Establishment of a Machine-Learning-Based Predaiction Model for the Timing of Dialysis in Patients with CKD Stage 4-5 Treated with the Method of"Yishen-qingli-Huoxue Therapy"
Objective We constructed a prediction model of the time point for CKD stage 4-5 patients to enter renal replacement therapy with the help of machine learning method,which can provide guidance for the selection of clinical treatment plan.Methods A retrospective cohort study was conducted to include patients with CKD stage 4-5 treated by Prof.Sun Wei with the"Yishen-qingli-Huoxue Therapy"from January 2010 to March 2021,Clinical data of patients with CKD stage 4-5 were collected,and relevant variables such as demographic data,laboratory test results,TCM symptoms,syndrome differentiation and use of Chinese medicine were screened.With renal replacement therapy as the end event,linear regression model combined with random forest model was used to reduce dimension of independent variables(predictors)in three stages.The variables with statistical significance(P<0.05)were screened,and a multi-linear prediction model was established based on symptoms,prescriptions,physical and chemical indexes,the model was evaluated by adjusted determination coefficient(Adjusted R-Square,Adjusted R2)and Bland-Altman plots.Results Five predictors were selected from the predictor variables and constructed with multiple linear model equation lnDay=5.058+0.031×albumin-0.004×creatinine+0.010×hemoglobin-0.412×using Centella-0.715×skin pruritus;the predicted value was evaluated using the Bland-Altman plot,showing that the scatter in the Bland-Altman plot was well distributed within the 95%normal value of the difference,and the consistency between the predicted value and the real value was good.Conclusion The multiple linear prediction model can be used to assist clinical prediction of the length of renal function progression,which is conducive to identify high-risk groups and provide reference for the selection of regimen before entering renal replacement therapy.
Timing of dialysisYishen Qingli Huoxue TherapyRandom forestPrediction model
NETL
NSTL
万方数据
