405 - Patients' and clinicians' perspectives about use of artificial intelligence for language services in pediatric healthcare settings

Publication Number: 405.5922

Gabriella Ficerai-Garland, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States; Peyton Groves, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States; Elena Puccio, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States; Henry Hoffman, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States; Sabrina Bruno, University of Pittsburgh School of Medicine, Pittsburgh, PA, United States; Nadia C. Pineda-Villegas, University of Pittsburgh, Jamaica plains, MA, United States; Maria Cecilia Q.. Dancisin, UPMC Children's Hospital of Pittsburgh, Pittsburgh, PA, PA, United States; Maya I. Ragavan, University of Pittsburgh, Wexford, PA, United States

Background: People who use languages other than English (LOE), defined as those who report speaking or understanding English less than very well, account for 8% of the U.S. population. Children and their caregivers who use LOE experience healthcare disparities and poorer overall health. Provision of linguistically equitable services is an evidenced-based solution to dismantle health disparities. Human interpreters and translators are the mainstay of communication, though gaps in interpretation and translation services exist. Evolution of large language models like ChatGPT have raised questions about future use of artificial intelligence (AI) when human services are not available.

Objective: We conducted a qualitative study to explore perspectives of pediatric clinicians and caregivers who use LOE on unmet needs in healthcare interpretation and translation and the use of AI to assist with medical communication.

Design/Methods: We included clinicians and caregivers of pediatric patients who use LOE. Clinicians were recruited via word-of-mouth email and listservs. Caregivers were recruited via referral by physicians or from our institution’s international department. Interviews focused on use of AI language technologies in healthcare. Interviews were 45-60 minutes and audio-recorded via Zoom, with a phone interpreter used for non-English interviews. Spanish interviews were conducted with a fluent team member. Recordings were transcribed, translated as needed, coded, and analyzed using an inductive thematic analysis approach.

Results: 22 clinicians (from 16 divisions) and 20 caregivers (who spoke 11 languages) participated. Participants described how technologies (e.g., Google Translate) are used in healthcare as a stop-gap when human services are not available, especially for written communication, emergencies, and inaccurate or poor interpreter connection. Despite use of Google Translate, participants were concerned about threats to accuracy, privacy, and empathy when using AI. Participants reported wanting to use AI in the future for improved real-time translation of documents, interpretation that can accommodate specific dialects, and in some cases to discuss sensitive topics. Ease of use for clinicians and caregivers, validation of the technology, and equitable creation and access to AI technology was described as critical for widespread, ethical use.

Conclusion(s): Unmet needs exist in medical translation and interpretation. Clinicians and caregivers are open to using AI to fill these gaps when a human interpreter or translator is not available, but validation of AI technology is needed to establish trust.

Table 1
Participant Demographics

Table 2
Representative Quotations