JMIR Publications

ABSTRACT

Background:

LLMs in secure messaging have been targeted as a potential solution to clinicians’ growing burnout related to EHR work outside of clinical encounters. Unfortunately, evaluations of this tool focus on the content of individual messages rather than how its use could influence patient and clinician perspectives of the therapeutic relationship.

Objective:

To understand how patients and clinicians view the use of LLMs in secure messaging and identify potential implications for the patient-clinician relationship.

Methods:

This qualitative study used semi-structured interviews and vignette-based prompts that mirrored real messaging exchanges between a patient and their clinician. The study setting included a purposive sample of patients and outpatient clinicians. Participants included patients and clinicians. A purposive sample of 20 adult patients was recruited from across the United States. Recruitment sought increased representation from minoritized racial and ethnic groups and older patients. A purposive sample of 11 clinicians were recruited from health systems concentrated in the San Francisco Bay Area. Data were collected between August 2024 and April 2025. One-time interviews conducted via videoconference. Interviews included questions about messaging use, perspectives on AI in healthcare, clinicians’ use of LLMs in patient messaging, and two vignette-based messaging scenarios. We examined patient and clinician perspectives on (1) the role of secure messaging and (2) the impact of LLM-drafted messages on the therapeutic relationship.

Results:

Participants expressed the importance of secure messaging for fostering the patient-clinician relationship and concerns that LLM-generated messages could detract from such relationships. They expressed concerns that, while these tools should help clinicians focus more on patient-centered care, they could instead increase work burdens and limit the care patients expected to receive with messaging. Some expressed further concerns that such tool use could harm rapport and trust.

Conclusions:

This qualitative interview study found that some patients and clinicians perceive LLM use in secure messaging could detract from the therapeutic relationship by harming patients’ rapport and trust. Implementation decisions should consider how patients’ expectations of their clinician and their use of messaging could be reshaped by the technology.