In the United States, Black individuals have suffered from 300 years of racism, bias, segregation and have been systematically and intentionally denied opportunities to accrue wealth. These disadvantages have resulted in disparities in health outcomes. Over the last decade there has been a growing interest in examining social determinants of health as upstream factors that lead to downstream health disparities. It is of vital importance to quantify the contribution of SDH factors to racial disparities in order to inform policy and social justice initiatives. This demonstration project uses years of education and white matter hyperintensities (WMH) to illustrate two methods of quantifying the role of a SDH in producing health disparities.

The current study is a secondary data analysis of baseline data from a subset of the National Alzheimer's Coordinating Center database with neuroimaging data collected from 2002-2019. Participants were 997 cognitively diverse, Black and White (10.4% Black) individuals, aged 60-94 (mean=73.86, 56.5% female), mean education of 15.18 years (range= 0-23, SD=3.55). First, mediation, was conducted in the SEM framework using the R package lavaan. Black/White race was the independent variable, education was the mediator, WMH volume was the dependent variable, and age/sex were the covariates. Bootstrapped standard errors were calculated using 1000 iterations. The indirect effect was then divided by the total effect to determine the proportion of the total effect attributable to education. Second, a population attributable fraction (PAF) or the expected reduction in WMH if we eliminated low education and structural racism for which Black serves as a proxy was calculated. Two logistic regressions with dichotomous (median split) WMH as the dependent variable, first with low (less than high school) versus high education, and second with Black/White race added as predictors. Age/sex were covariates. PAF of education, and then of Black/White race controlling for education were obtained. Subsequently, a combined PAF was calculated.

In the lavaan model, the total effect of Black/White race on WMH was not significant (B=.040, se=.113, p=.246); however, Black/White race significantly predicted education (B= -.108, se=.390, p=.001) and education significantly predicted WMH burden (B=-.084, se=.008, p=.002). This resulted in a significant indirect effect (effect=.009, se=.014, p=.032). 22.6 % of the relationship between Black/White race and WMH was mediated by education. In the logistic models, the PAF of education was 5.3% and the additional PAF of Black/White race was 2.7%. The combined PAF of Black race and low education was 7.8%.

From our mediation we can conclude that 22.6% of the relationship between Black/White race and WMH volume is explained by education. Our PAF analysis suggests that we could reduce 7.8% of the cases with high WMH burden if we eliminated low education and the structural racism for which Black serves as a proxy. This is an under estimation of the role that education and structural racism play in WMH burden due to our positively selected sample and crude measure of education. However, these methods can help researchers quantify the contribution of SDH to disparities in older adulthood and provide targets for policy change.