A Meta-analysis Of Risk Factors And Establishment Of Risk Assessment Model Of Hyperkalemia In Patients With Chronic Kidney Disease

Objective:The aim of this study was to screen the possible risk factors associated with hyperkalemia in patients with chronic kidney diseases(CKD)through meta-analysis.Based on a comprehensive list of risk factors for hyperkalemia in patients with CKD,machine learning methods such as decision tree,random forest(RF),Bayesian network(BN),support vector machine(SVM)and K-nearest neighbor(KNN)method were utilized to build a prediction model for the risk of hyperkalemia in the group of patients of our study using cross-sectional data,and the best performing prediction model was screened.In addition,the accuracy of the prediction models constructed with the addition of temporal factor was analyzed to provide more assistance in the prediction,diagnosis and treatment of hyperkalemia in patients with CKD,and to assist physicians in making clinical decision and developing follow-up strategies for personalized management of patients’ hyperkalemia risk.Finally,considering the widespread application of Renin-angiotensin-aldosterone system inhibitor(RAASi)and Sodium glucose cotransporter 2 inhibitor(SGLT2i)drugs in the current renal disease population,we performed a net meta-analysis of the effects of these drugs on hyperkalemia in the specific CKD population,the diabetic nephropathy population(DKD),both alone and in combination,as a reference for the use of the drugs mentioned above in the relevant population.Methods:1.The literatures on factors influencing hyperkalemia in patients with CKD were searched through Zhiwang,Vipshop,Wanfang,Pub Med,China Biomedical Literature Service,Web of Science,Embase,and Cochrane databases,and relevant data were extracted.Analyses were performed according to the heterogeneity of the data using random or fixed effects models,and OR values and 95% CI were used to present the results.2.A retrospective analysis was performed on 607 patients with CKD admitted to our nephrology department from July 2020 to December 2020.General information and laboratory test results of the patients were collected.Lasso regression was used for the factors that have a greater influence on hyperkalemia,and on this basis,models were constructed applied decision trees,RF,BN,SVM and KNN method,respectively,and the sensitivity,specificity and accuracy of each model and the hyperkalemia-related logistic regression model constructed by our nephrology department in the previous period were evaluated,and the area under the curve was plotted to evaluate the overall effectiveness of each model.3.Patients in the first part of the data were followed up,the data from the follow-up were collected,and the prediction model of hyperkalemia was constructed using random survival forest(RSF),and the model was evaluated comprehensively to assess its performance.4.Pub Med,Web of Science,Embase,the Cochrane Central Register of Controlled Trials,and Clinical Trials.gov were searched for the information on the risk of hyperkalemia in patients with DKD using SGLT2 i and/or RAASi.The literatures were searched and relevant data were extracted.Network meta-analysis was used to analyze single-drug use and drug combinations,and ORs and 95% CIs were used to present the results.Results:1.After screening,a total of 22 papers were included in the meta-analysis,in which the risk factors related to the development of hyperkalemia in CKD patients were listed as follow: ACEI/ARB,smoking,gender,tumor,CKD stage,diabetes mellitus,metabolic acidosis(hypocapnia),and peripheral vascular diseases.2.Based on the cross-sectional information we collected and Lasso regression analysis,corresponding hyperkalemia-related models of decision tree,RF,BN,SVM and KNN method were constructed,respectively.The results showed that the decision tree model had an AUC of 0.834,sensitivity: 0.877,specificity: 0.694,and accuracy: 0.718;the BN model had an AUC of 0.857,sensitivity: 0.827,specificity: 0.747,and accuracy: 0.758;the KNN model had an AUC of 0.898,sensitivity: 0.802,specificity.0.831,accuracy: 0.827;SVM model had an AUC of 0.811,sensitivity: 0.827,specificity: 0.795,accuracy: 0.799;Our nephrology department’s pre-constructed logistic regression model on hyperkalemia had an AUC of 0.793,sensitivity: 0.914,specificity: 0.327,accuracy: 0.705.RF model had an AUC of 0.905,sensitivity: 0.827,specificity: 0.833,and accuracy: 0.832.4.A RSF model was constructed based on the follow-up data.The results showed that the effects of CKD stage,starting hyperkalemia,acidosis,and herbal medicine on the occurrence of the endpoint event(hyperkalemia)were statistically different between the two groups.The RSFmodel constructed in this study had relatively high accuracy.5.Based on inclusion and exclusion criteria,this mesh meta ultimately included 27 studies with a total of 43589 participants.The use of Mineralocorticoid receptor antagonist(MRA)in addition to the application of angiotensin-converting enzyme inhibitors(ACEI)/angiotensin receptor blockers(ARB)significantly increased the incidence of hyperkalemia.Further subgroup analyses of different generations of MRA found that spironolactone had the strongest positive effect on hyperkalemia in DKD patients(OR 9.48,95% CI: 3.53 ～ 25.46),even higher than that of ACEI combined with ARB(OR 2.89,95%CI: 1.26 ～ 6.63).In addition,SGLT2 i significantly reduced the risk of hyperkalemia in DKD patients in both single-use and combination regimens.SGLT2 i significantly reduced the incidence of hyperkalemia in DKD patients compared to ACEI or ARB(OR 0.33,95% CI:0.12-0.91;OR 0.28,95% CI: 0.13-0.61).In addition,the combination of SGLT2 i,MRA and ACEI/ARB significantly reduced the incidence of hyperkalemia in DKD patients compared with MRA combined with ACEI/ARB(OR 0.31,95% CI 0.16-0.62).Conclusions:The model constructed using RF and RSF based on the risk factors for hyperkalemia in patients with CKD screened by meta-analysis was highly accurate and could be used to predict the risk of hyperkalemia in relevant populations.In a specific CKD population,the DKD population,where there was a potential risk of elevated blood potassium,MRA additionally increased the risk of hyperkalemia,while SGLT2 i showed the opposite effect and even could reverse the elevated blood potassium caused by the combination regimens including MRA.