Determining a range of false-positive rates for which ROC curves differ

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Accuracy

Background:

• Receiver Operating Characteristic (ROC) curves are vital for evaluating diagnostic/prognostic tool performance.
• Existing global summary measures for ROC curves may not be optimal for direct comparisons.
• Comparing two ROC curves often requires nuanced analysis beyond single summary indices.

Purpose of the Study:

• To introduce a method for directly comparing true-positive rates of two diagnostic, screening, or prognostic tools.
• To determine the specific range of false-positive values over which two ROC curves differ.
• To provide a flexible method applicable to both independent and dependent samples.

Main Methods:

• Direct comparison of true-positive rates between two ROC curves.
• Identification of the range of false-positive values where significant differences exist.
• Application of the method to independent and dependent sample data.

Main Results:

• The proposed method effectively identifies ranges where diagnostic tools exhibit differing performance.
• Demonstrated applicability using examples from medical imaging (gallium citrate) and critical care prognostics (ICU severity index).
• Quantified the specific false-positive rate intervals for which the compared ROC curves diverge.

Conclusions:

• The presented method offers a direct and effective approach for comparing ROC curves, especially when global measures are insufficient.
• It precisely pinpoints performance differences between diagnostic tools within specific operating ranges.
• This technique enhances the comparative analysis of medical diagnostic, screening, and prognostic tools.