Prognostic Model Establishment Of Diabetic Kidney Disease G3a-G4 Patients And Influence Of Chinese Medicine Intervention

Objective:The aim of this study is to explore Diabetic kidney disease through the combination of Random survival forest(RSF)and COX proportional risk regression.Influencing factors of renal replacement therapy and creatinine doubling in stage G3a-G4 patients,And the influence of traditional Chinese medicine intervention on the prognosis,and based on this,the risk prediction model was constructed,and the visualization column graph was further drawn.To improve the probability of accurate screening of high-risk groups,realize early individualized intervention,improve the survival rate of residual nephrons,and provide reference for enriching risk prediction in clinical work.Method: This study is a retrospective cohort study.From January 2019 to April 2022,patients diagnosed with diabetic nephropathy were retrieved through the integrated platform of scientific research in our hospital,and data preprocessing and database construction were carried out on the relevant case data of patients who met the exclusion criteria.The patients included in the study were divided into the traditional Chinese medicine intervention group and the Western medicine treatment group according to the treatment plan.Kaplan-Meier(KM)method was used for survival analysis,and log-rank test was used to compare the therapeutic effect between the two groups.R4.2.2software was used to establish the traditional COX regression model and RSF-COX model.Firstly,all the included variables were screened by COX single factor method,and meaningful single factors were included in COX regression for multi-factor analysis and the traditional COX regression model was established.Then,the beneficial variables selected by RSF were included in COX multi-factor regression to establish the RSF-COX model.The prediction ability of the two groups of models was evaluated by the concordance index(C-index).The results were expressed in the form of a line graph.At the same time,the consistency calibration curve was drawn for the line graph model to evaluate the accuracy of the prediction model.Result: A total of 91 diabetic nephropathy patients were included in this study,and a total of 40(43.96%)patients developed compound endpoints.The median follow-up time was 769 days,and the mean age was 63.51±10.66 years old.There were 69 males(75.82%)and 22 females(24.18%).In the traditional Chinese medicine(TCM)intervention group(n=49 cases),the average age was 63.22±9.63 years old,among which 37 cases(75.51%)were male.In the Western medicine treatment group(n=42 cases),the mean age was 63.84±11.87 years old,of which 32 cases(76.19%)were male.There was no statistical difference between the two groups(P<0.05).The random survival forest method was used to rank the importance of all 20 variables,and 11 variables with VIMP greater than 0 were selected by the VIMP method:Serum albumin,urinary protein quantification,erythrocyte sedimentation rate,hemoglobin,glycosylated hemoglobin,low density lipoprotein cholesterol,diabetic retinopathy or not,cystatin C,urea nitrogen,and glomerular filtration rate by treatment.Moreover,these variables conducive to model prediction were incorporated into COX multivariate analysis to construct the RSF-COX model.The influencing factors with obvious correlation were finally screened out as follows: Treatment regimen(P=0.033 < 0.05),urinary protein quantification(P=0.019 < 0.05),erythrocyte sedimentation(P=0.039 < 0.05),glycosylated hemoglobin(P=0.035 < 0.05).At the same time,univariate COX regression analysis was used,among which the P values of 10 variables including serum albumin,urine protein quantity,erythrocyte sedimentation rate,hemoglobin,glycosylated hemoglobin,age,triglyceride,treatment method,retinopathy and fasting blood glucose were < 0.1.COX multivariate analysis was conducted again,and the factors with significant correlation were finally screened out as follows: treatment plan(P=0.043 < 0.05),urinary protein quantity(P=0.032 < 0.05),erythrocyte sedimentation rate(P=0.045 <0.05).The prediction ability of the two groups of models was evaluated by C-index,and the score of the RSF-COX model was 0.794,while that of the traditional COX model was 0.782,both of which belonged to the medium accuracy,but the former was higher than the latter.Therefore,the variables screened by the former were used to construct a column graph.The model fitting curves of the two groups showed that the calibration curve and the standard curve basically matched,which proved that the probability of complex endpoint events in diabetic nephropathy patients predicted by the model was similar to the actual situation,and the model could obtain better clinical benefits.Conclusion: The KM curve of the treatment method in this study suggests that the cumulative survival rate of the TCM intervention group is higher than that of the Western treatment group.Although P > 0.05 in univariate analysis,the kidney survival rate of TCM intervention group was better than that of Western treatment group after multivariate analysis to reduce the influence of confounding factors,which was statistically significant.It can be seen that TCM intervention has certain favorable factors for the progression of DKD.RSF and COX proportional risk regression were used to screen out factors closely affecting the outcome events.Such as treatment,urinary protein quantification,erythrocyte sedimentation,glycosylated hemoglobin.The prognostic risk prediction diagram of DKD constructed by us has important clinical value for early prognostic prediction of DKD and early identification of high-risk patients.