Latest / CareTalk: Healthcare. Unfiltered. / The Ambient AI Inside 300 Health Systems w/ Dr. Shiv Rao, CEO, Abridge

Transcript

The Ambient AI Inside 300 Health Systems w/ Dr. Shiv Rao, CEO, Abridge · CareTalk: Healthcare. Unfiltered.

• 0:07
  Speaker: Welcome to Care Talk, America's home for incisive debate for healthcare, business, and politics.
• 0:12
  Speaker: Today, we welcome Dr. Shiv Rao, one of my friends and one of the true creative entrepreneurs in the AI physician ambient listening workflow space.
• 0:26
  Speaker: Shiv is going to walk us through where we are with AI, what the impact is on a healthcare process, and actually what the impact's gonna be on patients as well.
• 0:37
  Speaker: Um, Shiv, welcome to the show.
• 0:39
  Speaker 2: Thanks, John.
• 0:39
  Speaker 2: It's a privilege to be here.
• 0:41
  Speaker 2: I appreciate it.
• 0:42
  Speaker: So you're a cardiologist, but you started as a history and film theory major.
• 0:48
  Speaker: How did that- Can you just walk us through that path?
• 0:52
  Speaker 2: Uh, yeah, absolutely.
• 0:53
  Speaker 2: I think-- So I, I was in college, I think I was subconsciously rebelling against all the Indian doctors in my family and in the world.
• 0:59
  Speaker 2: And so, um, I was studying history, and I got to-- I got a scholarship year.
• 1:05
  Speaker 2: I got to do critical theory, film theory classes, got to make music.
• 1:09
  Speaker 2: I was programming virtual synthesizers in the School of Computer Science and School of Fine Arts.
• 1:13
  Speaker 2: So you can't go to Carnegie Mellon and not do a little bit of programming.
• 1:16
  Speaker 2: But that's around when I realized I wanted to be a doctor, and so I pivoted pretty hard and took all of those pre-med courses and eventually became a cardiologist.
• 1:25
  Speaker: You are a cardiologist.
• 1:27
  Speaker: I believe you're, you still every so often will actually see a patient or two.
• 1:31
  Speaker 2: Yeah.
• 1:31
  Speaker 2: About once a month.
• 1:33
  Speaker: So how, how did you transition from-- And then, and then you're, you're, you're a cardiologist, and then as I recall, I think you went to UPMC as an investor.
• 1:43
  Speaker 2: Yeah.
• 1:44
  Speaker 2: So prior to starting Abridge, I got to do, like, corporate venture lab work at UPMC.
• 1:51
  Speaker 2: And by that I mean there was the venture piece where we were investing UPMC's dollars into everything from startups to VC firms as limited
• 2:01
  Speaker: partners.
• 2:01
  Speaker: Just so people have context, Carnegie Mellon's in Pittsburgh, and UPMC is an incredible, innovative, classic academic medical center that has actually been one of the leaders in- I would, I would argue kind of ac- AMC, academic medical center venture capital, having founded and spun out a number of companies.
• 2:23
  Speaker: And so- Yeah … uh, that, that, that, uh, uh, that UPMC Medical Center is actually pretty much a force in healthcare- Yeah … early stage venture capital.
• 2:33
  Speaker 2: Yeah.
• 2:34
  Speaker 2: It's, it's a very entrepreneurial health system.
• 2:36
  Speaker 2: And at, at the time, the CEO at the time had created this venture lab of sorts where you could invest, leaders could, but they could also build.
• 2:44
  Speaker 2: And so we had over 200 people when I was there, mostly engineers, product people.
• 2:50
  Speaker 2: We were building, building solutions to the internal systems problems, but the aspiration was always to take them into the market and try to, try to actually build real commercial engines.
• 3:03
  Speaker: And so how did-- You know, that's an easy job.
• 3:07
  Speaker: Like, why would you go from venture capital to founding?
• 3:10
  Speaker 2: Um, well, that, that was an, a really exciting job.
• 3:13
  Speaker 2: I got to learn osmotically from all these really, really smart people around me, and got to get the, the, the sort of taste for what it could feel like to really, really, you know, truly build something from, from nothing.
• 3:27
  Speaker 2: And I think for me, it's not about building anything.
