MADISON, Wis. – A randomized trial from the University of Wisconsin School of Medicine and Public Health and UW Health showed that introducing ambient artificial intelligence scribes in UW Health clinics reduced health care practitioner burnout and time spent documenting patient notes.
The clinical trial also provided one of the first, most comprehensive pathways to date for other health systems to thoughtfully and successfully test and implement the technology.
The study was published in the New England Journal of Medicine Artificial Intelligencein two parts. The first article established a rigorous trial framework and protocols to design, monitor and evaluate ambient AI within routine care. The second article examined the impact of ambient AI on health care practitioner burnout and well-being.
The time spent documenting patient interactions in electronic health records, or EHR, is a major contributor to burnout, according to Dr. Joel Gordon, chief medical information officer, UW Health.
“Traditionally, providers spend a significant amount of time documenting their patients’ stories and their professional thoughts,” said Gordon, who is also an associate professor of family medicine and community health, UW School of Medicine and Public Health. “Ambient AI can draft notes securely in the background while the provider and patient interact directly.”
Ambient AI technology is a tool that can record, transcribe and analyze the discussion between a health care practitioner and patient during an appointment, creating a draft note that the provider reviews and uses as part of the patient’s visit documentation.
The goal of using ambient AI technology, such as the one created by Abridge and tested at UW Health, is to reduce the cognitive burden that EHR documentation places on health care practitioners, according to Gordon.
“This way, they can focus on diagnosis, treatment and bonding with the patient,” he said.
Eagerness to try the technology was accompanied by commitment to scientific rigor, as members of the medical informatics and clinical operations teams at UW Health also wanted to test the best way to implement it and to continually monitor whether it was working.
Researchers partnered with the University of Wisconsin–Madison Institute for Clinical and Translational Research’s Learning Health System, or LHS, program. The LHS program’s multidisciplinary, evidence-based approach supported the development of technical workflows, governance structures and a pragmatic randomized trial design. Program members provided expertise in implementation science, biostatistics, clinical trial design, data science and protocol development needed to generate the research-grade analytics to run the trial on an operational timeline.
The work resulted in the creation of a “playbook” on pragmatic trial operations for health systems to introduce this form of AI technology, and to test safety and effectiveness, according to Dr. Majid Afshar, associate professor of medicine, UW School of Medicine and Public Health, and director of the Learning Health System.
“Our LHS team became integrated with the operations teams, which allows us to learn from what we do and do what we learn,” said Afshar, who is also a critical care doctor at UW Health. “By initiating the partnership between UW Health and LHS, we employed best practices in implementation science, pragmatic trials and evaluation to ensure compliance with operational best practices. It’s about setting up the technology and the users for success, and continuing to monitor them to make sure everything is working as intended.”
The Pragmatic Trial Operations Playbook is now available as an open-source guide for other health systems wanting to implement, monitor and analyze the use of ambient AI.
“Ambient AI technology makes many promises, but just like when a pharmaceutical company says they have a great new drug, we need to do clinical trials to prove the drug works as intended,” Afshar said.
The second article focused on the pragmatic randomized trial testing the impact of ambient AI on health care practitioner well-being. Researchers created an individually randomized clinical trial involving 66 physicians and advanced practice providers, or APPs. The 66 providers were randomized into three waves of 22, and each provider was repeatedly surveyed and monitored for well-being and documentation metrics during and after using the new technology. Multiple questions on well-being were drawn from the Stanford Professional Fulfillment Index, which measures drivers of practitioner exhaustion and interpersonal disengagement. There were also questions about the time spent working after hours to catch up on clinical documentation.
The trial showed that the use of ambient AI technology correlated with a clinically meaningful reduction in burnout scores. The technology also reduced documentation time by 30 minutes per day per provider, improved the accuracy of the notes for diagnosis billing and improved other secondary measures on well-being, like task load.
The trial design allowed a balance between the desire for quick implementation and the need for careful testing to ensure medical record accuracy and whether the technology improved efficiency and clinician well-being, according to Gordon.
“Our mantra was doing research-grade analytics at an operational pace,” he said. “We needed to go fast, but not so fast that we forfeited the rigor of the trial. Most of the studies of ambient AI so far have been observational, but a trial like ours provides indisputable data.”
After the successful trial, which ran from August 2024 through March 2025, UW Health expanded the use of ambient AI across clinics and hospitals in Wisconsin and Illinois. Currently, about 800 physicians and advanced practice providers use the technology. A data dashboard allows continued monitoring of how the technology is performing in clinical operations.
Researchers are supporting additional studies on patient experience with their care providers using ambient AI, Afshar said.
“Patients, as well as physicians and APPs seem pretty happy with it,” he said. “We hope this work gives other health systems the tools and data they need to determine how they can use ambient AI successfully.”
The project described is supported by UW Health and the Clinical and Translational Science Award program, through the National Institutes of Health National Center for Advancing Translational Sciences grant UL1TR002373. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
