Artificial intelligence in clinical settings: a systematic review of its role in language translation and interpretation.

Abstract

Background: Addressing language barriers through accurate interpretation is crucial for providing quality care and establishing trust. While the ability of artificial intelligence (AI) to translate medical documentation has been studied, its role for patient-provider communication is less explored. This review evaluates AI's effectiveness in clinical translation by assessing accuracy, usability, satisfaction, and feedback on its use.

Methods: A systematic search was conducted on July 11, 2024, across Cumulated Index in Nursing and Allied Health Literature (CINAHL), Institute of Electrical and Electronics Engineers (IEEE) Xplore, PubMed, Scopus, Web of Science, and Google Scholar. Inclusion criteria required AI to translate clinical information for a real or theoretical consultation. Exclusion criteria included reviews, correspondence, educational materials, non-peer-reviewed or retracted reports, non-English translations, pre-2016 publications, and reports on sign language or patient education. Search strings representing AI, language interpretation, and healthcare were used. Two investigators independently conducted the screening, extraction, synthesis of results, and bias assessments using Risk Of Bias In Non-randomized Studies - of Interventions (ROBINS-I), Mixed Methods Appraisal Tool (MMAT), and the Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Qualitative Research. A third investigator resolved conflicts.

Results: Of 1,095 reports, 9 studies were analyzed, evaluating AI translation platforms Google Translate, Microsoft Translator, Apple iTranslate, AwezaMed, Pocketalk W, and the Asynchronous Telepsychiatry (ATP) App. Investigations occurred in the US, France, Switzerland, and South Africa, with publications from 2019-2024. AI medical translation shows promise, typically providing accurate translations for brief communications in limited languages, though human translation is often necessary. Accuracy scores ranged from 83-97.8% when translating from English, and 36-76% when translating to English. Usability scores were 76.7-96.7%. Patients were more satisfied than clinicians, with 84-96.6% and 53.8-86.7% satisfied, respectively. Clinicians were hesitant to use AI due to questions of respect, quality, reliability, and misunderstanding. AI is being used as a last-resort option, to assist fluent, non-certified providers and lay interpreters, and for brief communications.

Conclusions: Limitations include few languages tested, unidirectional translation, simulation, and evolving translation tools. AI shows promise in clinical translation, but the complexity of medical consultations requires a balanced approach combining AI and human translation services for quality care.