Current Epidemiology Of Bleeding Complication Of Percutaneous Renal Biopsy And Prediction Model Development

Objectives The population and the indications of percutaneous renal biopsy have changed in recent years.This article aimed to study the current epidemiology and risk factors of bleeding complication of percutaneous renal biopsy and develop prediction model.Methods Retrospectively analyzed the clinical and pathological data of adult patients who underwent ultrasound-guided percutaneous renal biopsy at the Department of Nephrology,Guangdong Provincial People's Hospital from June 13,2017 to December 27,2018.Based on the need to increase the amount and/or frequency of hemostatic drugs/infusion of fresh frozen plasma,severe bleeding after renal biopsy was defined as infusion of red blood cells,angiography/interventional therapy,renal surgery or death.Defined bleeding that requires clinical intervention after percutaneous renal biopsy as clinical bleeding,including the need to increase the amount or frequency of hemostatic drugs,infusion of fresh frozen plasma,infusion of red blood cells,angiography/interventional therapy,kidney surgery,or death.Analyzed the current situation and risk factors through the collected patient data,and excluded patients with missing data before constructing model.LAASO regression was used as screening variables in this study,because there were few positive events and only 16 cases of severe bleeding.According to the principle that the number of positive events:variables was at least 15-20:1,Logistic regression cannot be used for variables screening.The variables were selected by LASSO regression and combined with Logistic regression to construct the model.Besides,explored whether height-corrected kidney length can increase the predictive power of the model.The value of the model was evaluated through Bootstrap,receiver operating characteristic curve and calibration curve.Results The retrospective study collected data from 1036 patients.Among them,16 cases(1.54%)had severe bleeding,39 cases(3.76%)had clinical bleeding,and none required kidney surgery or death.Absolute kidney length(OR value 0.540,95%CI 0.310-0.940,P=0.029),unclear renal ultrasound cortical medulla demarcation line(OR value 4.323,95%CI 1.490-12.542,P=0.007),serum creatinine(OR value 1.005,95%CI 1.002-1.007,P=0.000),eGFR(OR value 0.980,95%CI 0.964-0.996,P=0.014),hemoglobin(OR value 0.947,95%CI 0.921-0.975,P=0.000)were the risk factors for severe bleeding after biopsy.The data of 1036 patients were analyzed retrospectively,109 patients with missing data were excluded,and a total of 927 renal biopsy patients were used to establish model.The absolute length of the kidney was used as an evaluation indicator of renal atrophy.Three selected variables,hemoglobin,unclear medulla of renal ultrasound,and absolute length of kidney were obtained by LASSO regression.Logistic regression analysis was performed to construct a prediction model of severe bleeding.The area under the ROC curve of the prediction model was 0.797(95%CI,0.682-0.913).After 1000 internal verifications of bootstrap iterations,corrected area under the ROC curve was 0.774,which showed a good discrimination of the model.The predictive model was used to draw the calibration curve,and Brier score was 0.016.The predicted results were consistent with the actual results,showing a good degree of calibration of the model.Replacing the absolute length of the kidney,the relative length of the kidney was used as an evaluation indicator of renal atrophy.The three selected variables were obtained through LASSO regression,hemoglobin,unclear dermal medulla of renal ultrasound,and relative length of kidney.The area under the ROC curve of this prediction model was 0.801(95%CI,0.689-0.912).After 1,000 internal bootstrap iterations,the corrected area under the ROC curve was 0.780.Using the prediction model to draw the calibration curve,and the Brier score was 0.016,showing that the prediction result was consistent with the actual result.The performance of the two models was similar,based on the principle of easy clinical operation,and the prediction model containing the absolute length of the kidney without height correction was selected to form a visual Nomogram.Besides,four selected variables were obtained by LASSO regression,including hemoglobin,eGFR<30ml/min/1.73m2,systolic blood pressure and unclear renal ultrasound dermal medulla boundary.Logistic regression analysis was performed to construct a clinical bleeding prediction model.The area under the ROC curve of the prediction model was 0.719(95%CI,0.644-0.794).After 1000 internal verifications of bootstrap iterations,the corrected area under the ROC curve was 0.699;the calibration curve was drawn using the prediction model and its Brier score was 0.039,The predicted result was consistent with the actual result;showing that the model has a certain degree of differentiation and calibration.Visualized the prediction model and drew Nomogram.Conclusion In this study,the incidence of bleeding complications was low,of which 16 cases(1.54%)had severe bleeding,39 cases(3.76%)had clinical bleeding,and none required kidney surgery or death.Absolute length of kidney,unclear boundary of medulla,creatinine,eGFR and hemoglobin were risk factors for severe bleeding after renal biopsy.This study successfully established predictive models of bleeding risk after percutaneous renal puncture and prediction models had certain clinical value.Kidney length corrected for height can only slightly increase the predictive power of the model.