• 3:31
  Speaker 2: It's not about being in an AI company.
• 3:34
  Speaker 2: It's not about, um, some new technology necessarily that, uh, I needed it to, to, to, to build with.
• 3:41
  Speaker 2: It was more about the problem, first and foremost.
• 3:44
  Speaker 2: And once we saw this problem, you couldn't sort of unsee the moment and what could happen with it.
• 3:52
  Speaker 2: And so we started Abridge.
• 3:54
  Speaker 2: I quit that job, so we didn't spin out of UPMC.
• 3:57
  Speaker 2: I quit the job to start Abridge.
• 3:59
  Speaker: But I thought, I thought you funded Zach Li- Your, your co-founder, Zach Lipton-
• 4:03
  Speaker 2: Yeah… Speaker: from UPMC.
• 4:05
  Speaker 2: Yeah.
• 4:05
  Speaker 2: So as an investor, we got to fund research at Carnegie Mellon University.
• 4:11
  Speaker 2: And so there was like a, a cohort of professors whose research we're funding.
• 4:15
  Speaker 2: But that was a different engagement than founding the company, than starting something from scratch.
• 4:20
  Speaker 2: And so in that sense, in some ways, like the company was a lot more clean, like a lot more straightforward from a cap table standpoint in terms of, um, having all the agency in the world, uh, to go figure this out from scratch, essentially.
• 4:35
  Speaker 2: But it was a, it was a really exciting m- time because we started the company in 2018, and it was, you know, when deep learning was really taking off.
• 4:44
  Speaker: What was that light bulb moment That you decided to, to, okay, your funding is research.
• 4:50
  Speaker: Carnegie Mellon, by the way, is a international center, machine learning, comp- computer science, and AI.
• 4:57
  Speaker: So it's really a fabulous kind of neighborhood of, of, of, of tech, tech thinking to fund great research.
• 5:05
  Speaker: But what, what was the light bulb moment when you decided- Yeah… "Okay, we're gonna go for this"?
• 5:09
  Speaker: Because even in 2018, while academically the runway was, you know, w- was getting closer, y- or we were closing to landing the plane of AI, it was still a little bit foggy as to what, where we, we, where we were gonna land.
• 5:24
  Speaker 2: Yeah, absolutely.
• 5:25
  Speaker 2: So we started the company in just a few months after Google Research had published their paper, "Attention Is All You Need." And transformers were certainly already being used in the research community, and these transformer-based models that we started with predated LLMs.
• 5:40
  Speaker 2: Um, and so we started with models like BERT and BioBERT, Longformer, Pegasus, T5.
• 5:45
  Speaker 2: We were spending millions of dollars on expert annotation so that we could fine-tune these models to be able to work for this use case.
• 5:55
  Speaker 2: But the aha moment or the light bulb moment for me from the problem standpoint was really twofold.
• 6:00
  Speaker 2: One is my, my dad's a cardiologist.
• 6:03
  Speaker 2: He retired 'cause he couldn't type fast enough.
• 6:06
  Speaker 2: Nothing crushes my soul as a doctor more than clerical work.
• 6:09
  Speaker 2: I just wanna focus on the person in front of me and deliver the best possible experience and outcome, and everything else is noise, and everything else is overhead, and it's an unfortunate part of the job, but can we find, could we find a way to build technology that could take, take a lot of that burden away?
• 6:26
  Speaker 2: But the other aha moment for me was a patient.
• 6:28
  Speaker 2: So in March of 2018 in my clinic, I saw this patient, 10-year history of breast cancer.
• 6:34
  Speaker 2: She was about to start doxorubicin, which could impact her heart muscle.
• 6:38
  Speaker 2: She needed clearance essentially from me as a cardiologist.
• 6:42
  Speaker 2: She was super nervous and anxious, crawling out of her skin.
• 6:45
  Speaker 2: So I asked her why at some point, and she told me that for the last 10 years, her husband had come to every single visit with a new doctor, except this one he just couldn't make it.
• 6:53
  Speaker 2: And I asked her, she's an English professor at the University of Pittsburgh, I asked her, "What does he do that's not obvious?" And she told me that he sits in the corner, he's quiet, he just takes notes.
