| Background:End stage renal disease(ESRD)is the final outcome of the progression of various chronic kidney diseases.Most patients with ESRD require renal replacement therapy to maintain their lives.In recent years,due to the increase of kidney disease,the number of patients entering ESRD has increased year by year,and it has become a major part of social security payments.Renal replacement treatment methods include hemodialysis(HD),peritoneal dialysis(PD),and renal transplantation.However,due to the lack of kidney sources,hemodialysis and peritoneal dialysis are still the current renal replacement treatments in our country.Studies have shown that peritoneal dialysis is significantly better than hemodialysis in medical costs,which can save valuable medical resources for social security.However,how to choose the best dialysis mode,the survival rate is one of the most important considerations.PD patients often have a variety of complications,and the longterm survival rate is low.Cardiovascular disease(CVD)is the most common complication.Compared with healthy people,the incidence and mortality of cardiovascular diseases are significantly higher.Studies indicate that more than 50%of PD patients die from cardiovascular diseases.Therefore,risk factor analysis of cardiovascular diseases for such patients and early prediction of patients at high risk of death is very important.Active early intervention to improve the survival PD patients has always been the focus of research.Traditional statistical models,such as correlation analysis,simple linear regression,logistic regression model,Cox regression,etc.,have found some predictive indicators related to clinical prognosis.In essence,these studies are often based on linear relationships.However,with the increase of massive medical data,traditional statistical models can not fully reveal the complex non-linear relationships between characteristic variables.Therefore,it is necessary to choose other more appropriate data analysis tools to further study the factors related to the survival of PD patients so as to find high-risk patients earlier and find more suitable intervention targets,and ultimately improve the survival prognosis of PD patients.On the other hand,it can also increase the willingness of patients to choose PD,thereby increasing the proportion of their renal replacement therapy and reducing the burden of social security expenditure.Part Ⅰ.Risk factors of cardiovascular events in single peritoneal dialysis centerPurpose:The incidence of cardiovascular events in peritoneal dialysis patients is significantly higher than that of the general population.Cardiovascular disease is the most common cause of death in patients.Research on the risk factors of cardiovascular events has positive significance for improving the survival rate and quality of life of patients.The clinical data of peritoneal dialysis patients in our center was retrospectively analyzed so as to improve the prevention,diagnosis,and treatment of cardiovascular events in peritoneal dialysis patients.Methods:We identified 980 peritoneal dialysis patients who were followed up in the First Affiliated Hospital of Wenzhou Medical University from January 2006 to December 2018.Cases were filtered by the including and exclusion criteria.The demographic and clinical data at the beginning of peritoneal dialysis and during the follow-up period were collected.Whether there were cardiovascular events(including acute myocardial infarction,unstable angina,non-fatal heart failure,ischemic stroke and hemorrhagic stroke),they were divided into cardiovascular event groups and non-cardiovascular event groups.Univariate COX regression was used to analyze the risk factors of cardiovascular events,The variables with P<0.10 in univariate COX regression analysis were included,and multivariate COX regression analysis was performed.Result:1.A total of 757 eligible CAPD patients were included,of which 417 were males and 340 were females,with a median age of 49(38-60)years and a median follow-up time of 44.5(25-69)months.Stratified by age,the number of people aged 50 to 60 was the largest,accounting for 24.17%,followed by those aged 40 to 50,accounting for 23.38%,and patients older than 60 years,accounting for 26.29%.Stratified by dialysis age,the proportion of peritoneal dialysis patients with dialysis age between 24 and 48 months was the highest,accounting for 30.91%followed by those 48-72 months,accounting for 23.78%.Among the causes of ESRD,chronic glomerulonephritis was the highest,accounting for 50.99%,followed by hypertensive nephropathy accounting for 13.61%,and diabetic nephropathy accounting for 13.47%.During the follow-up period,128 cases(16.9%)had cardiovascular events,of which 91 had cerebrovascular events,70 had cardiovascular