Machine learning model for in-hospital mortality prediction in myocardial infarction and heart failure patients post-PCI
- VernacularTitle:心肌梗死并心衰患者PCI术后院内死亡的机器学习预测模型的构建
- Author:
Huasheng LV
1
;
Fengyu SUN
;
Teng YUAN
;
Haoliang SHEN
;
LAZAIYI·BAHETI
;
Wei JI
;
You CHEN
Author Information
- Publication Type:Journal Article
- Keywords: machine learning; myocardial infarction(MI); heart failure(HF); percutaneous coronary intervention(PCI); in-hospital mortality
- From: Journal of Xi'an Jiaotong University(Medical Sciences) 2025;46(3):393-401
- CountryChina
- Language:Chinese
- Abstract: Objective To develop and validate a machine learning-based predictive model to assess the in-hospital mortality risk of patients with myocardial infarction(MI)complicated by heart failure(HF)undergoing percutaneous coronary intervention(PCI).Methods This retrospective study analyzed MI patients with HF who underwent PCI at The First Affiliated Hospital of Xinjiang Medical University from January 2019 to January 2023.Patient data,including demographic characteristics,vital signs,laboratory test results,imaging parameters and medication use,were collected and randomly divided into a training set(70％)and a validation set(30％).The extreme gradient boosting(XGBoost)model was used to identify variables significantly associated with in-hospital mortality,and the Shapley additive explanations(SHAP)model was applied to assess feature importance.A predictive model was then constructed using univariate and multivariate Logistic regression analyses.Model performance was evaluated using receiver operating characteristic(ROC)curves,area under the curve(AUC)values,calibration curves,and decision curve analysis.Finally,a nomogram was developed for intuitive risk assessment.Results A total of 1 214 MI patients with HF were included in the study,with a median age of 64 years.The in-hospital mortality rate was 7.41％(90 deaths).XGBoost feature selection identified ten key predictive variables:age,myoglobin,albumin,fasting blood glucose,N-terminal pro-B-type natriuretic peptide(NT-proBNP),diabetes mellitus,creatinine,cystatin C,procalcitonin,and left ventricular ejection fraction.Based on these variables,a Logistic regression model was developed,with seven final predictors:age,diabetes mellitus,creatinine,fasting blood glucose,cystatin C,NT-proBNP,and albumin.The model demonstrated high predictive accuracy,with AUC value of 0.869(95％CI:0.84-0.89)in the training set and 0.827(95％CI:0.79-0.85)in the validation set.The calibration curve indicated that the predicted probabilities were consistent with the actual observed outcomes,and decision curve analysis showed that the model had a high net benefit across various decision thresholds.Conclusion This study developed a machine learning-based predictive model incorporating Logistic regression to assess the in-hospital mortality risk of MI patients with HF undergoing PCI.The model demonstrated high predictive performance and clinical utility.The nomogram derived from this model provides an intuitive tool for individualized risk assessment,aiding clinicians in the early identification of high-risk patients,optimizing intervention strategies,and improving patient outcomes.
