Interest in using generative artificial intelligence to help doctors document patient visits has exploded in the past few years.
Technology companies, in a bid to capitalize on growing attention, say AI clinical documentation tools cut down on the amount of time and effort doctors spend on administrative work in their electronic health records — tasks that could impede patient care and contribute to burnout.
But doctors aren’t alone in this regard. Nurses are the “scaffolding” of the U.S. healthcare system, yet many have signaled they’re considering leaving the profession entirely, said Shiv Rao, CEO and co-founder of AI notetaking company Abridge.
Last month, Abridge announced it was working with the Mayo Clinic and electronic health record vendor Epic to develop a generative AI documentation workflow geared toward nurses. Abridge’s software aims to help nurses manage their wide range of day-to-day work, like patient and team communication, clinical notetaking and data capture.
Rao joined Healthcare Dive to discuss how the product will handle nurses’ work, the growth in generative AI in healthcare and how to build trust in AI tools.
This interview has been edited for clarity and length.
HEALTHCARE DIVE: How do you envision this AI documentation workflow for nurses?
SHIV RAO: When I think of nursing clerical work, the first image in my head is a flowsheet. That’s actually where they spend most of their time. That looks really different to where I spend most of my time as a doctor: I spend most of my time in note fields inside the medical record.
It’s a lot of discrete data that they need to capture. They are measuring vital signs, the patient’s blood pressure or their heart rate or their temperature. They’re also taking histories and actually conversing with the patient and trying to figure out what’s going on with them. They’re also having to check off boxes from a compliance perspective related to safety measures that they’re measuring around the patient. In the ICU, they’re turning the patient every so many hours to mitigate the risk of bed sores.
The amount of work and the complexity of the work that they have to do is really, really immense. And so the technology and the solution that we’re building has to seamlessly flex and context switch as well. For example, monologue mode and verbalizations to dialogue mode and conversations. It has to not just summarize information, it also has to structure a lot of data.
And that’s why we’re so thrilled about this announcement with both the Mayo Clinic and with Epic, because we really need all three pieces of this collaboration to make something that can really work. We need the nursing expertise. This needs to feel like it was built for nurses, by nurses. And that’s where our partners at the Mayo Clinic come in.
It also needs to involve the latest and greatest generative AI models, but other types of technology as well to pull off a technology solution that feels like really good air conditioning. When it’s set right, you’re not aware of it. But then the third piece is getting all of that data back into the medical record. And that’s where being able to co-design and co-develop with Epic is an incredible privilege for us.
What are some things that the Mayo Clinic and Epic are bringing to the development partnership?
It takes subject matter expertise. It takes leadership from from not just nursing executives at Mayo, but also all the nurses who are on the ground and leaning in to help us co-design this. It also takes technologists, it takes the professors, the PhDs, the postdocs — all of the computer science stuff that we’ve got at Abridge to be able to connect the dots and think outside the box and design something that can create a lot of value. And then there’s the third ingredient, the ability to actually get those notes back into the medical record.
I think what’s fascinating about the nurse space compared to the doctor space is that two out of three isn’t good enough. I think, in many segments of the doctor market, two out of three is okay, you can get by. And that’s where there are a lot of companies like trying to build in that space too, because they might perceive there to be a lower barrier to entry. But in the nursing space, if you can’t do all three, you’re nowhere.
Using AI for clinical documentation is a pretty popular use case. How far along is the market when it comes to adoption?
Over the last 18 months, two years maybe, the entire market has recognized that this technology can create immediate value for their clinicians. And so health systems are able to very quickly test this technology and recognize the value. I think the word has spread very quickly.
What’s happened alongside that is folks have gotten a lot more sophisticated about this technology as well. Executives and even end users are recognizing and asking what’s next. Can you go deeper on a note that is specialty-specific? Can you recognize the latest FDA medication? Does this work in other languages? Can it create summaries at the right reading level for patients? How deeply does it integrate into my workflow? And so that’s also a part of the exciting moment right now where everyone is evolving so quickly with this technology.
I think that safety and the responsible deployment of AI is something that’s top of mind for healthcare executives across the country. Trust, we think, is the ultimate currency in healthcare.
As a company, we try to figure out — how can we manifest transparency in the ethos of the company from a scientific perspective? So we publish papers. How can we instill trust in the actual product user experience? There we have a feature called linked evidence, where every doctor can highlight a word or a sentence fragment or an entire paragraph of our generated output, and we show them where the evidence came from.
Because imagine if you had one sentence show up in your note that you don’t remember from the conversation you had with your patient — maybe it’s something they said and you were preoccupied and you didn’t hear. But when you see that in your note, you start to lose trust in the entire technology stack.
We think that’s very critical, if not table stakes, soon enough for any type of technology that’s generative AI or AI-centered that’s going to be deployed in an industry like healthcare.
