Table 2 Individuals domain: focuses on the characteristics of individuals involved in implementation, including their knowledge, beliefs, and personal attributes
From: Operationalizing machine-assisted translation in healthcare
Areas that impact the implementation gap | Barriers to implementation | Potential solutions |
|---|---|---|
D. Translators | LLM-based MAT integration that disrupts established workflows | EHR/IT teams can partner with translation leads to add “auto-draft” capabilities directly into translators’ current workflows. This should include integrating existing translator tools into the MAT workflow, such as CAT software |
Design comprehensive training focusing on best practices when interacting with MAT tools, how to handle MAT outages, etc. | ||
E. Clinicians | Poorly structured or jargon-laden clinical notes cause translation errors and reduce MAT accuracy | Encourage clinicians to improve the quality of written discharge summaries |
Enable a two-prompt approach where the MAT LLM uses a “preparation” prompt to clean up messy notes before applying the “translation” prompt | ||
F. Patients | Mistrust or misunderstanding of AI-driven translations | Ensure transparency regarding the use of AI in translating patient documents |
Inform patients (via consent forms or discharge packets) that MAT is used, emphasizing final human verification | ||
Lack of patient engagement in refining MAT processes | Patient advisory boards can be leveraged to collect direct patient feedback on clarity and acceptability of translations | |
Tailor MAT deployment from patient feedback by offering MAT primarily in areas where patients feel comfortable (e.g., adult vs. pediatric settings), expanding usage as trust grows |
