At seven o’clock at night, you can find at least one doctor still bent over a screen, completing notes from a morning that ended hours ago, in practically every hospital wing in the United States. It’s a well-known scene that has almost become the norm at UToledo Health. Charts are accumulating. Half-written notes were left. The gradual deterioration of the ideal clinical day.
The numbers then changed. The number of open charts fell from over 400 to less than 30, giving those closest to the work a topic other than exhaustion. The chief medical informatics officer of the health system, Dr. Ryan Sadeghian, has been the one to publicly explain the change, and there is a subtle pride in his description. The figure may sound smaller than it actually is. That type of drop is uncommon in medical records.
| Profile | Details |
|---|---|
| Institution | University of Toledo Health |
| Location | Toledo, Ohio, United States |
| CMIO | Dr. Ryan Sadeghian |
| Technology Vendor | Nabla (Ambient AI clinical documentation) |
| EHR Integration | Epic Systems |
| Pilot Size | 40 providers, multiple specialties |
| Pilot Duration | 8 weeks |
| Encounters Captured | More than 3,000 |
| Chart Closure Time | Reduced by 29% |
| Open Charts | Reduced from 400+ to under 30 |
| Status (2026) | Scaled deployment after evaluation phase |
Nabla, a company that develops ambient AI tools that listen, parse, and produce structured notes that flow into the hospital’s existing Epic environment, is the source of the technology. There are no new login screens. No system in parallel. Clinicians simply converse with their patients as they have always wanted to, and the paperwork then shows up, ready for review.
For the first pilot, forty providers from a variety of specialties were selected. They recorded over 3,000 interactions with the platform over the course of eight weeks. There was a 29% decrease in chart closure time. Backlogs significantly decreased in a number of areas. UToledo Health announced its decision to scale beyond the pilot group earlier this year through a press release that was picked up by trade outlets because the early data was strong enough.
Speaking with healthcare IT professionals lately has given me the impression that ambient AI is approaching the point that Epic itself reached twenty years ago—the point at which a tool ceases to be novel and begins to be taken for granted. For years, the American Medical Association has documented how doctors are forced out of practice due to the burden of paperwork. The hours lost were quantified in studies published in the Annals of Internal Medicine. Therefore, it tends to spread quickly between health systems when something quantifiably lowers that load.
The qualitative aspect is more difficult to capture, which is where Sadeghian becomes more intriguing. He has stated that doctors claim to be more attentive to their patients. In response, patients observe that there is less screen distraction. Those are more subtle claims, the kind that defy precise measurements, but anyone who has sat across from a doctor who is typing rather than listening will recognize the difference right away.
There are still unanswered questions. A category of protected health information that HIPAA was not designed for is introduced by ambient audio in a clinical setting. Specific guidelines have not yet been released by the HHS Office for Civil Rights. According to the IBM Cost of a Data Breach Report, data breaches in the healthcare industry typically cost more than $10 million, making it the most expensive industry globally. No matter how beneficial, any increase in the PHI surface area has consequences.
Even so, it’s difficult to ignore how the conversation sounds different now than it did a year ago when observing this from the outside. In a sense, the pilot phase is over. Whether ambient AI is effective in a hospital is no longer a question. It’s the speed at which the rest of the nation catches up to what Toledo has already discovered.
