Machine learning model SVC predicted 30-day mortality in AMI-CS patients with AUC 0.959; key predictors included BUN, WBC, SBP, lactic acid, and BNP.
Can machine learning models accurately predict 30-day all-cause mortality in patients with acute myocardial infarction complicated by cardiogenic shock?
Machine learning models demonstrate high predictive accuracy for 30-day mortality in patients with AMI-CS, potentially aiding in clinical risk stratification.
Absolute Event Rate: 0% vs 0%
Abstract Background Patients with acute myocardial infarction complicated by cardiogenic shock (AMI-CS) are at high risk of death and current predictive tools are unsatisfactory. Machine learning (ML) is a promising method to predict the outcome in patients with AMI-CS. Purpose The present study aimed to analyze the utility of ML method to predict the outcome in patients with AMI-CS. Methods Using the single-center database of patients diagnosed with AMI-CS, eight ML predictive models were developed. The study endpoint was 30-day all-cause mortality. The predictive performance of the nine ML models was compared, and the importance of these features was quantified using SHapley Additive exPlanations (SHAP) values. The predictive performance was validated by MIMIC-IV database. Results A total of 268 consecutive patients with AMI-CS between 2013 and 2022 were included. The mean age of this cohort was 69 years and 179 (66.5%) were men. The 30-day all-cause mortality was 43.7% (117/268). 15 important features were identified through univariate Logistic regression, Boruta, and LASSO regression. Among the eight ML models based on these features, SVC had highest area under curve AUC (AUC=0.959, 95% CI 0.874-1.000), followed by LMP (AUC=0.951, 95% CI 0.874-1.000), LR (AUC=0.944, 95% CI 0.861-1.000), other ML models had AUC more than 0.850. The SHAP plot revealed that blood urea nitrogen, white blood cell counts, admission systolic blood pressure, admission lactic acid, and B-type natriuretic peptide level were the five most important features for predicting 30-day all-cause mortality. The efficacy of the ML was externally validated by MIMIC-IV database with AUC 0.841 (95% CI 0.784-0.881) with LMP model. Conclusion ML models can identify risk factors for 30-day mortality in patients with AMI-CS. ML established risk factors aid in risk stratification and guide clinical treatment.Features and ROC of ML models Importance of feature with SHAP
Huang et al. (Sat,) reported a other. Machine learning model SVC predicted 30-day mortality in AMI-CS patients with AUC 0.959; key predictors included BUN, WBC, SBP, lactic acid, and BNP.