Date of Award
2026
Document Type
Thesis (Ph.D.)
Department or Program
Computer Science
First Advisor
Sarah Preum
Abstract
In recent years, significant efforts have been made to integrate Large Language Models (LLMs) into healthcare. In this study, we explore how to improve patient-provider communication using LLMs --- with explicit focus on asynchronous communication through EHR-integrated web portals. Throughout this thesis, we describe three interconnected studies that allow us to adapt, train, and evaluate LLMs in asynchronous portal communications. The first study focuses on Patient Message Triage, which is the task of determining the medical urgency of patient-authored message content. We introduce a novel benchmark, alongside methods and training strategies producing a state-of-the-art and publicly available triage solution. Our next study focuses on Follow-up Question Generation, where we introduce a multi-agent system for automatically generating follow-up questions to patient messages --- reducing information gaps in patient messages. Finally, we introduce PatientEvent which is a novel event extraction ontology for understanding patient message data at scale. Our approach is a training-free, efficient solution to large-scale data characterization and can be made available to the broader research community.
Recommended Citation
Gatto, Joseph, "LANGUAGE MODELS FOR ENHANCED PATIENT-PROVIDER COMMUNICATION: METHODS, MEASURES, AND BENCHMARKS" (2026). Dartmouth College Ph.D Dissertations. 507.
https://digitalcommons.dartmouth.edu/dissertations/507
