Machine learning-enhanced hypoxia-age-shock index for predicting mortality in STEMI

What is it about?

Background Timely identification of high-risk patients with ST-segment elevation myocardial infarction (STEMI) in the emergency department remains a clinical challenge. Although the shock index (SI) and age-shock index (ASI) are commonly used to assess early hemodynamic compromise, they do not account for oxygenation status, which is a critical determinant of patient outcomes. Objectives This study aimed to evaluate the hypoxia-age-shock index Hypoxia Age Shock Index (HASI), a novel oxygenation-adjusted composite index, and compare its predictive performance with SI, ASI, and machine learning (ML) models for in-hospital mortality and emergency intubation. Methods In this retrospective cohort study, 711 adult patients with STEMI undergoing primary percutaneous coronary intervention from 2019 to 2022 were analyzed. HASI was calculated as ASI divided by SpO2. Predictive performance was assessed using area under the receiver operating characteristic curve (area under the curve [AUC]). Five ML models were trained using triage variables and evaluated for accuracy, sensitivity, specificity, and explainability via SHapley Additive exPlanations (SHAP). Results HASI outperformed SI and ASI in predicting mortality (AUC = 0.747) and intubation (AUC = 0.736). ML models, particularly XGBoost and random forest, achieved near-perfect mortality prediction (AUCs ≈ 0.99) but performed modestly for intubation (AUCs < 0.80). SHAP analysis identified SpO2 as the strongest predictor of mortality, whereas age dominated intubation risk. Conclusions HASI provides a simple, interpretable, and effective tool for early triage risk stratification in STEMI. Although ML models enhance mortality prediction, they offer limited improvement for airway events without additional respiratory data. Routine capture of SpO2 and further model refinement may enable precision-guided triage and resource allocation in emergency cardiac care.

Featured Image

Why is it important?

HASI provides a simple, interpretable, and effective tool for early triage risk stratification in STEMI. Although ML models enhance mortality prediction, they offer limited improvement for airway events without additional respiratory data. Routine capture of SpO2 and further model refinement may enable precision-guided triage and resource allocation in emergency cardiac care.