“This Is Real Innovation”: Q&A with Michael Hasselberg, CDHO, University of Rochester Medical Center

When Generative AI was first introduced in the fall of 2022, there was a lot of excitement and a lot of expectations among the healthcare community – including Michael Hasselberg. “When I first played with this technology, I thought, ‘this is amazing. This is really going to be transformative,’” he said. And although his organization (University of Rochester Medical Center) is still in the “experimentation and piloting” stage, he’s already seen a tremendous impact – but not necessarily with the technology itself.

“I think the biggest transformation is that it has everybody talking about data and the importance of data,” noted Hasselberg, who serves as Chief Digital Health Officer as well as Co-Director of the UR Medicine Health Lab. As a result, stakeholders are having critical discussions around the key pieces that need to be in place, such as privacy, governance, and data warehousing. “That, to me, has been the biggest win.”

During a recent interview with Kate Gamble, Managing Editor at HealthsystemCIO, Hasselberg talked about how his team hopes to leverage AI to improve efficiency and care quality, while enabling nurses to operate at the top of their license. He also discussed URMC’s groundbreaking initiative to expand patient care into the community; how they’re working to build buy-in among users; and why he believes it’s so critical to focus on nursing innovation.

LISTEN HERE USING THE PLAYER BELOW OR SUBSCRIBE THROUGH YOUR FAVORITE PODCASTING SERVICE. 

Podcast: Play in new window | Download (Duration: 21:24 — 11.7MB)

Subscribe: Apple Podcasts | Spotify | Android | Pandora | iHeartRadio | Podchaser | Podcast Index | Email | TuneIn | RSS

Bold Statements

That’s real innovation and it’s really been an ‘a-ha’ moment. We’ve already had several health systems reach out to us. They want to replicate this in their communities because it makes sense.

What I’m most excited about is where we are right now with generative AI and the large language models. We’ve done a lot of testing with these new AI models through our innovation team at the University of Rochester, and we’re really blown away with the potential of what this technology can do, especially on the administrative burden side.

It’s probably the first time in my career thus far that I’m hopeful that technology will actually make the lives of our clinicians better.

We also have to think about introducing this into workflows. This is one of those situations where technology is, in many ways, the easy part. The hard part is how do I fit it into our current operational workflows? How do I get cultural buy-in so that people trust this ‘AI co-pilot’ that’s eager to help out?

Now, everyone is talking about AI, and that’s allowing us to have discussions around what does our enterprise data warehouse look like? What about privacy and security and risks? How do we set up the appropriate governance structure? We’re now making investments on the data side, and that’s going to be key.

Q&A with Michael Hasselberg, CDHO, URMC

Gamble: Hi Michael, thank you for putting aside some time to speak, I appreciate it. I know things are really busy.

Hasselberg: We had a big announcement that came out recently. We have a novel, innovative project where we’ve put virtual care stations in banks in rural communities. We’re doing it in partnership with two startup companies, a global telecommunications company, and a bank. We’re the first health system in the country to offer primary care in a bank. There’s a rationale behind it. We’re really excited.

 

Gamble: That’s very cool. Can you tell me a bit more about that and how it came about?

Hasselberg: Sure. During the pandemic, like every health system in the country, we had to turn on telehealth overnight. And as discussions started around the public health emergency being lifted, we became concerned that telemedicine reimbursement would go back to the pre-Covid state. We did a deep dive into our data and found that where we were making the biggest impact was with our Medicaid patient population, especially in rural communities. But there was a caveat; in rural communities, patients were engaging in audio-only visits. They weren’t using video. Initially we thought it was a broadband issue, but it wasn’t. The infrastructure was there.

As it turned out, it was a poverty issue. Patients couldn’t afford broadband internet in their homes. And so, the only access they had to telemedicine was through the data plans on their cell phones. But video is very data intensive, and folks didn’t want to use up their plans. And we knew that audio only reimbursement was going to go away. And so, we had to think through where we could set up access points in these rural communities to continue providing services to these patients where we could do video.

Retail’s challenges

As you know, a lot of retailers have tried to do this, but they’ve struggled with the financial ROI, because they calculate per square footage based on how much product they have to sell. And if you use that space to deliver healthcare, it’s really hard to get the volumes that you would need to offset that product loss.

At that point, we went back to the drawing board and asked, what do they have in rural communities? They have a Dollar Store, traffic lights, and banks. And so, we did a deeper dive into the banking industry and some of the struggles they face from a brick-and-mortar standpoint, and in terms of their own digital transformation. We learned that small, rural communities still need a brick-and-mortar bank, because small businesses drop off their deposits on a daily basis. And so, banks are trying to reinvent themselves to figure out how we can get more foot traffic.

