JMIR Publications

ABSTRACT

Aims: Machine learning (ML) risk prediction models, although much more accurate than traditional statistical methods, were inconvenient to use in clinical practice due to its non-transparency and requirements of a big number of input variables. We aimed to develop a precise, explainable and flexible ML model to predict the risk of in-hospital mortality in patients with ST-segment elevation myocardial infarction (STEMI). Methods &

Results:

The study recruited 18, 744 patients enrolled in the China Acute Myocardial Infarction (CAMI) registry and 12, 018 patients in the China PEACE-Retrospective Acute Myocardial Infarction Study. The Extreme Gradient Boosting (XGBoost) was derived on the 9616 patients in CAMI registry (Year 2014, 89 variables) with 5-fold cross validation and validated on both the 9125 patients in CAMI registry (Year 2015, 89 variables) and the independent China PEACE cohort (10 variables). The SHapley Additive exPlanations (SHAP) approach was employed to interpret the complex relationships embedded in the proposed model. The Area Under the Curve (AUC) on the CAMI validation set was 0.896 (95% CI: 0.884-0.909) on the CAMI validation set, significantly higher than the previous models (GRACE: 0.809, 95%CI: 0.790-0.828; TIMI: 0.782 ,95% CI: 0.763-0.800). Although the China PEACE validation set only has ten available variables, the AUC on the China PEACE validation set still reached 0.840 (0.829-0.852), which improved substantially to the GRACE (0.762, 95% CI: 0.748-0.776) and TIMI (0.789, 95% CI:0.776-0.803]. Several novel and non-linear relationships such as the U-shape pattern of HDL-C were discovered between the patients’ characteristics and the in-hospital mortality. Conclusion: The proposed ML risk prediction model was highly accurate to predict the in-hospital mortality. The flexible and explainable characteristics makes the model convenient to use in clinical practice and could help to guide the patient management.