Nonparametric inference of the area under ROC curve under two-phase cluster sampling.

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Test Evaluation

Background:

• Nonparametric inference for the area under the ROC curve (AUC) is established for verification bias or clustering individually.
• Existing methods fail when both verification bias and clustering are present, common in two-phase studies with multiple test results per subject.
• Accurate AUC inference requires accounting for both verification bias and intra-cluster correlation.

Purpose of the Study:

• To develop a nonparametric method for AUC inference that simultaneously addresses verification bias and clustering.
• To propose an inverse probability weighting (IPW) AUC estimator robust to these combined biases.
• To derive a variance formula that correctly accounts for intra-cluster correlations.

Main Methods:

• Proposed an Inverse Probability Weighting (IPW) AUC estimator.
• Developed a variance formula incorporating intra-cluster correlations for disease status and test results.
• Utilized a simulation study to evaluate the proposed method's performance.

Main Results:

• Methods assuming independence between subjects and test results underestimate the true variance of the IPW AUC estimator when intra-cluster correlations exist.
• The proposed method provides a consistent variance estimate for the IPW AUC estimator.
• The simulation results demonstrate the superiority of the proposed approach in handling combined biases.

Conclusions:

• The proposed IPW AUC estimator and variance formula effectively address the challenges of verification bias and clustering in diagnostic accuracy studies.
• Accounting for intra-cluster correlations is essential for accurate variance estimation in such complex study designs.
• This method enhances the reliability of AUC inference in real-world clinical and epidemiological research.