“The high-dimensional propensity score (hdPS) algorithm has been shown to reduce bias in nonrandomized studies of treatments in administrative claims databases through empirical selection of confounders. Lasso regression provides an alternative confounder selection method and allows for direct modeling of the outcome in a high-dimensional covariate space through shrinkage of coefficient estimates. However, these methods have not been able to be compared, due to limitations in ordinary simulation techniques. In this talk, I will discuss a novel "plasmode" simulation framework that is better suited to evaluating methods in the context of a high-dimensional covariate space, and I will present a study in progress that uses this framework to compare the performance of hdPS to that of a lasso outcome regression model for reduction of confounding bias.”
The Department of Quantitative Health Sciences and the Quantitative Methods Core will conduct monthly seminars to explore statistical issues of general interest.
Please email with any questions.