• 7:02
  Speaker 2: And she told me that him taking notes meant that she could feel liberated to be more present, to build a relationship with the clinician, go back home and unpack all the notes, all the big words, rewrite their stories in words they understood using Google, you know, obviously pre-ChatGPT, and then go to the next clinician and feel like the main characters of their story retelling it as opposed to someone looking in from the outside.
• 7:24
  Speaker 2: And so the, the, for me, the aha moment was what she was struggling with in her chronic disease journey was agency.
• 7:30
  Speaker 2: It's, it's control.
• 7:31
  Speaker 2: It's being able to build relationships, be present, person in front of
• 7:35
  Speaker: her.
• 7:35
  Speaker: It's agency on both sides.
• 7:36
  Speaker: You know, the- Exactly.
• 7:37
  Speaker: Yeah … I think the cl- the clerical- Yes … the clerical piece is, can be 30% to 50% of a doctor's job.
• 7:43
  Speaker: Clerical meaning not just It's not just paperwork, it's not just or- order entry, it's not just dealing with computers.
• 7:50
  Speaker: It could be being on the phone- Yeah… or dealing with prior authorizations or all of the things that are taking you away from thinking, diagnosing, and caring for- Yeah … the, the, the- Yeah the, the patients.
• 8:01
  Speaker: But it's an interesting point of agency on both sides.
• 8:03
  Speaker: And just to go back just a tad, the transformers were the breakthrough paper, the "Attention Is All You Need" from Google- Yeah … that really allowed the acceleration that, that, that sort of laid out a clear path to accelerating accuracy, sort of reach and relevance of the early stage models that you were just talking about.
• 8:26
  Speaker: And sa- sadly, Google lost all of those researchers.
• 8:29
  Speaker: Mm-hmm.
• 8:29
  Speaker: But many of them ended up at the big, the big hyperscalers we now know of as Anthropic, uh, OpenAI, and hopefully soon Abridge.
• 8:37
  Speaker: Uh, the, the, the, the, um, that was really a light bulb moment for anybody who is in AI because un- up until then, the cost of compute plus the burden and the sort of irrelevance of the models was as heavy as, as, as, as, as the promise of intelligence was at the time.
• 8:56
  Speaker: And so what, what a fascinating time to, to start the company.
• 9:00
  Speaker: You've talked a little bit in other podcasts about how skateboard and music culture somehow affected your ability to build.
• 9:08
  Speaker: Now, I have talked to a lot of founders- Yeah … and no one has brought together skateboarding and music culture.
• 9:14
  Speaker: So can you explain yourself as a defendant on this question?
• 9:19
  Speaker 2: Yeah, it's a good question.
• 9:20
  Speaker 2: I, I, I think that the-- maybe, maybe if there is a thread Between founding a company and skateboarding, perhaps it's just around like risk-taking and risk tolerance and what you're willing to do.
• 9:34
  Speaker 2: And, um, I think the-- As a kid, I, I-- Skateboarding was my life in college, no question about it.
• 9:40
  Speaker 2: I spent more time doing that than I, than I did going to class.
• 9:44
  Speaker 2: And, um, for me, skateboarding was a chance to like get into a flow state.
• 9:50
  Speaker 2: When you're about to do a very dangerous trick, potentially, you're, you're not really thinking anymore.
• 9:55
  Speaker 2: And if you are thinking, you're gonna get hurt.
• 9:57
  Speaker 2: You have to find a way to just lose yourself in that moment.
• 9:59
  Speaker 2: And I think whether you're in the cath lab as a cardiologist, whether, you know, you're doing a procedure, whether you're just, you know, uh, rounding in the hospital and you're just in that flow state, those, those moments are so magical for all of us as professionals, and you wanna get as much of that as you possibly can.
• 10:16
  Speaker 2: And for me, um, I think that's, that's what I have right now with the company, but that's what I had with skateboarding at one point in time.
• 10:22
  Speaker 2: I also think that there's this refrain now that, that, um, w- what matters most in this new world with AI is agency and taste.
• 10:31
  Speaker 2: And taste is judgment, but taste is also sensibilities.
• 10:34
  Speaker 2: It's like what kind of, um… How do you see the world, and what, what kind of lens do you wanna put on top of all it?
• 10:41
  Speaker 2: And how do you wanna-- what kind of beauty do you wanna leave behind?