events,80(10.57%)died,102(13.47%)were transferred to hemodialysis,128 patients(16.91%)had kidney transplantation,91 patients(12.02%)were lost to follow-up,and 356 patients(47.03%)continued peritoneal dialysis.Among the causes of death,49 cases(61.25%)died from cardiovascular events,13 cases(16.25%)died from infection,11 cases(13.75%)died from tumors,and 7 cases(8.75%)died from other causes.According to cardiovascular events,they were divided into cardiovascular event groups and non-cardiovascular event groups.2.Compared with the group without cardiovascular disease,the group with cardiovascular events has a higher age(P<0.001),a higher proportion of diabetes mellitus(P<0.001),and longer dialysis age(P=0.028).The creatinine value was lower(P=0.02),serum phosphorus(P=0.028)was lower,serum iPTH(P=0.004)was lower,and diastolic blood pressure was lower(P<0.001),the difference was statistically significant.3.Univariate COX regression analysis showed:older age(1.07,95%CI:1.06-1.09,P<0.001),high body mass index(1.19,95%CI:1.12～1.27,P<0.001),low serum albumin(0.97,95%CI:0.94～0.99,P=0.018),low serum phosphorus(0.45,95%CI:0.28～0.75,P=0.002),low serum iPTH(0.998,95%CI:0.997～0.999,P<0.001),high systolic blood pressure(1.009,95%CI:1.002～1.017,P=0.017),low diastolic blood pressure(0.984,95%CI:0.973～0.995,P=0.005),combined diabetes(2.71,95%CI:1.61～4.31,P<0.001)is a risk factor for cardiovascular events in PD patients.4.With reference to the test level of α=0.1,age,body mass index,residual urine volume,residual renal urea clearance index(Krt/V),serum albumin,serum phosphorus,serum iPTH,systolic blood pressure,diastolic blood pressure and diabetes mellitus were included as independent variables in multivariate COX regression analysis.The results showed that older age(1.057,95%CI:1.035-1.078,P<0.001),low serum albumin(0.972,95%CI:0.936-0.998,P=0.041),low serum iPTH(0.998,95%CI:0.998-0.999,P=0.005)and diabetes mellitus(1.23,95%CI:0.98～1.89,P=0.038)were independent risk factors for cardiovascular events in PD patients.Conclusion:The incidence of cardiovascular events in PD patients in our center is as high as 16.9%.Cardiovascular events are the most common cause of death in PD patients in our center.Old age,low iPTH levels,low albumin and diabetes are independent risk factors for cardiovascular events in PD patients,and clinical attention should be paid to the management of iPTH and albumin.Part Ⅱ.Role of Parathyroid hormone and vitamin D supplementation in stroke among patients on peritoneal dialysisPurpose:The high incidence of stroke in peritoneal dialysis patients is often accompanied by elevated levels of intact parathyroid hormone(iPTH)and insufficient vitamin D.This study investigated the incidence of stroke in CAPD patients and the effect of iPTH levels and vitamin D use on the incidence of stroke in peritoneal dialysis patients.Methods:Retrospectively enrolled 980 peritoneal dialysis patients with long-term follow-up in the First Affiliated Hospital of Wenzhou Medical University.The demographic and clinical data recorded at the beginning of CAPD and during follow-up were collected.According to the time of occurrence of stroke,they were divided into stroke group and nonstroke group.To evaluate the role of iPTH levels and vitamin D use in stroke in CAPD patients.The primary endpoint was defined as the first stroke,and the composite endpoint was defined as death or conversion to hemodialysis.Result:1.A total of 757 eligible peritoneal dialysis patients were included,of which 417 were males and 340 were females,with a median age of 49(38-60)years and a median follow-up time of 44.5(25-69)months.91 cases(12%)of patients had stroke events,including 74 cases of ischemic stroke(83.1%)and 23 cases(25.8%)of hemorrhagic stroke.The median age of stroke was 61.5 years,and the median annual incidence of stroke was 18.9‰.The incidence of stroke was high at initial peritoneal dialysis,peritoneal dialysis at 5 years,and peritoneal dialysis at 10 years.2.According to whether a stroke occurs,it was divided into a stroke group and a nonstroke group.Compared with the stroke group,the median age of the stroke group was older(P<0.001),serum albumin(P=0.002),serum phosphorus(P=0.001),iPTH(P=0.001),diastolic blood pressure(P<0.001)was lower in the stroke group,.Compared with the nonstroke group,The proportion of patients with chronic heart disease(P<0.001)and the proportion of patients with diabetes(P<0.001)were higher in the stroke group.Compared with the non-stroke group,the composite endpoint event(P<0.001)was higher in the stroke group.Compared with the non-stroke group,the use rate of active vitamin D in the stroke group(P=0.002)was lower,and the difference was statistically significant.3.There was a significant difference in the probability density distribution