Can AI cure doctor burnout?

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

SHIV RAO: William McDonough, the sustainable architect, had visited Carnegie Mellon in my junior year, and he told this story about how we’re all designers of our world. And one of his examples was this ophthalmologist in India who designed this revolving platform that he sits on and he does cataracts. These patients come in, they’re sitting on these chairs around him and he does a cataract, five minutes, he says spin, and his staff spins him and he does the next cataract spin and he just spinning all day long. By the time I’d heard this lecture, he’d given eyesight to over a million people.

RANA EL KALIOUBY: Shiv Rao went on to become a cardiologist, and has now founded the AI healthcare company Abridge. But it was this moment – back as an undergrad at Carnegie Mellon – that really started his journey.

RAO: So I remember leaving that lecture thinking that that’s like insanely incredible world changing impact that he was creating. That was my hard pivot to, okay, I think I want to be a part of something that number one is in healthcare and can do something so profound. But also, I wanted to be a part of something that could really create value at scale.

EL KALIOUBY: Shiv’s company Abridge is valued at 5.3 billion dollars. And while you may not recognize the company’s name – your doctors probably do. Abridge has partnerships with over 200 major health care systems. Among other things, it automates medical note-taking using AI. Basically their platform cuts out the time-consuming bureaucratic work doctors do, so that they can spend more time with patients.

To me this is a really powerful AI use case. Abridge deploys AI to do important yet mundane tasks – so that we can build stronger human connections. And they’re doing so on a massive scale. There’s so much to talk about with Shiv – from how AI can be a solution to doctor burnout to why vanity matters when building a business. So let’s dig in.

I’m Rana el Kaliouby – and this is Pioneers of AI, a podcast taking you behind-the-scenes of the AI Revolution.

[THEME MUSIC]

Hi, Shiv. Welcome to Pioneers of AI. I am so excited for our conversation.

RAO: Hey, Rana, it’s such a privilege to be here. Thank you.

Copy LinkFrom medicine to building impact at scale

EL KALIOUBY: So you have a long career in health. What was your path to entrepreneurship?

RAO: Yeah, it wasn’t that I was in college thinking I wanna start a company, was though thinking about wanting to be a part of a team that could create this impact, and wanting to create impact at scale. I sort of pivoted pretty late to wanting to go to medical school. Went to medical school, felt like it was rote memorization.

Didn’t see a lot of room for creativity, and I was kind of a little bit disillusioned on what I’d signed up for, to be honest with you. But then went to residency. I went to University of Michigan in Ann Arbor and I just fell in love with patient care. ’cause you’re just thrown right into the trenches as a resident and you’re in these life and death moments.

But while I was in residency, I programmed something in Ruby on Rails for the program director, and I remember experiencing some peasant level of product market fit because he said like, oh, this is, this is useful. Like, how should we pay you for this? And I remember thinking, well, all that clinical research I was thinking about or that I was doing, I’m just gonna focus all of my energy at this new thing, this new intersection of software and healthcare. And that’s where I think I intuited this idea of being able to create the kind of impact that ophthalmologist or some estimation of the impact for myself.

Copy LinkWhy clinician burnout has become a public health emergency

EL KALIOUBY: Yeah. I wanna talk about Abridge. So you saw this big problem in medicine, which is basically doctors burning out because of all of the administrative burden and paperwork that they have to deal with. How serious is this problem anyway?

RAO: Yeah, it’s a public health emergency. It’s that serious. It’s impacting all of us in some way, shape, or form, and we have to find a way to leverage technology to solve for it.

EL KALIOUBY: Shiv says that part of this public health emergency is burnout. According to one study, one in five doctors want to leave the profession in two years. Other studies show that a quarter of all nurses plan on leaving the field.

RAO: And when you ask them what is the reason, they’ll tell you that they went to medical school or nursing school to serve people, to creatively sit down, build relationships, and help folks in their most vulnerable moments.

Help them be the healthiest versions of themselves. And so much of the business of healthcare actually pulls them away from what they want to do. Instead it pulls them into processes that are important and are necessary, but can really create overhead — processes around billing, processes around revenue cycle, processes around clerical work essentially.

