Chris Bodimeade
East Surrey Hospital, Redhill, United Kingdom
Abstract
As part of any surgery a detailed discharge letter is required to give information to the patient and their GP. Artificial intelligence (AI) has the capability to fulfil this repetitive task, however is not currently embraced within the health system. We investigated the potential of this technology to generate discharge letters that were non inferior to those written by a doctor for a common elective surgical procedure.
Using an online AI programme “Chat GPT” we generated two discharge letters for hypothetical elective tonsillectomy patients. We then asked the ENT department of a district general hospital to blindly compare the AI letters to two anonymised discharge letters for real patients. The participants were asked to rate the quality of the medical information, the ease of reading and the length of each of these letters. As well as guessing which letter was written by AI.
We had 19 respondents to the survey - 7 consultants, 5 registrars and 7 SHOs. Our results showed the AI letters were on average thought to contain better medical information and were easier to read than the real discharge letters. The difference was statistically significant (p=0.001) in the latter case. The reviewers had a 34% sensitivity in correctly identifying the letters written by AI. We conclude that AI writes better discharge letters than humans for elective tonsillectomies and these are indistinguishable from real discharge letters according to ENT professionals. There is therefore great potential to expand the use of AI and free doctors to complete other tasks.