• 10:44
  Speaker 2: And, um, for me, art is a big deal.
• 10:47
  Speaker 2: And, and being able to think about music culture, um, being able to reflect upon all those influences, they find a way in somehow.
• 10:54
  Speaker 2: We just
• 10:55
  Speaker: did this- As, as the, as the tech founder between the Noguchi sculpture and the, and the, and the, and the tower of, of- … of a leaning tower of books.
• 11:02
  Speaker 2: Yeah, they look like they're gonna fall down.
• 11:04
  Speaker 2: But we just created a video, for example, to try to kinda convey our, our mission.
• 11:11
  Speaker 2: It's called "The Doctor Will See You Now." And the, the music that we used in it is like a, a, a beat that, you know, I g- a lot of folks who listen to hip hop sort of grew up on.
• 11:21
  Speaker 2: It's by Feature Premiere, and it's a, it's a beat that was used by Nas on his, uh, on one of his albums.
• 11:26
  Speaker: Oh, you're old school.
• 11:28
  Speaker 2: Yeah, yeah.
• 11:29
  Speaker 2: Yeah, absolutely.
• 11:31
  Speaker: Well, I, I, I, I think, I, I think if I was a venture funder of yours, I might be a lit… I would ask how many orthopedic incidents you had as a skateboarder before I underwrote your, your willingness to- Sure … 'cause with a son who spent much more time on skateboards than in school, than, than at, than at, than at school during different periods of his life.
• 11:50
  Speaker: Yeah.
• 11:50
  Speaker: Um, but I wanna, I wanna just- poke a little bit at the taste point because for those of us, for those of our listeners who aren't following, you know, the, the, the divide there, what, what, what taste often refers to is what's-- what makes-- What, what about the interaction with models, and tell me whether this is what you mean, remains what is human unique and sort of unique, I guess uniquely human.
• 12:18
  Speaker: Yeah.
• 12:18
  Speaker: When I think of the taste conversation, which is you can, you can… We've got almost infinite compute.
• 12:25
  Speaker: We've-- We're-- We don't have infinite information, but we certainly have access to more.
• 12:28
  Speaker: The machine has more access to information than anyone could ever have imagined, except probably the early founders of AI.
• 12:37
  Speaker: But what's unclear is how we retain what's human, and for me, that's inter- that, that ties into the conversation about taste.
• 12:44
  Speaker: Is that sort of what you're thinking about there?
• 12:47
  Speaker 2: Yeah.
• 12:48
  Speaker 2: I think it's that.
• 12:49
  Speaker 2: I think it's also to, to make it even more technical, it's how you actually point these models, how you decide to leverage them in the wild.
• 12:57
  Speaker 2: The
• 12:57
  Speaker: weights.
• 12:58
  Speaker: Yeah.
• 12:58
  Speaker 2: Yeah, the weights, and it's also the evaluations.
• 13:01
  Speaker 2: So, so for example, you know, the distance between a party trick which anybody can create right now and an actual enterprise-grade product of, you know, even in healthcare, um, even more so, is very, very vast Anybody can prompt a model to get an output, but actually building that into a solution that can scale is a very, very different story.
• 13:25
  Speaker 2: And a part of how you traverse that huge distance, which is the entire game anymore, is with evals.
• 13:32
  Speaker 2: And evaluations are basically how you convey your judgment, how you convey your values.
• 13:38
  Speaker 2: So if you're building a product around, for example, let's say, um, oncological, uh, prior authorizations, then you're gonna need to look through the stack and make sure that you are able to evaluate how good you recognize all the oncological terms, symptoms, medications, diagnoses, and procedures.
• 13:59
  Speaker 2: You're gonna wanna make sure that you
• 14:02
  Speaker: can- And cancer's complicated.
• 14:03
  Speaker 2: Cancer is complicated, and these words are complicated.
• 14:06
  Speaker 2: These prior authorization criteria are complicated.
• 14:08
  Speaker 2: What's important to you?
• 14:10
  Speaker 2: Is it, uh, recall?
• 14:11
  Speaker 2: Is it precision?
• 14:12
  Speaker 2: And how are you measuring that stuff?
• 14:14
  Speaker 2: How will you know when to sort of pull on the escape hatch because it's not working quite well?
• 14:19
  Speaker 2: When, when do you know that you actually need to improve?