So it shouldn’t be a surprise that we’re losing people and they don’t want to come back to this profession. And it shouldn’t be a surprise that there are rural health systems that are shutting down because they can’t staff them anymore. And that means patients are having to drive five, six hours into the inner city hospitals to see the rheumatologist or other specialists who can save their life.

So we have to do something about this, and this moment I think is so exciting because I think the entire industry these last few years has woken up to this idea that technology is our way out. There’s actually no other way out. We have to find a way to assist and in spots to actually responsibly automate and to unburden people and help the system treat patients the way they need to be treated.

Copy LinkHow reducing admin work improves care for patients and doctors

EL KALIOUBY: Yeah. This clearly affects the physician’s ability to provide quality patient care. But this administrative burden also affects the patient experience. And one of the things that comes to mind is the series of studies where they showed the doctors who communicate with empathy are less likely to be sued, right?

So I would love to hear how this whole problem affects the patient experience as well.

RAO: Yeah. So our entire thesis is that healthcare is about people, and we don’t think that’s gonna change in the next five to 10 years. We think that no matter what happens with AI, there will still be a place for clinicians, for doctors, for nurses. My father-in-law recently was in the hospital.

I spent some time there. It was very eye-opening to see how important that role is, that nurses play, for example, on a daily basis. And no matter what kind of technology you might have in the operating room, you’ll still probably want someone overseeing that technology. So if that is true, then conversations, dialogues are one of the original signals in healthcare.

They’re one of the most historically untapped signals, and we believe you can build a platform on top of it. You can create all kinds of value on top of it. But she told me that him taking notes meant that she could feel free to be fully present, to make eye contact, to build that relationship with me, knowing that when she got home, she wouldn’t have to be paranoid about how much she remembered.

When someone asked her, what did the doctor say? What they would do when they got home, they would actually unpack all of his notes. They would Google all the big words and rewrite their stories in words they could understand so that they could go to the next doctor and retell it, feeling like the main characters as opposed to someone looking in from the outside.

So much of what she shared was about agency. Agency for herself, for her family, to feel like that main character. But then you think about the other side of the room. And clinicians burning out, doctors and nurses leaving the profession. And really what they’re wanting is agency as well.

Agency to serve their patients as best as they can, to do the best possible job. And every competing force burns them out. And so I remember when we first raised our seed round in 2019, our deck, which sort of looks like a high school yearbook photo now. The design ethos has leveled up for sure, 10 times since then, but there was a slide that still is like our true North slide.

It’s just like a clinician on one side of the room, the patient on the other. But as people, they’re engaging, and we’re creating value for both sides. The clinician is unburdened getting agency. The patient’s getting a summary that can help them better understand and follow through.

Copy LinkWhy Abridge bet early on AI for healthcare

EL KALIOUBY: Yeah. When did you recognize that AI is gonna be a key component of the solution for both sides.

RAO: Yeah, absolutely. We’re very much an AI native company, an AI first company. We started with that as a big part of our Why now thesis and why us.

So prior to starting this company, I was an executive at a large health system called UPMC in Pennsylvania. And I joke, but it’s true, I sort of violated Peter’s principle 10 times.

In five years I got to lead their provider facing innovation portfolio and also their investment portfolio. We were putting money into companies, startups of different stages, but also putting money into venture capital funds. And so I was getting exposed to a lot. I was learning osmotically about the different dimensions of business and healthcare.

And one of our investments was into Carnegie Mellon. We put a lot of money into Carnegie Mellon CS department to start a machine learning and health program. We got to fund professors doing different types of research at the intersection of AI and healthcare. The first person I funded is my co-founder and CTO now, his name Zach Lipton.

He’s an associate professor at Carnegie Mellon and being able to ride shotgun with those professors, I don’t even know if it was shotgun, but to ride along and to learn from them. It was an incredible kind of privilege for me. It helped me kind of sense and feel in my bones, have that intuition that something is gonna happen, but also that something was happening in 2017, 2018 that maybe was worth the bet, was worth taking a risk. So we started our company just months after, for example, Attention is All You Need was published.

EL KALIOUBY: That’s the paper that basically spurred all these transformer models that are now at the core of all of the large language models.

RAO: Yeah. And in day one of our company, we were leveraging pre-trained models that we were fine tuning, like BERT, BioBERT, Longformer, Pegasus, T5. We published a paper in late 2021. Zach, as first author, published a paper about what you could do with those models to actually unburden clinicians with a clinically useful note.

