Athenahealth CMO Nele Jessel, M.D., on How Ambulatory Clinicians Are Adopting AI Tools

Health IT vendor athenahealth recently published the results of a survey of ambulatory physicians and practice managers about their use of and comfort level with artificial intelligence tools. Nele Jessel, M.D., the company’s chief medical officer, recently spoke with Healthcare Innovation about the results of the survey and some of the challenges EHR vendors face in rolling out new AI features. 

Healthcare Innovation: Before we talk about the results of this survey, could you describe your role as chief medical officer at athenahealth? Does it involve a lot of communicating with clinicians about their experience and relaying that back to product development teams?

Jessel: I joined athena as chief medical officer four and a half years ago. I was actually a former customer. I'm a pediatrician and clinical informaticist, and I worked for athena enterprise customers for almost a decade, so I am very familiar with the product from a physician perspective as well as a practice administrator perspective. 

I work directly with our product development team, and I consider myself the translator in chief. Physicians and IT — and engineers in particular — speak very different languages. I take our customer feedback and translate it into product features, to make sure that our product closely aligns with our customers’ needs. We have a very customer-centric product development approach that's somewhat unique to athena, where we partner closely with our customers on the development of our product features and functionality. Because we're a single instance platform, we can test our product features very easily in real life, so to speak, with many customers at the same time. That makes it easier for us, compared to some other vendors who deal with locally installed solutions, to actually test at scale and get lots of feedback on our product development from our customers.

HCI: Before we dive into the survey results, I wanted to ask more broadly if for a company like athena and other EHR vendors, there's a challenge with calibrating the pace of AI feature adoption in the platform with the needs and the acceptance of the user base — figuring out how fast people want to go and which features it makes sense to prioritize first. 

Jessel: That is an excellent question, and I completely agree that is exactly the challenge, because people adopt things at varying speeds. There are the early adopters, then there's the broad middle, and then there are the laggards who probably never adopt something. If you're talking about product functionality, especially in a product like athena where everybody is on the same instance of the platform and everybody's looking at the same product, it becomes a little bit more challenging to figure out: how do you make it available to all but allow people to adopt it at their speed? We incorporate our AI features in a way that allows people to test them out if they're so inclined, but they don't have to use them. They're available, but if you don't want to use it and you want to do it the traditional way, we're not forcing you to use it. We give you both options. And the benefit of that is that you can more gradually build physician and clinician trust in general, because maybe after having some of your colleagues try it out and give you great feedback, you can try it out at your own pace and decide if you like it or not.

HCI: Athena is known for having a strong marketplace program of other vendors that integrate with the platform. Is there a challenge in deciding which things to build into the core EHR itself and which things to allow that array of vendors in the marketplace to innovate around and then people can pick and choose from those in the integrations?

Jessel: We are a firm believer in offering variety, because we serve lots of different customers with very diverse needs, and what works for one may not work for the other. We've always prided ourselves on our marketplace. In our open ecosystem, we have almost 600 marketplace partners that offer various solutions — some that we offer, but in a slightly different version, some that we don't offer, and that may add to our existing functionality for a select subset of customers. 

What makes our marketplace unique is that all our marketplace partners are also fully integrated with our software, so there is no integration work required. It's as simple as saying, “Oh, I like this. I want to try it out,” and signing a contract with the vendor. The switch gets turned on, and you can use it. So you can use our own athena integrated functionality or if that doesn't quite meet your needs yet, or you would like to try something different, you can try something else that's offered by the marketplace.

HCI: Let’s use the the uptake of ambient scribe technology as an example. There are a lot of vendors vying in that space. Do you have an experience of rapid uptake of that type of tool by your user base?

Jessel: Ambient mode is probably the No. 1 feature that providers currently talk about, and there's very good reasons for that, right? For years providers have complained about administrative burden. But it’s not just administrative burden, it’s also a lot of cognitive burden. If you think about a physician who's in a room with a patient, not only are they the meeting facilitator and have to run the show, but they also have to take the notes at the same time, and they have to make real-time decisions. The cognitive burden is really high to try to do all these things at the same time. Ambient removes the documentation burden, but it also removes that need to be the meeting facilitator and the note taker at the same time. So it's actually more about reducing cognitive load than true administrative burden, in my mind, and that's why it's become such a huge success, because physicians are realizing they can be fully present with their patients.