• 14:21
  Speaker 2: How are you also fine-tuning and post-training with user edits and feedback that you're getting every day?
• 14:27
  Speaker 2: And then how are you evaluating those improvements to the model to make sure that it's-- they're not like the model's not veering in some direction that you c- you didn't anticipate in the first place?
• 14:36
  Speaker 2: That's really-
• 14:37
  Speaker: Humans still have a role.
• 14:38
  Speaker: I'm very excited about that
• 14:40
  Speaker 2: High level, our thesis is that healthcare is human, and we wanna bring those humans closer together.
• 14:44
  Speaker 2: And the most important humans are professionals, but, you know, obviously, even more importantly, it's patients at the center of everything.
• 14:51
  Speaker 2: And there's so much work that gets between them, and some of that work is clerical.
• 14:56
  Speaker 2: It's notes, it's orders, it's visit diagnoses, it's inpatient versus op status, it's prior authorization criteria, it's risk adjustments, clinical trial recruitment.
• 15:04
  Speaker 2: There's infinite number of jobs.
• 15:06
  Speaker 2: There was an American Journal of General Internal Medicine article that was published that suggested that doctors need 30 hours a day to just get all of those clerical jobs done.
• 15:14
  Speaker 2: But then there's the clinical jobs.
• 15:16
  Speaker 2: There, there's how difficult it is to keep track, to keep up to date with all the guidelines that are coming out, the, all the new research, and how do you keep up with, you know, all, all of the new expert opinions that are out there as well to help you feel like you're delivering the best possible outcome over time too.
• 15:33
  Speaker 2: And so high level, how can you make clinicians feel like superheroes?
• 15:37
  Speaker 2: So that's what we're all about.
• 15:38
  Speaker 2: So we try to basically build technologies that approximate a team of assistants, agents, if you will, in the background that can work 24 hours a day, seven days a week on behalf of every single clinician out there.
• 15:50
  Speaker 2: Maybe they're prepping charts for my clinic on Friday right now.
• 15:54
  Speaker 2: Maybe they're actually looking through to see what care gaps there are that I should absolutely address when I see those patients on Friday.
• 16:00
  Speaker 2: Maybe some of these agents are listening right now to an encounter that I'm having with a patient and not necessarily interrupting me, but creating a scorecard so that when I look down, I can see what gaps are still there that I should potentially address based on what the patient just said.
• 16:15
  Speaker 2: So our high level mantra in the company is, one, save time, two, save money for the system at large.
• 16:21
  Speaker 2: We wanna be a deflationary force that can make healthcare cheaper, better, faster.
• 16:25
  Speaker 2: But also, how do we save lives?
• 16:27
  Speaker 2: And so we have a clinical intelligence feature.
• 16:29
  Speaker 2: It's like clinical decision support basically, but souped up with all the context on who the patient is that we're pulling from any number of systems, including the medical record, um, but also pulling from the, the conversation if you already had one.
• 16:42
  Speaker: And, and the way I think about it, and correct me if I'm wrong, is that hospitals and doctor groups buy this platform or system that they-- that is built into workflow-
• 16:53
  Speaker 2: Yeah… Speaker: that then allows-- that i- I've, I, I, I've heard you kind of nudge back at this, an ambient scribe, as you first described it, a sort of an intelligent assistant that initially can just play that role of your patient's husband to listen carefully document accurately and perha- perhaps actually add some, some intelligence to it, um, as well as hopefully some heart.
• 17:22
  Speaker 2: And you've had a great deal of success.
• 17:24
  Speaker 2: I mean, your system is in some of the largest and most successful medical centers in the country.
• 17:30
  Speaker 2: I mean, can you just give people a sense of the scale of your installed base and kind of maybe some of the names that are actually using, uh, Ubridge right now?
• 17:39
  Speaker 2: Yeah, absolutely.
• 17:39
  Speaker 2: We're, we're incredibly privileged.
• 17:42
  Speaker 2: We have scaled across well over 300 large health systems in the country.
• 17:46
  Speaker 2: They represent the care of over 250 million Americans, those systems.
• 17:50
  Speaker 2: They represent over half the doctors in the country.
• 17:53
  Speaker 2: We are live at large systems like Kaiser.
