Healthcare professionals currently dedicate an average of more than 16 minutes to managing Electronic Health Records (EHRs) for each patient visit, totaling between 3.5 to 6 hours per day (Fierce Healthcare, Medical Economics). Reduction in admin tasks allow health care professionals to focus on providing the best patient care.
This summer, Deloitte Government & Public Services sponsored a University of Virginia School of Data Science Masters capstone project designed to enhance the accuracy and comprehensiveness of retrieval, summarization, and analysis of medical information.
The students developed a Generative AI proof of concept from open-sourced and de-identified data sets from Medical Information Mart for Intensive Care (MIMIC) III and IV. The impact was aimed at accelerating the process of identifying, analyzing, and abstracting EHRs.
The patient record summaries produced by Google MedLM were semantically closer to electronic health records, boasting a median cosine similarity score of 0.61, higher than Gemini Pro’s median of 0.53, and other LLMs median score of 0.57.
The conclusion of this capstone project suggests that training Language Learning Models (LLMs) within specialized domains yields improved outcomes. The impact of this work creates the underlying blueprint and framework for training LLMs to interact with medical and clinical data to perform healthcare-specific Q&A, statistical analysis, and record summarization, freeing up our critical front-line health professionals to dedicate more time to patient care.
Building on this progress, the next phase involves broadening the preliminary study to expand AI solutions for Federal Health and Life Science clients. Thanks to our student research partners Britny Hopwood, Austin Rivera, Karis Roberts, Sophia Williams
Thank you to our Deloitte Consulting LLP team
Vivien Bonazzi, Managing Director, UVA SDS Advisory Board
Rice Tyler, Senior Consultant, UVA SEAS ‘19 | UVA SDS ‘20
Sarah Burinsky, Manager
Stefan Lehman, Consultant
Will Mulquin, Consultant UVA SEAS ‘22
Saahithi Budharaju, Consultant UVA CLAS ‘20
Lastly, special shout out to our Deloitte principals, Ed Van Buren (AI Strategic Growth Offering), (Deloitte AI Institute for Government), and Mark Urbanczyk (Federal Health AI)