The common teaching that “the key to diagnosis is in the HPI” holds true for many of the patients we care for in the ED.
But, what if ChatGPT was tasked with obtaining the initial HPI rather than the provider? How accurate/reliable can a ChatGPT-driven history prove to the physician, or seem to the patient?
Researchers in the field of AI applications are beginning to tackle these questions as emergency departments pilot AI-driven history platforms to leverage time spent in the waiting room with a goal of improving efficiency and provider workload. What do you think?
Click the link to see a summary of some interesting findings in this pilot study conducted in a large urban pediatric ED waiting room.
Additional Information
- Physician reviewers rated ChatGPT-derived HPI summaries favorably across accuracy, completeness, efficiency, readability, and overall satisfaction.
- Patients and caregivers reported high satisfaction and ease of use with the process of providing an HPI through ChatGPT and described the interaction as similar to speaking to a clinician.
- Some patient's felt that the AI-interaction helped them prepare for the actual conversation with the clinician, reminding them of important symptoms to mention.
- No hallucinations were reported.
- There were some key omissions and errors which were caught when summaries were reviewed by the patients/caregivers and clinicians.
- Trade-offs included decreased patient-clinician interaction which may impede the relationship-building aspect of history-taking and fewer nonverbal cues between the patient and the clinician.
References
Morley-Fletcher A, Raghavan V, Geanacopoulos A, et al. A Pilot Study to Evaluate Artificial Intelligence-Driven Early Retrieval of Medical Histories in the Emergency Department. Annals of Emergency Medicine, 2026; 88, 135-143.