Construction of prediction model and risk stratification for heart failure complications in acute myocardial infarction patients based on big medical test data

CAO Wei; YANG Zhi-ning; JI Bing; WANG Li-juan

doi:10.3969/j.issn.1672-5301.2025.09.005

Abstract：Objective To mine big medical test data through machine learning algorithms and construct a prediction model and risk stratification for heart failure(HF)complications in acute myocardial infarction(AMI)patients.Methods Clinical data of 1,500 AMI patients admitted to Shanxi Cardiovascular Hospital from October 2020 to October 2024 were collected through the electronic medical record system.The patients were divided into the training set(n=1050)and the validation set(n=450)at a ratio of 7∶3.According to whether the patients had HF complications,the patients in the training group were further divided into the heart failure group(n=425)and the non-heart failure group(n=625).Logistic regression,support vector machine and random forest training models were used to screen the risk variables for HF complications in AMI.The best subset of each algorithm was selected through cross-validation,and the intersection was taken as the final input variable.The results of the model intersection were included to construct a model for predicting HF complications in AMI patients.The intersection results of the models were included to construct a model for predicting HF in AMI patients,and the predictive efficacy of the model was tested.Results There were statistically significant differences between the heart failure group and the non-heart failure group in terms of age,hypertension,Killip cardiac function classification,soluble growth stimulated gene 2 protein(sST-2),high-sensitivity cardiac troponin I(hs-cTnI),N-terminal pro-brain natriuretic peptide(NT-proBNP),blood urea nitrogen(BUN),serum creatinine(Crea),high-density lipoprotein cholesterol(HDL-C)and low-density lipoprotein cholesterol(LDL-C)levels(P<0.05).Logistic regression,support vector machine and random forest machine learning algorithms were used to analyze the 10 statistically significant risk variables mentioned above.The intersection variables screened by the three machine algorithms were Killip cardiac function classification≥gradeⅡ,sST-2,hs-cTnI,NT-proBNP,HDL-C and LDL-C.Based on these results,a nomogram model for predicting HF complications in AMI patients was constructed.The areas under the ROC curves(AUC)of the model in the training set and the validation set were 0.837 and 0.804 respectively(95%CI 0.751-0.960 and 0.729-0.883).The Hosmer-Lemeshow test curves showed that the calibration curves of the model in the training set and the validation set fitted well with the ideal curve(χ2=1.443 and 1.234,P=0.276 and 0.203).Conclusion The prediction model for HF complications in AMI patients constructed based on big medical test data has good predictive efficiency and can provide a reference for clinical practice.