JMIR Publications

ABSTRACT

Background:

Numerous studies have identified risk factors for physical restraint use in older adults in long-term care facilities. Nevertheless, there is a lack of predictive tools to identify high-risk individuals.

Objective:

We aimed to develop machine learning–based models to predict the risk of physical restraint in older adults.

Methods:

This study conducted a cross-sectional secondary data analysis based on 1026 older adults from six long-term care facilities in Chongqing, China, from July 2019 to November 2019. The primary outcome was the use of physical restraint (yes or no), identified by two collectors’ direct observation. Candidate predictors (older adults’ demographic and clinical factors) were used to build nine independent machine learning models: Gaussian naïve Bayesian (GNB), k-nearest neighbor (KNN), decision tree (DT), logistic regression (LR), support vector machine (SVM), random forest (RF), multilayer perceptron (MLP), XGBoost, and Lightgbm, as well as stacking ensemble machine learning. Performance was evaluated using accuracy, precision, recall, F-score, a comprehensive evaluation indicator (CEI) weighed by the above indicators, and the area under the receiver operating characteristic curve (AUC). The decision curve analysis (DCA) was performed to evaluate the clinical utility. Feature importance was interpreted using Shapley Additive exPlanations (SHAP).

Results:

A total of 1026 older adults (mean [SD] age, 83.5 [7.6] years; 586 [57.1%] male older adults), with 265 restrained older adults, were included in the study. All machine learning models performed well, with an AUC above 0.905 and F score above 0.900. The two best independent models are RF (AUC: 0.938, 95%CI: 0.914～0.947)and SVM (AUC: 0.949, 95%CI:0.911～0.953). The DCA demonstrated that the RF model displayed better clinical utility than other models. The stacking model combined with SVM, RF, and MLP performed best with AUC (0.950) and CEI (0.943) values, as well as the DCA curve indicated best clinical utility. The SHAP plots demonstrated that the significant contributors to model performance were related to cognitive impairment, care dependency, mobility decline, physical agitation, and indwelling tube.

Conclusions:

RF and stacking model had high performance and clinical utility. Machine learning prediction models for predicting the probability of physical restraint in older adults could offer clinical screening and decision support, which could help medical staff in the early identification and physical restraint management of older adults. Clinical Trial: Not applicable.