Prediction models for cognitive aging measures have largely evaluated demographic variables and APOE carrier status in populations of European ancestry. To comprehensively assess prediction models among Hispanic/Latinos, we considered 12 models (6 predictor sets and 2 methods) for global cognitive score change (GCSC) and mild cognitive impairment (MCI) in the Study of Latinos-Investigation of Neurocognitive Aging (SOL-INCA) (N = 5856). Based on the average mean squared error (MSE) for GCSC or average area under the curve (AUC) for MCI across 100 randomly split testing and training sets, performance was similar across models, but slightly better for the following models: the chronic conditions model and genetic model (mean MSEs = 0.2464) using gradient-boosted trees for GCSC prediction and the chronic conditions model (mean AUC = 62%) and metabolite model (mean AUC = 60%) using logistic regression for MCI prediction. Using the Shapley Additive Explanations (SHAP) method, age at baseline, time between exams, and sex were the most important predictors for GCSC, followed by diabetes and global ancestral proportions. Diabetes and the metabolite ribitol had the highest influence on prediction of MCI. Although prediction performance was not especially high and did not vary greatly across models, incorporating information on diabetes, ancestry and metabolites may help improve prediction of GCSC and MCI.

Polygenic risk scores (PRS) are commonly used to quantify the inherited susceptibility for a trait, yet they fail to account for non-linear and interaction effects between single nucleotide polymorphisms (SNPs). We address this via a machine learning approach, validated in nine complex phenotypes in a multi-ancestry population. We use an ensemble method of SNP selection followed by gradient boosted trees (XGBoost) to allow for non-linearities and interaction effects. We compare our results to the standard, linear PRS model developed using PRSice, LDpred2, and lassosum2. Combining a PRS as a feature in an XGBoost model results in a relative increase in the percentage variance explained compared to the standard linear PRS model by 22% for height, 27% for HDL cholesterol, 43% for body mass index, 50% for sleep duration, 58% for systolic blood pressure, 64% for total cholesterol, 66% for triglycerides, 77% for LDL cholesterol, and 100% for diastolic blood pressure. Multi-ancestry trained models perform similarly to specific racial/ethnic group trained models and are consistently superior to the standard linear PRS models. This work demonstrates an effective method to account for non-linearities and interaction effects in genetics-based prediction models.