Estimation of the optimal surrogate based on a randomized trial.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Determining valid surrogate outcomes is crucial for efficient clinical trial design.
• Current methods often lack the ability to optimally predict long-term outcomes from intermediate data.
• Randomized trials aim to infer treatment effects on final outcomes, but collecting long-term data can be resource-intensive.

Purpose of the Study:

• To define and estimate an optimal surrogate outcome that accurately predicts long-term treatment effects.
• To develop statistical methods for identifying surrogates that satisfy the Prentice definition and optimize prediction.
• To provide a framework for selecting surrogates that can streamline data collection in future randomized studies.

Main Methods:

• Utilized observational data with baseline covariates, treatment assignment, intermediate surrogate outcomes, and final outcomes.
• Defined an optimal surrogate as a conditional mean that meets surrogate endpoint criteria and maximizes prediction of the final outcome.
• Employed super-learner and targeted super-learner algorithms for estimating the optimal surrogate.

Main Results:

• Demonstrated that the optimal surrogate is a conditional mean.
• Developed and presented novel super-learner and targeted super-learner estimators for practical application.
• Showcased desirable statistical properties of the proposed optimal surrogate and its estimators through simulations.

Conclusions:

• The proposed methodology effectively identifies optimal surrogate outcomes for predicting long-term treatment effects.
• The developed estimators offer a robust approach for selecting surrogates in clinical trial settings.
• This work has direct implications for improving the efficiency and design of future vaccine efficacy trials, such as those for dengue.