Explainable Machine Learning for Early Detection of Mild Cognitive Impairment, Fall Risk, and Frailty Using...

Area of Science:

• Gerontology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Aging populations face increased risks of mild cognitive impairment (MCI), falls, and frailty.
• Early identification of these conditions is crucial for timely intervention and management.
• Current screening methods may not fully capture the complex interplay of motor and cognitive decline.

Purpose of the Study:

• To design and evaluate an explainable machine learning (ML) framework for early detection of MCI, fall risk, and frailty in older adults.
• To integrate sensor-based motor assessments with demographic and clinical data for enhanced predictive accuracy.
• To develop a unified multilabel model capable of predicting all three conditions simultaneously.

Main Methods:

• Eighty-three community-dwelling older adults (≥60 years) underwent multimodal motor assessments using the Mizzou Point-of-Care Assessment System (MPASS).
• Sensor-derived features from gait, balance, and sit-to-stand tasks were combined with demographic and clinical data.
• XGBoost, Decision Tree, and AdaBoost algorithms were employed for predictive modeling, with SHapley Additive exPlanations (SHAP) for interpretability.

Main Results:

• The ML model for MCI achieved 94% accuracy (AUC=0.88), fall risk 94% accuracy (AUC=0.90), and frailty 82% accuracy (AUC=0.77).
• A unified multilabel XGBoost model demonstrated the highest performance (73% accuracy, sensitivity, F1 score).
• SHAP analysis highlighted stride length, balance measures, and knee velocity during sit-to-stand as key predictors.

Conclusions:

• A novel, explainable ML framework effectively integrates sensor-based motor data for predicting MCI, fall risk, and frailty.
• The framework enables transparent early screening of multidomain cognitive and physical decline in aging.
• This approach holds promise for proactive healthcare strategies in geriatric populations.