Regularized Regression Versus the High-Dimensional Propensity Score for Confounding Adjustment in Secondary Database...

Area of Science:

• Epidemiology
• Biostatistics
• Health Data Science

Background:

• Confounder selection is crucial for accurate analysis in non-randomized studies.
• Variable selection for confounder adjustment presents practical challenges, especially in secondary data analyses.

Purpose of the Study:

• To compare the high-dimensional propensity score (hdPS) algorithm with regularized regression methods (ridge, lasso) for confounder selection.
• To evaluate method performance based on bias and mean squared error in estimating treatment effects.

Main Methods:

• A simulation study using the plasmode simulation framework based on two pharmacoepidemiologic cohorts.
• Generation of realistic simulated datasets with thousands of potential confounders.
• Comparison of hdPS, ridge regression, and lasso regression for confounder adjustment.

Main Results:

• High-dimensional propensity score approaches generally demonstrated superior performance compared to regularized regression.
• Regularized regression approaches showed good performance when selected variables were used in a standard propensity score model.

Conclusions:

• High-dimensional propensity score methods are effective for confounder selection in complex datasets.
• Lasso regression followed by standard propensity score adjustment is a viable alternative for variable selection in confounder adjustment.