And so we were sort of early, and there is a refrain in Silicon Valley that being early is being wrong. But we are here to say that not if you don’t die, if you stay standing. It can sometimes lead to not just first mover advantage, but also it can lead to just being a more prepared mind with all this potential energy on what this opportunity actually is. So when the sky opens, you can run right into it.

EL KALIOUBY: In a minute, Shiv walks us through why his product is so innovative and how they provide real day-to-day value to clinicians. Stay with us.

[AD BREAK]

Copy LinkHow Abridge turns doctor patient conversations into usable notes

EL KALIOUBY: So walk us through how Abridge works. Say I’m a patient, I’m coming in to see you. What is Abridge taking care of and what do I get as a patient? What do you see as a physician?

RAO: Yeah. So if I saw you in clinic, the first question I would ask is, hey Rana, do you mind if I use this note taking solution called Abridge?

It’s gonna help me focus on you instead of all the distractions that get between us. The backdrop of our company is really trust and that is the currency that matters.

And so even at that individual user level, it’s about trust. It’s about telling, getting consent, and you’re about to hit record. And so it makes total sense that you’d want to do this. But then once the patient consents, and by the way, almost no one says no anymore. The zeitgeist, the culture has shifted so much over these last few years.

You hit record and you have a conversation about anything in whatever order. So it could be about family. It could be about the games this last weekend, the weather and your hypertension, and then about something else, and then your congestive heart failure. But regardless of how you talk, when you hit stop, you should be able to swivel your chair and your note’s there.

It’s waiting for you inside the medical record for you to trust and verify, edit, adjust as you see fit.

EL KALIOUBY: Amazing. Was there a lot of pushback when Abridge went to market early on?

RAO: Yeah. In 2019, we weren’t yet ready to deliver an enterprise-grade solution. Like the bar for good enough there is high. There’s so many boxes you’ve gotta check off. But we were ready to put something out there on the other side of the room. So we actually led with a consumer facing app, believe it or not.

EL KALIOUBY: Interesting. I didn’t know.

RAO: Yeah. So totally different go to market. Totally different motion.

But it’s been fascinating. These last three to four years, it hasn’t been hockey stick for us. It’s been more like telephone pole growth. And we’ve had to build a lot of muscles across different parts of the organization almost overnight, and generally speaking it should take a little bit more time to develop.

And it’s been awesome. Like we have a saying in our company, pressure makes diamonds and we’re under tons of amazing pressure. It’s a privilege. But over these last four years, what we’ve also seen is just AI move so quickly. So even that scenario I described to you, the real differentiation and the hard part is in the details and oftentimes the invisible details. It’s what we’re doing behind the scenes, like how we might be pulling information from your chart to understand who is Rana. Like, what’s Rana’s story? What are her allergies? What are her medications, what’s her family situation like?

Also pulling maybe information from your insurance plan, to better understand what this note needs to look like so that you can get the procedure you need as fast as you need it. Pulling information from guidelines, from medical textbooks to help the clinician hopefully ask the right question.

So if I’m in front of you, I would love to feel like I’m a superhero, like I know everything. And anymore, both of us, all of us have access to these new AI primitives that allow us to be that much more informed and equipped to help me think about what the right diagnosis could be. So a lot of this stuff ends up being invisible and that’s where a lot of the hard work happens.

Copy LinkWhat it takes to build reliable healthcare AI behind the scenes

EL KALIOUBY: I wanna go behind the scenes. This is an AI podcast and we love to go behind the scenes and kind of unpack the AI powering Abridge.

So are you building foundation models from scratch or are you fine tuning these foundation models?

RAO: For us, it’s about fine tuning. It’s about post training, it’s about being able to leverage what’s out there, being able to distill into our own models as much as we possibly can. We’re here to put the best possible product out, and so that generally involves a playbook that involves distillation, it involves fine tuning, it involves post training. And now that we’re at scale, so next year maybe we’ll do close to a hundred million conversations, let’s say. And if we touch a hundred million patients, well, for all those clinicians and patients who are involved, the edits that they might make, the adjustments a doctor might make to a note in the orthopedic surgery space.

