The Washington-based Windfall well being system continues to publish research on the impression of ambient scientific intelligence (ACI) on documentation workload and burnout. Lately, two Windfall researchers who revealed new analysis on the subject in JAMA Community Open spoke with Healthcare Innovation about their findings.
Again in August 2025, Healthcare Innovation spoke in depth with Windfall’s Maulin Shah, M.D., chief medical info officer; Scott Smitherman, M.D., M.B.A., affiliate vp, CMIO – Windfall Medical Community; and Staci Wendt, Ph.D, director of the Windfall Well being Analysis Accelerator, concerning the well being system’s first research on ambient scientific intelligence.
Windfall has now expanded on earlier research by performing a complete analysis of the associations between an ambient AI system (Dragon Ambient eXperience [DAX]; Nuance) and clinician productiveness and effectivity. The analysis workforce assessed the affiliation between ambient AI use and goal documentation burden, after-hour documentation, and work quantity amongst clinicians utilizing retrospective EHR encounter metadata between July 1, 2023, and March 31, 2025. By the top of their research interval, roughly 8% of clinicians throughout the well being system have been thought of lively customers.
In a dialog with Healthcare Innovation, the analysis workforce mentioned their findings.
“These ambient applied sciences are instruments. They don’t seem to be a alternative for care, they are a instrument for clinicians. They cannot be anticipated to unravel all the issues of doctor administrative burden,” mentioned Canada Parrish, Ph.D., M.S.P.H., senior scientific analysis scientist at Windfall. “In our work, we wish to quantify what that impression appears to be like like. The anecdotes and the lived expertise of clinicians matter, however what additionally issues is, can we truly see this impression? There are a number of instruments on the market. Is sustained funding in a single instrument over one other warranted? There’s goal knowledge to take a look at that, however then there’s additionally the experiential knowledge from these clinicians.”
Of their paper, the researchers discovered productiveness outcomes demonstrated statistically important variations between the pre-active and post-active ambient AI use durations. For example, they discovered a major decline in imply time spent on notes throughout the first month of ambient AI use.
The Medical Effectivity Profile (CEP) is an effectivity metric that’s generated by Epic itself. The researchers discovered that the clinicians’ imply CEP scores had no rapid or sustained affiliation with ambient AI use. “I believe that it was necessary with the research to take a look at effectivity and productiveness and administrative burden from quite a lot of completely different lenses, as a result of there’s not a method and even one customary on how one can quantify or operationalize these constructs,” Parrish mentioned. “For us, it was necessary to take a broad purview on this analysis to see the place we did see proof of enchancment in these metrics, and perhaps ones the place we don’t.”
Ambient AI use additionally was not related to an instantaneous decline in after-hours documentation time, however a statistically important sustained decline in minutes spent documenting after hours was noticed.
“There was an preliminary lower within the time spent in notes throughout the workday, however we did not see that rapid decline in hours post-workday,” mentioned Robyn Husa, Ph.D., senior scientific analysis analyst at Windfall’s Healthcare Analysis Accelerator. “We suspect that is as a result of as clinicians have been considering that the AI wrote the notes throughout the workday, so now I’ve acquired to go assessment them, they usually have been spending extra time in that assessment interval, however because it integrates extra into their workflow over time, they spend much less and fewer time reviewing it outdoors of their work hours. It’s extra of a testomony to the advantages of the gradual improve in use, and the combination into the workflow.”
There have been no associations between ambient AI use and appointments per day, however there was an instantaneous improve in imply RVUs following lively ambient AI use.
“We noticed an instantaneous improve of about seven Relative Worth Items,” Husa mentioned. “Usually, increased RVUs equate to extra providers akin to labs, imaging, referrals, and follow-ups. Some sufferers are extra advanced of their healthcare wants, they usually require extra time for these providers. So our discovering there means that clinicians who use the ambient AI scribe can maybe see sufferers like that extra effectively, or deal with these extra advanced instances, translating to extra providers. Nonetheless, there was no change in affected person quantity per day, that means clinicians weren’t being pushed to see extra sufferers, so that they have been simply spending much less time within the documentation, permitting for the billing of extra providers for every of those sufferers. I do wish to admit that one fear concerning the introduction of such a expertise is that the system would possibly punish suppliers for being extra environment friendly by rising the variety of sufferers they see and we didn’t see that occur right here. It simply allowed them to be with their sufferers extra.”
The researchers famous that there are completely different workflows and issues about deploying these instruments in specialist workplaces than in main care. “The sort of instrument works rather well for main care suppliers or clinicians who’ve a templated workflow that they should put into the EMR. But when they’ve a specialised sort of service that they should present and doc, then the healthcare system would wish to work extra with these instruments to pinpoint how one can assist these particular suppliers,” added Husa.
Parrish defined that the interrupted time sequence design of the research permits people to function their very own controls. “Except you are randomizing otherwise you’re forcing folks into utilizing the instrument, it may be troublesome to quantify the impact that is unbiased of those traits that will drive somebody to make use of the instrument,” she defined. “We all know that early adopters do look completely different than those who got here on later, however selecting an applicable research design helps guard towards a few of these issues.”
Husa talked about just a few different areas for potential analysis. Future research might evaluate the a number of ambient AI instruments in the marketplace to see which options work greatest, not solely as an entire, but additionally for several types of clinicians, like a main care physician versus a surgeon. They might have completely different documentation wants, she mentioned. “One other space could be the impression of ambient AI instrument use on affected person experiences and notice high quality, not simply effectivity. Lastly, the present research checked out goal productiveness measures. Canada talked about there are all kinds of how to measure productiveness. Right here we targeted on some goal ones, however we’re planning on engaged on an examination of extra subjective outcomes — what physicians themselves report about the advantages and disadvantages of AI use.”