HCI: So across the board you're seeing widespread adoption?

Jessel: Pretty much, yes. People are trying it. I will say that in some ways it's not quite there yet, and there are two reasons for that. The models are all still learning, especially specialty-specific documentation. It's been pretty widely used in primary care — internal medicine, family medicine, and pediatrics, to some degree. It’s a little bit more challenging in pediatrics for various reasons. 

Specialties are still, with certain exceptions, not quite as happy as primary care physicians, because the needs of specialists are very different. Primary care physicians need to document a lot about a lot. Specialists need to document succinct, comprehensive information about very little but very specific things, and the models aren't great at that yet.

The other challenge is that specialists also tend to place a lot of orders and tests, and the models or the ambient solutions aren't quite there yet. They're starting to get to the point where they can actually queue up orders on behalf of a provider, similar to a real human scribe, but they're not quite there yet to do this in a way where the provider just has to quickly review and sign off on it.  But I think once we get to that point, that will be like the Holy Grail, right? 

What we're trying to do is leverage ambient to basically mimic a human scribe, which has been a lifesaver for many physicians for many years. The challenge was that a human scribe, of course, is expensive and difficult to train. If they leave, you have to train someone new — and you have the intrusion of another person in the room, and you don't have that with ambient.

HCI: Well, let's turn to this survey results. First, were there any surprises in the results to you about about how rapidly people are adopting AI tools or what they say they're working on? The white paper says 62% of practices reported using four or more AI-enabled tools. Does it seem surprising that it's that high?

Jessel: No, it did not surprise me, because that's what we're observing. It's like after that initial slow start, it’s taking off like wildfire. I think people really are starting to see the benefit, and most of us have become much more comfortable leveraging AI tools in some way, shape or form in our own personal daily life. So it's not surprising that it's being increasingly adopted in healthcare.

HCI: We’ve already talked about the ambient scribes. Are there some others areas where you are seeing high adoption — around coding or documentation?

Jessel: What I most frequently hear from providers, in addition to the ambient, which we already talked about, is the desire to have the AI find information in the chart and surface critical information at the point of care. We have made major advances in interoperability over the last couple of years. The EHRs have gotten much better at talking to each other. We've gotten better at information exchange. However, the downside to that is that it actually has become very challenging to make sense of all that information and curate it quickly enough in those 5 to 10 minutes that you have with a patient per visit to leverage it and make appropriate decisions with regard to the patients under your care. 

So providers have started to complain about information overload. Everybody finds it's great to have access to all that information, but they have expressed a strong desire to have help in curating this information, and we're finding that that is an excellent use case for AI, especially large language models, because they are excellent at quickly consuming vast amounts of information and distilling it into a digested TL/DR version. The second use case that providers have mentioned to us is: can you help me sift through the chart and surface information to me, rather than me having to go and look for it? We're finding that is very quickly becoming a popular use case for AI bots or chart bots to actually curate the information in a way that makes sense and can be easily consumed at the point of care.

HCI: So is this actually re-envisioning what clinical decision support looks like? One of the things the report talks about is having these summaries be context-aware about a specific patient’s experience. Are we seeing tools that are already doing that or is that on the horizon?

Jessel: I think that's on the horizon. But definitely the future that we're all hoping we'll be able to achieve is really re-imagining what the EHR could look like. Traditionally, EHRs have basically just been large information reservoirs — a little bit better than the paper chart, because we can get information more easily. We can also more easily send things electronically, such as prescriptions. So it's lightened the workload in that regard, right? But it's still a tool that today requires a lot of manual work, finding things, queuing up things, clicking on things. We're hoping that AI tools will help completely shift this paradigm, so that the EHR of the future will not require anyone to click around in it, but it will surface context-specific information in a fashion that will allow people to have a normal conversation and then maybe quickly stop and confirm that they didn't miss anything or there's nothing else they need to pay attention to — without having to do all the clicking and the hunting and pecking themselves. So that's the future vision. I think it's not that far away. 

