Predicting low premorbid cognitive ability with social determinants: A machine learning approach

Background: Social determinants of health and biological processes are shaped by the exposome, which provides a framework for understanding how social adversity drives molecular and cellular mechanisms underlying Alzheimer's disease risk. Individuals with low premorbid intellectual ability (pIQ ≤70) may be particularly vulnerable to adverse social determinants of health due to reduced cognitive reserve, yet this relationship is understudied. Methods: = 2691) were analyzed. Participants were classified as low pIQ (IQ ≤70) or average pIQ (IQ 90-100) via word reading scores. Using a machine learning approach, an XGBoost model evaluated education, income, Area Deprivation Index (ADI), social support, stress, health status, and worry in prediction of pIQ grouping. Results: The model achieved and AUC of 0.72 [0.64, 0.81]. Top predictors included worry, ADI, income, high school completion, and tangible support. Low pIQ was associated with greater neighborhood deprivation, lower income, and reduced support resources. Conclusion: Low pIQ, when combined with SDoH factors reflects a vulnerable psychosocial-cognitive phenotype that may accelerate pathways to cognitive decline potentially through inflammatory mechanisms.