Analyzing heterogeneity in biomarker discriminative performance through partial time-dependent receiver operating...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Prognostic Biomarkers

Background:

• Biomarker performance evaluation is crucial for clinical decision-making.
• Traditional metrics like Area Under the Curve (AUC) may not fully capture performance relevant to population screening.
• Partial AUC (PAUC) offers a more focused evaluation for diagnostic utility.

Purpose of the Study:

• To develop and validate a regression model for assessing the heterogeneity of biomarker discriminative performance using Partial AUC (PAUC).
• To investigate variations in biomarker predictive accuracy across diverse patient subgroups for time-to-event outcomes.
• To apply the developed methods to real-world data for early Alzheimer's disease diagnosis.

Main Methods:

• Proposed a novel regression model specifically designed for Partial AUC (PAUC) estimation.
• Developed two distinct estimation procedures for discrete and continuous covariates.
• Employed a pseudo-partial likelihood method for robust inference.
• Conducted simulation studies to evaluate the proposed methods under various scenarios.

Main Results:

• The proposed regression model and estimation procedures effectively evaluated biomarker performance heterogeneity.
• Simulation studies demonstrated the reliability of the methods across different covariate types and scenarios.
• Application to the Alzheimer's Disease Neuroimaging Initiative dataset revealed significant heterogeneities in biomarker performance based on patient characteristics.

Conclusions:

• The developed statistical framework provides a robust approach to assessing biomarker discriminative performance heterogeneity.
• Understanding performance variations is essential for accurate biomarker interpretation and application in clinical practice, particularly in early disease diagnosis.
• This methodology enhances the evaluation of biomarkers for time-to-event outcomes, offering valuable insights for personalized medicine and population health strategies.