High-dimensional propensity scores for empirical covariate selection in secondary database studies: Planning,...

Area of Science:

• Pharmacoepidemiology
• Health Informatics
• Biostatistics

Background:

• Real-world evidence (RWE) requires robust epidemiological methods to minimize confounding in non-randomized studies.
• Propensity score (PS) analysis is a key technique for adjusting measured preexposure covariates.
• The high-dimensional propensity score (hdPS) method enhances traditional PS by incorporating a large number of covariates.

Purpose of the Study:

• To provide an overview of the high-dimensional propensity score (hdPS) approach.
• To offer recommendations for the planning, implementation, and reporting of hdPS in healthcare databases.
• To support causal treatment-effect estimations using RWE.

Main Methods:

• The hdPS method is an automated, data-driven approach for covariate selection.
• It empirically identifies preexposure variables and proxies for inclusion in the PS model.
• This extends traditional PS covariate selection to handle large numbers of covariates.

Main Results:

• hdPS can potentially reduce confounding bias in the analysis of longitudinal healthcare databases.
• The article provides a checklist to aid investigators in implementing and reporting hdPS.
• It also assists decision-makers in understanding studies that use hdPS.

Conclusions:

• The hdPS method is a valuable tool for causal inference from RWE.
• Transparent planning, implementation, and reporting are crucial for reliable hdPS application.
• This approach enhances the validity of RWE for regulatory, payer, and clinical decisions.