The Early Mortality Risk Prediction Model For Neonates With Congenital Heart Disease After Surgery Based On Machine Learning

Objective:To explore and establish the early mortality(< 30 days)risk prediction model for neonates with congenital heart disease after surgery.Methods:The medical records of neonates who underwent congenital heart disease surgery in The Second Xiangya Hospital of Central South University from January 2012 to December 2021 were retrospectively collected.According to the exclusion and inclusion criteria,171 neonates were included and divided into survival group(n=142)and death group(n=29)according to the early postoperative survival.The preoperative risk factors associated with early postoperative death were screened by univariate analysis and multivariate regression analysis.The approach of machine learning with R language was used to establish the Logistic regression,LASSO regression and decision tree models of neonatal congenital heart disease after operation,and the prediction effect of these models was compared according to the Area Under Curve(AUC).Results:The early mortality rate of neonatal congenital heart disease in our hospital was 16.96%.According to the univariate analysis between the two groups,20 predictors were screened out to be statistically different:Height,weight,surface area,Quetelet Index(QI),Polock Index(PI),Preoperative percutaneous arterial oxygen saturation,1 minute Apgar score,10 minutes Apgar score,fetal heart defects classification layered long-term prognosis risk score,fetal heart defects classification layered long-term prognosis risk grades,serum albumin,C-reactive protein,calcitonin,mean pulmonary artery pressure,Emergency surgery or not,external surgery or not,American Society of Anesthesiologists Physical Status(ASA-PS)grading,Aristotle’s Basic Score,Aristotle’s Basic Grade,congenital heart disease surgery complexity and risk grading.The result showed that the LASSO regression model has the highest AUC(AUC=0.8515),and its prediction ability is better than Logistics regression(AUC=0.7327)and decision tree model(AUC=0.8393)in this study.Conclusion:All of the three early mortality risk prediction models for neonates with congenital heart disease after surgery established in this study have good predictive efficacy,among which LASSO regression model has the highest predictive efficacy.