TPT 7.16 Can ChatGPT Accurately Mimic a Clinician When Analysing ECGs?

This is the authors' abstract. We don't add key points for this paper.

Aims This study aimed to compare clinicians’ interpretations of ECGs to that of ChatGPT to assess for potential automation. Methods In a retrospective study, 61 twelve lead ECGs arising from surgical patients were analysed by a clinician before being interpreted by ChatGPT. Both evaluations were then scrutinised by an independent clinician for verification. Results Within the sample, 40 ECGs had documentation by the clinicians. ChatGPT was found to have formulated the same diagnosis in 7.5% of cases when compared to that of the clinicians. For the total sample of 61 ECGs, the independent clinicians agreed with ChatGPT 10% of the time. Conclusion At the moment, ChatGPT cannot mimic a clinician when analysing ECGs. Not only it often interpreted the ECGs inaccurately, but it also often over-diagnosed normal ECGs to have ischemia features.