Receiver operating characteristic analysis under tree orderings of disease classes

Area of Science:

• Biostatistics
• Medical Diagnostics
• Machine Learning

Background:

• Receiver Operating Characteristic (ROC) curves and Area Under Curve (AUC) are standard for binary diagnostic accuracy.
• Existing extensions like ROC surfaces and naive AUC (NAUC) have limitations for ordered multi-class problems.
• Current methods like Umbrella Volume (UV) are restricted to three classes and have drawbacks.

Purpose of the Study:

• To propose a novel ROC framework for tree or umbrella ordering (TROC).
• To introduce the Area Under TROC curve (TAUC) as a superior diagnostic measure for ordered multi-class classification.
• To develop and evaluate methods for estimating TAUC confidence intervals.

Main Methods:

• Introduced the TROC framework and TAUC diagnostic measure.
• Explored both parametric and nonparametric approaches for TAUC confidence interval estimation.
• Conducted simulation studies to compare method performance under various settings.

Main Results:

• The proposed TROC and TAUC demonstrate desirable properties similar to traditional ROC and AUC.
• Simulation studies validated the performance of the proposed TAUC estimation methods.
• The methods were successfully applied to a real-world microarray dataset.

Conclusions:

• The TROC framework and TAUC offer a valuable advancement for diagnostic accuracy assessment in ordered multi-class settings.
• TAUC provides a more appropriate and robust measure compared to NAUC and UV.
• The developed methods for TAUC estimation are effective and applicable to biological data analysis.