Driven by the capacity to automate medical record documentation, nearly one billion healthcare conversations between clinicians, patients and families will soon be analyzed using AI every year in the U.S.

In a recent issue of The New England Journal of Medicine AI, Bob Gramling, M.D., D.Sc. and colleagues at Dartmouth College with complementary experience studying clinical conversation using machine learning methods offer suggestions to mitigate unintended but foreseeable harms of routinely analyzing clinical visits with AI. The authors raise awareness about the value that time savings from AI visit documentation can offer for meaningful face time with patients and families, that AI methods in other domains of complexity science can help promote algorithmic fairness when analyzing situationally and culturally complex clinical interactions, and the need for protections to commoditization of algorithms and the conversation data upon which they are trained. 

Doing so, the authors propose, can present opportunities for cancer care that go beyond medical record documentation, including epidemiological understanding about the variability, complexity and impacts of clinical conversations on patient and family cancer experience, supporting patients with tailored visit summaries and personalized decision-support, and offering systematic feedback for educational and quality improvement initiatives.

This paper was made possible by a Vermont Conversation Lab team of interprofessional faculty, staff and study scientists who each contribute to our shared understanding of communication, relationship and the possibilities of AI. Dr. Gramling is a past recipient of an American Cancer Society Research Scholar Award and currently PI (with Dr. Donna Rizzo of CEMS) of a National Cancer Institute R21 that have both been instrumental for catalyzing this work.

To learn more, read the full publication here.