TOP 13 - Are We Ready for AI? Pediatric Emergency Providers and Discharge Innovation

Publication Number: 4715.TOP 13

Leslie Kaufmann, NewYork Presbyterian - Weill Cornell Medicine, New York, NY, United States; Nicole Gerber, NewYork Presbyterian - Weill Cornell Medicine, New York, NY, United States; Deborah A. Levine, Weill Cornell Medical College - New York, NY, Port Washington, NY, United States; Yaffa M.. Vitberg, Weill Cornell Medicine, New York, NY, United States; Shari Platt, Weill Cornell Medicine, New York, NY, United States; Brady Rippon, Weill Cornell Medicine, Jackson Heights, NY, United States; Alexander Stephan, Weill Cornell Medicine, New York, NY, United States

Background: Artificial intelligence (AI) has seen rapid adoption in healthcare, yet its role in generating emergency department (ED) discharge instructions remains largely unexplored. While prior studies have examined AI’s application for summarizing inpatient hospital courses into the discharge summary, there is a lack of research in the use of AI to generate patient-friendly discharge instruction in pediatric emergency medicine (PEM).

Objective: The primary aim of this study is to examine the likelihood of PEM providers adopting and distributing AI-generated discharge instructions in the emergency setting. The secondary objectives are to extend this comparison to three additional domains—comprehensiveness, accuracy, and readability—and to examine PEM providers’ attitudes toward, and readiness to incorporate, AI-assisted tools in clinical practice.

Design/Methods: We distributed a REDCap survey to PEM providers via an international listserv of 2,986 subscribers. We asked participants to review ChatGPT model 4.0-generated discharge instructions for three common pediatric ED conditions – croup, concussion, and ankle sprain – and to compare them to existing discharge documents (Healthwise® Epic template-generated, expert task force-generated, and individual clinician-written, respectively). We evaluated the documents across four key domains: comprehensiveness (sub-categorized into diagnosis overview, homecare plan, follow-up recommendations, and return guidelines), accuracy, readability, and likelihood that the provider would distribute each exact document to patients. We also asked respondents about their current use of AI tools and their concerns regarding the integration of AI into their discharge process. We collected demographic data, and this study was approved by our Institutional Review Board.

Participant demographics will be described using frequency (%) for categorical variables. Participants’ likelihood to distribute AI-generated discharge instructions for each ED condition compared to a preference for physician-generated instructions or no preference will be assessed by chi-squared test as our primary aim. The same statistical approach will be used to assess feedback for comprehensiveness, accuracy, and readability. A secondary analysis will explore differences in these responses by provider demographics. Qualitative analyses will examine providers’ current use of and concerns for AI implementation in the discharge process. All p-values will be evaluated with statistical significance at the 0.05 alpha level. All analyses will be performed in R version 4.5.1.