But to your point, with more personalized clinical decision support, the AI tools could read the entirety of the chart, including perhaps some social identifiers — where does the patient live, what's the ethnicity, and then create customized recommendations. Today's clinical decision support sees the patient has an elevated blood pressure and says “consider the diagnosis hypertension. You should start medication X, Y and Z.” We could fine-tune that and make it much more personal. For this particular patient, you should really consider medication A vs. medication B. That is slowly starting, but I think it's a little a ways away still.

HCI: One of the issues that the white paper brings up is building trust, and it describes what it calls an AI trust ladder. Could you describe why that matters?

Jessel: It matters greatly for two reasons, One is patient safety. If providers can't trust that the solution is safe for patients, they are very unlikely to adopt it. The second reason is adoption. In order to be helpful with your clinical work, you actually have to use the tool. And we know from experience that it's not that easy to get physicians and clinicians to change their ways and adopt technology solutions. If we had spoken a year ago about our physician sentiment survey at that time, I would have said that the physicians were cautiously optimistic but very hesitant because of their experience to date with technology. They were very fearful that it would add yet another thing on their plate that they have to learn without any benefit, just more administrative burden. 

So the crucial thing was to get quick wins, where you give someone something, they try it out, and they find it helpful, and that sparks their interest. Then they want to try additional things. So the quick win in this case was the ambient. As soon as that started spreading, that sparked the curiosity, and people started to think of additional use cases. 

That leads us to that second stage of experimentation: How can you quickly confirm that something actually is helpful? We talked earlier about our alpha and beta testing process. We actually added a new thing, what we call a pre-alpha experiment where we very quickly started to test out new AI feature that we think are kind of cool with actual users to have them immediately give feedback. Is this worth further investing in? They can tell us if they could see this make a difference in their clinical realm. Six to 12 months ago, people weren't really quite sure what good clinical use cases were. But after the ambient success, everybody wanted more AI. But when you ask people what exactly, there was a lot of silence. There was not clear agreement what those valuable use cases would be. So with the experimentation, people will gain good experiences with the additional testing and experimentation, which increases trust and that makes them much more likely to adopt new features and also start thinking about what
else could be helpful. 

HCI: One other thing mentioned in the strategic implications section of the white paper involved effective AI governance frameworks. When I interview people about that topic, it’s at large health systems where they're setting up some kind of formal governance framework for AI. But I'm thinking of athena customers at a large or mid-size physician practice, and I am wondering if they have the capacity to create an AI governance committee to ask the vendors questions about transparency and what they need to be monitoring to be more confident in each of the solutions that comes along.

Jessel: Many of our customers are groups like you just described — small to medium-sized physician practices. And to your point, while large healthcare systems have many of these resources to do a procurement process, to have an AI governance council, our small groups don't have those resources. Our philosophy has always been that we are a software and service provider. We partner with our customers, and we take on many of those services on behalf of our customers. We have an AI governance council. We extensively test our solutions. We have a process for our marketplace partners. Because we want our customers to know that what we offer is safe for them to adopt, especially if they don't have the resources to do this due diligence themselves.

HCI: Is this report one that you do annually? Will we see another one next year? 

Jessel: Yes, we tend to do these annually, because it is cool to see the progression. It was so impressive to see the change in physician sentiments, especially from last year to this year, and it coincides with what we are observing on our platform, right? That’s the benefit of having direct access to our customers, because everyone is on the same instance of the software, so we can see what people use and don't use. We have a very comprehensive customer feedback process. They can give us feedback directly in the application. We have a customer success community where they can post it and we monitor that. Then, of course, we have customer success managers who partner directly with our customers and bring feedback verbally back to us. So we collect feedback from numerous avenues, and the annual survey is just an additional avenue, but it's very cool when it all matches up.