A classic argument for self-driving cars is that they don’t need to be perfect; just better than humans. That is a pretty low bar, although one that AI-driven cars are still having trouble with. The same goes with ECG interpretations – AI needs not be perfect, just better than human doctors. This begs the question – just how good is the average doctor at reading ECGs?
Not as good as you’d think
A US-led meta-analysis examined 78 studies that assessed the accuracy of physicians’ ECG interpretations in a test setting. The results? Somewhat mediocre. The average non-cardiologist resident got the diagnosis correct only 55.8% of the time. For practising non-cardiologist physicians, the number rose to a still-middling 68.5%. Cardiologists got ECGs right 74.9% of the time, which is far from a perfect score.
ECG interpretation – a dying art?
Difficult to master (and even harder to teach), the area of ECG interpretation has spawned an entire learning industry devoted to the topic. In the modern, fast-paced world of clinical decision-making, it looks like many doctors are content with learning the basics about ECG and forego lengthier, more in-depth analysis.
AI ECG assistants
In the past, we covered AI that could make impossible deductions by looking at ECGs, such as whether a patient has heart failure or even whether they are male or female. AI development teams may have been hesitant to create neural networks whose task would be to antagonize doctors directly. Looking at the results of this meta-analysis, even an imperfect AI ECG assistant could really help doctors wrap their head around ECGs!