ROC Estimation from Clustered Data with an Application to Liver Cancer Data

Area of Science:

• Biostatistics
• Medical Imaging Analysis
• Diagnostic Test Evaluation

Background:

• Comparing diagnostic methods with clustered ordinal outcomes presents statistical challenges.
• Existing models may not adequately address the correlation within clustered data.
• Accurate performance assessment is crucial for selecting optimal diagnostic strategies.

Purpose of the Study:

• To propose a novel regression model for comparing diagnostic methods with clustered ordinal test outcomes.
• To provide a flexible framework for analyzing correlated ordinal data in diagnostic research.
• To facilitate the selection of the most effective diagnostic methods.

Main Methods:

• A regression model treating ordinal outcomes as grouped-survival time data.
• Inclusion of random effects to account for correlations within clusters.
• Use of covariates to represent different diagnostic methods for performance comparison.
• Development of a model defining a Lehmann family and a location-scale family of receiver operating characteristic (ROC) curves.

Main Results:

• The proposed model effectively handles clustered ordinal test outcomes.
• It allows for direct comparison of diagnostic method performances through coefficient analysis.
• The model framework integrates with receiver operating characteristic (ROC) curve analysis.
• Demonstrated applicability using magnetic resonance imaging (MRI) for liver lesion detection.

Conclusions:

• The developed regression model offers a robust approach for comparing diagnostic methods with clustered ordinal data.
• It provides a unified framework for analyzing diagnostic test performance and correlation.
• The model is practical, estimable with standard software (SAS, SPSS), and applicable to real-world medical imaging scenarios.