Leveraging ChatGPT for improved decision-making in interfacility transfers for plastic surgery emergencies based on United States guidelines

Abstract

Plastic surgery emergencies-such as burns, facial trauma, and upper extremity injuries-often require timely specialist consultation. Yet, interfacility triage decisions are frequently made by frontline providers without direct access to plastic surgeons, leading to unnecessary transfers that burden healthcare systems and patients. Artificial intelligence (AI) platforms like ChatGPT may offer a novel way to support these decisions by aligning recommendations with established clinical guidelines. This study evaluated ChatGPT-4.0's ability to assess the need for interfacility transfer across 60 standardized clinical scenarios representing common plastic surgery emergencies. Scenarios were based on U.S. guidelines or expert consensus and entered into ChatGPT using three different prompt formulations, or ``primers,'' yielding 180 total recommendations. These were compared against gold-standard transfer decisions. Overall, ChatGPT aligned with guideline-based decisions in 77% of cases. Performance varied by prompt: primers 1 and 3 achieved 77% alignment, while primer 2 achieved 73%. Primer 1 tended to under-refer, whereas primers 2 and 3 over-referred (<i>p</i> < 0.001). By clinical domain, alignment was highest for upper extremity trauma (83%), followed by burns (75%) and facial trauma (68%). These findings suggest ChatGPT can offer generally appropriate transfer recommendations, though its accuracy depends on both scenario type and prompt phrasing. While promising, such tools must be used with clinical oversight, given variability in performance and lack of explainability. With further validation, AI platforms like ChatGPT could support emergency triage for plastic surgery referrals, particularly in low-resource settings where timely specialist input is limited.

Ähnliche Arbeiten