Efficient confidence bounds for ROC curves

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• The accuracy of quantitative diagnostic markers relies on selecting appropriate cut-off points.
• Receiver Operating Characteristic (ROC) curves are crucial for visualizing and evaluating marker performance across all possible cut-offs.
• Existing methods for constructing confidence intervals for ROC curves can be imprecise.

Purpose of the Study:

• To present a more efficient statistical method for calculating confidence intervals for ROC curves.
• To improve the precision of confidence bounds for sensitivity and specificity.
• To provide a formula for sample size calculation for desired confidence interval length.

Main Methods:

• The study utilizes a statistical test introduced by Greenhouse and Mantel.
• Confidence intervals for sensitivity and specificity are combined using this novel approach.
• The method is applied to evaluate a tumor marker for diagnosing bone marrow metastases.

Main Results:

• The proposed method yields confidence intervals that are up to 40% smaller than traditional methods.
• This increased precision allows for a more refined assessment of diagnostic marker performance.
• A practical formula for sample size determination is provided.

Conclusions:

• The Greenhouse and Mantel-based method offers a significant improvement in the efficiency and precision of ROC curve analysis.
• Smaller confidence intervals enhance the ability to select optimal cut-off points for diagnostic markers.
• This approach is valuable for the development and validation of new diagnostic tools.