Bayesian Estimation of Combined Accuracy for Tests with Verification Bias

Area of Science:

• Medical diagnostics
• Biostatistics
• Health technology assessment

Background:

• Verification bias, where not all subjects undergo the gold standard, complicates the assessment of diagnostic test accuracy.
• Accurate estimation of combined test accuracy is crucial for clinical decision-making and healthcare resource allocation.

Purpose of the Study:

• To develop and present a Bayesian methodology for estimating the combined accuracy of two or more diagnostic tests in the presence of verification bias.
• To extend this methodology to both binary and ordinal diagnostic tests.

Main Methods:

• A Bayesian approach is utilized, estimating test accuracy based on the posterior distribution of relevant parameters.
• For binary tests, "believe the positive" or "believe the negative" rules are applied to compute true and false positive fractions.
• For ordinal tests, the accuracy is assessed using the Receiver Operating Characteristic (ROC) area under the risk function.

Main Results:

• The study demonstrates the application of the Bayesian method to estimate the combined accuracy of CT and MRI for lung cancer diagnosis, addressing verification bias.
• An example using mammography with two readers and significant verification bias illustrates the successful estimation of combined accuracy for ordinal tests.

Conclusions:

• The proposed Bayesian framework effectively estimates the combined accuracy of diagnostic tests under verification bias for both binary and ordinal data.
• This approach provides a valuable tool for evaluating the performance of multiple diagnostic strategies in real-world clinical scenarios.