Likelihood-Based Non-Parametric Receiver Operating Characteristic Curve Analysis in the Presence of Imperfect...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• Diagnostic studies often use imperfect reference standards, leading to misclassified labels.
• Using these imperfect standards as gold standards can introduce bias in receiver operating characteristic (ROC) curve analysis.

Purpose of the Study:

• To develop a novel statistical method for reliable ROC curve analysis in the presence of imperfect reference standards.
• To enable accurate estimation of ROC curve, area under the curve (AUC), partial AUC, and Youden's index.

Main Methods:

• Proposed a likelihood-based method utilizing a non-parametric density ratio model.
• Developed an efficient expectation-maximization (EM) algorithm for method implementation.
• Evaluated finite-sample performance through extensive simulations.

Main Results:

• The proposed method demonstrates favorable statistical properties for estimating ROC curve, AUC, partial AUC, and Youden's index.
• Simulations showed smaller mean squared errors compared to existing methods.
• The approach was successfully applied to a malaria diagnostic study.

Conclusions:

• The novel likelihood-based method provides a robust solution for ROC analysis with imperfect reference standards.
• This method enhances the accuracy and reliability of diagnostic test performance evaluation.
• The developed EM algorithm ensures efficient implementation and practical applicability.