- What are the different ways to account for SES in an analytic model when investigating racial/ethnic health disparities? (Hint: you should have three options). Discuss the interpretations/implications of each approach as it relates to the interest in understand health disparities by race/ethnicity. Draw a DAG for each option and reference it in your response (you do not have to post this!).
There are several different ways to account for SES when investigating racial/ethnic disparities.
- Confounder: The Headen article acknowledges that it can be difficult to control for SES because it is highly correlated with race. They propose to deal with this issue by using “detailed” SES data from the NLSY79 survey, including past-year employment, mother’s years of education, and family income accounting for family size. By adjusting for SES in this model, Headen can determine the direct effect of race/ethnicity on GWG. **I am not sure I agree that having “detailed” information alone is enough to account for the difficulty in controlling for SES in regards to race. The authors did conduct sensitivity analyses to determine any bias due to missing data (do they mean that detailed = lack of missing data?). Instead, it would seem that they should explore the extent to which these SES indicators are collinear and try to maximize measures of SES that are not collinear (instead of “detailed”). However, I am unsure to what extent collinearity is an issue with GEE analysis as compared to linear and logistic regressions.
- Mediator: Lorch explore the role of SES as a mediator along the pathway between race and risk of fetal death. Lorch used a mediation anlaysis to quantify the direct effects of race on fetal death and the indirect effects of the mediators on fetal death. By using mediation analysis Lorch was able to identify which causal pathways between race and fetal death may be most amenable to intervention depending on which mediator accounted for the greatest percentage of fetal death risk. For instance, they conclude that antepartum/intrapartum factors may be more amenable to intervention than pre-term delivery.
- MLRA: Merlo discusses exploring the unequal clustering of health in a neighborhood unit not as a statistical “nuisance” to be explained away by confounding or mediation analysis, but as a “key concept in social epidemiology that yields important information itself.” By using MLRA to explore how health within and across neighborhoods is alike or different, we can identify contextual factors (such as SES and race) that determine health disparities. The benefit of this approach is that it draws our attention to specific geographic units to target for interventions.
- Think about multilevel influences on a health outcome of interest to you. Discuss how you would study this, including measurement and analytic approaches you would use to account for exposures across multiple levels.
One way that multi-level analysis could apply to something I am studying is differences in hospital cesarean rates, which in the U.S can be wide ranging from 10-70% of all deliveries. These differences cannot be explained by clinical factors alone. MLRA may be applicable to explore the variance at the levels of the individual, the hospital, and region. At the individual level, there are clinical factors, but also a woman’s level of acceptance around cesarean and her provider’s attitudes/beliefs about cesareans. At the hospital level, there are factors that make certain places more prone to cesareans—say a particular history of malpractice, labor and delivery unit volume, nursing staffing, provider staffing, etc. At the state level, reimbursement practices, malpractice caps, and provider (midwife/doctor) practice regulations could all be relevant. Thus, the factors that create low, average, and high cesarean rates are highly contextual and vary significantly across and within states.
If I am understanding the Merlo article correctly, MLRA could be used to explore at which level the variance around cesarean rates in order to identify the hospital and state factors that contribute to inappropriate variation in the cesarean rate. I would construct my MLRA model using measures of the clinical risk of the patient population, the variation in provider attitudes and beliefs, hospital history of malpractice and staffing models (one on one nursing, in-house obstetrics coverage), and state factors (independent midwifery licensing, malpractice caps).