Photo Credit: Ipopba

The following is a summary of “Assessing the precision of artificial intelligence in ED triage decisions: Insights from a study with ChatGPT,” published in the April 2024 issue of Emergency Medicine by Paslı, et al.

The escalating number of emergency department visits globally presents challenges in efficiently managing patient care. In response, various strategies aim to streamline patient management, with artificial intelligence (AI) emerging as a promising solution due to its consistent performance and rapid data interpretation capabilities. Introducing a robust AI tool like ChatGPT, based on GPT-4 developed by OpenAI, holds the potential to enhance patient care by improving the speed and accuracy of resource allocation in emergency departments. For a study, researchers sought to assess ChatGPT’s ability to predict triage outcomes based on local emergency department protocols.

The single-center prospective observational study included all patients presenting to the emergency department with various symptoms who consented to participate. Data collection occurred over three non-consecutive days, totaling 72 hours. Patient chief complaints, vital signs, medical histories, and triage destinations assigned by the triage team were recorded. Simultaneously, an emergency medicine physician inputs the same data into a pre-trained GPT-4 model following local protocols. Triage decisions generated by GPT-4 were recorded, and a gold standard was established by an emergency medicine specialist’s triage decisions based on collected data. Cohen’s kappa test evaluated the agreement between the triage team, GPT-4, and the gold standard. The receiver operating characteristic (ROC) analysis assessed the patient’s triage accuracy by the triage team and GPT -4.

The study included 758 patients, with 54.9% male and 45.1% female participants. Evaluation of primary endpoints revealed almost perfect agreement between the triage team and the gold standard, as well as between GPT-4 and the gold standard (Cohen’s Kappa 0.893 and 0.899, respectively; P < 0.001 for each).

GPT-4 demonstrated exceptional predictive capabilities in triaging patients in an emergency setting. Its accuracy and reliability can effectively support the triage process, potentially enhancing efficiency and patient outcomes in emergency departments.

Reference: sciencedirect.com/science/article/abs/pii/S0735675724000408