Systematic evaluation of the application effect of machine learning models in the prediction of frailty risk of community-dwelling older adults

CHENG Yuxin; LIU Yongfen; HUANG Xiaorong; YIN Jie; HU Dieer

doi:10.3969/j.issn.2096-9058.2025.01.010

Abstract：Objective To systematically evaluation the predictive value of machine learning(ML)for the frailty risk of community-dwelling older adults.Methods PubMed,Embase,Web of Science,WanFang Data,CBM and CNKI da-tabases were electronically searched to retrieve all ML studies on predicting the frailty of community-dwelling older adults from the time of database construction to July 2024.Two investigators independently conducted literature identification,screening,data included,and bias risk assessment for the studies that were included.The predictive value of each model was determined by examining the area under the receiver operating characteristic curve(AUC)and the accuracy.Re-sults A total of 13 studies were included.Regarding data sources,13 studies were obtained from the community,4 stud-ies were based on public database,and 7 from China.A total of 20 models had been adopted,among which the most popu-lar ML methods included Random Forest(n=5),followed by Support Vector Machine(n=4)and Extreme Gradient Boost-ing(n=4).The most frequently used input predictor was age(n=11),followed by education level(n=7),depression(n=5),and physical activity(n=5).A total of 4 studies compared the predictive value of ML and traditional statistical mod-els,2 of which concluded that the AUC value of the ML model was higher than that of the traditional statistical model,and it was noteworthy that 2 studies took the opposite view,suggesting that logistic regression had the highest predictive effica-cy.A study compared the predictive value of different ML models and concluded that the Random Forest had the best pre-dictive efficacy.Conclusions ML is widely used in predicting frailty risk among community-dwelling older adults,accu-rately predict the onset and progression of frailty.In specific scenarios,ML models outperform traditional models in frailty prediction,although some studies argue that logistic regression is more effective.Further validation is needed to assess the predictive performance of ML models for frailty.