Regional confidence bands for ROC curves

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• Diagnostic test performance relies on sensitivity and specificity, which are influenced by cut-off points.
• Receiver operating characteristic (ROC) curves visualize test performance across all possible cut-off values.
• Estimating ROC curves from clinical data requires methods to select optimal cut-off points while considering sample variability.

Purpose of the Study:

• To propose a nonparametric asymptotic approach for constructing regional confidence bands for ROC curves.
• To provide a method applicable to any specificity interval of interest for ROC curve analysis.
• To offer a tool for selecting optimal cut-off points in quantitative diagnostic tests.

Main Methods:

• Development of a nonparametric asymptotic approach based on empirical and quantile processes.
• Conducting a Monte Carlo study with normal and log-normal data to assess small sample properties.
• Presentation of a sample size calculation method for ROC curve analysis.

Main Results:

• The proposed method allows for the construction of regional confidence bands for ROC curves.
• The study investigated the small sample properties of different approaches through simulation.
• A practical method for sample size determination in ROC analysis was presented.

Conclusions:

• The developed nonparametric approach provides a valuable tool for ROC curve analysis and optimal cut-off selection.
• The method is flexible, allowing for confidence bands within specific specificity intervals.
• The findings are applicable to diagnostic test evaluation, including the analysis of a tumor marker for bone marrow metastases.