|Friday, February 16|
|PS2 Poster Session 2 and Refreshments||
Fri, Feb 16, 5:15 PM - 6:30 PM
Estimating the Relative Excess Risk Due to Interaction in Clustered Data Settings (303606)
Paige Williams, Harvard T.H. Chan School of Public Health
Keywords: RERI, additive interaction, log binomial regression, clustered data
There is wide agreement in the epidemiology community that the risk difference scale is often of greater interest than relative differences when evaluating intervention effects on binary outcomes, particularly in public health settings. Despite recommendations that interaction results be presented on both the multiplicative scale and the additive scale, very few studies have presented findings in terms of additive interaction, likely because the typical models used for binary outcomes implicitly measure interaction on the multiplicative scale. One measure to assess additive interaction from multiplicative models is the relative excess risk due to interaction (RERI). The RERI literature to date has focused on rare outcomes in an independent data setting. In an effort to further advance the reporting of additive interaction measures for binary outcomes, we evaluate the RERI metric in clustered data settings. We apply frequentist and Bayesian implementations of log binomial regression for clustered data to two examples from the medical field, one involving a national dataset on in-vitro fertilization cycles and the other assessing antiretroviral therapy in pregnant HIV+ women.