811: AI-Translate: Harnessing Artificial Intelligence for Multilingual Communication in Critical Care

Introduction: Effective communication is vital in critical care, especially for patients with non-English language preference (NELP), however access to professional interpreters is limited. While machine translation has been studied in healthcare, most research focuses on translation of written content, and existing speech tools are not sufficiently trained for ICU needs. Real-time, bidirectional interpretation tools for complex ICU conversations remain underexplored. To address this gap, we developed AI-TransLATE (AI-enhanced Transition to Language-Agnostic Transcultural Engagement), a speech-based tool for real-time multilingual communication in critical care. Methods: This two-phase study evaluated the performance and usability of AI-based interpretation tools. In Phase 1, we created three standardized ICU scripts for shared decision-making on ECMO, tracheostomy, and comfort care, and had bilingual team members simulate dialogues using AI-TRANSLATE. We tested the scripts in 4 languages-Turkish, Spanish, Arabic and Chinese. Two interpreters scored interpretation on a 5-point Likert scale across four domains—fluency, adequacy, meaning accuracy, and error severity. In Phase 2, clinicians completed 3 simulated ICU encounters using AI-Translate, ChatGPT, and Gemini. After each session, participants completed a survey assessing communication quality, usability, and satisfaction, including the System Usability Scale (SUS). Results: AI-TransLATE achieved acceptable composite scores (≥16/20) across all languages. Spanish and Turkish performed well, while Chinese and Arabic showed variability due to omissions and errors. For phase 2, we have recruited 6 participants to date. AI-TransLATE had highest ratings, with 100% finding it potentially helpful and 83% willing to use it clinically; ChatGPT followed with 83% and 67%, respectively. Gemini had lowest scores, with 33% and 17% respectively. Overall, 94% of sessions were rated as realistic. SUS scores supported these findings, with AI-TransLATE scoring 73.3 (SD 6.8), ChatGPT 70.4 (SD 4.6), and Google Translate 48.3 (SD 21.8). Conclusions: AI-TransLATE shows promise as an interpretation tool, but further evaluation in real-world ICU settings is needed.