Analysis Of Risk Factors Related To Death In 64 Patients Undergoing Single-center Peritoneal Dialysis

Objective:BP neural network and logistic regression model were used to analyze and explore the risk factors related to death in peritoneal dialysis patients,so as to provide a theoretical basis for clinical intervention of related risk factors,and to improve the prognosis and survival rate of peritoneal dialysis patients.Methods:263 patients with PD hospitalized in nephrology department of our hospital from January 2012 to August 2018 were selected.All patients were divided into death group(64 cases)and survival group(199 cases).The basic conditions of patients were analyzed,including age,sex,height,weight,blood pressure,primary disease,smoking and Kt/V within 1-3 months after dialysis,infection during observation,cardio-cerebrovascular disease and the cause of death at the end of observation;Biochemical indicators including HB,ALB,TC,TG,LDL,HDL,CRP,Scr,UA,Ca,P,Mg and iPTH were detected,and BMI,eGFR and Kt/V were calculated by corresponding formulas.Chi-square test and t-test were used to compare the differences of clinical indicators between the two groups.Logistic regression and BP neural network were used to analyze and explore the risk factors affecting the death of PD patients.Results:1.General situationA total of 263 PD patients were enrolled in this study.The average age was(51.81 +12.94)years,143 males(54.37%)and 120 females(45.63%).The overall dialysis time was(32.13 +16.52)months.The average dialysis time in the dead group was(28.13 +17.31)months.81 cases(30.80%)were complicated with cardiovascular and cerebrovascular diseases.Primary diseases included chronic glomerulonephritis in 113 cases(42.97%),diabetic nephropathy in 87 cases(33.08%),hypertensive renal damage in 44 cases(16.73%)and other causes in 19 cases(6.46%).By the end of the study,64 patients(24.33%)died.The causes of death included 18 cases of CVD(28.13%),13 cases of cerebrovascular diseases(20.31%),19 cases of infection(29.69%)and 14 cases of other causes of death(21.88%).2.The comparison of clinical data between death group and survival groupCompared with the survival group,the combination of death and cardiovascular and cerebrovascular diseases,infection,Scr,BUN,P,UA,iPTH were significantly higher,ALB and eGFR were significantly lower(P < 0.05);there were no significant difference in gender,age,smoking,primary disease,BMI,MAP,Hb,TC,TG,LDL,HDL between the two groups(P > 0.05).3.Risk factors associated with death in PD patients(1)Logistic regression analysis Taked death as the dependent variable and adjusted the confounding factors,the two-category Logistic stepwise regression was used to analyze the included indicators.The results showed that hypoproteinemia,combined with cardiovascular and cerebrovascular diseases,infection,inadequate dialysis,and hyperphosphatemia were independent risk factors for death of PD patients(P<0.05).The above five risk factors and constant terms were corporated into the logistic regression equation,and the predictive equation based on the model parameters was:logit(P)= 3.303+(-2.488)X5+(-0.118)X3+(-1.596)X4+1.581X2+0.052X1(2)BP neural networkBP neural network was used to rank the five significant variables in logistic stepwise regression analysis.The results showed that the most important factors affecting the mortality of PD patients were the combination of cardiovascular and cerebrovascular diseases(0.284),and the rest in order were infection.(0.197),hypoproteinemia(0.196),hyperphosphatemia(0.176),and inadequate dialysis(0.147).The establishment of BP neural network prognostic model: The input unit was five significant variables in Logistic stepwise regression analysis,and the outcome(survival or death)of PD patients was taken as the output unit.The final verification of ANN architecture included a hidden layer,which included three nodes H(1:1),H(1:2)and H(1:3).4.Comparison of Model Prediction Ability-Area under ROC Curve(AUC):The AUC of logistic regression prediction equation was 0.701 and 95% CI is 0.581-0.821.The AUC of BP neural network prediction model was 0.839 and 95% CI is 0.768-0.910.The AUC of the two models was significantly different(Z > 1.96,P < 0.05).The results show that compared with the Logistic regression model,the model constructed by BP neural network has significant advantages in prediction performance.Conclusion:1.Cardiovascular and cerebrovascular diseases,infection,hypoproteinemia,hyperphosphorus and insufficient dialysis are independent risk factors for the death of PD patients in our center.Active improvement of these conditions may help to improve the survival rate of PD patients.2.Compared with the Logistic regression model,the BP neural network has certain advantages in predicting the death risk of PD patients.