A University of Pittsburgh-led team developed an ECG-based AI model capable of detecting occlusion myocardial infarction, showing that it improves both OMI detection and assessments, and finding that it has real clinical potential.
Up to 35% of non-STEMI patients have total coronary occlusion, also known as occlusion myocardial infarction (OMI). These patients require emergency catheterization, but ECG assessment challenges often lead to diagnostic and treatment delays, and greater mortality risks.
To address these OMI challenges, the researchers developed an AI model using 12-lead ECGs from 4,026 consecutive patients with chest pain, all sourced from three Pittsburgh-area hospitals.
To externally validate the model, they analyzed ECGs from 3,287 patients from two other hospital systems, finding that their AI-based model…
- Classified patients with and without OMI with an 0.87 AUROC, well above assessments from experienced clinicians (0.80) and a commercial ECG algorithm (0.75).
- Correctly reclassified one in three patients with chest pain versus HEART score classifications, flagging 66% more patients as “low risk” for OMI and 50% fewer patients as “intermediate risk” (with similar or better false-negative rates)
- Detected patients with ACS more accurately than both a commercial system and practicing clinicians (AUROCs: 0.79 vs. 0.68 & 0.72)
The study’s authors and online commenters were bullish about these results, with the authors suggesting that it could help EMS and ED personnel improve OMI detection and care decisions, while reducing unnecessary care among low-risk patients.
Unlike many AI academic research initiatives, this ECG model appears to be destined for clinical use. The team has already partnered with the City of Pittsburgh Bureau of Emergency Medical Services to plan for future deployment, and is developing a cloud-based system to analyze incoming ECGs and route assessments to care teams.
The Takeaway
Although many AI insiders will want to wait to see how this model operates in the real world before predicting its clinical impact, this study suggests that it might be able to address current challenges with OMI detection, helping to improve care decisions and potentially outcomes. It’s also a great example of how AI could expand the value of ECG, in addition to the imaging modalities that are more commonly associated with AI-based disease detection and assessments.
