Confidence intervals for differences between volumes under receiver operating characteristic surfaces (VUS) and...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Machine Learning

Background:

• Comparing diagnostic accuracy of biomarkers is crucial for clinical decision-making.
• Existing methods for multi-class classification accuracy assessment have limitations.
• Receiver operating characteristic (ROC) surface analysis offers a framework for evaluating diagnostic tests.

Purpose of the Study:

• To explore and develop methods for constructing confidence intervals for differences in diagnostic accuracy indices.
• To address methodological gaps in parametric and non-parametric approaches within the ROC surface framework for three-class problems.
• To provide practical guidance on the selection and implementation of these statistical methods.

Main Methods:

• Development of novel and existing statistical methods for confidence interval construction.
• Application of parametric and non-parametric approaches to ROC surface analysis.
• Extensive simulation studies under various distributional and sample size scenarios.
• Illustration of methods using real-world data from the Alzheimer's Disease Neuroimaging Initiative.

Main Results:

• The study provides a comprehensive comparison of different methods for assessing biomarker accuracy differences.
• Simulation results offer insights into the performance and appropriateness of each method across diverse scenarios.
• The proposed methods are demonstrated to be applicable to complex, real-world clinical data.

Conclusions:

• The article fills critical methodological gaps in the statistical analysis of diagnostic accuracy for three-class classification.
• The findings enhance the toolkit for researchers and clinicians evaluating competing biomarkers.
• The study promotes more robust and reliable comparisons of diagnostic test performance.