A Bayesian hierarchical approach to multirater correlated ROC analysis

Area of Science:

• Biostatistics
• Medical Imaging Analysis
• Diagnostic Test Evaluation

Background:

• Receiver Operating Characteristic (ROC) studies are common for evaluating diagnostic tests.
• Analyzing ROC data involves multiple readers and modalities, introducing complex sources of variation.
• Existing methods may lack efficiency in estimating performance differences.

Purpose of the Study:

• To present a Bayesian hierarchical model for ROC study analysis.
• To explicitly model the three inherent sources of variation in ROC data.
• To improve the precision and reduce the mean squared error (MSE) of performance difference estimates.

Main Methods:

• Developed a Bayesian hierarchical model tailored for ROC study designs.
• The model explicitly accounts for reader variability, case variability, and modality effects.
• Utilized simulation studies to assess model performance.

Main Results:

• The proposed model significantly reduces posterior uncertainty in estimating differences between areas under the ROC curves (AUCs).
• Mean squared error (MSE) for these estimates is substantially reduced, often by a factor exceeding five.
• Coverage intervals for AUC differences are narrower, indicating increased precision.

Conclusions:

• The Bayesian hierarchical model offers a powerful approach for ROC data analysis.
• This methodology enhances statistical power for detecting differences in diagnostic test performance.
• The findings have significant implications for clinical trial design and diagnostic accuracy research.