A new study by Siru Liu, PhD, Adam Wright, PhD, and colleagues explores using large language models to help patients craft more effective messages to their health care providers through patient portals.
In a blinded test reported in the Journal of the American Medical Informatics Association, the team found that their custom-trained model, CLAIR, produced follow-up questions with similar clarity and conciseness, and higher utility, than actual follow-up questions written by members of the care team. By comparison, the well-known GPT-4 model from OpenAI, variously operating under prompts of low and higher complexity, generated more complete but less clear questions for the scenarios tested.
The idea is that in real time before patients hit “send,” AI could prompt them to clarify their portal messages. Such a tool, by reducing all too common back-and-forth messaging between patients and providers, could streamline communication and improve care efficiency.
In a previous study the team established that AI is better than doctors at responding to patient messages. In the new study, the authors note that auto-prompting better patient questions could prove more advantageous than strategies resting on auto-generated clinician responses. The team next plans to evaluate the clinical impact of implementing this AI-guided patient messaging.
Also on the study were Aileen Wright, MD, MS, Allison McCoy, PhD, Sean Huang, MD, Julian Genkins, MD, Josh Peterson, MD, MPH, Yaa Kumah-Crystal, MD, MPH, MS, William Martinez, MD, MS, Babatunde Carew, MD, Dara Mize, MD, and Bryan Steitz, PhD. The study was supported by the National Institutes of Health (grants R00LM014097, R01AG062499, R01LM013995).
