Purpose: Language barriers and limited health literacy contribute to poor patient understanding and increased risk of adverse healthcare outcomes. Artificial intelligence (AI) tools are increasingly used to generate and translate patient education materials, including discharge instructions and surgical procedure explanations; however, their accuracy and clinical safety remain variable.
Methodology: A structured literature review was conducted across ten major databases (including PubMed, Embase, and CINAHL), identifying 417 candidate studies. Inclusion criteria targeted studies evaluating AI-generated adult patient education materials and discharge instructions. Studies published prior to 2020, as well as pediatric, psychiatric, duplicate, and review articles, were excluded. Eighteen studies met inclusion criteria for final analysis. Extracted data were analyzed to identify trends in translation accuracy, types of errors, and potential clinical impact.
Results: Across studies, AI-generated translations demonstrated inconsistent accuracy, with substantial variation by language and clinical context. Errors affecting linguistic fluency were the most frequently reported across all languages and commonly reduced comprehensibility of translated materials. Clinically significant errors, including omissions, mistranslations, and altered meanings were more prevalent in less commonly supported languages, encompassing most languages aside from Spanish.
Conclusion: These findings highlight the potential risks associated with unreviewed AI translation of medical information and highlight the need for human oversight to ensure patient safety. Further research focusing on under-represented languages in AI translation is essential to mitigate healthcare disparities and improve equitable access to safe, comprehensible patient education.
Methodology: A structured literature review was conducted across ten major databases (including PubMed, Embase, and CINAHL), identifying 417 candidate studies. Inclusion criteria targeted studies evaluating AI-generated adult patient education materials and discharge instructions. Studies published prior to 2020, as well as pediatric, psychiatric, duplicate, and review articles, were excluded. Eighteen studies met inclusion criteria for final analysis. Extracted data were analyzed to identify trends in translation accuracy, types of errors, and potential clinical impact.
Results: Across studies, AI-generated translations demonstrated inconsistent accuracy, with substantial variation by language and clinical context. Errors affecting linguistic fluency were the most frequently reported across all languages and commonly reduced comprehensibility of translated materials. Clinically significant errors, including omissions, mistranslations, and altered meanings were more prevalent in less commonly supported languages, encompassing most languages aside from Spanish.
Conclusion: These findings highlight the potential risks associated with unreviewed AI translation of medical information and highlight the need for human oversight to ensure patient safety. Further research focusing on under-represented languages in AI translation is essential to mitigate healthcare disparities and improve equitable access to safe, comprehensible patient education.