Primary care team testing AI-assisted patient messaging offers lessons learned

Innovations in Primary Care 

Primary Care Team Testing AI-Assisted Patient Messaging Offers Lessons Learned

Background: Primary care clinicians spend a growing amount of time responding to patient messages through electronic portals,  a task that contributes to burnout. Some health systems are piloting using large language models (LLMs) to generate draft responses to patient messages.

The Innovation: At West Virginia University, the authors tested an artificial intelligence tool called Augmented Response Technology (ART) that generates a draft reply as soon as a patient message arrives. Nurses review each draft and decide whether to send, edit, or forward it to a physician. Early versions of ART produced responses that were not very useful, often focusing on triage rather than directly answering patients. The team improved the tool by overhauling the prompt input to enforce tone, safety, and content delivery by grouping messages into categories by type (results review, medication refills, paperwork, general symptoms) and adding a symptom severity library to best help tailor responses.

Implications: The authors conclude that tools like ART may support patient messaging, but only with careful, adaptable prompt design. Different specialties, message types, and patient populations require different approaches. 

Lessons Learned From the Front Line of AI-Augmented Patient Messaging 

Joseph E. Capito, MD, et al 

Department of Family Medicine, West Virginia University, Morgantown, West Virginia

PRE-EMBARGO LINK (Temporary)