Back

Assessing clinical acuity in the Emergency Department using the GPT-3.5 Artificial Intelligence Model

Williams, C. Y. K.; Zack, T.; Miao, B. Y.; Sushil, M.; Wang, M.; Butte, A. J.

2023-08-13 health informatics
10.1101/2023.08.09.23293795 medRxiv
Show abstract

This paper evaluates the performance of the Chat Generative Pre-trained Transformer (ChatGPT; GPT-3.5) in accurately identifying higher acuity patients in a real-world clinical context. Using a dataset of 10,000 pairs of patient Emergency Department (ED) visits with varying acuity levels, we demonstrate that GPT-3.5 can successfully determine the patient with higher acuity based on clinical history sections extracted from ED physician notes. The model achieves an accuracy of 84% and an F1 score of 0.83, with improved performance for more disparate acuity scores. Among the 500 pair subsample that was also manually classified by a resident physician, GPT-3.5 achieved similar performance (Accuracy = 0.84; F1 score = 0.85) compared to the physician (Accuracy = 0.86, F1 score = 0.87). Our results suggest that, in real-world settings, GPT-3.5 can perform comparably to physicians on the clinical reasoning task of ED acuity determination.

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.