The scalability factor

One of the nice things about banks is they’re set up in a branch distribution model, which makes them more scalable. As a health system, I only have to negotiate with one bank and potentially have access to all their branches versus doing this with libraries or barbershops or community centers — they’re all one-offs.

And so, we figured out which bank had the largest market penetration and approached them. We said, ‘we’d love to provide healthcare in your banks.’ And they loved the idea because they had real estate that wasn’t being used. They’re not selling a product. And it’s a trusted location in these communities. Patients trust banks.

 

Forming strategic partnerships

Gamble: Who were the partners you’re working with?

Hasselberg: First, we partnered with a publicly traded community bank called Five Star. We also partnered with Higi, which makes the biometric stations you see at Walmart and RiteAid, and DexCare, a software platform provider. We integrated Dex Care into the Higi stations. Whereas before you could only get biometrics at those stations, now you can do telemedicine as well. We worked with Verizon to overlay the connectivity to those stations.

We now have virtual care stations in three rural banks, where we can provide primary care with no need for a healthcare provider on the other end.

 

Gamble: Do patients need to schedule the appointments? How does that work?

Hasselberg: They don’t. It’s on-demand, urgent care. Patients can get their biometrics collected, and the data comes back to us at the health system for primary care providers to see. We’re really excited.

We think that this may help solve some of the rural access problems. At the same time, we hope to address financial insecurities. For example, if a patient screens positive for financial insecurity on the social determinants of health questionnaire, we can refer that patient to one of the financial advisors in the bank. Our hope is to address physical and financial wellness at the same time.

On ‘real’ innovation

Gamble: There’s certainly potential there for a win-win situation. Very interesting. I had originally contacted you to talk about some other topics, like innovation, and this teed it up quite nicely, because you’re talking about solving problems with tools that exist.

Hasselberg: That’s real innovation and it’s really been an ‘a-ha’ moment. We’ve already had several health systems reach out to us. They want to replicate this in their communities because it makes sense. There’s a bank in every rural town on Main Street by the traffic lights.

 

Nurses @ViVE

Gamble: Absolutely. I also wanted to talk about ViVE. I noticed you’re presenting as part of the Nurses @ViVE program, which is new. That’s so great to see, because we know there are so many challenges for nursing, and it doesn’t always get as much attention at conferences. Why is this so important to you?

Hasselberg: First and foremost, I’m a nurse by background. And so, I was really excited to see ViVE and HLTH make an investment to support nurses and draw attention to how technology can make the lives of nurses better. That’s really exciting. When they reached out and asked me to join this inaugural nursing initiative, I said, ‘yes, absolutely.’

“Excitement” around AI

What I’m most excited about is where we are right now with generative AI and the large language models. We’ve done a lot of testing with these new AI models through our innovation team at the University of Rochester, and we’re really blown away with the potential of what this technology can do, especially on the administrative burden side.

One of the things we’re looking at is using AI to triage patient messages. Right now, our nurses are doing that. To me, having nurses manage messages and triage them to the appropriate person is using a nurse at the lowest scope of practice. And it’s not very rewarding for them. We’ve already developed a tool build off of GPT4 that can triage messages just as well, if not better, than nurses can, and it’s much more efficient.

We’re also spending some time with ambient documentation tools. That’s really exciting; not just for nurses, but for providers as well. We’ve been dabbling in virtual nursing for a while. As you probably know, the early virtual nursing models were essentially telesitting models; the idea was to have a central nurse watch 20 or so rooms using video conferencing so they could identify patients who might be at risk for a fall and call for help. We’ve been doing that for quite some time.

The progression we’re seeing now in that space is really exciting. Hospital rooms already have cameras and other sensors; by overlaying AI on top of that telesitting model, you could have one nurse watching 200 rooms and use AI to detect potential issues. That’s something we’re really interested in.

“That’s something AI can fix”

Another area is the operating room. Nurses in the OR have to document so many things: what time everyone washes their hands, what time patients roll in, when the first cut happens, etc. That’s another example of utilizing nurses at the lowest scope of their practice.

And beyond that, the data that they’re entering into the computer isn’t consistent because it’s being done by a human. And so, oftentimes, there’s noise in that data. And when we try to use the data for modeling down the road, it’s very difficult. That is something that AI can fix right now by using computer vision to see when people wash their hands and timestamp that, and then consistently enter that data into the record the same way, every single time. That in turn allows nurses to do patient care, which is what they went to school for and what they want to be doing. It also structures my data in a way that I can develop further models that may be able to support clinicians while also someday being able to predict patient outcomes.

