基于机器学习算法构建急性心肌炎患者发生暴发性心肌炎的预测模型及其价值分析OA

Objective To explore the influencing factors for fulminant myocarditis(FM)in patients with acute myocarditis(AM),and to analyze the predictive value of prediction models based on machine learning algorithms for FM in AM patients.Methods A total of 200 AM patients admitted to Quzhou People's Hospital of Zhejiang Province from March 2020 to March 2024 were consecutively and retrospectively selected.According to the occurrence of FM,patients were divided into an occurrence group(100 cases)and a non-occurrence group(100 cases).The influencing factors of FM in AM patients were analyzed.Machine learning algorithms of Logistic regression,decision classification and regression tree(DCRT),and back propagation neural network(BPNN)were used to construct prediction models of FM in AM patients.Receiver operating characteristic(ROC)curve was used to compare the predictive value of the models constructed by the three methods for the occurrence of FM in AM patients.Results Univariate analysis showed statistically significant differences between the two groups in C-reactive protein(CRP),cardiac troponin I(cTnI),creatinine,albumin,and the incidence of sinus tachycardia,ventricular tachycardia/ventricular fibrillation(VT/VF),third-degree atrioventricular block,sinus arrest,QTC interval prolongation,and left ventricular ejection fraction(LVEF)(t/χ2=3.806,3.795,3.571,2.046,4.196,11.060,3.907,8.865,10.526,2.159;all P<0.05).Multivariate Logistic regression analysis identified the following as risk factors for FM in AM patients:elevated levels of CRP,cTnI,and creatinine,presence of VT/VF and sinus arrest,and QTC interval prolongation(OR=1.422,47.154,1.033,30.891,34.478,3.229;all P<0.05).The prediction model constructed using the DCRT method showed that the levels of cTnI,creatinine,LVEF,CRP,and albumin,and the incidence of sinus tachycardia and VT/VF were classification factors for FM in AM patients.According to the standardized importance of independent variables in the BPNN model,the top five influencing factors for FM in AM patients were CRP level(100.00%),presence of VT/VF(89.10%),creatinine level(81.90%),presence of sinus arrest(81.20%),and cTnI level(59.40%).The models constructed by the three machine learning algorithms all achieved an area under the curve(AUC)greater than 0.800,demonstrating good predictive accuracy(all P<0.05).Among them,the DCRT model had the best performance in predicting FM in AM patients.ROC curve showed an AUC of 0.880,a sensitivity of 85.00%,and a specificity of 75.00%.Conclusion The prediction models of FM in AM patients based on machine learning algorithms have good predictive efficiency.Among them,the DCRT model demonstrates the best diagnostic efficacy,which can be further applied to verify the efficacy of the prediction model.