| Study objectives:The incidence of chronic kidney disease is increasing year by year worldwide and is receiving more and more attention.When it progresses to end-stage renal disease,most patients need to receive renal replacement therapy,of which hemodialysis becomes the choice of most patients.Therefore,clinicians’awareness and early intervention of cardiovascular disease risk factors in hemodialysis patients have a key role in prolonging the survival time and improving the quality of life of patients.Risk prediction models,which are based on multiple risk factors for the occurrence of a disease and are established by statistical methods,have been widely used in the medical field,but there is a lack of risk prediction models related to cardiovascular disease death in hemodialysis patients,based on which this study was conducted.The objectives of this study were(1)to explore and analyze the risk factors for death due to cardiovascular disease in maintenance hemodialysis patients;(2)to construct a nomogram prediction model based on the regression coefficients of the screened factors and to evaluate its discrimination,calibration,and clinical utility ability;(3)to construct a decision tree prediction model using the CART machine learning algorithm,to evaluate the predictive ability of the model and to assess the importance of the variables.It provides a basis for clinicians in making judgments about patients’conditions and medical decisions.Materials and Methods:Patients who received hemodialysis in the hemodialysis center of the First Affiliated Hospital of Chengdu Medical College from January 1,2016 to December 30,2021 were retrospectively collected.Patients who died due to cardiovascular disease were the death group,and patients who were hospitalized due to cardiovascular disease but did not die in the same time period were considered as the non-death group(if the same patient was repeatedly hospitalized,each hospitalization was counted as one record).Basic information,clinical data such as dialysis time,primary disease,whether they had diabetes,whether they had pulmonary hypertension,and blood pressure before dialysis,and test information such as blood routine examination,liver function,kidney function,electrolytes,parathyroid hormone,cardiac troponin I,brain natriuretic peptide,C-reactive protein,procalcitonin,and cardiac ultrasound index were collected from both groups.t-test,rank sum test,chi-square test,and Fisher’s exact probability method were used for comparative analysis of the data information of the two groups.Univariate logistic regression analysis and forward-backward stepwise regression analysis were used to screen the variables for constructing the nomogram prediction model,and then the"rms package"in R software was used to construct the prediction model based on screened factors regression coefficients,and the discrimination,calibration and clinical utility of the model were evaluated.The original dataset was randomly divided into training and validation sets in the ratio of 6:4,and the CART decision tree prediction model was constructed in the training dataset using the"rpart package"of R software.The constructed decision tree model was evaluated in the training and validation sets,and the importance of each variable in the model was assessed.Results:1.Univariate logistic regression analysis suggested that a total of 14 variables with statistical differences(P<0.05)were screened.The 14 variables were further utilized to forward-backward stepwise regression analysis,and the results suggested that systolic pressure control before dialysis,diabetes,uric acid,total cholesterol,myoglobin,serum albumin,and procalcitonin were the variables finally included in the construction of the nomogram prediction model.Among them,myoglobin(OR=1.033,95%CI=1.008～1.064,P=0.016),serum albumin(OR=0.808,95%CI=0.675～0.938,P=0.010),and procalcitonin(OR=1.920,95%CI=1.302～3.050,P=0.003)were independent risk factors for cardiovascular disease mortality risk in MHD patients(P<0.05).2.The nomogram prediction model was constructed based on the regression coefficients of the final screened 7 variables with an area under the ROC curve-AUC of0.947(95%CI:0.903～0.991)and a cut-off value of 0.3736 when the Youden index was0.770,with a sensitivity of 81.5%and specificity of 95.5%.The joint prediction of the seven variables had better discrimination than each variable individually.Internal validation using the Bootstrap method showed a C-index of 0.919 after correction for fit bias,accuracy of 0.897,and kappa value of 0.634.The Hosmer-Lemeshow goodness of fit test showed:χ~2=0.336,P=0.845(P>0.05).3.The analysis of the clinical utility of the nomogram prediction model suggested that the use of the nomogram model to predict the risk of death due to cardiovascular disease in MHD patients and to decide whether to receive the relevant treatment measures had a greater net effect than the"all-treatment"or"no-treatment"strategy.When the probability threshold reaches 80%,the model predictions were generally consistent with the actual situation.4.The decision tree prediction model was constructed using the CART algorithm,and the results indicated that the total patient population could be divided into four categories(three categories with higher risk of death and one category with lower risk of death)based on three variables:serum albumin,procalcitonin,and myoglobin(cutoffs of30g/L,750g/L,and 3.9ng/m L,respectively).5.The ROC curve was plotted in the training set based on the CART decision tree prediction model with an optimal cutoff value of 0.348,which had a sensitivity of 88.2%,specificity of 97.0%,AUC of 0.933(95%CI:0.851～1.000),and the confusion matrix was constructed using the cutoff value with an accuracy of 0.952(95%CI:0.881～0.987),and the Kappa value was 0.852.In the validation set,the AUC was 0.774(95%CI:0.612～0.936),with a sensitivity of 70.0%and specificity of 84.8%,and the confusion matrix suggested an accuracy of 0.821(95%CI:0.696～0.911)with a Kappa value of 0.474.Conclusions:1.Risk factors for cardiovascular disease death in patients on maintenance hemodialysis were:blood pressure control before dialysis,diabetes,uric acid,total cholesterol,myoglobin,serum albumin,and procalcitonin,with myoglobin,serum albumin,and procalcitonin being independent risk factors for cardiovascular disease death in MHD patients(P<0.05).2.The nomogram prediction model constructed in this study had a high degree of discrimination and calibration,and had a higher predictive ability compared with each individual risk factor,can predict the risk of death due to cardiovascular disease"in MHD patients individually,and the clinical utility analysis suggested that it had a certain net benefit rate and had clinical application value.3.Serum albumin,procalcitonin,and myoglobin were screened as predictors by the CART decision tree prediction model.The total population was classified into four categories according to the cut-off points of the three variables.The prediction model had good prediction ability,differentiation ability,and consistency in both the training and validation sets. |
PDF Full Text Request