Bayesian ROC curve estimation under verification bias.

Area of Science:

• Medical Statistics
• Diagnostic Test Accuracy
• Biostatistics

Background:

• Receiver operating characteristic (ROC) curves are vital for assessing diagnostic test accuracy against a gold standard.
• Gold standard tests are often impractical (invasive, costly, unavailable), leading to verification bias in subject selection.
• Verification bias occurs when disease status verification is non-random, complicating accurate ROC analysis.

Purpose of the Study:

• To propose a new Bayesian approach for estimating ROC curves with continuous data.
• To address the challenge of verification bias in diagnostic accuracy studies.
• To accurately estimate binormal model parameters and the area under the ROC curve (AUC) despite missing data.

Main Methods:

• Utilized a semiparametric binormal model for continuous data.
• Employed a rank-based likelihood and Gibbs sampling techniques.
• Imputed missing disease status labels using Markov Chain Monte-Carlo (MCMC) iterations to compute posterior distributions.

Main Results:

• Successfully computed the posterior distribution of binormal parameters (intercept and slope).
• Estimated the area under the curve (AUC) by imputing missing labels.
• Established the consistency of the resulting posterior distribution under mild conditions.

Conclusions:

• The proposed Bayesian method effectively handles verification bias in ROC analysis.
• The new estimator demonstrates strong performance and accuracy compared to existing methods.
• This approach offers a robust solution for estimating diagnostic accuracy when gold standard data is incomplete.