Is probabilistic bias analysis approximately Bayesian?

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Case-control studies are prone to differential exposure misclassification, especially when exposure is assessed after case identification.
• Existing methods like probabilistic bias analysis (PBA) adjust effect estimates using sensitivity and specificity.
• PBA's iterative sampling shares similarities with Bayesian adjustments but lacks a formal theoretical framework, complicating interpretation.

Purpose of the Study:

• To theoretically and empirically assess the extent to which probabilistic bias analysis can be viewed as approximately Bayesian.
• To compare the performance of PBA and Bayesian approaches for handling exposure misclassification in case-control studies.

Main Methods:

• Theoretical comparison of PBA's iterative sampling with Bayesian adjustment methods.
• Empirical evaluation of PBA and Bayesian approaches under various misclassification scenarios.
• Analysis of situations where differences between PBA and Bayesian methods become substantial.

Main Results:

• Probabilistic bias analysis (PBA) exhibits a strong resemblance to Bayesian adjustment methods.
• In most practical situations, PBA and Bayesian approaches to exposure misclassification perform comparably.
• Substantial differences arise mainly with unrealistic prior specifications, which are detectable.

Conclusions:

• Probabilistic bias analysis can be largely understood as an approximately Bayesian approach for exposure misclassification in case-control studies.
• The methods yield similar results in typical scenarios, enhancing the interpretability of PBA findings.
• Researchers should be aware of specific conditions, like unrealistic priors, where deviations may occur.