of treatment weighting, and doubly robust estimation) and explain how each method may affect the estimation of average treatment effect. Results are applied to a study of the effect of breast cancer chemotherapy treatment regimen on subsequent development of cardiotoxicity.
|
