DisC²o-HD: Distributed causal inference with covariates shift for analyzing real-world high-dimensional data.

Area of Science:

• Health Informatics
• Biostatistics
• Machine Learning

Background:

• High-dimensional healthcare data (EHR, claims) present challenges: numerous variables, multi-site data consolidation, and covariate shift.
• Estimating treatment effects in such data requires robust methods to handle heterogeneity across sites.

Purpose of the Study:

• To propose a novel distributed learning algorithm, DisC²o-HD, for estimating the average treatment effect (ATE) in high-dimensional healthcare data.
• To address covariate shift and data heterogeneity across multiple clinical sites.

Main Methods:

• Developed DisC²o-HD, a distributed learning algorithm utilizing surrogate likelihood.
• Employs propensity score and outcome model calibration to achieve covariate balancing and account for covariate shift.
• Demonstrates that the distributed estimator approximates the pooled estimator.

Main Results:

• The proposed estimator is consistent if either the propensity score or outcome regression model is correctly specified.
• Achieves semiparametric efficiency when both models are correctly specified.
• Simulation studies and real-world data application validate the algorithm's performance and readiness.

Conclusions:

• DisC²o-HD offers a valid and implementable solution for estimating ATE in distributed, high-dimensional healthcare data with covariate shift.
• The algorithm provides a robust approach to leveraging multi-site data while maintaining statistical validity.