Coronary disease & ACSDiagnostic or prognostic study
Machine Learning for Myocardial Infarction Compared With Guideline-Recommended Diagnostic Pathways
- Diagnostic accuracy study
- multicentre (12 sites, 5 countries)
Authors' conclusion
CoDE-ACS performs consistently irrespective of the timing of serial cardiac troponin measurement, identifying more patients as low probability with comparable performance to guideline-recommended pathways for MI.
Study context
- Population
- Patients with possible myocardial infarction (MI) without ST-segment elevation, serial high-sensitivity cardiac troponin I at 0, 1 and 2 hours575 (14%) type 1 MI
- Participants
- 4,105 patients
- Comparison
- Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome (CoDE-ACS) machine learning model vs guideline-recommended European Society of Cardiology (ESC) 0/1-hour, ESC 0/2-hour and High-STEACS pathways
- Outcome
- Diagnostic performance for index type 1 MI
Performance
As reported in the abstract
Type 1 MI · after serial measurements
| Test or model | Ruled out (low probability) at 1 or 2 hours | NPV | Ruled in (high probability) at 1 or 2 hours | PPV at 1 hour | PPV at 2 hours | Intermediate probability at 1 or 2 hours | Ruled out (low risk) | Ruled in (high risk) | PPV | Further observation |
|---|---|---|---|---|---|---|---|---|---|---|
| CoDE-ACS | 65% or 68% | 99.7% (99.5%-99.9%) | 19% or 18% | 64.9% (63.5%-66.4%) | 68.8% (67.3%-70.1%) | 16% or 14% | ||||
| ESC 0/1-hour | 100% (99.9%-100%) | 49% | 20% | 61.5% (60.0%-63.0%) | 31% | |||||
| ESC 0/2-hour | 100% (99.9%-100%) | 53% | 19% | 65.8% (64.3%-67.2%) | 28% | |||||
| High-STEACS | 99.7% (99.5%-99.8%) | 71% | 29% | 48.3% (46.8%-49.8%) | 0% |
Further findings
| Outcome | CoDE-ACS | ESC 0/1-hour | ESC 0/2-hour | High-STEACS |
|---|---|---|---|---|
| Type 1 MI at presentation (single troponin) | ||||
| Ruled out (low probability) | 56% | Not reported in the abstract | Not reported in the abstract | Not reported in the abstract |
| NPV | 99.7% (95% CI, 99.5%-99.9%) | Not reported in the abstract | Not reported in the abstract | Not reported in the abstract |
| Sensitivity | 99.0% (98.6%-99.2%) | Not reported in the abstract | Not reported in the abstract | Not reported in the abstract |
| Ruled out | Not reported in the abstract | Not reported in the abstract | Not reported in the abstract | 35% |
- Type 1 MI at presentation (single troponin): ESC 0-hour Ruled out 25%
Original publication
Machine Learning for Myocardial Infarction Compared With Guideline-Recommended Diagnostic Pathways. Circulation. Published 12 February 2024. doi:10.1161/circulationaha.123.066917
Summarised from the published abstract; every number is checked against it