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

Comprehensive Summary

This mixed-methods pilot study examined the integration of ChatGPT into the pediatric emergency department (ED) waiting room to improve documentation quality and streamline clinical workflows. Conducted in a high-volume urban academic hospital with approximately 62,000 annual visits, the study enrolled a convenience sample of English-speaking patients aged 12 or younger, excluding those with mental health concerns. Most participants were lower-acuity cases (72% Emergency Severity Index scores of 3–5). A total of 31 participants consented, with a median age of 9.7 years; one-third identified as Hispanic, and most presented with respiratory or gastrointestinal complaints. Patients interacted with ChatGPT—configured as a medical assistant operating at a sixth-grade reading level—to complete a history of present illness. Following this interaction, attending physicians reviewed ChatGPT-generated summaries and compared them to traditional clinician notes across several domains, including accuracy, completeness, readability, and efficiency. Both patients and clinicians rated their experiences using 11-point REDCap scales. Patients reported highly positive experiences with ChatGPT, noting its ease of use (median 10), satisfaction (median 8), and helpfulness in preparing for the visit (median 9). Participants valued being able to “type in my own responses” and appreciated the tool’s objectivity and consistency, though some felt the process was more time-consuming than a traditional encounter. Attending physicians found ChatGPT notes more detailed yet occasionally missing demographic or historical details. Despite limitations related to sample size, generalizability, and novelty effects, this study demonstrated the feasibility and acceptability of AI-assisted patient histories in ED environments, highlighting potential for broader EHR integration and scalability.

Outcomes and Implications

Integrating ChatGPT into emergency department triage supports a more patient-centered process by helping patients better understand ED workflows and potentially reducing wait times. The system can ease documentation demands on clinicians while mirroring a clinician’s history-taking summary, allowing patients to describe symptoms they might otherwise overlook. However, this approach may lessen opportunities for direct patient–clinician interaction, leading to less nuanced and empathetic care. Additionally, because ChatGPT lacks access to prior medical records, its histories may be less comprehensive than traditional clinician notes. While the tool can enhance transparency and consistency for patients, it may not suit the preferences or needs of all patients and providers.