• 17:56
  Speaker 2: We are also live at the VA health system.
• 17:59
  Speaker 2: We're also live on the other end of the spectrum with federally qualified health centers like AltaMed, for example, in California, and, and we wanna be able to cover everything in between over time.
• 18:10
  Speaker 2: Um, and so we've, we've earned this right.
• 18:12
  Speaker 2: And you mentioned, like, the notes is like, was the wedge, and, and a part of the reason why notes is the wedge is what you just said.
• 18:20
  Speaker 2: You can help both sides of the room.
• 18:22
  Speaker 2: We can help the clinician, but we can also help the patient with an after-visit summary, essentially, with patient instructions.
• 18:29
  Speaker 2: But what's been so fascinating is every time you think, every time I think I know everything that there is to know about how to create even a clinically useful and complete and compliant note, we learn something else because the ground's always moving.
• 18:43
  Speaker 2: Some new requirement just comes out from the American Hospital Association about how to code this new acute diagnosis in the inpatient hospital system, or perhaps AHIMA puts out a new requirement about what compliant documentation looks like for revenue cycle.
• 18:58
  Speaker 2: Like, the, the, the rules are always shifting, and now clinicians don't have to attend lunch and learns to learn revenue cycle and keep up with those rules.
• 19:07
  Speaker 2: AI can keep up with those rules so that they- Oh,
• 19:09
  Speaker: don't worry.
• 19:09
  Speaker: They w- you know they weren't going anyways.
• 19:11
  Speaker 2: Yeah, I know.
• 19:13
  Speaker: They-- Let's be honest.
• 19:14
  Speaker: Right.
• 19:15
  Speaker: But the problem you're trying-- I mean, the, the-- between the time that a, a, a new medical practice or- Yeah … innovation is validated by the physician authorities, the academies, the best research institutions, it takes anywhere between 15 and 17 years before it becomes broadly sha- broadly utilized across healthcare.
• 19:33
  Speaker: Healthcare, the information and innovation just on the clinical side, and you know this better than I as a doc, is-- increases probably two or three X every three years, and there's just no way that a, that a, that a, that a- Yeah … a 60 to 70-hour-a-week doctor who's, who's working Putting many of those hours into clinical care, plus all the documentation and insurance stuff, plus trying to think about new stuff has the ability to pull that in.
• 20:03
  Speaker: So that is a- Yeah.
• 20:04
  Speaker: That is really giving-- That, that's the- Yeah.
• 20:06
  Speaker: I think that's a great classical use case for the augmented AI- Yeah … where, where we're actually accelerating the relevance and reach of what a doctor and a, and a clinician can do.
• 20:17
  Speaker 2: 100%.
• 20:18
  Speaker 2: And my, my-- So my dad, I mentioned, re-retired 'cause he couldn't type fast enough.
• 20:21
  Speaker 2: But now the way I describe Abridge to my dad, I saw him yesterday, is imagine the, the smartest fellow you've ever worked with.
• 20:29
  Speaker 2: It's actually like the smartest attending you've ever worked with.
• 20:32
  Speaker 2: Uh, and imagine you've got a team of them, and you can just ask them a question before you walk in the room: "Who am I about to see?
• 20:38
  Speaker 2: What should I talk about?
• 20:40
  Speaker 2: What, what's important that I should remember, you know, about their condition or even about their, their, their personal story?" And then afterwards, that team will get all of the work done, and you can always ask it more questions, and it can even cue you with things you should be thinking about because it knows the patient's story.
• 20:56
  Speaker 2: So it's not just pull, it can even push to you.
• 20:59
  Speaker 2: So funny story, um, Abridge didn't make-- it didn't call me an idiot recently, but when I was rounding in the hospital, it very, very gracefully like cued me after a conversation for chest pain.
• 21:10
  Speaker 2: That was like the-- that's my bread and butter consult, obviously, as a cardiologist.
• 21:14
  Speaker 2: I r- I ordered a nuclear stress test.
• 21:16
  Speaker 2: Afterwards, it didn't call me an idiot, but it said, "Hey, you know, based on the information about this patient in the chart, based on prior nuclears that have had a lot of artifact when they've gotten those stress tests before, and based on their BMI, do you wanna consider a MIBI this time, a PET scan?" And it was 100% right.
