P046: CHATGPT AS A TOOL FOR PERIOPERATIVE SAFETY: EVALUATING DRUG INTERACTION DETECTION IN ANESTHETIC REGIMENS
Sarah Simon; Da Young Lee; Kiana Foster; Rose Berkun, MD; Noel Barengo, MD, PhD, MPH
Florida International University Herbert Wertheim College of Medicine
Background: Drug-drug interactions (DDIs) represent a major contributor to perioperative complications. With the increasing popularity and usage of large language models by healthcare professionals, their performance in high-risk settings should be assessed. No prior studies have assessed the accuracy of large language models such as ChatGPT in this context. This study evaluated the accuracy of ChatGPT (GPT-5) in identifying DDIs in anesthetic regimens, as compared to the gold-standard reference Lexicomp.
Methods: An analytical cross-sectional study was performed using 40 synthetic perioperative vignettes created by an anesthesiologist. Each vignette was classified by the presence of clinically significant DDIs according to Lexicomp’s “Interactions” feature. ChatGPT was queried twice with standardized prompts, and responses were graded as either Correct or Incorrect. The overall identification rate and sensitivity were calculated.
Results: ChatGPT accurately identified 76 out of 80 (95%) clinically significant DDIs across both trials. Each Incorrect response recognized at least one major interaction, but they omitted additional clinically relevant interactions or failed to provide comprehensive management recommendations. Overall sensitivity was 95% (95% CI, 87.69-98.62%), reflecting occasional minor omissions in clinical detail. Notably, ChatGPT often supplemented its responses with pertinent contextual information, including age-specific considerations and perioperative monitoring strategies.
Conclusion: ChatGPT demonstrated high accuracy in detecting clinically significant DDIs in perioperative anesthetic regimens. Its contextual insights may enhance clinical decision-making; however, occasional omissions and limitations in sourcing warrant cautious integration into practice. Further studies in real-world perioperative settings are needed.
Keywords: drug-drug interactions, perioperative care, anesthetic regimens, polypharmacy, medication safety, artificial intelligence, ChatGPT, clinical decision support, drug safety, large language models