Epic President says EHRs are not walled off, integrations are just hard

At this year’s HIMSS conference in Las Vegas, Epic President Carl Dvorak sat down for an interview to discuss the Verona, Wisconsin-based vendor’s approach to digital health, startups and artificial intelligence.

Dvorak joined the EHR company in 1987 and has held the role of president since 2013.

This exchange has been lightly edited.

MedCity: Integration with the EHR seems to be a big challenge for new technologies that aim to disrupt the healthcare industry. How can Epic make life easier for startups that have the exciting new tech but find EHRs walled off?

First off, I don’t really think EHRs are truly walled off in any way. What a lot of startups that I watch experience is coming to grips with the real complexity of healthcare — the real issues around patient safety, the real issues about a notification process should you find your app has a problem in it.

I don’t think it’s walled off, I just think it’s hard. For many people, their first realization of how hard it is comes when they’re trying to put a product into place.

The other part of your question is important though, in that we have taken pretty significant measures to make room in the ecosystem for third parties to thrive, be they startups, be they established companies. One of the documents that Meghan [Roh, Epic’s director of public affairs] gave you is the current participants in App Orchard. These are startups to significant firms that have now been putting products into App Orchard. So it gives them full access to APIs, documentation, a test harness. Then we put them through the rigors of patient safety notification protocol. We put them through scalability and sizing. We give them some coaching on data stewardship.

[Currently, there are more than 200 vendors participating in App Orchard. Fifty-two apps are listed on the App Orchard site.]

The general consensus of EHRs is that they were conceived as billing tools. What is Epic doing to change the functionality so they help in decision support and in improving population health?

Nothing, because it’s a false premise to start with. When I joined Epic 30 years ago, I worked on clinical systems. We’re a clinical company first, which may be different than others. Our history was clinical and clinical research. Only later did we add billing. And it wasn’t until 2002 that we added hospital billing, and we had well over 150 EMR clients by then. It’s a false premise. It’s a narrative. It’s not real.

That said, to actually build clinical systems, we do significant immersion programs. Our programmers spend time every single year out on site shadowing caregivers. Our developers are required to learn the domain because you can’t design good clinical systems without good clinical knowledge and people. And we work with thousands of doctors and nurses and therapists and rehab specialists every year to make sure we’re designing software that makes their life better.

In so far as the workflow and making it easier, or helping them more broadly?

Everything. Connecting, first and foremost. So much of healthcare is disconnected even within the walls of an institution. So, first and foremost, helping them understand the upstream and downstream impact and implications.

Helping everyone see a holistic picture of the data and then at the same time remembering what’s important to a therapist might be different than what’s important to an anesthesiologist than a surgeon. So, constructing views that make sense to them.

That’s one of the areas that we’re doing a lot of AI work in. Much of AI discussion right now is a little bit speculative. But one of the areas that we’re focused on is something we call relevance. We’re looking at millions and millions of clinical encounters to understand, what was the information that the person looked at? And marrying that up, with what did they order? What did they talk about? What did they put as diagnoses? We’re using machine learning to scour what people have been doing with the technology and using that to have the computer give them what they’re likely to want before they ask for it.

What is Epic’s vision for AI going forward?

Relevance is one of the big ones. We have about 10, 20 predictive models in production already, where the computer is able to peek into the future, and based on what we’ve learned so far, we can relatively accurately predict several things that are important to caregivers: readmission risk, patient deterioration, some socioeconomic things.

Clinical surveillance of subtle issues like sepsis and patient deterioration. The computer can sense it before the person can see it. Small signals can be meaningful.

With our group of users, we have a lot of academics that are focused on genomics. We’re doing a lot of genomics-based research. We know what the gene is, we just don’t know what it always means yet. So correlating the phenotypic data from the electronic health record with the genomic information.

We’re focused on making the possible practical in these domains.

Can you give a few examples of successful integrations of digital health into your EHR at U.S. hospitals?

Any particular domain of digital health you’re curious about?

Just broadly.

I think of this roughly in four buckets.

The first is content. Companies like healthfinch are doing a great job of having the computer figure out which prescriptions need a clinician to authorize refill versus which we can refill by protocol. They are one of the most popular App Orchard participants so far. We’re working with Mayo to do content for virtual care for chronic conditions also through the App Orchard and in partnership with some of our development teams.

Another category is visualization. Some of the SMART on FHIR [Fast Healthcare Interoperability Resources] growth charts [are] helping people see data not just have to interpret it in a table format.

Another bucket is forms or data collection devices. AJCC [American Joint Committee on Cancer] is a good example of a cancer staging standard for the country. We work with them to develop the cancer staging forms that come through to customers. [Cancer staging involves figuring out how much cancer is in the body and where it’s located.]

The last bucket is analytics. We do predictive models ourselves. We believe there’ll be thousands if not tens of thousands of them. So it’s a world where everyone should participate, and anyone who can make a better predictive model should. We use industry standards to incorporate them in. In App Orchard, you see a lot of FHIR, but also on the predictive model side, the format is known as R. It allows us to include other people’s predictive models into production in a standardized way.

How is Epic utilizing FHIR standards and how many hospitals have turned on that capability?

We’ve been longtime supporters of HL7. [HL7 is a set of international standards used for the transfer of medical data.] We also founded the Argonaut Project, which is helping take the raw FHIR standard and put it into Meaningful Use cases that simplify using it.

Our work with Apple on the HealthKit integration is FHIR based.

We have FHIR in production. It’s available to all of our hospital sites at this point. Many of the patient-facing apps that use FHIR have already been tested with Epic. We have sandboxes up there, free and open for people to test their apps against.

It certainly gets a lot of fanfare, but there’s a meaty part to it as well that people are beginning to use. We’ll continue to foster its development. It’s as important to us to use as it is for third parties to use with us.

What is Epic up to in the world of telehealth? Are you working with telemedicine vendors as far as integration is concerned?

There are two kinds of telehealth companies that you routinely hear of in healthcare.

Some are infrastructure and technology providers, and others are actually providing doctors. So MDLive, Teladoc — their business is to organize physicians who have time on their hands and to allow them to use their time to serve patient needs through their platform. Those organizations are actually selling doctor time to maybe a 300-doctor group practice who doesn’t want to make somebody stay up overnight in case there’s a phone call. We work with those organizations and our customers to integrate information to do interoperability.

On the other side of the fence, we work on telemedicine infrastructure and technology. For example, we work with Mercy Virtual. They’re actually private labeling technology from Epic to serve other Epic sites and other non-Epic sites to be a Teladoc to them. We do the technology, the infrastructure to allow our healthcare organizations to do telemedicine themselves with their own doctors.

Photo: Pixtum, Getty Images