• 21:33
  Speaker 2: That was absolutely the right thing to do.
• 21:35
  Speaker 2: I changed my mind.
• 21:36
  Speaker 2: I said, "Okay, change my note and my order to reflect that." That's how easy it is to make the changes.
• 21:41
  Speaker 2: But we probably sh- you know, saved a day and a half, two days of, of, of admission, you know, time for that patient, and we probably got to a better outcome, you know, sooner than we would have otherwise.
• 21:53
  Speaker: You know, you, you've said that, um, you s- you've talked about how docs aren't paid for care, they're paid for documentation, which is accurate, but the promise of, it sounds like of Abridge is you really want it to be that intelligent partner for doctors.
• 22:09
  Speaker: It's a pretty crowded space you're playing in right now, productively so, I would guess.
• 22:14
  Speaker: I mean, you've got Ambience, you know, ki- uh, which is another AI scribe with intelligence that I think has some pretty strong, uh, folks on, you know, d- leading that.
• 22:26
  Speaker: They seem to focus more on coding accuracy, the documentation piece, and at least w- from what you've talked about, you talk a lot about the care improvement, but there is a, there, the, and, and frankly, time saved.
• 22:40
  Speaker: I mean, you, you, Abridge has actually-- I mean, you can speak to it, but, you know, ou- outside parties have already shown that Abridge is actually saving doctors time, which honestly probably leads to doctor sanity.
• 22:52
  Speaker 2: Yeah, absolutely.
• 22:54
  Speaker 2: We've gotten a lot of feedback about saving time, saving marriages.
• 22:58
  Speaker 2: We've seen, we've seen clinicians tell us stories about for the first time going to soccer practice with their kids.
• 23:04
  Speaker 2: And certainly that's a wedge, and you can measure that as well, and you can evaluate that as well.
• 23:08
  Speaker 2: And there's ways to get very granular and detailed, and ways to build dashboards to demonstrate the value proposition, even for the CFO on, on a lot of that value that we, we, and, you know, started this journey on, with.
• 23:21
  Speaker 2: But then even beyond that, you know, on the revenue cycle side, yeah, we ha- you have to go deep there.
• 23:26
  Speaker 2: And you're right, like clinicians, we don't get compensated for the care that we deliver, we get compensated for the care that we document that we deliver.
• 23:32
  Speaker 2: But it's not just what you put in a note.
• 23:34
  Speaker 2: It's how the note is concordant and consistent with what end up in the claim, in the 837.
• 23:41
  Speaker 2: It's how the note and the orders and the, the claim need to be-- 'cause the way things work today is that it's like sort of a Swiss cheese model.
• 23:50
  Speaker 2: I write a note, I do a terrible job'cause I haven't gone to all of these revenue cycle, you know, sort of, um, sessions, and I haven't gotten a degree on it, and nobody expects me
• 24:00
  Speaker: to.
• 24:00
  Speaker: Or you're busy.
• 24:00
  Speaker: Or- I'm busy … or, or some, or, or, or someone codes in the room next door.
• 24:04
  Speaker: I'm like, like- Yeah … it, th- th- we, we've, we've set up an insurance system that's inconsistent with care.
• 24:10
  Speaker 2: Yeah, exactly.
• 24:11
  Speaker 2: And everybody's pulling in a different direction.
• 24:13
  Speaker 2: And so the opportunity right now is to use AI to sort of abstract away these things, um, and all of that disharmony and, and just make it about the patient.
• 24:23
  Speaker 2: And so in this world, in the world with Abridge, well, the models have gone to revenue cycle risk adjustment school.
• 24:29
  Speaker 2: They understand inpatient versus op status.
• 24:31
  Speaker 2: They understand also the patient's story, their prior diagnoses.
• 24:35
  Speaker 2: And so when I say, "You got swelling in your legs, let's go up on your diuretic, your Lasix," it understand that's, uh, that the problem is acute on chronic systolic heart failure secondary to sarcoidosis, and increase the Lasix to 40 milligrams from 20 twice a day.
• 24:50
  Speaker 2: Whatever the case may be, but it can kind of bridge that gap 'cause it's got context and it also understands the landscape from a rev cycle standpoint.
• 24:58
  Speaker 2: But getting all of that done right, getting all of that done consistently means that the CFO is gonna see that bottom line improvement that you were speaking to, and every startup has to have that value proposition.
