Providence health system in Washington has released another study examining how ambient AI tools affect clinician workload and burnout, adding to a growing body of research on the technology’s real-world impact. The latest paper, published in JAMA Network Open, looks at whether an ambient clinical intelligence system actually changes how doctors spend their time.
Two researchers from the health system involved in the study spoke with the outlet about what they found. The system has been studying ambient AI since at least August 2025, when its first paper on the topic was published.
The new research evaluated associations between an ambient AI system — Dragon Ambient eXperience, or DAX, from Nuance — and clinician productivity and efficiency. The team used retrospective EHR encounter metadata from July 1, 2023, through March 31, 2025. By the end of that period, about 8% of clinicians in the system were active users.
What the data actually showed about documentation time
Productivity outcomes showed statistically significant differences between the periods before and after doctors started using the tool. In the first month of use, there was a notable decline in mean time spent on notes.
Related: Which are the Important Excel Functions and Formulas Needed for Business Analytics
But the Clinical Efficiency Profile — an efficiency metric generated by Epic — showed no immediate or sustained association with the tool’s use. “I think that it was important with the study to look at efficiency and productivity and administrative burden from a variety of different lenses,” said Canada Parrish, Ph.D., M.S.P.H., senior clinical research scientist at the system.
“There’s not one way or even one standard on how to quantify or operationalize these constructs,” Parrish added. “For us, it was important to take a broad purview in this research to see where we did see evidence of improvement in these metrics, and maybe ones where we don’t.”
After-hours documentation didn’t drop right away
Its use was not associated with an immediate decline in after-hours documentation time. But the team did observe a statistically significant sustained decline in minutes spent documenting after hours over time.
“There was an initial decrease in the time spent in notes during the workday, but we didn’t see that immediate decline in hours post-workday,” said Robyn Husa, Ph.D., senior clinical research analyst at the system’s Healthcare Research Accelerator. “We suspect that’s because as doctors were thinking that the AI wrote the notes during the workday, so now I’ve got to go review them, and they were spending more time in that review period.”
Related: VPS For E-Commerce: Is It The Right Choice?
As the tool integrated more into workflow over time, doctors spent less time reviewing notes outside work hours, Husa added.
RVUs went up, but patient volume didn’t
There were no associations between the tool’s use and appointments per day. But there was an immediate increase in mean Relative Value Units following active use — about seven RVUs, according to Husa.
“Generally, higher RVUs equate to more services such as labs, imaging, referrals, and follow-ups,” Husa explained. “Some patients are more complex in their healthcare needs, and they require more time for these services. So our finding there suggests that doctors who use the tool can perhaps see patients like that more efficiently.”
Husa noted a common worry about such technology: that systems might punish doctors for being more efficient by increasing patient volume.
Related: PAFI Kota Pandeglang: Enhancing Community Engagement and Healthcare Access
“We did not see that happen here,” she said.
“It just allowed them to be with their patients more.”
Different tools for different specialties
The team pointed out that workflows vary significantly between specialist offices and primary care. “This type of tool works really well for primary care providers or doctors who have a templated workflow that they need to put into the EMR,” Husa remarked. “But if they have a specialized type of service that they need to provide and document, then the healthcare system would need to work more with these tools to pinpoint how to help those specific providers.” Parrish explained that the study’s interrupted time series design allowed individuals to serve as their own controls, which helps address the fact that early adopters of technology often differ from later users. “Unless you’re randomizing or you’re forcing people into using the tool, it can be difficult to quantify the effect that’s independent of these characteristics that may drive someone to use the tool,” she noted.
Husa mentioned several areas for future research, including comparing multiple ambient AI tools on the market to see which features work best for different types of doctors. She also pointed to the need for studies on patient experiences and note quality, not just efficiency. The current study focused on objective productivity measures, but the team plans to examine subjective outcomes — what physicians themselves report about the benefits and drawbacks of AI use.
