Ambient AI can help clinicians complete notes, but health systems are also looking for evidence that better documentation improves coding, revenue cycle performance and clinician capacity.
In a featured session sponsored by Nabla at Becker’s 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health on Sept. 14, leaders from KeyCare and Codman Square Health Center described how those connections are taking shape in two different care settings.
The panelists were:
- Ed Lee, MD, MPH, chief medical officer at Nabla (moderator)
- Renee Crichlow, MD, FAAFP, former chief medical officer at Codman Square Health Center in Boston and vice chair of the family medicine department at Boston University Medical School
- Cory Ogden, MD, MPH, senior medical director of product and medical informatics at KeyCare
1. Consistent documentation
KeyCare provides virtual care through Epic for health system partners. Their clinicians spend only a few seconds selecting a code, Dr. Ogden said. Because the code does not always match the visit documentation, the organization has a manual process to review charts afterward.
Required note templates brought some consistency, but ambient AI has made a larger difference.
“One of the things that actually has created the most consistency in the way our documentation is done is using ambient AI,” he said. More predictable notes make it easier for KeyCare to assess whether the submitted code reflects the care documented.
2. Reducing coding errors
Dr. Crichlow said many physicians receive little coding instruction in medical school and may use the same codes for years without realizing what they are missing. Suggestions shown at the end of a note give clinicians guidance while the visit is still fresh.
In her view, that can be more useful than hearing later that they coded something incorrectly. “They’re getting help as opposed to just getting feedback that they’re doing something wrong,” she said.
KeyCare tested Nabla’s coding suggestions with a small group that included both strong coders and clinicians whose charts often needed correction. Dr. Ogden said error rates fell by roughly 15% to 20% for some clinicians with higher error rates, while the strong coders did not become less accurate.
The pilot also exposed a workflow problem: clinicians had to open another tab to see the suggested code. After KeyCare raised the issue, Nabla moved the suggestion to the bottom of the note. Dr. Ogden said KeyCare was preparing to expand the tool to the full group.
3. Codman Square emphasizes teamwork
Codman Square is a community health center in Boston. Dr. Lee cited a reduction in Codman Square’s days in accounts receivable from more than 40 to seven, which Dr. Crichlow confirmed. She described changes beyond the introduction of an ambient AI. When she arrived at the health center, claims more than 90 days old were being written off and clinicians received little useful feedback about denials.
“Bringing in the ambient AI was actually just an excuse to get our systems together,” she said. Clinical directors and revenue cycle staff began meeting every Tuesday to review denials and cycle time. Clinicians committed to improving documentation, while the revenue cycle team provided clearer, faster feedback on claims that were not going through.
Dr. Crichlow also viewed clinician capacity as a measure of whether support was improving. She told physicians she would know the organization was making progress when they stopped cutting their hours. During her tenure, Codman Square was able to hire primary care physicians despite recruitment challenges faced by peers serving low-resource communities.
4. Human in the loop
An audience physician raised a concern about errors that can appear in AI-generated notes, particularly when a clinician is moving quickly between patients. Dr. Crichlow said the change in how notes are produced does not change who is responsible for their accuracy. “You don’t write notes anymore. You edit notes, and if you’re not editing notes, you’re not being a good clinician,” she said.
Dr. Ogden agreed that clinicians must review and correct AI drafts. He added that generating a note immediately after a visit, while the conversation can still be checked, may help clinicians avoid errors that arise when they finish documentation from memory days later.
