Comparison of value of risk assessment models based on three machine learning algorithms in predicting frailty risk among maintenance hemodialysis patients
WANG Dandan
YAO Kanfei
ZHU Xuehua
Abstract：Objective:To compare the value of risk assessment models based on Logistic regression,decision tree and random forest machine learning algorithms in predicting frailty risk among maintenance hemodialysis patients.Methods:From October 2021 to March 2022,a total of 485 patients receiving maintenance hemodialysis treatment in two tertiary grade-A hospitals in Hangzhou were selected and treated according to 7∶3 ratio was randomly divided into training set(n=341)and test set(n=144).Logistic regression,decision tree and random forest were used to establish frailty risk prediction models for maintenance hemodialysis patients.Accuracy,sensitivity,specificity,positive predictive value,negative predictive value,Kappa and AUC value were used to compare the predictive performance of the three models.Results:In the training set,the accuracy of Logistic regression,CART,and random forest were 91.79%,91.50%and 97.95%,the specificity was 96.84%,92.11%,and 96.91%,and the sensitivity was 85.43%,90.73%,and 99.32%,respectively.The positive predictive value was 95.56%,90.13%,96.05%,the negative predictive value was 89.32%,92.59%,99.47%,the Kappa value was 0.832,0.828,0.958,and the AUC value was 0.971,0.954,0.998.The AUC values of the three models were tested,and the results showed that the random forest model was significantly different from the other two models(P<0.05).Age,gender,Charlson Comorbidity Index and nutritional risk screening score were common predictors of the three prediction models.Conclusion:Random forest model is the best model in predicting frailty risk among maintenance hemodialysis patients.
Keywords：maintenance hemodialysisfrailtyprediction modelLogistic regressiondecision treerandom forest
Publication Date：2024-01-10
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：9( 8-16 )