Prediction of cognitive performance by demographics, sleep, and brain morphometry: machine learning findings from ENIGMA-Sleep Working Group

Group-level studies have highlighted the roles of aging, poor sleep, and brain atrophy in cognitive performance (CP) but have overlooked inter-individual variability. We predict CP from feature sets (demographic, subjective/objective sleep parameters, and regional brain morphometry) using multisite ENIGMA-Sleep data (n = 2,372). Linear and non-linear machine learning models were trained on the largest cohort (n = 845), and the best-performing models were validated on independent cohorts. Subsequently, based on the best-performing model on the largest cohort, we characterized feature importance and interactions across all cohorts. We observed that a combination of demographic, sleep, and brain parameters moderately predicted CP, with age emerging as the key predictor. Model explanations further suggested that age was the primary driver of prediction models, while sleep played a smaller role that varied across subgroups. These findings endorsed inter-individual variability and complex interaction between aging, sleep, brain, and CP.