W-42: Performance of Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies
Poster Presenter
Kun Nie
Director of Biostatistics
Clindata Insight Inc United States
Objectives
Assess the choice of covariates, development methods, and applications of propensity score for reducing the effects of confounding in observational studies
Method
Simulation studies motivated by a real cross-study comparison analysis
Results
Propensity score method is effective in reducing confounding effect.
Conclusion
There is a growing interest in using observational studies to estimate the effects of treatments on outcomes. In observational studies, treatment selection is often in?uenced by subject characteristics. Therefore, one must account for systematic differences in baseline characteristics between treated and untreated subjects when estimating the effect of treatment on outcomes. The propensity score allows one to design and analyze an observational (nonrandomized) study so that it mimics some of the particular characteristics of a randomized controlled trial.
While propensity score adjustments are widely used in observational studies, there is a lack of consensus as to which variables to include in the propensity score model. Simulation studies are conducted to compare the performance of different strategies to choose covariates. We recommend to include real confounders in the derivation of propensity score to reduce the bias of treatment effect estimates, while include prognostic covariates that do not correlate with treatment selection in regression model for outcome to reduce the variance of estimates.
Propensity scores are generally estimated using logistic regression. However, parametric models require assumptions regarding the functional form. If the assumptions are incorrect, covariate balance may not be achieved by conditioning on the propensity score, which may result in a biased effect estimate. We examine the use of machine learning methods such as random forest as one alternative to logistic regression.
Finally, we discuss 4 commonly used propensity score methods in analyses: matching, strati?cation, inverse probability of treatment weighting using the propensity score, and covariate adjustment using the propensity score. The performances of those 4 different propensity score adjustment methods are evaluated via simulation studies. We recommend the use of IPTW method when variables are appropriately selected to balance baseline characteristics.
