Nearly half of patients surveyed by UC Davis Health in Sacramento, Calif., said an AI tool that records and transcribes medical visits would improve care, though concerns about accuracy, privacy and being recorded remain.
The survey was conducted as the health system prepared to implement an AI-powered scribe aimed at reducing physicians’ documentation burden and allowing greater focus on patients during clinical visits. Leaders said in a Dec. 16 news release that the results helped shape how the technology was introduced and communicated to patients.
UC Davis Health emailed the survey to more than 9,000 patients and received about 1,900 responses.
Here are five key findings from the survey:
- While 73% of respondents said they felt heard during clinical visits, 23% said their physicians appeared more focused on notetaking than on them.
- When asked about AI transcription, 48% said it would be a good solution, 33% were neutral and 19% expressed concerns. Patients ages 18 to 30 were more skeptical of the technology than older respondents.
- The top concern was accuracy of AI-generated notes, cited by 39% of respondents. Privacy and security concerns were mentioned by 13%, and another 13% said they were uncomfortable being recorded. About 10% said the technology could negatively affect clinicians or staff.
- Patients said they preferred to be informed early that an AI tool would be used — ideally during appointment scheduling or check-in. A majority, 57%, preferred face-to-face notification, while 45% said email was acceptable.
- Based on those preferences, UC Davis Health prioritized in-person discussions, required vendors to use domestic servers and set recordings to be deleted within 10 days. Clinicians review and edit AI-generated notes before they are added to the EHR, and patients can review notes, request corrections or opt out of the technology.
The health system also reviews AI tools through a dedicated analytics oversight committee focused on responsible and ethical implementation, according to the study, which was published in JMIR Medical Informatics.
