Development and Refinement of a Toolkit for Large Scale Covariate Adjustment in the Electronic Health Record and Claims Data Partner Network

A key focus for the Innovation Center has been on evaluating the feasibility of improved confounding adjustment from ultra-high-dimensional data structures that include Electronic Health Record (EHR) based variables through a combination of automated feature generation algorithms and advanced statistical and machine learning approaches for causal inference. Specifically, there has been recent interest in the use of regularized machine learning tools (e.g., Least Absolute Shrinkage and Selection Operator (LASSO) based models) combined with targeted learning methods for improved large-scale covariate adjustment.

This Innovation Center project aims to develop a reusable analysis toolkit to implement confounding adjustment using various adaptations of LASSO-based models for Propensity Score (PS) estimation in combination with targeted learning approaches for routine use in Sentinel investigations. This project has two specific aims:

• Development of an open-source toolkit in the form of reusable, user-friendly R packages to implement large-scale confounding adjustment with EHR data elements.

• Pilot deployment and user acceptance testing of the developed toolkit for empirical cohorts in at least two partner sites from the Development Network or Data Partner network to inform toolkit refinement.