Combining dependent tests to compare the diagnostic accuracies--a non-parametric approach

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Accuracy Research

Background:

• Comparing diagnostic accuracies across multiple readers and tests is crucial in medical research.
• Existing non-parametric methods may face challenges with reader heterogeneity and correlated data.
• Receiver Operating Characteristic (ROC) analysis is a standard tool for evaluating diagnostic performance.

Purpose of the Study:

• To propose a novel non-parametric approach for comparing diagnostic accuracies in multi-reader ROC studies.
• To develop a method robust to reader heterogeneity and applicable to correlated ROC data.
• To provide a flexible statistical framework for analyzing diagnostic test performance.

Main Methods:

• Constructing reader-specific tests by extending conventional non-parametric methods.
• Combining individual test statistics to derive an overall conclusion on diagnostic test accuracy.
• Handling both continuous and ordinal data types within the ROC framework.

Main Results:

• The proposed method demonstrates robustness against reader heterogeneity.
• The approach effectively analyzes correlated ROC studies.
• Simulation studies and a real-world example validate the method's performance.

Conclusions:

• The developed non-parametric approach offers a robust and versatile tool for comparing diagnostic accuracies.
• This method improves upon existing techniques by addressing reader variability and data correlation.
• The findings have significant implications for the statistical analysis of diagnostic accuracy in clinical research.