They might want a different style, a different structure. They might need something different for revenue cycle. We might get edits from a coding expert — coders or folks who are creating the bills that end up getting to the insurance company.

EL KALIOUBY: Right.

RAO: And so we can learn from all of that. And so that’s I think the playbook that probably all our other AI native companies across industries are pursuing as well.

EL KALIOUBY: Yeah. So do you use these a hundred million conversations to train or kind of post train the models? What do you do with this data?

RAO: Yeah, so the edits absolutely — the post training piece — we’ll always work with these health systems to have permission. It actually took us a long time, maybe perhaps 18 months end to end, to build a de-identification pipeline that allowed us to even normalize data to the point where we could leverage these edits to deliver a better product more dynamically, more continually to any given user.

So the industry needs this technology so desperately. There are necessary barriers to be able to do this in a safe way. And some of those barriers, they just take time and we’ve been able to navigate them. And I think that’s also part of our early years story.

EL KALIOUBY: That’s very cool. So it sounds like you have many models that need to come together to work accurately and effectively, and that’s the contextual reasoning engine. Can you tell us a little bit more about how that works?

RAO: Yeah, it’s essentially context engineering, or what people call that now. And it’s being able to pull information context from seemingly disparate sources, but then orchestrate that information through pipelines so that we can create the best possible output. So just to maybe give you a sense of the complexity here, let’s say there’s a quality metric in healthcare called the PSI nine, and that looks at perioperative bleeding after you have a procedure. And historically, if a patient was on a blood thinner, the only way to make sure that they weren’t misclassified was to use a specific diagnosis code.

And that described a hemorrhagic, a bleeding disorder code that never really matched what doctors would say, like how they would speak to the patients in front of them. So then these folks, downstream of the doctor patient interaction — those are the folks we were talking about earlier. They might be called clinical documentation improvement or coders.

Exactly. They’d have to chase that stuff down and it would create so much friction. ‘Cause then they’d send me a tech, like an email or what’s called a query and be like, clarify this. What actually happened. In August of this year in healthcare, there were new diagnosis codes that went live — Z79.01 — you don’t wanna know this stuff, but these new codes came out, and now all of a sudden when a doctor just says, hey, this patient’s on this blood thinner, or they take aspirin every day. What Abridge can do dynamically in the conversation is map that spoken language directly to that new, cutting edge diagnosis code that just came out. So no more workarounds, no more queries, no more paper cuts back to me. We’re shifting a lot of workflows upstream into the conversation itself.

EL KALIOUBY: But by the way, has Abridge, and kind of helping with these workflows, has that affected the job of the coders? I’m sure you get that question all the time.

RAO: Yeah, healthcare is one of those industries where, as a system, we can’t hire humans fast enough to do all the jobs that are out there. So I would argue that it’s the most kumbaya industry, at least for the foreseeable future, for AI to be implemented. Like everybody wins when we have more automation.

I would say there aren’t too many casualties of war now. Certainly there were people who would literally, all they would do is what they call scribe. They would stand in the corner and write notes for people. And yeah, that job certainly has seen a pretty dramatic change over these last few years. I also think that there are so many jobs to do and so many new places to focus human creativity in healthcare that it’s gonna be a while until we see this be controversial in any way.

EL KALIOUBY: Yeah, very interesting. So I’m trying to think of the environment, like where, you know, a doctor’s visit, right? Like sometimes the environment is noisy. The doctor’s saying these big and complex words, you’ve got different accents. I’m Egyptian and my PCP’s Egyptian, and I love her. So we often banter back and forth in a mix of Arabic and English. How does Abridge deal with all these complexities?

RAO: Yeah. So that’s a great point. I was listening to an interview with the Shopify CEO this morning where he talks about how lucky you feel when you choose a problem you’ll never have the perfect solution for, ’cause you can just work on it forever.

And I think part of what AI companies do, especially AI companies who are ambitious and want to really build for years — and we tell our health system partners that we want to be their partner for the decades to come — is you choose these problem spaces where you know you’re always gonna be on a treadmill.

