Bayesian methods of confidence interval construction for the population attributable risk from cross-sectional...

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Population attributable risk (PAR) quantifies the public health impact of removing a risk factor.
• Calculating confidence intervals for PAR estimates is crucial for assessing uncertainty but lacks standardized methods in the literature.
• Existing frequentist methods for confidence intervals may not consistently provide accurate coverage.

Purpose of the Study:

• To implement and evaluate a fully Bayesian approach for constructing confidence intervals for population attributable risk in cross-sectional studies.
• To compare the performance of the Bayesian method against standard frequentist approaches (delta, jackknife, bootstrap).
• To investigate the influence of prior selection on the coverage properties of the Bayesian confidence intervals.

Main Methods:

• A fully Bayesian framework was developed for confidence interval construction of population attributable risk.
• Cross-sectional study data was utilized to implement and test the proposed methodology.
• Performance was assessed by comparing percent coverage rates against frequentist confidence interval methods.

Main Results:

• The Bayesian approach demonstrated superior percent coverage compared to delta, jackknife, and bootstrap frequentist methods in most scenarios.
• The choice of prior distribution significantly affects the coverage of the Bayesian confidence intervals.
• Alternative prior specifications were explored for different study situations.

Conclusions:

• The Bayesian approach offers a robust and often superior method for confidence interval estimation of population attributable risk in cross-sectional studies.
• Understanding and selecting appropriate priors is essential for optimizing the performance of the Bayesian method.
• This work provides a valuable tool for more accurate public health impact assessment of risk factors.