Construction Of A Risk Prediction Model For The Frailty In Community-dwelling Older Adults Based On Machine Learning

BackgroundWith the accelerated process of population aging,frailty,as an important geriatric syndrome,has gradually become a research hotspot in the field of geriatrics at home and abroad.The study showed that the prevalence rate of older people in the community ranged from 4.0% to 59.1%,and the prevalence of combined frailty was 10.7%.Frailty is a state in which the reserve and function of multiple physiological systems are reduced due to age growth,resulting in the impairment of body stability and the susceptibility to external pressure,which can lead to serious adverse consequences such as falls,disability,hospitalization and even death of older people.Machine learning is an important branch of artificial intelligence learning.It can use a large number of data samples for autonomous learning,and has good data processing and prediction ability.It has been more and more used in the medical field to predict disease risks and adverse outcomes.Applying machine learning to the prediction of the frailty risk can identify frailty early,accurately and effectively predict the risk of frailty,and improve the quality of life of the older people.ObjectiveTo describe and analyze the current situation of frailty and related risk factors of the older people in the community,and use a variety of machine learning algorithms to build a prediction model for older people in Chinese communities,and select the model algorithm with the best prediction performance through the model evaluation indicators.MethodsThis study is a cross-sectional study,which was conducted from October 2021 to June 2022.A total of 533 older persons in Changchun community who met the inclusion and exclusion criteria were selected by convenient sampling method.The survey tools were general information questionnaire,Mini-nutritional Assessment-shortform(MNA-SF),Pittsburgh Sleep Quality Index(PSQI),Activity of Daily Living(ADL),Short Falls Efficacy Scale-International(Short FES-I),Geriatric Depression Scale-15(GDS-15),Self-Rating Anxiety Scale(SAS),Social Support Rating Scale(SSRS),Tilburg frailty indicator(TFI).SPSS 24.0 software was used for statistical analysis.The categorical variables are described by frequency(n)and constituent ratio(%),and continuous variables are described by mean ± standard deviation((?)±s).The potential risk factors of frailty are analyzed by single factor analysis.The continuous variables subject to normal distribution are tested by utilizing independent t-tests and the continuous variables subject to abnormal distribution are tested by non-parametric test.The Chi-squared test is used for dichotomous variables and the non-parametric test is used for more than three categories of variables.The variables with statistical significance(P<0.05)in the single factor analysis are used to build the prediction model of frailty.Use Python 3.7 and Logistic regression,decision tree,naive Bayes,random forest,and XGBoost algorithm in machine learning to build a prediction model for the frailty in the community-dwelling older adults,and the confusion matrix and ROC curve are used as the evaluation indicators of the model.Results1.In this study,a total of 550 questionnaires were distributed in 7 communities in Changchun,533 of which were valid,with a response rate of 96.9%.The prevalence of frailty among community-dwelling older adults was 28.0%.2.The results of univariate analysis showed that there were statistically significant differences in age,education level,monthly family income,marital status,living status,comorbidity,multiple medication,fall history and surgery history in the past six months,self-evaluation of health status,fear of falling,sleep status,nutritional status,activity of daily living,depression and anxiety(P<0.05).3.The evaluation indicators of the prediction model built by a single algorithm showed that the prediction performance of the Logistic regression model was better than that of the other two models.The AUC value of Logistic regression model was0.7833,F1 value was 0.6053.The AUC value of Bernoulli Bayesian model was 0.7827,F1 value was 0.5679.The AUC value of Decision Tree model was 0.7192,F1 value was 0.5122.4.The performance of the integrated algorithm model was higher than that of the single algorithm prediction model.The results showed that the maximum AUC value of XGBoost model is 0.8126,F1 value is 0.5352,and the AUC value of Random Forest model was 0.8006,F1 value was 0.4127.Considering the AUC value and F1 value,the performance of XGBoost model was superior to all other models in predicting the frailty in community-dwelling older adults.5.In the XGBoost model,the importance order of the independent variables of the frailty risk of the older people in the community was fear of falling,self-evaluation of health status,age,activity of daily living,depression,sleep status,comorbidity,living status,family monthly income,nutritional status,education level,multiple medication,surgery history and fall history in the past 6 months.Conclusion1.The prevalence of frailty among older people in Changchun community was high,and the attention of community medical staff to frailty should be increased.2.The risk factors of frailty of the older people are multi-dimensional,involving physical,psychological and social factors.Therefore,a comprehensive assessment of multi-dimensional risk factors may be more conducive to the prediction of the risk of frailty.3.XGBoost is the best algorithm for predicting the risk of frailty in communitydwelling older adults.In the future,it can be combined with the electronic health records of the older people in community to timely screen the risk of frailty and provide personalized intervention measures.