Robust bayesian sensitivity analysis for case-control studies with uncertain exposure misclassification...

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Exposure misclassification in case-control studies biases odds ratio estimates.
• Current methods to adjust for misclassification require prior distributions for probabilities, which can be problematic with high uncertainty.
• This uncertainty leads to overly wide posterior intervals and subjective prior dependence.

Purpose of the Study:

• To propose a robust Bayesian approach for handling exposure misclassification in case-control studies.
• To provide an alternative to methods relying on elicited prior distributions for misclassification probabilities.
• To enable more useful sensitivity analyses by considering ranges of values for unknown parameters.

Main Methods:

• A robust Bayesian framework is proposed, utilizing a feasible region for misclassification probabilities instead of prior distributions.
• An inequality constrained optimization algorithm is employed to find the extrema of posterior inference within this region.
• This approach avoids fixing all unknown parameters to specific values, allowing for consideration of parameter ranges.

Main Results:

• The proposed method offers a more informative approach when exposure misclassification is highly uncertain.
• It facilitates sensitivity analyses by allowing exploration of a range of misclassification probabilities.
• Posterior inference becomes less dependent on subjective prior elicitation.

Conclusions:

• The robust Bayesian approach provides a valuable alternative for adjusting bias and quantifying uncertainty in the presence of uncertain exposure misclassification.
• This method enhances the utility of sensitivity analyses in epidemiological studies.
• It addresses limitations of existing methods when dealing with high uncertainty in misclassification probabilities.