Related Experiment Video
Updated: May 24, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Development of an interpretable machine learning model-based online tool for risk prediction of falls and
Background:
This study aimed to establish and validate interpretable Machine Learning (ML) models for predicting falls and fall-related injuries in middle-aged and older adults with depressive symptoms (DS) and to develop relevant online computational tools.
Methods:
Using data from the China Health and Retirement Longitudinal Study (CHARLS) survey from 2015 to 2018, 32 predictor variables related to the risk of falls and fall-related injuries in middle-aged and older adults with DS were included based on five dimensions of the health ecology model, and the important predictor variables were screened using Principal Component Analysis and LASSO regression at the same time. We further developed eight ML algorithms-Logistic Regression(LR), Support Vector Machine(SVM), Gradient Boosting Machine (GBM), Neural Network (NN), eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Light Gradient Boosting Machine (LightGBM) and Categorical Boosting (CatBoost)-to construct the risk prediction model, and selected the best predictive variables based on grid search and 10-fold cross-validation. SHapley Additive exPlanations (SHAP) was used for personalised interpretation of the models. In addition, we further performed stratified analyses by dividing participants into two age groups: 45-59 years and 60 years and older.
Results:
Among 3,664 middle-aged and older adults with DS, the incidence rate of falls and fall-related injuries after three years of follow-up was 20.36% and 8.92%, respectively. Among all models, LightGBM had the best performance. LightGBM performed the best, with an area under the curve (AUC) of 0.821 (95% CI: 0.802-0.841) for the fall risk test set and an AUC of 0.905 (95% CI: 0.892-0.919) for the fall injury risk test set. We identified important risk factors for falls and fall-related injuries in middle-aged and older adults with DS. The optimal predictive model and risk predictors differed from those identified before stratification by age. SHAP visualises the specific contributions of these risk factors, thereby enhancing the model's value for application. Online tools to implement the model are available at https://riskpredictiontool.shinyapps.io/falls_prediction_tool/ and https://riskpredictiontool.shinyapps.io/fall_related_injuries_prediction_tool/.
Discussion And Conclusions:
The results can help predict risk of falls and fall-related injuries among middle-aged and older adults with DS. These findings provide an important guide for the development of public health strategies.
