Adjusting for verification bias in diagnostic test evaluation: a Bayesian approach.

Area of Science:

• Biostatistics
• Medical Diagnostics
• Health Services Research

Background:

• Estimating diagnostic test performance requires representative patient samples.
• Non-representative samples lead to biased performance estimates, known as verification bias.
• Gold standard procedures for verification can be invasive and carry risks.

Purpose of the Study:

• To introduce a Bayesian method for adjusting diagnostic test performance estimates for verification bias.
• To address non-ignorable verification bias.
• To evaluate the clinical utility of the proposed statistical findings.

Main Methods:

• Developed a Bayesian approach treating verification bias as a missing data problem.
• Addressed non-ignorable verification bias within the Bayesian framework.
• Compared Bayesian estimates with likelihood-based approaches.

Main Results:

• The proposed Bayesian method provides consistent test performance estimates.
• The approach effectively adjusts for verification bias.
• Statistical findings demonstrate value in clinical decision-making contexts.

Conclusions:

• Bayesian methods offer a robust solution for adjusting diagnostic test performance estimates affected by verification bias.
• The developed methodology enhances the reliability of test performance evaluation.
• The study underscores the importance of addressing bias for accurate clinical application.