| Objective:This study intends to collect the relevant diagnosis and treatment information of patients with stage 4-5 CKD treated by "Yishen Qingli Huoxue Therapy",To construct a prediction model for the time point of renal replacement therapy for CKD4-5 patients,and the progress of renal function was analyzed,so as to provide guidance for the selection of clinical treatment plan.Methods:In this study,retrospective cohort study was used to include patients in stage 4-5 CKD who were regularly followed up by Professor Sun Wei with the Yishen Qingli Huoxue Therapy on the "China Kidney Disease Big Data Application innovation Alliance Founding Conference" from January 2010 to December 2020,Demographic data,laboratory examination results,TCM symptoms,syndrome differentiation and usage of TCM were collected comprehensively,The time(days)from the fourth stage to the endpoint event was set as the main outcome indicator.and use the duration as the dependent variable to use the machine learning method of random forest combined with linear regression to perform three-stage dimensionality reduction on the variables,Taking the time(days)as the dependent variable,machine learning method which linear regression model combined with random forest model was used to reduce the dimensionality of independent variables(predictors)in three stages.The variables with statistical significance(P<0.05)were selected to establish a multi-dimensional linear prediction model based on symptoms,prescriptions,physical and chemical examination indexes.The model was evaluated by Adjusted R-square and Bland-Altman graph.In addition,the final predictors were stratified to observe the trend of glomerular filtration rate in stage 4 patients over the course of 1 year.Results:(1)99 patients,were included in the analysis,including 61 males and 38 females;The mean age was 47.40± 13.51 years old,and the age range ranged from 17 to 78 years old;lnDay from baseline EGFR to endpoint EGFR was 5.99±0.91;There were 27 cases of primary glomerulonephritis,16 cases of diabetic nephropathy,31 cases of hypertensive nephropathy,and 6 cases of other diseases(including 1 case of obstructive nephropathy,I case of polycystic kidney disease,2 cases of systemic lupus erythematosus,2 cases of allergic purpura)with unknown causes.Family hypertension accounted for 28.28%,and family diabetes accounted for 8.08%;The majority of patients with hypertension(72.73%),anemia(66.7%),diabetes(33.3%)and metabolic acid(18.2%)were combined.(2)In the distribution of syndrome types,the main deficiency of CKD stage 4 patients is spleen and kidney qi deficiency(89.9%),followed by qi and yin deficiency(6.06%),spleen and liver and kidney yin deficiency(6.06%),kidney yang deficiency(3.03%);The empirical evidence is blood stasis syndrome(63.64%),damp turbidity syndrome(58.59%),water-qi syndrome(38.38%),damp-heat syndrome(20.20%),wind-movement syndrome(1.01%);the patients in this study did not see yin and yang deficiency certificate.(3)According to the four diagnostic information,the pulse pattern of CKD4 patients was mainly pulse-string or small-string,with a frequency greater than 50%;The frequency of blood stasis of sublingual veins in tongue images was more than 65%;The tongue is light or reddish,the tongue coating is mainly white or greasy white,The main clinical manifestations are fatigue,soft waist and knee,blurred vision,much urine foam and frequent nocturnal urination(4)Univariate regression analysis screened out 21 variables related to the main outcome indicators:marital status,primary disease of diabetic nephropathy,hemoglobin,total protein,albumin,alanine aminotransferase,baseline eGFR level,urea,serum creatinine,calcium;centella asiatica,dizziness,skin itching,white sputum,heart palpitations,gastric fullness,red tongue,blood stasis of veins under the tongue,slippery pulse,blood stasis syndrome and Chinese medicine prescription compatibility with Ptereophyllum,these variables have statistically significant relationships with time(P<0.05).(5)In the random forest model,9 important variables including albumin,creatinine,hemoglobin,primary disease,Centella assifolia,urea,baseline EGFR,serum calcium,and epigastric distend pain were screened out for the duration of CKD4 progression(days);Considering that the information of the four diagnoses of Chinese medicine and the dialectical classification of Chinese medicine may have the ability to predict the progress,the important variables of random forest screening as well as the statistically significant variables of Chinese medicine symptoms,tongue,pulse and Chinese medicine syndrome type in the univariate analysis were added into the final multi-linear model one by one.The analysis results showed that albumin(β=0.031,P=0.024),serum creatinine(β=-0.004),P=0.001),blood red white eggs(β=0.010,P=0.013),use of Centella Asiatica(β=-0.412,P=0.008),skin pruritus(β=-0.715,P=0.006)and time statistics.The multiple linear model equation was constructed as LNDAY=5.058+0.031*albumin-0.004*creatinine+0.010*hemoglobin-0.412*Centella asiatica-0.715*pruritus;The predicted values were evaluated by Bland-Altman diagram,and the results showed that the scattered points in the Bland-Altman diagram were well distributed within the 95%normal range of the difference values,which proved that the predicted values were in good agreement with the actual values.Conclusion:Using random forest combined with linear regression machine learning method,the statistical association between clinical information characteristics and the time to progression of CKD4 to renal replacement therapy was explored,Multiple linear models were constructed using statistically correlated predictors(albumin,hemoglobin,serum creatinine,treatment with Centella asiatica and clinical symptoms of pruritus).And Bland-Altmant graph indicates that the model has a good consistency evaluation.This model can be used to predict the duration of renal function progression in clinical practice,and can provide a reference for the selection of alternative therapy before entering renal replacement therapy. |
PDF Full Text Request