Confidence intervals of the difference between areas under two ROC curves in matched-pair experiments

Area of Science:

• Biostatistics
• Medical Diagnostics

Background:

• Comparing diagnostic tests often involves analyzing correlated areas under the ROC curve (AUCs) in matched-pair studies.
• Accurate confidence interval (CI) construction is crucial for reliable comparisons of diagnostic test discriminatory ability.

Purpose of the Study:

• To address challenges in constructing confidence intervals for the difference between two correlated AUCs in matched-pair experiments.
• To propose and evaluate novel hybrid and bootstrap-based CI methods.

Main Methods:

• Development of 13 hybrid CIs using maximum likelihood estimation, Delong's statistic, Wilson score statistic (WS), modified Wald statistic (MW), and Agresti-Coull statistic, with and without continuity corrections.
• Inclusion of three Bootstrap-resampling-based CIs for comparison.
• Simulation studies to assess empirical coverage probabilities, interval widths, and noncoverage probability ratios.

Main Results:

• The hybrid Agresti-Coull CI with empirical estimation (EAC) exhibited the most satisfactory performance.
• EAC CI demonstrated coverage probability close to the nominal level with a narrow interval width.
• Comparison with traditional parametric and nonparametric CIs highlighted the advantages of the proposed hybrid methods.

Conclusions:

• The hybrid Agresti-Coull CI with empirical estimation (EAC) is recommended for comparing correlated AUCs in matched-pair studies due to its excellent performance.
• The proposed methodologies provide robust tools for evaluating diagnostic test accuracy in clinical research.
• Further application of EAC CI in clinical studies is encouraged for reliable diagnostic test comparisons.