Machine-Learning for Phenotyping and Prognostication of Myocardial Infarction and Injury in Suspected Acute Coronary Syndrome
Authors' conclusion
Machine learning models can digitally phenotype suspected ACS patients at index presentation and predict subsequent events within 30 days.
Study context
- Population
- Suspected acute coronary syndrome (ACS) patients presenting to South Australian hospitals (training 6, testing 26 hospitals)training 6,722
- Comparison
- XGB vs deep-learning (DL) diagnostic models; XGB event prediction model
- Outcome
- Phenotyping of myocardial injury and infarction (Fourth UDMI) and prediction of 30-day death or MI
- Follow-up
- 30 days
Methodological detailsmachine learning, training and testing data sets, multicentre, Index presentations and 30-day death and myocardial infarction (MI) adjudicated using the Fourth Universal Definition of MI
- Machine learning
- Training and testing data sets
- Multicentre
- Index presentations and 30-day death and MI adjudicated using the Fourth Universal Definition of MI
Performance
As reported in the abstract
Diagnostic prediction AUC
| Test or model | AUC, acute myocardial injury pattern vs no injury or chronic injury | AUC, type 1 MI vs type 2 MI or acute nonischemic myocardial injury |
|---|---|---|
| XGB | 99.2% ± 0.1% | 95.5% ± 0.2% |
| DL | 98.8% ± 0.2% | 94.6% ± 0.9% |
AUC, acute myocardial injury pattern vs no injury or chronic injury
XGB99.2% ± 0.1%
DL98.8% ± 0.2%
AUC, type 1 MI vs type 2 MI or acute nonischemic myocardial injury
XGB95.5% ± 0.2%
DL94.6% ± 0.9%
Further findings
30-day death/MI event prediction model (XGB)
- AUC for 30-day death/MI
- 88.5% ± 0.5%
- Training n = 6,722; testing n = 8,869
Original publication
Machine-Learning for Phenotyping and Prognostication of Myocardial Infarction and Injury in Suspected Acute Coronary Syndrome. JACC Adv. Published 19 June 2024. doi:10.
Summarised from the published abstract; every number is checked against it