关键词: 患者, 跌倒风险, 决策树, 预测模型

Abstract: Objective　The decision tree algorithm in data mining was applied to build a fall risk prediction model. The influencing factors of fall risk for hospitalized patients were analyzed, and the model was integrated into the hospital information system for clinical application, providing a reference for the comprehensive management of preventing falls for patients. Methods　The patients who suffered falls while hospitalized at Guangzhou First People's Hospital from January 2013 to December 2020 were selected as the study subjects. The clinical data of the patients (including past medical history, nursing classification and activity ability status, etc.) were collected. Independent sample t test, and χ² test were used for statistical analysis. A single-factor logistic regression analysis was conducted to identify the influencing factors of falls. A decision tree model was constructed using the R programming language and the rpart package. The model was optimized by adjusting the decision tree parameters, and the performance of the prediction model was evaluated by drawing the receiver operating characteristic curve and calculating the area under the curve. Based on the characteristics of the constructed decision tree model, improve the fall risk monitoring indicators and the functions of the hospital nursing information system, formulate comprehensive fall prevention management strategies and implement them (from January 2023 to December 2024), and continuously track patients to compare the implementation effect of the model. Results　This study included a total of 2 040 hospitalized patients. Among them, 414 patients experienced falls, while 1 626 patients did not. There were 1 083 male patients and 957 female patients. The age of the fallers was (76.88±9.07) years, and the age of the non-fallers was (62.49±9.89) years. The results of the single-factor logistic regression analysis showed that age (OR=1.125), gait during exercise (OR=594.435), history of falls (OR=1 807.398), sleep condition (OR=16.772), vision condition (OR=6.710), hearing condition (OR=42.385), defecation ability (OR=34.771), Mini-Mental State Examination (MMSE) score (OR=1.575), nutritional score (OR=1.347), nursing classification (OR=1.493), and Activities of Daily Living (ADL) (OR=0.981) were all factors influencing the occurrence of falls (all P<0.05). The decision tree model was verified using the test set, and the evaluation indicators of the training set and test set were obtained. The accuracy of the training set was 0.983 9, the recall rate was 0.958 6, and the F1 value was 0.960 2; the accuracy of the test set was 0.978 7, the recall rate was 0.935 5, and the F1 value was 0.946 9. The analysis of feature importance in the decision tree model shows that gait during exercise, MMSE score, history of falls, and age are all important features for predicting the risk of falls in hospitalized patients. Through practical application, the decision tree model constructed in this study has increased the assessment rate for patients at high risk of falls and reduced the incidence of falls. Conclusions　Gait during exercise, MMSE score, history of falls, and age are four important factors for predicting the risk of falls in hospitalized patients. They are conducive to the rapid screening of high-risk patients with falls and the comprehensive management of fall prevention in clinical settings.

Key words: Patient, Fall risk, Decision tree, Prediction model