The purpose of this SBIR study is to develop and evaluate a real-time speech translation system to facilitate Medication Therapy Management (MTM) of patients with diabetes who primarily speak Spanish. We plan to accomplish this by (1) Collecting 1000 MTM practitioner-patient interactions as development data in two Community Health Clinics that provide MTM for diabetes, (2) using the data, research approaches to handling large domains and long utterances, and build the system for the domain, and (3) test and evaluate the resulting system on 100 practitioner-patient interactions from the same clinics.
The specific aims i n this SBIR research proposal are to: 1) Handle large domain with high accuracy. In Phase I, we showed that we can build very accurate systems (98 %+) for small domains. In Phase II we will investigate whether that high level of accuracy can be achieved for a large domain that covers the full range of MTM interactions. 2) Handle very long utterances. In phase I we showed success with utterances that contained 2 concepts (10-15 words). In Phase II our goal is to extend that capability to handle a sequence of at least 5 concepts. 3) Patient education support. We observed in Phase I that pharmacists spend considerable time going over basic concepts or ideas in the domain (use of insulin, balanced meal, exercising, etc.). The translation system will be extended to support the patient education component of the MTM approach. This project addresses the growing problem of diabetes management among Hispanics with Limited English Proficiency (LEP). The development and testing of the proposed translation system will achieve: deployment of a lower-cost translation alternative for community clinics;opening of community clinic doors to non-Spanish- speaking English language health professional volunteers to participate in diabetes patient care;enhancement of the Teach-back component of the practitioner visit by ensuring comprehension of MTM recommendations during the visit;and demonstrate the viability of large scale clinical speech translation systems creatin the possibility of systems for many additional domains.
In this proposal, Fluential, Inc. and UCSF propose to develop and evaluate a comprehensive real-time speech translation system to facilitate Medication Therapy Management (MTM) of Limited English Proficient (LEP) patients with diabetes patients who primarily speak Spanish.