Construction of a Predictive Model of Risk Factors for Kidney Replacement Therapy after Heart Transplantation

Background: Acute kidney injury (AKI) is one of the common complications after heart transplantation (HT), which seriously affects the quality of life and survival rate of patients after heart transplantation. Severe acute kidney injury requires renal replacement therapy (RRT), and early identification and intervention of risk factors can improve the prognosis, survival and quality of life of heart transplant patients. Methods: The study enrolled 288 heart failure patients who underwent HT at Guangdong Provincial People's Hospital (Jan 2010-Feb 2023) and Zhongshan City People's Hospital (Jul 2017-Mar 2023).Risk factors were screened using machine learning and one-way analysis, the optimal coefficients were determined using Lasso regression model combined with cross-validation, non-zero coefficient variables were selected to construct a column-line graph prediction model, and prediction equations were established by multifactor logistic regression. The differentiation of the prediction model was assessed by the Area under Curve , the accuracy of the model was tested by the calibration curve, and the clinical applicability was analyzed by the Decision Curve Analysis. Results: The incidence of severe acute kidney injury requiring RRT after HT was 28.13%. Six risk factors were identified and integrated into a column-line diagram: history of intra-aortic balloon counterpulsation use, preoperative serum albumin, preoperative blood creatinine, intraoperative red blood cell transfusion volume, duration of HT surgery, and intraoperative mechanical circulatory support. The area under the ROC curve for the predictive model was 0.765, and the calibration curve for the column-line diagram model was close to the ideal curve. The clinical decision curve showed that the column-line diagram prediction model performed optimally in the 0.1-0.6 probability interval. Internal validation was performed using Bootstrap method with 500 repetitive samples, yielding a mean AUC of 0.757. Conclusion: In this study, we analyzed the risk factors for severe acute kidney injury requiring RRT after HT, and established a column-line graph prediction model and its equations. Internal validation proved that the model was stable and effective in predicting the risk of needing RRT after HT, which can help early intervention and improve prognosis.