Subscribe
Share
Share
Embed
Ambient AI speeds chart closing up to 39% at Jefferson Health, study finds
Ambient AI speeds chart closing up to 39% at Jefferson Health, study finds
The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.
Clinicians at Jefferson Health who adopted Abridge's ambient AI cut their time to close charts by 25.3% to 39.3%, according to a study published Oct. 23 in NEJM Catalyst Innovations in Care Delivery.
The study covered 5,234 clinicians with access to the tool, about 34% of whom actively recorded visits, and 283,028 ambulatory encounters in the second half of 2025, Becker's Hospital Review reported. The authors described the results as associations and noted that adopters may have differed from clinicians who did not use the tool. Abridge had no role in the study and provided no funding, according to the report.
Elsewhere, Inova Health System said it had drawn up plans to switch off three AI features whose performance declined, a Lancet Digital Health experiment found that radiologists and other physicians both followed wrong AI advice and ignored correct advice on head CT scans, and a JAMA Psychiatry study found that adding imaging, genetics and microbiome data did not improve a model's prediction of depression relapse.
In this episode
How technological innovation is changing medicine for patients and the people who practice on the front lines.
Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Sunday, October 25th, 2026.
Voice 1: Clinicians at Jefferson Health who started using Abridge's ambient AI cut their time to close charts by between about 25% and 39%, according to a study published Friday in NEJM Catalyst Innovations in Care Delivery.
Voice 2: That's a wide range. What's driving the spread?
Voice 1: The study, led by Dr. Jennifer Stephens, compared Epic usage before and after adoption among 637 clinicians at Jefferson and its Lehigh Valley Health Network. Time spent in notes per appointment fell about 10% at one site and about 17% at the other, and after-hours documentation fell by similar margins.
Voice 2: Did everyone benefit the same amount?
Voice 1: No. Advanced practice providers gained about 10 minutes of documentation time a day, physicians gained about 4. Patient experience scores looked equal or better on ambient visits, but that edge mostly vanished once you compared each clinician's own ambient and nonambient visits.
Voice 2: So it's association, not proof the tool itself caused the gain.
Voice 1: Right, the authors call these associations and note adopters may have simply differed from clinicians who skipped the tool. And Jefferson's earlier claim of about 1 million hours returned to patient care comes from executives, not from this study.
Voice 2: Worth separating the press release number from the peer-reviewed one.
Voice 1: Abridge had no role in the study and provided no funding, according to the report.
Voice 2: Inova is going the other direction with some of its tools.
Voice 1: What's happening there?
Voice 2: Inova's latest quarterly safety and reliability review flagged three AI features whose performance was declining, and the system has built decommissioning plans for all three. It now reviews more than 12 features a month, up from about one a month two years ago.
Voice 1: Did they say which three features?
Voice 2: No, the report didn't name them. Jon McManus, Inova's chief data and AI officer, said the point is that they're turning AI off when it stops meeting performance standards. He also said Inova is preparing governance to oversee as many as 500 features.
Voice 1: A system willing to shut things off is a better sign than one that just keeps adding.
Voice 2: A new experiment pokes at how doctors actually use AI advice, not just whether health systems keep it running.
Voice 1: This is the head CT study?
Voice 2: Yes. Radiologists and other physicians read scans for intracranial hemorrhage with no AI help, with an AI prediction alone, or with a prediction plus annotations. Twenty percent of the advice was deliberately wrong.
Voice 1: And did they fall for the wrong advice?
Voice 2: Both things happened. Some followed wrong advice, others ignored correct advice, and the researchers found under-reliance did more harm to accuracy overall. Correct annotated advice did improve diagnostic performance compared with no advice at all, though it didn't speed up reading time.
Voice 1: So the tool can help, but only if people actually trust it when it's right and doubt it when it's wrong.
Voice 2: Which is apparently harder than it sounds, for radiologists and non-radiologists alike.
Voice 1: A psychiatry study out this week found that piling on more kinds of data didn't actually help.
Voice 2: More data usually sounds like a good thing. What went wrong?
Voice 1: German researchers followed 564 patients with major depressive disorder for about two years. Two hundred seventy relapsed. They built models using nine kinds of baseline data, including four types of brain imaging, genetics, blood markers and gut microbiome.
Voice 2: And adding all that didn't beat a simpler model?
Voice 1: No. Balanced accuracy across the models ranged from about 47% to about 64%, and the best-performing model used only a handful of clinical risk factors. The authors wrote the results suggest, in their words, limited scope for guiding individual-level clinical decision-making.
Voice 2: That's a plain statement from the people who ran the numbers.
Voice 1: It is. Sometimes the expensive data just doesn't add anything the basics didn't already tell you.
Voice 2: The study doesn't go further than that, so we'll leave it there.
Voice 1: Sources and corrections are at physicianintheloop.org, where you can sign up for the free email briefing. This briefing was read by AI voices.