An alternative trial-level measure for evaluating failure-time surrogate endpoints based on prediction error.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Cancer Research

Background:

• Meta-analysis is crucial for validating time-to-event surrogates against established endpoints such as overall survival.
• Current correlation approaches using Kendall's tau and the coefficient of determination (R²) have limitations due to estimation errors in treatment effects.

Purpose of the Study:

• To introduce and validate a novel prediction error-based approach for evaluating time-to-event surrogate endpoints.
• To assess the robustness of this new method compared to traditional R² estimation, particularly in meta-analytic settings.

Main Methods:

• Developed a prediction error measure based on the weighted difference between observed and model-predicted treatment effects.
• Estimated these measures using cross-validation within meta-analyses and external validation on trial data.
• Conducted simulation studies varying trial number, size, Kendall's tau, and R²; applied methods to gastric cancer trial data.

Main Results:

• The proposed distance-based prediction error measures demonstrated robustness across various simulation scenarios.
• These measures were effective even with differing numbers and sizes of clinical trials and varying correlation levels.
• Application to gastric cancer meta-analysis data confirmed the practical utility of the prediction error approach.

Conclusions:

• The absolute prediction error serves as a robust and reliable alternative to the trial-level R² for evaluating candidate time-to-event surrogates.
• This method enhances the validation process for surrogate endpoints in clinical research, offering improved accuracy and stability.