A Machine Learning Model for Predicting In-Hospital Mortality in Chinese Patients With ST-Segment Elevation
Jingang Yang1, Yingxue Li2, Xiang Li2
1State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China.
Journal of Medical Internet Research
|July 30, 2024
Summary
This study developed an accurate and explainable machine learning model to predict in-hospital mortality in ST-segment elevation myocardial infarction (STEMI) patients. The new model outperforms traditional methods, offering practical clinical utility.
Area of Science:
- Cardiology
- Medical Informatics
- Machine Learning
Background:
- Traditional statistical models for predicting in-hospital mortality in ST-segment elevation myocardial infarction (STEMI) are less accurate and less transparent than machine learning (ML) models.
- ML models often require numerous input variables and lack transparency, limiting their clinical applicability.
Purpose of the Study:
- To develop a precise, explainable, and flexible ML model for predicting in-hospital mortality in STEMI patients.
- To improve upon the accuracy and usability of existing risk prediction tools for STEMI.
Main Methods:
- Utilized Extreme Gradient Boosting (XGBoost) on data from the China Acute Myocardial Infarction (CAMI) registry and the China Patient-Centered Evaluative Assessment of Cardiac Events (PEACE) registry.
- Validated the XGBoost model using 5-fold cross-validation and an independent cohort, employing the Shapley Additive Explanations (SHAP) approach for model interpretability.
- Recruited 18,744 patients from CAMI and 12,018 from PEACE, with model derivation on 9,616 CAMI patients and validation on 9,125 CAMI patients and the PEACE cohort.
Main Results:
- The XGBoost model demonstrated high accuracy in predicting in-hospital mortality, with an Area Under the Curve (AUC) of 0.896 on the CAMI validation set, significantly outperforming GRACE and TIMI models.
- Even with only 10 variables, the model achieved an AUC of 0.840 on the China PEACE validation set, surpassing GRACE and TIMI scores.
- Identified key predictors of mortality including age, left ventricular ejection fraction, Killip class, heart rate, creatinine, blood glucose, white blood cell count, and use of ACEIs/ARBs. Discovered novel nonlinear relationships, such as a U-shape pattern for HDL-C.
Conclusions:
- The developed ML risk prediction model is highly accurate for predicting in-hospital mortality in STEMI patients.
- The model's explainability and flexibility enhance its clinical utility, aiding in patient management decisions.
- This approach offers a significant advancement over traditional risk scores, providing a more precise and interpretable tool for clinicians.
