Empirical likelihood inference for the area under the receiver operating characteristic (ROC) curve with verification...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Epidemiology

Background:

• The area under the receiver operating characteristic curve (AUC) is a key metric for evaluating diagnostic test performance.
• Verification bias, arising from incomplete disease status verification, can compromise AUC estimation validity.
• Existing bias correction methods for AUC estimation have limitations, particularly in constructing confidence intervals.

Purpose of the Study:

• To propose novel, robust methods for constructing bias-corrected confidence intervals for the AUC.
• To address the challenge of partially verified disease status in diagnostic accuracy studies.
• To provide valid and precise interval estimates for AUC in the presence of missing verification data.

Main Methods:

• Development of two bias-corrected confidence interval methods for AUC: one using bootstrap resampling and another using empirical likelihood.
• Leveraging Alonzo and Pepe's bias-corrected ROC estimators to handle missing disease verification data.
• Utilizing simulation studies and real-world data analysis to validate the proposed methods.

Main Results:

• The proposed bootstrap and empirical likelihood methods provide valid and precise confidence intervals for AUC.
• The methods effectively correct for verification bias under the missing-at-random assumption.
• Performance was evaluated across diverse clinical scenarios.

Conclusions:

• The novel methods offer a significant advancement in estimating AUC with confidence intervals when disease verification is incomplete.
• These approaches enhance the reliability of diagnostic test performance evaluation in the presence of missing data.
• The findings are applicable to various medical diagnostic studies facing verification bias.