Predicting Cognitive Decline in Motoric Cognitive Risk Syndrome Using Machine Learning Approaches

Background: Motoric Cognitive Risk Syndrome (MCR), defined by the co-occurrence of subjective cognitive complaints and slow gait, is recognized as a preclinical risk state for cognitive decline. However, not all individuals with MCR experience cognitive deterioration, making early and individualized prediction critical. Methods: This study included 80 participants aged 60 and older with MCR who underwent baseline assessments including plasma biomarkers (β-amyloid, tau), dual-task gait measurements, and neuropsychological tests. Participants were followed for one year to monitor cognitive changes. Support Vector Machine (SVM) classifiers with different kernel functions were trained to predict cognitive decline. Feature importance was evaluated using the weight coefficients of a linear SVM. Results: Key predictors of cognitive decline included plasma β-amyloid and tau concentrations, gait features from dual-task conditions, and memory performance scores (e.g., California Verbal Learning Test). The best-performing model used a linear kernel with 30 selected features, achieving 88.2% accuracy and an AUC of 83.7% on the test set. Cross-validation yielded an average accuracy of 95.3% and an AUC of 99.6%. Conclusions: This study demonstrates the feasibility of combining biomarker, motor, and cognitive assessments in a machine learning framework to predict short-term cognitive decline in individuals with MCR. The findings support the potential clinical utility of such models but also underscore the need for external validation.