Table 5 Outer setting domain: examines external influences on implementation, including policies, incentives, and interorganizational networks
From: Operationalizing machine-assisted translation in healthcare
Areas that impact the implementation gap | Barriers to implementation | Potential solutions |
|---|---|---|
K. Regulatory oversight | Need for regulatory oversight of MAT practices in healthcare | Regulatory bodies, like the Joint Commission, should extend their evaluations to examine whether MAT meets the same standards required for traditional translation methods |
There is no standard benchmark for comparing LLM-generated versus human-generated translations | Expand existing Section 1557 compliance checks to include accuracy benchmarks for LLM outputs that include evaluation metrics described in the “Implementation Process Domain” | |
L. Data | There is a lack of centralized, high-quality bilingual corpora for training and validation with emphasis on sufficient examples for digitally underrepresented languages | A consortium-based dataset could be built via state health agencies, academic medical centers, and smaller hospitals that pool resources to build a shared de-identified translation repository |
M. National Policy on LLM use | Outdated national standards (e.g., CLAS) that do not address LLM-based translation needs | Update CLAS standards to include specific guidelines and mandates for LLM use in medical translation |
Need for amendments to section 1557 of the Affordable Care Act and its implications on MAT | Revise Section 1557 to incorporate detailed requirements for LLM use, ensuring guidelines on safeguarding patient data privacy, acceptable MAT error rates and types, baseline qualifications an LLM must meet before it can be used for MAT, and clear roles and conditions when MAT can be used independently and when human review is mandatory |
