HealthMarshall: A Path through the Forest behind Phone Trees in Healthcare

Presentation Time: 02:15 PM - 02:45 PM

Abstract Keywords: Large Language Models (LLMs), Telemedicine, Natural Language Processing, Human-computer Interaction, Health Equity, Interoperability and Health Information Exchange
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

HealthMarshall is an LLM powered question answering chatbot which integrates into the patient portal of an OpenEMR instance. It can answer basic questions about a patient’s prescriptions using retrieval augmented generation (RAG). The RAG process operates by extracting information about the patient's prescriptions from the patient's medical notes, and linking this information to prescription knowledge from the database of MicroMedex. This information is injected into the LLM's prompt to ground its responses in real knowledge and prevent hallucinated responses to the patient's questions. HealthMarshall was designed to help reduce the burden that complex call trees place upon patients. Complex call trees can make it quite complex and tedious for a patient to access their care provider to get basic questions about their care answered. Technology that can answer the most basic of these questions would provide significant help in saving both patients’ and providers’ time and increase healthcare accessibility for vulnerable populations with limited digital literacy.

Speaker(s):
Jacob Solinsky, Masters of Science
University Of Minnesota

Author(s):
Changye Li, PhD - University of Minnesota; Martin Michalowski, PhD, FAMIA - University of Minnesota; Serguei Pakhomov, PhD - University of Minnesota; Jacob Solinsky, Masters of Science - University Of Minnesota;