Addressing subject heterogeneity in time-dependent discrimination for biomarker evaluation

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Biomarker Research

Background:

• Accurate biomarker discrimination is crucial for disease monitoring and early detection.
• Biomarker accuracy for time-to-event outcomes changes over time, necessitating time-dependent measures like the time-dependent receiver operating characteristic curve and area under the curve (AUC).
• Existing measures do not account for subject heterogeneity, limiting the understanding of how covariates influence biomarker performance.

Purpose of the Study:

• To propose and develop a novel measure, the covariate-specific time-dependent AUC, for covariate-adjusted discrimination.
• To investigate how covariates influence biomarker performance in terms of magnitude and effect.
• To provide a statistical framework for estimation, inference, and asymptotic properties of the proposed measure.

Main Methods:

• Development of a regression model for the covariate-specific time-dependent AUC.
• Construction of a pseudo partial-likelihood for estimation and inference.
• Establishment of asymptotic properties of the proposed estimators and provision of variance estimation.

Main Results:

• The proposed covariate-specific time-dependent AUC method effectively assesses covariate-adjusted discrimination.
• The regression model successfully elucidates the influence of covariates on biomarker performance.
• Simulation studies and application to real-world data (AIDS Clinical Trials Group 175) validate the method's utility.

Conclusions:

• The covariate-specific time-dependent AUC is an informative tool for evaluating biomarker predictive discrimination.
• This method enhances understanding of biomarker performance by incorporating subject heterogeneity and time-dependent effects.
• The approach is valuable for clinical applications requiring accurate biomarker assessment in diverse patient populations.