Primary Submission Category: Biomarkers or biological pathways
How much can biomarkers explain sociodemographic inequalities in cognitive functioning? Results from a machine learning model in the Health and Retirement Study
Authors: Eric Klopack Eileen Crimmins
Presenting Author: Eric Klopack*
Large studies, like the Health and Retirement Study (HRS), are gathering extensive blood-based biomarker data (e.g., cytokines, DNAm, blood lipids), potentially presenting a unique opportunity to investigate population-level cognitive aging effects driven by social inequalities. These biomarkers are believed to represent pathways by which social adversity gets “under the skin” to affect health outcomes like cognitive aging and Alzheimer’s disease. However, it is unknown how much these biomarkers can explain sociodemographic differences in cognitive functioning, and thus, their utility for understanding the biological underpinnings of social inequalities.
We utilized data from the 2016 HRS Venous Blood Study with DNA methylation data (N = 4018), including 57 blood-based biomarkers, cognitive functioning measured using the Telephone Interview for Cognitive Status (TICS), and sociodemographic variables (viz., age, race/ethnicity, sex/gender, and educational attainment). Participants were randomly separated into a training and testing set. eXtreme Gradient Boosting (xgboost) was used to create a variable of predicted cognitive functioning using the 57 biomarkers. According to this model, the most important features were markers of aging (DNAm age), neuropathology (viz., GFAP, NfL, Aβ 42:40 ratio), diet (vitamin D3 epimer), and genetic risk (a PGS for Alzheimer’s disease).
We performed mediation analysis in the testing data. Biomarker-predicted functioning significantly mediated the effects of sociodemographic variables. The biomarkers explained a large portion of the total effect of age (50.8%), a moderate portion of the total effect of sex/gender and race/ethnicity (between 38.2% and 26.1%) and a modest portion of the total effect of education (between 11.3% and 9.7%).
Findings suggest that these biomarkers are useful in explaining biological processes underlying sociodemographic inequalities in cognitive functioning. However, the majority of variance remains unexplained.