of iPTH between the two groups.The stroke group has a significant peak shift to the left,suggesting that the stroke group had a lower serum iPTH level compared with the no-stroke group.The relationship between the baseline iPTH level and the relative incidence of stroke showed a J-shaped curve.4.According to the baseline iPTH level,the groups were divided into 4 groups,namely iPTH≤150 pg/ml group,iPTH 150-300 pg/ml group,iPTH 300-600 pg/ml group,and iPTH>600 pg/ml group.Kaplan-Meier survival analysis shows a significant difference in the cumulative risk of stroke between the groups(Log-rank test,P<0.001),and the cumulative risk of stroke in the iPTH≤150 pg/ml group was significantly increased.The comparison between the paired groups showed that the cumulative risk of stroke between the ≤150pg/ml group and the 150-300pg/ml group and the 300-600pg/ml group was significantly different(P 0.002 and<0.001,respectively).the cumulative risk of stroke had no significant difference compared with iPTH>600 groups(P=0.1).5.Univariate Cox regression analysis found that increasing age,decreased DBP and iPTH levels,combined with chronic heart disease and diabetes,receiving antiplatelet drugs,and not using active vitamin D were risk factors for stroke and composite endpoint events.After adjusting for the interaction between baseline iPTH and age and other confounding factors,multivariate Cox regression analysis showed that low serum iPTH levels were still a risk factor for stroke,and the use of active vitamin D was a protective factor for stroke and the composite endpoint.Conclusion:CAPD patients had a higher incidence of stroke,especially at the threetime points of initial peritoneal dialysis,peritoneal dialysis at 5 years,and peritoneal dialysis at 10 years.Lower iPTH levels were significantly associated with an increased risk of stroke,especially the baseline iPTH level,which was an independent risk factor for stroke.Lower iPTH level increased the risk of stroke,and vitamin D supplementation was an independent protective factor for stroke in peritoneal dialysis patients.Part Ⅲ.Prediction of all-cause mortality in patients receiving peritoneal dialysis using modified artificial neural networksPurpose:The first part of our study showed the clinical indicators of PD patients were closely related to the adverse prognosis of cardiovascular events and all-cause mortality.To find characteristic variables closely related to death and explore the impact of early intervention on the prognosis of PD patients.In this part,Our center’s registered PD followup data for up to 14 years was analyzed.A robust early prediction risk model for PD patients’death finally was constructed using an artificial neural network(Artificial Neural Network,ANN)based on deep learning.Subjects and methods:A retrospective analysis of 1,241 PD patients in the First Affiliated Hospital of Wenzhou Medical University from January 2006 to December 2019 were followed.Demographic data and clinical data at the beginning of PD were collected,and data cleaning was performed on the collected data.Bootstrap’s method was used to split the data set at the beginning of PD,and divide it into training set,validation set and test set.The training and validation set had partial intersections,but the test set and the other two groups had no intersection,data division were performed 100 Cycles to construct 100 data sets.The Tensorflow framework was used to build a classic ANN and a new architecture ANN based on the Python3.7 environment.at the same time,A Logistic regression model was constructed in the training set.All feature variables in the construction process were included.The test set and the total data were used to test the predictive performance of the classic ANN model,the new architecture ANN model and the Logistic regression(Logistic regress,LR)model.The differences between the three model was compared using multiple evaluation indicators.In addition,the ranking importance analysis method was used to analyze and rank the included variables to explore the relationship between related variables and prognosis.the eigenvalue perturbation method was also used to explore the relationship between eigenvalue changes and predicted value changes.Result:1.859 PD patients who met the inclusion criteria were included in the study.Among them,483 were males and 376 were females.The median age was 49.5(38-60)years,and the median follow-up time was 40.5(21,66).82 patients died during follow-up,which was defined as PD-related death.According to whether death occurred,the patients were divided into death group and no death group.2.The clinical and laboratory characteristics of the two groups at baseline:Compared with the no-death group,the death group had a higher proportion of diabetes and chronic