A comparison of single- and double-threshold ROC plots for mixture distributions

Area of Science:

• Biostatistics
• Medical Diagnostics
• Clinical Research

Background:

• Single-threshold receiver operating characteristic (stROC) analysis is standard for bimodal test results.
• stROC demonstrates poor performance with three-modal distributed test results.
• Clinical diagnostic accuracy is limited by existing methods for complex data distributions.

Purpose of the Study:

• To introduce and evaluate a double-threshold receiver operating characteristic (dtROC) analysis.
• To demonstrate dtROC's superior discriminative performance for three-modal distributed test results.
• To compare dtROC against stROC for enhanced diagnostic accuracy.

Main Methods:

• Developed a dtROC plot by replacing the single threshold with a double threshold.
• Optimized sensitivity and specificity coordinates to maximize sensitivity for specific specificity values.
• Conducted simulation studies with lognormal, Poisson, and Weibull distributions.
• Performed secondary data analysis on clinical palpation test results (C7 spinous process).

Main Results:

• dtROC analysis significantly outperformed stROC analysis across all simulated distributions.
• Area under the ROC curve (AUROC) improved substantially: lognormal (0.436 to 0.983), Poisson (0.676 to 0.752), and Weibull (0.674 to 0.804).
• Clinical application demonstrated improved discrimination using the dtROC method.

Conclusions:

• Double-threshold ROC analysis offers superior discriminative performance for three-modal distributed data.
• dtROC analysis provides a more effective tool for clinical settings with complex test result distributions.
• This method enhances diagnostic accuracy and clinical decision-making.