Study evaluates potential of GPT-4 in prioritizing medical waiting lists
Click Here to Manage Email Alerts
A study showed moderate agreement between GPT-4 and retina specialists in assessing need for treatment and urgency of treatment in patients with diabetic retinopathy, according to a poster presentation at Real World Ophthalmology.
Because diabetic retinopathy (DR) cases are on the rise and patient wait times can be long, language model (LM)-based patient prioritization tools (PPTs) could help support the prioritization decision process, Jericho Morcilla and co-authors wrote.
Image: Adobe Stock
As a proof of principle for a broader applicability of LM-based PPTs, the study compared the patient priority ratings (PPR) and treatment urgency ratings (TUR) performed by GPT-4 and two board-certified retina specialists in four sets of anonymized DR patient profiles sorted by DR severity, central subfield thickness (CST), modified best corrected visual acuity and complete randomization.
There was moderate agreement between GPT-4 and the retina specialists in prioritizing TUR and PPR, which improved when the specialists’ decisions were averaged. The highest agreement occurred when the data set was organized by CST, showing that both GPT-4 and the specialists consider CST a key factor in their decisions. In cases with intentionally modified BCVA, however, only the human evaluators were able to detect the unrealistic scenarios, showing that GPT-4 does not have critical thinking abilities.
The inability of GPT-4 to acknowledge potential clinical improbabilities is, according to the authors, “a significant drawback” and suggests that “human evaluators may be more appropriate to guide management prioritization related to extraordinary patient cases.”
Collapse
Morcilla J, et al. A potential role for AI: Evaluating ChatGPT’s efficacy in prioritizing medical waiting lists. Presented at: Real World Ophthalmology; Sept. 30, 2023 (virtual meeting).
Read more about
Click Here to Manage Email Alerts