Semiparametric inference for surrogate endpoints with bivariate censored data

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Surrogate endpoints are increasingly used in clinical research to expedite drug development.
• Analyzing right-censored surrogate endpoints presents statistical challenges, particularly with standard proportional hazards models.
• Existing methods for assessing surrogacy with censored data are often complicated and may not be robust.

Purpose of the Study:

• To propose and evaluate the accelerated failure time (AFT) model as a robust alternative for analyzing surrogate endpoints in the presence of right censoring.
• To develop novel estimation and inference procedures for measures of surrogacy based on the AFT model.
• To address the complication of simultaneous censoring in both independent and dependent variables.

Main Methods:

• Application of the accelerated failure time model to surrogate endpoint analysis.
• Adaptation of the Theil-Sen estimator to handle doubly censored data (both independent and dependent variables).
• Development of asymptotic results and a novel resampling-based technique for variance estimation.

Main Results:

• The proposed AFT model-based approach provides a viable alternative to proportional hazards models for censored surrogate endpoints.
• The adapted Theil-Sen estimator and resampling technique demonstrate reliable performance in simulation studies.
• The methodology is successfully applied to real-world data from an acute myelogenous leukemia clinical trial.

Conclusions:

• The accelerated failure time model offers a flexible and effective framework for analyzing surrogate endpoints with right-censored data.
• The developed estimation and inference procedures provide robust tools for quantifying surrogacy in complex clinical trial settings.
• This approach enhances the statistical rigor for evaluating surrogate endpoints, potentially improving the efficiency of clinical research.