ABSTRACT
Background: Health management for elderly in the community is more difficult, especially under the impact of the COVID-19 pandemic. This study aims to build a frailty assessment platform and form a frailty early warning model by Machine Learning (ML) methods that community health workers can use to carry out overall evaluation and health management of the elderly. Methods:The intra-group correlation coefficient (ICC) was calculated to test the platform's reliability and conduct a user satisfaction survey. The frailty assessment results were taken as the dependent variable, and the comprehensive assessment results, the elderly capability assessment results, and the basic information of the elderly were taken as independent variables to train and develop a frailty risk model by ML methods. The model performance was evaluated by Precision, Recall, F1-score, Accuracy, area under the ROC curve (AUC value), Macro avg, Weighted avg and other indicators. The visual image SHapley Additive exPlanations (SHAP) method was used to analyze the features of the effective model, the risk factors, and the population distribution of frailty, thus forming a frailty early warning model. Results: The reliability test of frailty assessment and early warning platform showed high consistency with ICC of 0.966, and a 95% confidence interval (95% CI) of (0.888, 0.990), P<0.001. The user satisfaction survey showed the highest score for necessity. The cross-sectional survey showed that the frailty rate of the elderly reached 34.5%,and fitting support vector machine with RBF kernel function (SVM RBF) had the best performance in ML methods with AUC=0.862 in the Training set and AUC=0.865 in the Test set. The SHAP analysis showed that positive sarcopenia, high risk of falls, potential and above anxiety, and moderate and above impairment of daily living ability were more likely to cause frailty. The elderly with all four features accounted for 11.56% of the entire elderly population. Conclusions: The frailty assessment and early warning platform constructed in this study have good reliability and high user satisfaction. By running the SVM RBF model in the platform background thread, the platform named “I can help”, was developed for community elderly and health workers.