Enhancing Fall Risk prediction In Chinese Older Adults Using Explainable Machine Learning

PURPOSE: This study aims to integrate explainable ensemble learning techniques for improving fall risk prediction among the elderly population in the Tsinghua University community. The focus is on developing a predictive model that is not only accurate but also interpretable, aiding in better understanding and prevention of falls in older groups. METHOD: A mixed-methods approach was employed, gathering data from 1500 participants from 2019 to 2022, aged 65 years and above, living in the Tsinghua University community. The data collection included comprehensive health assessments, lifestyle questionnaires, and environmental risk evaluations. We applied the least absolute shrinkage and selection operator (LASSO) method was applied to determine the optimal predictors and then five commonly used ensemble learning algorithms: Random Forest (RF), Light Gradient Boosting Machine (LightGBM), categorical boosting (Catboost), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost) were used to build a predictive model. RESULTS: The study identified that the XGBoost model provided the highest accuracy. It showed the largest AUC value in the 10-fold cross-validation (0.8592; 95% CI: 0.8337-0.8726), testing (0.8542; 95% CI: 0.8347-0.8660) as well as the smallest Brier score in the training (0.1534; 95% CI: 0.1482-0.1609) and testing set (0.1601; 95% CI: 0.1565-0.1707). Notably, SHAP value analysis revealed that indoor environmental hazards, mobility patterns, experience of falling, medication history, cognitive function as well as changes in weight were significant predictors. CONCLUSION: The integration of explainable ensemble learning models in fall risk assessment offers a powerful tool for identifying and understanding risk factors among the elderly. Particularly in the Tsinghua University community, this approach enables the creation of targeted and effective fall prevention strategies. The study underscores the value of explainable machine learning in geriatric care, ensuring that predictive models are both accurate and understandable to clinicians and community caregiver. This paper is supported by the "Tang Scholar" project of the Cyrus Tang Foundation in China