Norfolk, Va.-based integrated health system Sentara Health has been working for three years with a company called Regard, whose AI-based technology performs comprehensive chart reviews within the Epic EHR, with the dual goals of supporting clinicians and improving documentation accuracy. Joe Evans, M.D., Sentara’s vice president and chief health information officer, recently described in an interview how after initial pilots the solution has been deployed across all 12 Sentara hospitals. 

Healthcare Innovation: Could you start by talking about some of the challenges that your clinicians face that this work with Regard can help with?

Evans: I think one of the biggest assets that Regard brings to the table is their ability to do a very comprehensive chart review within seconds, right within the patient's chart, and pull out key insights and then really create a draft assessment of that patient, providing citations along with that. That would give a hospitalist, for instance, a much better grasp of a patient in front of them, without having to spend a lot of time pouring through the charts, going to the dreaded media tab to have to look at PDF documents that were scanned in, and then paint the picture based on the available evidence, to give an assessment of the patient.

HCI: So previously the clinicians would have to do that time-consuming digging for information on their own?

Evans: Exactly. We have varying levels of how people use Regard. It creates a draft assessment that you can go through item by item and say, yes, that makes sense, or no, that doesn't make sense, and then it drops from the list. If you just want to push that over into your note, you already have a starting point of the problems and supporting evidence. There's time saving on the note composition side. So not only is there benefit with the chart review, I'm getting help with the documentation and charting. 

Then there's the value of getting better coding specificity based on how you document. For instance, Regard provides the information for why it might be acute on chronic systolic heart failure rather than just congestive heart failure, which can lead to improved DRGs [Diagnosis-Related Groups]. This improves not only compensation for the work that was done based on the complexity of the patient; it also helps adjust some of the observed to expected variables around length of stay, inpatient mortality, et cetera.

HCI: So which of those is a bigger selling point to the health system — the chart review for clinical support or the improvement with coding and documentation?

Evans: My answer to that question would be it depends on which audience we're speaking to. I would stay the comprehensive chart review that might reveal things that they weren't aware of or maybe they haven't thought of, and then being a starting point for their note, honestly, is what wins the hearts and minds of our clinicians to help with adoption. 

So for that audience, it is the efficiency and the peace of mind of knowing that this AI solution provides that comprehensive chart review that it would take hours for a human to do well.

HCI: Can I interrupt to ask a follow-up before we get to the coding aspect?  The example you've described was about hospitalists. Does it make just as much sense to use this in primary care or in a specialist’s office, or is it in the hospital setting — where they have to do that chart review fast — that this has the most impact?

Evans: Hospital medicine is where Regard got started. We're just now starting to move outside of hospital medicine to other acute care needs, and then to primary care and specialists’ offices. I think there's great potential there. To be as effective on the ambulatory side, we need to get integration of external data through one of the interoperability frameworks, and Regard is working on that. For example, one of our specialists might get a patient referred from a local community hospital where they might have had some evaluation, and they would be able to pull all that in so before they go in to see the patient, they could have a much better understanding of the history, and then be able to add some clarifying questions. 

In acute care, we have definitely started having some other user types use it within the hospital, and I think that's going to grow over the near term. But the shorter answer to your question is that 95% of our experience at this point has been in the hospital medicine space.

HCI: OK, thanks. So I interrupted you when you were talking about the benefits from the administrative side.

Evans: So from the view of our CFO and hospital operations leaders, the pitch to them is to be able to show the hard ROI and the benefit of capturing the CCS and MCCs [Complications or Comorbidities and Major Complications or Comorbidities] to help with DRG upgrades, which helps with hospital reimbursement. And it's easy to map out to them, and that's what we did after the pilot. We could say this is what we spent on this solution, and this was our hard return on investment. And those results are what helped us be able to spread it through all 12 hospitals.

HCI: I think I saw a note that said this created a four time return on investment per user.

Evans: It depends on the level of adoption. I would say we have consistently been in the twos, but we have had some up into the fours, and that's recognizing that only about 40% of our patients are seen by the hospitalists. So as we spread this to other specialties throughout the hospital, presumably that will at least stay the same, if not improve, especially as we target specialties like cardiology or general surgery. They may not always look as holistically at the patient as they focus more on the surgical problem or the cardiovascular problem, and they may not capture the complexity of the other comorbidities.

HCI: For the hospitalists, is it required that they use it, or do you have to convince them to add it to their workflow? 

Evans: Well, first I would say it's hard to require a physician to do much, you know. It was really getting some of those stories, creating some enthusiasm around it, having some physician champions who might be informal leaders. I'll be honest, adoption has been variable from some hospitals to the other, and it depends a little bit on the culture of some of the hospital medicine leaders there. But in general, thinking about all the things that we have taken from implementation to adoption, this has been one of the easier asks because it is so impressive. 

Also, Regard has been very responsive to the users. From the application inside Epic, the hospitalists can type in, ‘Hey, this doesn't make sense because of X, Y and Z and submit it, and within a brief time, they get a response. It might say you’re right, that doesn't make sense. There might be a tweak to one of the algorithms, and then that rapid cycle gets put into production. So I think the fact that they get rapid feedback really helps with the adoption and inspires confidence. 

HCI: Does Sentara have an AI governance committee that reviews solutions like this before deployment?

Evans: We have what we call the Sentara AI oversight program, which is an advisory body to look at what we call trustworthy AI, and there are eight components to that. After that review, we take it to a group of four executive leaders to make the final determination: Our CIO, our chief physician leader, our chief consumer officer and our chief legal officer, so they're the final body that says, yes, this makes sense. Let's move forward with it. 

Then once we put something in our production environment, our AI oversight program also has a responsibility to make sure it's not only performing as it was expected, but it's also not drifting or changing over time based on information that we're feeding it.

HCI: Are there other pain points or challenges in the health system that you think AI solutions might help with?

Evans: I think the next big opportunity for the players in this space is to create not only context-specific patient assessments, but then through clinical guidelines be able to offer the next best step in care, whether it be a medication or further testing. 

HCI: So kind of what clinical decision support guidelines do now, but boosted by AI and making that even more context-specific to the patient? 

Evans: All of our clinical decision support is mainly reactive — for instance, we realize we don't have enough patients on guideline-directed therapy for heart failure so we create decision support based on the guidelines. But that would just be the one problem we fixed, whereas this would be aggregating the American College of Cardiology and all the different big society guidelines and putting them in this corpus of knowledge. Then the AI has the ability to give that patient context, and be able to point to the corpus of knowledge that it was trained on and offer that next best step of
care. 

HCI: Are there vendors working on that right now?

Evans: Yes, there are several. I'm not sure that everybody has a comprehensive solution set for that. They're starting with a handful of conditions, but I think that's really where the puck is headed and where I think all these folks should be skating toward.