Doubly robust evaluation of high-dimensional surrogate markers

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Direct efficacy measures are often costly or time-consuming.
• Surrogate outcomes offer a more feasible alternative for evaluating interventions.
• Existing methods for surrogate marker evaluation are limited to single markers and randomized studies.

Purpose of the Study:

• To propose a robust and efficient method for evaluating high-dimensional surrogate markers.
• To extend surrogate marker evaluation to observational studies without randomization.
• To connect surrogate marker utility with causal inference and average treatment effect estimation.

Main Methods:

• Developed a novel method for evaluating a set of high-dimensional surrogate markers.
• Utilized causal inference principles for robust estimation of average treatment effects.
• Employed machine learning for estimating nuisance functions and relaxing model dependence.

Main Results:

• The proposed method effectively evaluates multiple surrogate markers in high dimensions.
• Demonstrated performance in observational studies, not requiring randomization.
• Established connections between the proposed approach and mediation effects.

Conclusions:

• The new method provides a powerful tool for assessing high-dimensional surrogate markers.
• Applicable to observational data, broadening the utility of surrogate markers.
• Successfully illustrated by assessing gene expression as a surrogate for immune activation in an Ebola study.