Committee reviews bill to allow telehealth recordings as UVM Health demonstrates AI note‑taking

A legislative committee on Tuesday revisited H.84, a proposal to allow recordings of telemedicine and audio‑only telephone appointments if both the patient and the provider consent. Jen Carvey of the Office of Legislative Council told members the bill’s change is a single added line to existing statute to permit recordings when both parties agree, and that the bill as introduced does not add separate rules about how recordings may be used.

Why it matters: Committee members pressed witnesses on whether the limited statutory change could allow vendor use of voice data for artificial‑intelligence training, emotion analysis or other purposes beyond the medical record. Those concerns cut to whether federal and state privacy laws, vendor contracts and internal health‑system governance provide adequate protection for patients and clinicians.

UVM Health’s chief health information officer, Justin Smith Donnelly, told the panel that his system has been using an ‘‘ambient’’ audio‑based tool (Abridge) to generate draft clinical notes. Donnelly described a staged rollout after pilots in 2024 and said the workflow requires provider permission to record, a short retention period for raw audio and provider review and signature before any draft becomes part of the permanent medical record. "Recordings and transcripts are retained for 30 days," he said, and "a subset can be de‑identified" and used under contract to improve the vendor’s system.

Donnelly said the pilot—50 primary‑care providers in February–April 2024—found note quality that was, by the study’s scoring method, "non‑inferior" to human notes. He also reported improvements in clinician well‑being among the pilot group, telling the committee that the project showed what he described as "a 48 percent reduction in burnout rates" for that cohort at the one‑year point. He cautioned that the product is not appropriate for all specialties and that final coding and billing decisions remain the responsibility of clinicians and, when used, professional coders.

Committee members asked whether the tool’s coding recommendations could drive upcoding and whether any observed claims surges could be traced to ambient AI. Donnelly said the system issues recommendations based on MEAT criteria but that clinicians make the final coding choices and that, in UVM’s view, the large claims surge raised previously in committee was more attributable to volume and service counts than to an automated jump in coding intensity.

On privacy and governance, Donnelly described UVM’s internal AI council, a contract review and security intake process for any AI product, and said the system limits raw audio retention to a transitory period to allow provider edits. He also said the signed clinical note—the provider‑edited, authenticated record—becomes part of the patient's medical record and is available to the patient via the portal; the raw audio itself is not kept in the permanent chart.

Unidentified committee members flagged unresolved questions the bill does not answer: what affirmative disclosures patients must receive about AI processing; whether vendors could combine voice data with other datasets to track individuals outside the medical record; and whether additional statutory guardrails or reporting requirements are warranted if recording in telehealth is allowed. Carvey and Donnelly both said the introduced bill does not include such guardrails and recommended that members identify specific follow‑up topics for staff and witnesses if they want changes before any vote.

The committee scheduled continued consideration of H.84 for the next day at 8 a.m., asking members to review a documents list and return with priorities for additional information or amendments. The bill may come up for a vote at that meeting.

Sources: Testimony and demonstration on the record from Jen Carvey (Office of Legislative Council) and Justin Smith Donnelly (chief health information officer, UVM Health).