Statistical inference for the difference between two maximized Youden indices obtained from correlated biomarkers

Area of Science:

• Biostatistics
• Medical Diagnostics
• Cancer Research

Background:

• Accurate cancer biomarkers are crucial for clinical decisions, often relying on maximized Youden index for optimal sensitivity and specificity.
• Correlated measurements from multiple classification criteria within individuals necessitate specialized statistical approaches.

Purpose of the Study:

• To develop and evaluate hypothesis tests and confidence intervals for comparing two correlated receiver operating characteristic (ROC) curves based on their maximized Youden indices.
• To provide robust statistical tools for assessing the comparative performance of diagnostic biomarkers.

Main Methods:

• Proposed hypothesis tests and confidence intervals using delta-based techniques under parametric assumptions and power transformations.
• Examined nonparametric kernel-based methods for comparing correlated ROC curves.
• Validated the proposed methods through extensive simulations.

Main Results:

• The developed statistical methods enable reliable comparison of correlated ROC curves.
• Demonstrated the utility of the methods in a real-world metabolomic study for pancreatic cancer detection.

Conclusions:

• The proposed statistical framework offers a valuable tool for biomarker discovery and validation in oncology.
• These methods can enhance the comparative evaluation of diagnostic tests, particularly when dealing with correlated data.