Curbside Consult with Dr. Jayne 6/9/25

People often ask me about the kinds of things that excite me within healthcare IT. I have to admit that despite the amount of money that has poured into the industry over the last few years, I don’t run across things that I think are cool as often as I would like.

Although I’m enthusiastic about new developments, a lot of companies appear to be trying to jump on a bandwagon. Plenty are hawking solutions in search of a problem, while ignoring the real problems that clinicians face each and every day.

I was glad to see that Stanford Medicine is going after a solution that could be a game changer for clinicians. Their new ChatEHR platform is getting a lot of buzz, and rightfully so. The ability to effectively query the medical record and find information quickly would create a tremendous advantage for clinicians.

Back in the days of paper charts, we thought a hospital stay was complicated if the patient’s visit documentation expanded into a second chart. Sometimes patients who had been there for a while even had a third or fourth chart. I cared for quite a few patients who were long-term residents of the inpatient units. I once dictated a discharge summary for a pediatric patient who had been hospitalized for 18 months. I was extremely grateful to the different residents who had created transition summaries whenever one of them rotated off that particular medical service. It allowed me to draw the overall summary from those interim summaries rather than having to dig through 550+ days of documentation.

It should also be mentioned that good or bad, hospital notes were shorter in those days. Although an admission History and Physical or a Discharge Summary might have been a couple of pages, the average daily note was a couple of inches long on the page and included much less regurgitated information than notes do today. Sometimes they were borderline illegible, which I agree is a patient safety risk, but they cut to the chase.

I always enjoyed the notes of a particular infectious disease consultant who wrote his notes in bullet format and put the truly important items in all caps. Now, even a simple daily progress note can be several pages long. It feels increasingly difficult to find the information that’s important.

EHR vendors have tried to combat this by creating various summary screens, tables, dashboards, and other elements. Although some of them are truly awesome (hip, hip, hooray for graphing and trending of lab values and vital signs data) they don’t do well at capturing narrative information that is still frequently found in providers’ notes. Often it’s the narrative comments that really tell the story of what is going on with the patient. This is where using AI to better harness that information can deliver real value.

When I read the initial description of the Stanford tool, it reminded me of working with a human scribe in the emergency department. Our scribes were phenomenal and did a great job of anticipating the attending physician’s questions and having the answer ready by digging through the different screens while we were talking with the patient. Their ability to multitask was much appreciated, although not every scribe is that proficient. Many physicians don’t have scribes, so their thought processes were fragmented while they’re trying to simultaneously hunt for information and also talk to the patient, their family, and the care team. Stanford leadership called out the importance of having this functionality in the clinician’s workflow.

It should be noted that several EHR vendors have been working on this, but there are some limitations to a vendor-driven approach, at least in my experience.

I’ve worked with more than a dozen EHRs over the years, and many different instances of the same two or three EHRs. Despite the idea of vendor-driven standardization, when you’ve seen one installation of a big EHR, you’ve seen one installation of a big EHR. Unless the vendor is strict about preventing customization, care delivery organizations have been known to customize themselves into a corner in the name of trying to enable their own unique workflows.

With the health system driving the AI search and summary efforts, not only can those local customizations be addressed, but it would also seem easier to incorporate source material from other systems. That could be a different EHR, legacy records, HIE information, or state registry information.

The Stanford team has been working on their solution since 2023, so it’s not something that an organization can just throw together overnight at this point. The model has limited use, with just over 30 clinicians at Stanford Hospital working with it and providing feedback on its performance and usability. Their goal is to roll it out to other clinicians at the facility as well as those at other facilities within the larger organization. It will be interesting to see how that timing looks and how quickly they can have more distributed utilization.

The team is also developing automated tasks within the tool, including one that looks at the records of potential transfer patients to determine whether they can be received and others that could help evaluate patients for hospice placement.

As I was reading about the solution, I assumed that it would have metadata or citations to identify the origin of the data in the summaries. It sounds like that is a feature on the “coming soon” list, but I personally think that’s an essential piece that is needed to gain clinicians’ trust. I know plenty of physicians that don’t trust their support staff to take a patient’s blood pressure properly, which results in the clinician rechecking it on every patient, so doing the change management tasks that are needed to create buy-in from end users will be important.

Seeing expensive solutions in place that clinicians don’t use is one of the most frustrating things I saw regularly as a healthcare IT consultant, but I know that the “AI” label will create a lot of clinician interest right off the bat regardless of how robust the solution might be.

I’d be interested in hearing from other organizations who might be working on similar projects, or from EHR vendors that are also trying to make this happen. What information is the easiest to access, and what ended up being more challenging than you think? How are clinicians receiving the solution, and what kinds of enhancements are they asking for right away? If you’re a clinician, I’d be interested in your thoughts on this kind of tool and what you would need to feel that it was reliable. As always, leave a comment or email me.

Email Dr. Jayne.