Less-trained medical staff — likely the first to lose to AI.
Nature Medicine (2025) study finds that human ECG licensed technicians vs. AI had a 14x higher risk of missing(!) critical arrhythmias in analyzing ECG data:
Key findings:
- Techs missed 44.3 critical arrhythmias per 1,000 patients
- AI missed only 3.2 per 1,000 patients
- Techs had 14.1x higher relative risk of missed diagnosis vs. AI
The important tradeoff:
- AI also had 2.4x higher false positive rate than technicians, i.e., AI identified more high-risk diagnosis when there was truly none.
- Note: This is typically considered an acceptable risk, particularly since these cases would get escalated & reviewed again, anyways.
The numbers:
- 14,606…ECGs
- 2,236…Critical arrhythmia events
- 167…Certified ECG technicians
- 17…Cardiologist consensus panels
- 1… AI model
AI won.
With AI models aggressively improving, cardiology (and many other fields of medicine) will need to aggressively move towards:
AI-augmented, human-centered care.
And the sooner we embrace this reality, the better we can re-train, re-deploy and re-imagine care by all levels of our healthcare teams — and ideally avoid the risk of ruthless displacement.
Written February 11, 2025.