You’re never gonna be able to get off this treadmill once you get on. And languages are a really good example of that. We have to support all the languages out there. So today there will be tens of thousands of conversations using Abridge, but spoken in Farsi in Los Angeles, in Mandarin, in Vietnamese, in Haitian Creole, and Brazilian Portuguese in Boston. So we have to support all these different languages. We benchmark how good we are on as many as we can find the bandwidth to do that for. And I think we’re over 30 on that front, but we can support over 90. And so Arabic is a really good example, but once you’re in these conversations, they can be polyglot conversations.

You can be jumping in and out of multiple languages and we’ll still be able to create that clinically useful but also compliant output on the other side.

EL KALIOUBY: And that’s fascinating. We’re going to take a short break. Stay with us.

[AD BREAK]

Copy LinkHow mission driven culture fuels fast innovation

EL KALIOUBY: I wanna ask you about the culture at Abridge. ‘Cause it sounds like, I love what you said that you’re in this space where the problems aren’t gonna be done anytime soon, and you’re gonna be continuously innovating, kind of shipping new products or new features. How does the culture at Abridge allow you to do that? How have you built it in a way that you’re continuously innovating.

RAO: Yeah. It’s a really magical moment in the company where the mission is so visceral.

I think that values do underpin how we behave. Like we have a value, which is you have to taste good things to have good taste. And I think that applies as much to the machine learning engineer who just read a really amazing paper from an AI lab and is taking inspiration from it and pulling it into something that we could productize. It factors into how we think about what being premium means, like how do we need to demonstrate the best possible customer service and what we want to be known for in the industry.

So we have a Slack channel in our company called Love Stories. And every day, yeah, every day people are funneling in new love stories that are coming in from end users. My favorite one comes from this rural health PCP, and she wrote to us some time ago, she wrote, I was sitting at dinner last week and my son asked me, mommy, why aren’t you working right now?

I literally took my phone out and explained to him that Abridge is a new tool that lets mommy come home early and eat dinner with her family. I started to tear up and looked over at my husband, who then said, mommy’s gonna be able to eat dinner with us every night now. And we get stuff like that every day.

And it’s not just dopamine, there’s a more powerful neurochemical involved that’s longer lasting. There’s like oxytocin involved here. And so when you can put together being on the cutting edge, like surfing the latest amazing waves with AI, with this kind of impact that we’re creating in people’s lives. Clinicians, but also these are the most vulnerable moments of patients’ lives — there’s a lot of magic. And I think it gives us the ability to build culture, but also be very dynamic with how that culture needs to evolve over time.

EL KALIOUBY: Yeah, that’s very much like Simon Sinek. Like start with the why. It’s really powerful. Because you brought this up, there is a lot of push for this 996 culture in Silicon Valley and in tech. And I am curious what you think about that.

RAO: Yeah. I think that the most powerful pushes are bottoms up, where folks just get so much energy from what they’re doing that there’s work life integration. And that definitely describes me in my life and I’m so privileged that I get to feel that way. That there is no boundary between work and life at all. It’s all the same thing. And it all gives me inspiration. And so I was joking with our CFO Saga recently. It’s like, I don’t know, if we were to quantify this, what is this? What are we? And it’s like 6, 12, 7.

I think it’s because everybody is feeling the dynamism of the moment. And they’re feeling the pressure of the moment and they’re feeling the excitement of the moment that you can build the future so quickly.

And I think that there’s also the paranoia of the moment that we should all be feeling in certain industries that the window won’t be open for this long. Like, you just don’t wanna take this moment for granted. Every second matters.

Copy LinkWhat gives Abridge a durable edge in healthcare AI

EL KALIOUBY: Oh yeah. There are a lot of companies in this kind of healthcare automation using AI that are popping up like every day. What would you say is Abridge’s competitive moat, but also long term defensibility?

RAO: One important aspect of our product and our roadmap is this reality that in healthcare, clinicians, doctors like me — I still see patients once a month.

I’ll do a weekend shift and when I see patients, I know that I’m not compensated for the care that I actually delivered in the room in the hospital. I’m compensated for the care that I documented that I delivered. So these notes are bills. And so when you get and you start to think about billing and coding and hospital diagnosis, coding and revenue cycle and healthcare, you start to realize this is really, really complicated stuff that is really high stakes in a different way.

It’s not high stakes necessarily from the clinical outcome, life and death way, but it is a different type of existential issue, which is that health systems have the narrowest of margins, and so the difference between an Abridge note and some other note can be the difference between you getting credit and compensated for the care you delivered versus losing money.

