Doubly robust matching estimators for high dimensional confounding adjustment.

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Accurate estimation of treatment effects from observational data is crucial but challenged by confounding.
• High-dimensional covariate spaces, common in real-world data, render traditional methods for controlling confounding infeasible.
• Existing methods like variable selection or penalization can reduce dimensionality but may not always suffice.

Purpose of the Study:

• To propose a novel matching strategy for valid treatment effect estimation in the presence of numerous covariates.
• To address the challenge of high-dimensional confounding in observational studies.
• To develop a method robust to potential model misspecification in score estimation.

Main Methods:

• Proposed a new matching estimator based on both propensity scores and prognostic scores.
• Derived asymptotic properties of the proposed matching estimator.
• Investigated the estimator's performance through simulations and applied it to real-world insurance claims data.

Main Results:

• The proposed matching estimator demonstrates effectiveness in controlling for confounding, even with a large number of covariates.
• The estimator is shown to be 'doubly robust,' requiring only one of the two score models (propensity or prognostic) to be correctly specified for consistent estimation.
• The method shows potential for addressing complex, nonlinear confounding structures.

Conclusions:

• Matching on both propensity and prognostic scores offers a robust and effective approach for estimating treatment effects in high-dimensional observational data.
• The doubly robust nature of the estimator enhances its reliability in practical applications.
• This method provides a valuable tool for analyzing complex health data, such as the impact of gender on prescription opioid use.