heart disease at baseline.compared with the no-death group,The age(P<0.001)was elder,and LDL-cholesterol(P<0.001)was higher in the death group.Compared with the no death group,the death group had lower baseline diastolic blood pressure(P<0.001)and albumin(P=0.004),the differences were statistically significant.3.The clinical and laboratory characteristics of the two groups at 6 months:Compared with the no-death group,the age was elder(P<0.001).Compared with the no-death group,LDL-cholesterol was higher in the death group(P=0.045).Compared with the no-death group,the proportion of diabetes(P<0.001),and chronic heart disease(P<0.001)were higher,Compared with the no-death group,diastolic blood pressure(P<0.001),hemoglobin(P=0.04),serum creatinine(P=0.03),albumin(P<0.001),iPTH(P=0.01)and HDL-cholesterol(P=0.01)were lower in the death group,and the differences are statistically significant.4.The clinical and laboratory characteristics of the two groups at 12 months:Compared with the no-death group,the age(P<0.001)was elder in the death group,Compared with the no-death group,LDL-cholesterol(P=0.04)was higher in the death group.Compared with the no-death group,the ratio of diabetes(P<0.001)and chronic heart disease(P<0.001)was higher.Compared with the no-death group,diastolic blood pressure(P<0.001),hemoglobin(P=0.01),blood creatinine(P=0.04),albumin(P<0.001),iPTH(P=0.01)and HDLcholesterol(P=0.03)were lower in the death group,the difference is statistically significant.5.The multivariate logistic regression model based on baseline,PD 6 months and PD 12 months data sets further confirmed that older age,chronic heart disease,decreased serum albumin levels,and elevated LDL-cholesterol levels were independent risk facto for allcause deaths in PD patients.6.The artificial neural network was constructed using the Tensorflow framework based on the Python language,and two artificial neural networks were constructed respectively,namely the classic ANN model and the new architecture ANN model.The classic ANN model added all the data at the same time,which did not distinguish between numerical variables and categorical variables,and included 1 neural network and 12 hidden layers.The new architecture of the ANN mixed model included 2 sub-neural networks:numerical variables(9 hidden layers),categorical variables(11 hidden layers),and the sub-neural network was finally merged into a new neural network(2 hidden layers).7.In the baseline test set,the LR model(AUROC:0.77)was slightly better than the ANN model(AUROC was 0.7),and the 0 month(overall data set)the ANN model was better(the new architecture ANN model was 0.85,the classic ANN model was 0.87,and the LR model was 0.81.);PD 6 months,the new architecture ANN model was the best(new architecture ANN model 0.8,classic ANN model 0.48,LR model 0.46);PD 12 months:the new architecture ANN model was the best(new architecture ANN model 0.79,classic ANN Model 0.48,LR model 0.46),suggesting that internal verification showed that the predictive performance of the new architecture model was similar to that of the classic ANN model,but when 6 month and 12month data were used as external verification data,the results showed the performance of the classic ANN model and the LR model significant declined,and the prediction performance of the new architecture model was more robust.8.Using disturbance-related variables to calculate the changes in the predicted value,the results showed that in the new architecture ANN model,the evaluation of the feature value was more robust.As the age increases,the predicted value increased.In the new architecture ANN model also showed that as the level of LDL-c increases,the predicted value increased.However,in the classic model,this feature was not significant,which also indicated that the classic model could not effectively extract the global feature of the feature value,and may be limited to local feature interpretation.9.We used the importance of rearrangement to judge the impact of the included variables on the performance of the model,and indirectly reflect the relationship between the included variables and prognosis.The results showed that age was the most robust predictor.Diastolic blood pressure,LDL-cholesterol,triglycerides and Systolic blood pressure were also closely related to prognosis.Conclusion:Through this research,we had collected a large sample of PD data of our center for 14 years,and constructed an ANN model based on the new architecture,which had shown good scalability in external verification.In addition,our research also showed that age,diastolic blood pressure and LDL-cholesterol were closely related to the premature death of PD patients.In particular,diastolic blood pressure and LDL-cholesterol were required more attention. |
PDF Full Text Request