Making clinicians’ lives better

That’s where I’m pretty bullish and pretty excited right now. It’s probably the first time in my career thus far that I’m hopeful that technology will actually make the lives of our clinicians better. When I think back on all the other technology we’ve introduced into healthcare, the intention was good — to make the lives of clinicians better. But most technology hasn’t lived up to those expectations. Our experiments and pilots with these new AI tools leads me to believe we’ll really be able to transform the lives of our workforce.

 

“Safety and trust” with AI models

Gamble: That’s very powerful, what you just said. And honestly, I’ve had a lot of people say that technology has not made things easier. That certainly wasn’t the intention, but it’s the reality. That’s very telling.

Hasselberg: One hundred percent. Now, there is still a lot we need to work through. There are nuances to this, and there’s a lot we don’t understand. A lot of these models are essentially a black box. We don’t know how they’re coming up with some of these outcomes. And some of the models fabricate things or hallucinate, which can have detrimental outcomes. With some of our pilots that we run outside of a production environment, we’re seeing the models start to degrade over time, and things are starting to get caught in the filters.

Truly understanding the safety and trust around these models takes a lot of effort, and we want to make sure we get that right before we deploy these at a large scale. And so, we’re spending a lot of time thinking through ethical concerns. We ethicists on our team who are helping us think through those types of implications.

“Technology is the easy part”

We also have to think about introducing this into workflows. This is one of those situations where technology is, in many ways, the easy part. The hard part is how do I fit it into our current operational workflows? How do I get cultural buy-in so that people trust this ‘AI co-pilot’ that’s eager to help out? Those are some things we’re still working through.

 

Building buy-in through low-hanging fruit

Gamble: And what are some of the ways in which you’re addressing that and trying to build buy-in?

Hasselberg: We’re targeting some of the low-hanging fruit situations where, if the model gets it wrong, the consequences aren’t as significant. For example, using generative AI to help generate a response back to a patient. There’s still a human to double-check and approve it before it goes out. Clinicians are excited about that because writing a letter to a patient is time-consuming. To have AI tackle the first draft and clinicians simply have to read through it has been great.

Interestingly, in some ways AI is even more empathetic in its responses, and the clinicians appreciate that. That’s helping to get buy-in. If we can get wins there, we can start thinking about more complex applications for AI. That’s been our approach: where’s the low-hanging fruit where we can remove something from clinicians’ plates while still keeping them in the loop? We want to take some of those tasks away so they can spend more time with patients.

The other key is transparency. We need to think at a system level about what our AI governance looks like and who sits on that. For me, having representation from ethics is really important because as clinicians, we hold ethics very strongly. Having that representation on our governance team can go a long way toward building buy-in.

 

“It’s got everyone taking about data”

Gamble: It’s interesting to me how the discussions on this have evolved during the past. A year ago, we were still in the excitement phase with things like ChatGPT. Now it seems like people are really looking at the implications. I think it has taken a good direction.

Hasselberg: It is. When I first played with this technology a year ago, I thought, ‘this is amazing. This is going to be really transformative.’ Reflecting on the past year, I think we’re still in the experimentation and pilot stage; we haven’t done the type of broad transformation that we hope to do. I think the biggest transformation for my health system is that it has everybody talking about data and the importance of data.

When I think back to a year and a half ago, the idea of getting buy-in from the entire system on the importance of robust data governance and data structure — that was hard, because we were so busy. But now, everyone is talking about AI, and that’s allowing us to have discussions around what does our enterprise data warehouse look like? What about privacy and security and risks? How do we set up the appropriate governance structure? We’re now making investments on the data side, and that’s going to be key. If health systems want to move to being value-based, personalized, and patient-centered, you have to have good data to drive that. To me, that has been the biggest win with Generative AI — forcing systems to invest in the data side of things.

Gamble: I hope that will be reflected as we enter conference season and see how those discussions go.

Hasselberg: I agree. It’ll be interesting. Over the past year, I feel like every panel has been about Generative AI. We almost have Generative AI fatigue at this point. I’m really interested to see what else people will be talking about.

From a ViVE standpoint, I’m really excited about the nursing component to it. We’ve had a nursing shortage for as long as I’ve been in the profession, and it has just become exacerbated. And so, to highlight nursing innovation and the fact that nurses are on the forefront of disrupting and transforming healthcare is so important. I also think giving nurses the opportunity to see what’s happening outside of their discipline is going to be huge, because nurses really are at the heart of healthcare. They’re the true drivers of healthcare. We need nurses to be leading the way. And so, I’m excited that ViVE is making that investment, and I hope other organizations do the same.