Construction Of Fall Risk Prediction Model For Community Elderly People Based On Machine Learning

Objective:Understand the occurrence of falls of the elderly in the community,analyze the influencing factors of the fall of the elderly in the community,describe and analyze the relationship between the fall of the elderly in the community and its influencing factors,construct a fall risk prediction model based on three machine learning algorithms according to its influencing factors,and select the optimal model for model evaluation,so as to provide an evaluation tool for the early prevention and intervention of falls of the elderly in the community.Methods:(1)Determine the search terms,search the Chinese and English databases through the method of literature review,obtain the influencing factors of falls of the elderly in the community from the previous literature.(2)The summarized influencing factors are designed as a correspondence scale,expert correspondence is carried out through the Delphi method,and the entries are processed on the basis of expert positivity coefficient,expert authority coefficient,expert coordination coefficient and entry retention standards,and a questionnaire on fall risk factors for the elderly in the community is formed.(3)383 elderly people in some communities in Changchun City were evaluated and obtained information through questionnaire survey method,a data set was established,and the collected data was statistically described and analyzed using SPSS 25.0,and the variable with P <0.05 was obtained as the predictor variable of the model after one-way analysis.(4)With the fall as the ending event,machine learning algorithms were used to construct a fall risk prediction model for the elderly in the community based on logistic,support vector machine and random forest through Anaconda,and the optimal parameters were searched by the 5-fold cross-validation method,and the AUC value(area under the ROC curve),accuracy,accuracy,sensitivity,The specificity and F1 values evaluate the model and select the optimal model.Results:(1)The literature review shows that the influencing factors of falls of the elderly in the community include 63 influencing factors in six dimensions: sociodemographic characteristics,disease factors,drug factors,psychological factors,physiological factors,and home environment factors.(2)The positive coefficient of the first round of experts is 73.3%,and the positive coefficient of the second round of experts is 100%: the expert authority coefficient is 0.96;the first round of expert coordination coefficient is 0.428,and the second round of expert coordination coefficient is 0.477.By adding,deleting and modifying entries,a questionnaire on fall risk factors for the elderly in the community was formed,including 58 factors such as age,gender,and vision.(3)Among the 383 elderly people in the community included in this study,75 cases of falls occurred,and the incidence of falls was 19.6%,and 17 factors such as age,endocrine metabolic diseases,and fear of falling were statistically different through univariate analysis and included in the characteristic variables of the prediction model.(4)The AUC value of the RF model for the classification of fall risk of the elderly in the community reaches 0.70,the accuracy = 70.0%,the accuracy = 22.2%,the sensitivity is the recall = 16.0%,the F1 value = 0.19,the AUC value of the SVM model reaches 0.68,the accuracy = 73.0%,the accuracy = 33.0%,the sensitivity is the recall = 24.0%,and the F1 value = 0.28;The AUC value of the LR model reached 0.69,the accuracy rate = 70.0%,the accuracy = 38.2%,the sensitivity was recall = 60.0%,and the F1 value = 0.47.The distribution of feature importance showed that neurological diseases,age,hearing and other factors showed obvious importance in the process of predicting the risk of fall in the elderly in the community by logistic regression model.Conclusion:(1)The incidence of falls among the elderly in the community is 19.6%,and the incidence of falls is high,and community medical staff and family members of the elderly should pay close attention to the occurrence of falls in the elderly.(2)Through the analysis of the optimal model,it was found that the fall of the elderly in the community was mainly closely related to 17 factors such as neurological diseases,age,hearing,weakness,nocturia,circulatory diseases,nutrition and supplementation,number of medications,and the number of diseases,among which neurological diseases had the greatest impact on the occurrence of falls in the elderly in the community.