Evaluating time-to-event surrogates for time-to-event true endpoints: an information-theoretic approach based on...

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Causal Inference

Background:

• Rigorous statistical evaluation is crucial for surrogate endpoints in clinical trials.
• Existing frameworks for surrogate endpoint evaluation have limitations.
• Time-to-event endpoints present unique challenges in causal inference.

Purpose of the Study:

• To extend information-theoretic causal inference for time-to-event surrogate endpoints.
• To introduce and evaluate a rank-based metric for the individual causal association (ICA).
• To address identifiability issues in causal inference using sensitivity analysis.

Main Methods:

• Utilized D-vine copulas for modeling joint distributions of time-to-event outcomes.
• Employed mutual information to define the individual causal association (ICA).
• Developed a rank-based metric as an alternative to the standard ICA for composite endpoints.
• Implemented sensitivity analysis to quantify uncertainty and ignorance in ICA estimation.

Main Results:

• Proposed methods effectively evaluate surrogate endpoints for time-to-event data.
• The rank-based metric provides a robust alternative for composite endpoints.
• Sensitivity analysis yields intervals of ignorance and uncertainty for ICA.
• The R package 'Surrogate' implements the developed techniques.

Conclusions:

• The extended causal-inference approach offers a robust framework for surrogate endpoint validation.
• The proposed methods enhance the statistical rigor of surrogate endpoint selection in clinical trials.
• The 'Surrogate' R package facilitates the application of these advanced statistical techniques.