Estimating the relative excess risk due to interaction: a bayesian approach.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical modeling

Background:

• Relative excess risk due to interaction (RERI) and relative excess odds due to interaction (RERIOR) are crucial metrics in epidemiologic data analysis.
• Existing frequentist methods for estimating confidence intervals for RERI/RERIOR include nonparametric bootstrap, variance estimate recovery, and profile likelihood.

Purpose of the Study:

• To propose a Bayesian approach as an alternative for estimating RERIOR in case-control studies and RERI in cohort studies.
• To demonstrate the advantages of the Bayesian approach, including ease of computation and flexibility in handling parameter constraints and confounders.

Main Methods:

• A Bayesian approach is proposed using a linear additive odds-ratio model for RERIOR estimation in case-control studies.
• A linear additive risk-ratio model is employed for RERI estimation in cohort studies.
• Posterior computation is performed using free software, accommodating inequality constraints.

Main Results:

• Bayesian credible intervals for RERI/RERIOR are often easier to obtain than frequentist confidence intervals.
• The Bayesian approach facilitates straightforward extension for adjusting analyses to account for confounding variables.

Conclusions:

• The proposed Bayesian approach provides a practical and computationally efficient alternative for estimating interaction effects in epidemiologic studies.
• The ease of implementation and extension for confounder adjustment make this Bayesian method valuable for applied research.