Johns Hopkins has big plans for AI in epic chart summary

Yesterday, inside part one of our in-depth interview with dr. Johns Hopkins Medicine’s Brian Hasselfeld, the senior medical director of digital health and innovation and associate director of Johns Hopkins inHealth, discussed the role of artificial intelligence in healthcare overall.

Today, Hasselfeld, who is also a resident physician in internal medicine and pediatrics at Johns Hopkins Community Physicians, is turning his attention to Johns Hopkins itself, where he and a number of teams within the organization have implemented AI in environmental registration and patient portal applications. . They are working with EHR giant Epic to use AI for graph summarization – a major step forward.

Q. Let’s turn to AI at Johns Hopkins Medicine. You are using ambient scribe technology. How does this work in your workflows and what results are you seeing?

A. Certainly a very current space. We see that a number of products use a wide range of strategies. We are similar to many who took some early steps in this area, recognizing that healthcare technology really hasn’t done what it was supposed to do.

Most of the data shows that technology, at least for the physician, has in some ways done more harm, at least for our own healthcare workflows and experiences. So we’re trying to think about some of those parts where we can put technology back at the center and make it more fun.

Again, many have recognized the documentation burden placed on our physicians by the explosion of EHR content, both from regulatory requirements and from the general workflow in many large systems. So for most of our systems that do have that picked up by ambient AI, a listening device, the ambient part of it listens to a clinical encounter, whether it’s an outpatient visit, an ER history, or inpatient rounds.

And on the back end, the AI ​​tool, usually what is now known as a large language model, such as GPT, takes the spoken word between the multiple parties and constructs it into a new generative paragraph.

It uses the actual function of those large language models to generate a paragraph of content, usually around a specific prompt. Given that model, “Please write a history based on this medical background.” And we’ve currently implemented that in a number of ambulatory or outpatient clinics, in a number of different specialties, currently with our first product and probably thinking about how we use more than one product to understand the different levels of functionality.

Personally, I just had a clinic this morning and was fortunate enough to be able to use the ambient AI technology using a device, my own smartphone, with our EHR on the phone, and launch the ambient AI product, which goes to the meeting listens and a concept note, for which I am of course responsible and which I must review and edit myself to ensure clinical accuracy. It really makes that clinical interaction much better.

The ability to take their hands off the keyboard, look directly at the patient and have an open conversation about a very intimate topic, their own personal health, and really take their eyes off the computer and back to the patient, in in my thoughts, is the biggest benefit so far.

Q. Johns Hopkins Medicine also uses AI for draft responses to patient portal messages. Explain how doctors and nurses use this and what results they achieve.

A. This enterprise tool is intended for early adopters. It is probably well known to many who follow HIMSS Media content that patient emails or basket messages, messages generated through the patient portal, have exploded during the pandemic.

Here at Hopkins, we saw a nearly threefold increase in the number of messages sent from patients to our physicians, from before the COVID-19 crisis in late 2019 to our run rate we are seeing now. And some of it is very good. We want our patients to be involved with us. We want to know when they are feeling good or not and help with triage.

But again, the clinical workflow, including payment models and clinical care models, is not built for this constant communication, this constant contact. It is built around visits. We have done something well-intentioned: increase connectivity with our patients. It’s a very easy modality, something we all do every day: emailing and texting.

We are used to what we would call asynchronous or written communication. But we haven’t actually changed the other side. The unintended consequence was to dump all that volume into an unchanged clinical practice system.

Now we’re all trying to figure out how to accelerate improvement in that meaningful area of ​​physician burnout, while maintaining the benefit to our patients through freer contact with their clinical team.

So a message comes in. Some things are excluded, especially if they contain attachments and the like, because these types of messages are more difficult to interpret. And once the message reaches a member of the clinical care team, those accessing the pilot implementation of the AI ​​draft responses will see an option to select a draft response based on the content of the original message and then select the draft of the large language model. answer, based on some instructions given to it to try to interpret it appropriately.

As a doctor, I can choose to start with that concept or start with a blank message. Stanford just published an article about this, and he articulates some of the pros and cons quite well, namely that one of the benefits is the reduced cognitive load of thinking about responses to very routine types of messages.

We have also seen that physicians who have picked up this tool and are using it regularly report significantly reduced burnout in their basket and decreased physician well-being. But at the same time, I think there is minimal time saved at this point because the draft responses are only really applicable and really useful to the patient’s message a minority of the time. In the article published by Stanford this was 20% of the time.

We see our clinics range from low single digits to 30-40% depending on the type of user, but still much less than half. The tool isn’t perfect, the workflow isn’t perfect, and it’s going to be part of that rapid but iterative process to figure out how we can apply these tools to the most useful scenarios right now.

Q. I understand that Johns Hopkins Medicine is working on AI-enabled chart summarization, with an initial focus on hospital course summarization. How will AI work here and what are your expectations?

A. Of all projects, this is in the earliest phase. It is a good example of the differences in the application of the technology across the continuum of care and the depth of the problem being addressed.

In the previous examples, mood and concept answers in the basket, we are really working on a very brief transactional component of the clinical continuum. The single visit and the accompanying discussion, the single message and the drafting of a response. That is very limited data.

When we start thinking about that broader topic of graph summarization, the sky is the limit, unfortunately or fortunately, in the problem that needs to be addressed: the depth of data that needs to be understood. And again, that needs to be extracted from unstructured to structured.

The work we do as physicians each time we interact with the diagram is that we move through the diagram in different ways, extract what we think we need to know, and summarize it again. It is a complex task. We try to work on the most targeted area: during an inpatient admission you are essentially more time-constrained than in other versions of the chart summary.

It may be necessary at the outpatient clinic The chart summarizes 10 years of information, depending on the reason you come to that doctor or the reason for your visit. I had a new patient earlier today. I needed to know everything about their medical history. That’s a huge task for summarizing graphs.

In inpatients we have the option to create a certain time limit around what needs to be summarized. So not to even start with everything about the hospital admission – which can actually include the reason for admission, which can then go back to the rest of the chart.

During an admission, we have the daily course of your trip, your hospital stay and interval changes. These are covered in daily progress notes and in handouts between clinical teams. And we can limit the information that needs to be summarized to the things that change and happen from yesterday to today, even if it’s a lot of potential things – images, labs, primary team notes, consultant notes, nursing notes. team.

It is much more time-sensitive and still provides meaningful efficiency for the inpatient teams, and certainly identifies a known area of ​​risk, which is transmission. Whenever your clinical team changes during your hospital stay, which is common because we don’t ask doctors to work 72 hours straight in most cases, we have the opportunity to help support those high-risk areas.

So if we’re trying to limit range, and even here in this very range-bound case, a lot of work needs to be done to get a potential tool ready for actual use in nature. the clinical workflow, given, frankly, the breadth and depth of the data available. We’re just starting this journey, working with our EHR partners at Epic, and look forward to seeing what’s possible here.

To watch a video of this interview with BONUS CONTENT that is not in this story, Click here.

Editor’s Note: This is the seventh in a series of articles from top voices in healthcare IT discussing the use of artificial intelligence in healthcare. To read the first part, about Dr. John Halamka from the Mayo Clinic: Click here. To complete the second interview with Dr. To read Aalpen Patel at Geisinger, click here. To read the third, with Helen Waters of Meditech, click here. To read the fourth, with Epic’s Sumit Rana, click here. To read the fifth, with Dr. General Brigham’s Rebecca G. Mishuris, click here. And to read the sixth, with Dr. Melek Somai of the Froedtert & Medical College of Wisconsin Health Network, click here.

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.