1. What are 3 different ways to account for SES in an analytic model when investigating racial/ethnic health disparities? (describe a fourth for extra bonus points!). Briefly discuss the interpretations/implications of each approach as it relates to the interest in understand health disparities by race/ethnicity.
- Effect modification is a common manner to evaluate SES between races. This can be done by creating an interaction term between race/ethnicity and SES. Odds ratios will be reported as the odds of the outcome within each level of race/ethnicity by SES and will evaluate interaction on the multiplicative scale. Post-estimation using the margins command can identify effect modification on the additive scale.
Another method is to perform a mediation analysis, in which the research performs three separate regression analysis. 1) regression evaluating the odds of having the exposure on the outcome controlling for the potential mediator 2) regression analysis evaluating the odds of the exposure on the mediating factors 3) regression analysis of the mediating factors on the odds of having the outcome.
Lastly, and most rudimentary, would be to sequentially add covariates to regression models in order to see if the effect of a certain exposure would be mitigated (decrease in odds ratio) with the addition of a potential mediator.
2. Describe a potential effect modifier, mediator, or contextual variable (for definition of contextual variable, see first page of option Merlo reading) for an association of interest to you and relevant to health disparities. For example, for investigating the association between SES and maternal mortality, I might be interested in the contextual variable of exposure to violence in the neighborhood. Describe how you would study whether this relationship exists.
Currently, I am performing an analysis to evaluate the impact of primary care visit adherence (percent of kept to total scheduled primary care visits) on acute care utilization. To quantify the impact of poor visit adherence on acute care visits at different levels of unstable housing status, we used a negative binomial regression models to evaluate the unadjusted and adjusted incident rate ratio of total ACVs by housing status including an interaction term between housing status and poor visit adherence. Adjusted models will control for age by decade (<30, 30-40, 40-50, >50), race (black, white, latino or other), gender (m/f), viral load >200 (y/n) and CD4 cell count <200 (y/n). We will use post-regression analysis (margins command) to evaluate for differences in total acute care visits (difference in incident rate ratios) between those with poor visit adherence housing status and visit adherence on both the multiplicative and additive scales to calculate the impact of poor visit adherence on acute care visits.