Outcome-adaptive variable selection for causal inference
Inverse probability weighting is a common approach to account for confounding in observational studies. As the amount of data available from electronic medical records and other data sources increases, statistical learning methods for selecting variables to include in weight models becomes increasingly important. In this talk we will introduce a novel method for selecting which variables to include in the propensity score model that is used for inverse probability of treatment weighted estimates. We will introduce propensity scores and inverse probability weighting, lasso and adaptive lasso, and then describe our outcome-adaptive variable selection method. We will provide simulations to show the importance of variable selection in this setting as well as simulations on the performance of our method.