A Pilot Study to Evaluate Artificial Intelligence-Driven Early Retrieval of Medical Histories in the Emergency Department

Alessio Morley-Fletcher; Vidya R Raghavan; Alexandra T Geanacopoulos; Molly Maher; Mark Waltzman; Yuval Barak Corren; Andrew M Fine

doi:10.1016/j.annemergmed.2026.01.022

A Pilot Study to Evaluate Artificial Intelligence-Driven Early Retrieval of Medical Histories in the Emergency Department

Ann Emerg Med. 2026 Mar 18:S0196-0644(26)00070-3. doi: 10.1016/j.annemergmed.2026.01.022. Online ahead of print.

Authors

Alessio Morley-Fletcher¹, Vidya R Raghavan², Alexandra T Geanacopoulos², Molly Maher³, Mark Waltzman², Yuval Barak Corren⁴, Andrew M Fine⁵

Affiliations

¹ Effective AI Committee, Department of Pediatrics, Boston Children's Hospital, Boston, MA.
² Division of Emergency Medicine, Boston Children's Hospital, Boston, MA.
³ Division of Emergency Medicine, Boston Children's Hospital, Boston, MA; Boston College, Morrissey College of Arts and Sciences, Chestnut Hill, MA.
⁴ Cardiology Department, Schneider Children's Medical Center of Israel, Petah Tikva, Israel; Predictive Medicine Group, Computational Health Informatics Program, Department of Pediatrics, Boston Children's Hospital, Boston, MA.
⁵ Division of Emergency Medicine, Boston Children's Hospital, Boston, MA. Electronic address: Andrew.fine@childrens.harvard.edu.

PMID: 41854578
PMCID: PMC13004171 (available on 2027-03-18)
DOI: 10.1016/j.annemergmed.2026.01.022

Abstract

Study objective: To assess the feasibility and acceptability of using ChatGPT to obtain histories of present illnesses directly from patients or caregivers in a pediatric emergency department waiting room in a pilot study.

Methods: In a prospective mixed-methods pilot study, patients (n=31) with Emergency Severity Index scores 3 to 5 used a HIPAA-compliant instance of ChatGPT-4 to generate histories of present illnesses which were collected in parallel with usual care and were not used to inform clinical decision-making. Participants used an 11-point scale to rate their experiences around usability and satisfaction and rated the quality of the generated history of present illness. Thematic analysis of open-ended responses was performed. Pediatric emergency physicians reviewed summaries for accuracy, completeness, efficiency, readability and overall satisfaction, using an 11-point rating scale.

Results: Participants reported high usability (median 10, interquartile range 8 to 10), satisfaction (8 and 7 to 10), and overall quality (9 and 8 to 10) of the summary history of present illness. Participants particularly commented on the ease of use, the objectivity of the system, and the accuracy of the summary. Physician reviewers gave favorable ratings across all 5 domains of accuracy, completeness, efficiency, readability, and overall satisfaction, with the highest for readability (9 and 7 to 9), and all others with a median score of 8/10. Clinicians noted that the summaries sometimes lacked important features of prior visits. They noted no hallucinations in the final transcripts.

Conclusion: This pilot study of ChatGPT-enabled patient history taking in the pediatric emergency department waiting room was feasible, well accepted, and produced accurate, complete, readable, and efficient summaries. This approach has potential to reduce documentation burden, enhance patient engagement, and support more streamlined triage.

Keywords: Artificial intelligence; Emergency medicine; Pediatrics.

Grants and funding

T32 HD040128/HD/NICHD NIH HHS/United States