Copula modeling of receiver operating characteristic and predictiveness curves

Area of Science:

• Biostatistics
• Medical Informatics
• Genomics

Background:

• Receiver operating characteristic (ROC) and predictiveness curves are crucial for evaluating binary classification models.
• Current methods for analyzing these curves can be computationally intensive or limited in scope.

Purpose of the Study:

• To develop a flexible and computationally efficient copula-based framework for constructing and analyzing ROC and predictiveness curves.
• To assess the discriminative and predictive capabilities of continuous markers in binary outcome prediction.

Main Methods:

• A copula-based joint density construction utilizing a Gaussian copula and customized marginal distributions.
• Maximum likelihood and resampling-based estimation techniques.
• Randomized quantile residuals for model adequacy assessment and outlier detection.

Main Results:

• The proposed method offers a numerically feasible approach to plotting and analyzing both ROC and predictiveness curves.
• Simulation studies demonstrate the robust performance of the estimators across various dependence structures and sample sizes.
• The methodology effectively analyzes gene expression data for breast cancer diagnosis and prediction.

Conclusions:

• The copula-based approach provides a powerful and adaptable tool for evaluating diagnostic markers.
• This method enhances the understanding of marker performance in clinical and research settings.
• The framework is applicable to various biomarker studies, including gene expression analysis for cancer prediction.