And no money, no mission, as mission driven as these health systems are. But then even beyond revenue cycle, what can you do in the moment? You can actually bend their trajectory for the decisions, for the thinking, for the caring that’s actually taking place. You can help clinicians realize that the MRI is a better choice than the CT or vice versa.

You can help them realize that this patient in front of them sounds like 10,000 other patients in California getting treated right now, and they’re all getting diagnosed with this new virus. Do you wanna consider that? You can bend the trajectory for outcomes. So that roadmap allows us to think about data.

We also get to think about going deep. We mentioned post training earlier, but how do you personalize this? How do you start to recognize and deliver the note for the neurosurgeon that looks different than the note for me, the cardiologist or the primary care clinician. How do you start to learn from people’s edits and that personalization piece too?

It starts to feel like this technology knows you.

EL KALIOUBY: Yeah. I’ve been thinking a lot about memory too, as a moat. And it sounds like, right, if Abridge is helping you write your note every single day for various patients, it learns your style and it remembers. Are you —

RAO: That’s exactly right. And it speaks to how so much of the hard work is behind the scenes. If you’re using ChatGPT and you ask this question and it gives you a more informed answer because you’ve been working with it for a while, you’ve been going back and forth with it, and now it’s hard to leave, right? And so I’m reminded frequently of this Scott Belsky quote where he says that in the first interaction or aha moment with a product, you have to feel vanity. You have to invoke vanity, selfishness, and laziness in the end user. And vanity is actually a really interesting category of opportunity for us because these notes, they’re kind of like your business cards as a doctor in the hospital. You wanna sound thoughtful. You want it to reflect all that you are considering when you saw that patient. So yeah, there’s a lot that we can do that we are investing in to make this invoke all those things at the same time.

Copy LinkThe long term vision from saving time to saving lives

EL KALIOUBY: Amazing. What’s the moonshot you hope Abridge will achieve someday?

RAO: Well, if there are three acts to what we’re doing, one is to save time, improve the experience, but save time.

Act two is around being deflationary and saving time for the system at large. And it’s interesting, healthcare is one of those industries where there are so many middleman entities, there’s so much bloat, there’s a lot of waste, there’s a lot of inefficiency. And so how can this technology start to play a part in changing things and making it more efficient? And then there’s a third act and we wanna be able to save lives. And I say that as a doctor, and so I say that with utmost humility, but we believe we can. We believe that these are the most sacrosanct moments in healthcare. These conversations.

EL KALIOUBY: Amazing. That’s very inspiring. Alright. Last question, and this is a question I ask of all my guests on the show, and it’s a question I think a lot about — what does it mean to be human in the age of AI?

RAO: There’s this saying in healthcare that we want all of our doctors and nurses to operate at the top of their license.

And oftentimes that’s meant take yoga breaks, spend a little bit more time on rest and relaxation and you’ll be more present for your patient. And it’s easier said than done, but I think we are in a moment where AI is allowing us to move higher and higher up in the stack.

We’re allowed now to think about the things that truly matter. And I think it’s allowing us to re-evaluate what our values are, like what does truly matter in any given moment as a human. And I think it’s fascinating, but I think we don’t know. And I think we’re trying to discover this, like figure this out.

And hopefully it means we’re gonna be so high up in the stack that we’re able to live our values in a way that we haven’t before.

EL KALIOUBY: I love it. Super awesome. Thank you, Shiv for joining me in this conversation.

RAO: Thank you, Rana. It’s been awesome. It’s an honor.

EL KALIOUBY: Strong values are at the core of Abridge. And in my experience as an early-stage founder, they’re critical in any successful startup. But like Shiv said, there are so many other factors that contribute to the success of an AI company. Things in AI are moving SO fast and it’s pretty competitive out there!

To differentiate you need a defensible moat AND you need to see where innovation is headed. I often say to founders, if you wake up every day worried that the next version of ChatGPT will render your product obsolete, then it’s not really a sustainable idea. You want your product to get better as the underlying AI model advances.

Shiv’s moonshot is to save lives through Abridge. It’s an ambitious goal – but he believes with AI’s potential, it’s one he can achieve.