• 25:08
  Speaker: Care and claims.
• 25:10
  Speaker: This is a crowded sp- And AI.
• 25:12
  Speaker: This is a crowded space.
• 25:14
  Speaker: You've got Epic breathing down your neck as they do breathe down every-- the, the, the dominant, the, uh, uh, the, uh, the, the, the empire of, of healthcare IT and the electronic medical record space.
• 25:26
  Speaker: You've got some great, uh, talent in, in the, in the ambient listening, but adding more features, Ambient, Suki and others.
• 25:34
  Speaker: How do you think about competing and winning here on the terms that you wanna win, which are both clinical and financial for your investors and, and, and for your practitioners?
• 25:44
  Speaker: Like, how, how are you, how are you gonna cut through the clutter and, and, and beat back either the, the, the, the aged but, but, but dominant incumbent or some of these, you know, well, well-managed and well-led and creative competitors in your lane?
• 25:59
  Speaker 2: I think if all of us do this right, we'll find a way to collaborate everywhere we can.
• 26:05
  Speaker 2: We'll find a way to be complements to each other, and we'll find a way to use this technology to really make a difference here.
• 26:14
  Speaker 2: And it's not just in terms of the end user experiences that we can create, but it's also the business models that we need in healthcare to, to, to make this, uh, any of this sustainable.
• 26:25
  Speaker 2: And, uh, to, to truly innovate on that level, it's gonna take all of us.
• 26:28
  Speaker 2: And I, I am, I'm very optimistic that we're all coming together to make that happen.
• 26:32
  Speaker 2: So, for example, we serve well over 300 systems.
• 26:36
  Speaker 2: We work together with Epic to make sure that we can serve them.
• 26:38
  Speaker 2: We have to.
• 26:39
  Speaker: Oh, I'm not sure Epic is as focused on your health as you are focused on the health of your- Well- of your patients.
• 26:45
  Speaker: Just, just one person's perspective.
• 26:47
  Speaker 2: Well, we are focused on serving, and, um, you know, I, w- I, I do think that the best companies in this industry are also very service-centered, and we take inspiration from them.
• 26:57
  Speaker 2: And so we gotta come together.
• 26:59
  Speaker 2: And we also are gonna compete in certain spots, and that's where we need to demonstrate as a company that's, I think, in a Goldilocks zone, where we've got a lot of capital.
• 27:09
  Speaker 2: We've got incredible resources.
• 27:11
  Speaker 2: We're headquartered in San Francisco, where all of this is happening.
• 27:14
  Speaker 2: We're recruiting those types of talents that you mentioned from the labs, whe- whether it's, like, the Googles or the OpenAIs or in the Anthropics.
• 27:22
  Speaker 2: And so we can ride and actually create the bleeding edge, and we can then deliver and scale that bleeding edge to the majority of healthcare very, very quickly, and that's a role that we embrace.
• 27:35
  Speaker 2: It's, it's incredibly exciting.
• 27:36
  Speaker: And correct me if I'm wrong, but I think you've said historically that, you know, 60, 70% of your models are actually models, AI models that you've built yourself- Yeah.
• 27:44
  Speaker: That's right … and you're, you're not only 30-ish percent using frontier models.
• 27:48
  Speaker: These guys have infinite capital, the frontier models being the ChatGPT, Anthropic, uh, Mistral is sort of, uh, you know, and others are, but Gemini is certainly getting better every day.
• 27:58
  Speaker: How do you, how do you compete against those m- mega dollar hyperscalers?
• 28:04
  Speaker 2: Yeah.
• 28:04
  Speaker 2: We're playing a very different game than them.
• 28:06
  Speaker 2: We're, we're not competing at the foundation model layer the way they are.
• 28:10
  Speaker 2: They're trying to create these all-purpose tools that can power the AI economy.
• 28:17
  Speaker 2: And so they can deal those tools to us, and we'll use them when it makes sense for us to use them.
• 28:23
  Speaker 2: But any more, the difference between an AI native company In a vertical industry like ours and a company that's basically just sort of hitting a model off the shelf and hoping for the best in terms of like not just the value they can create, but also the P&L that they can sustain, is back to what we were talking about.
• 28:40
  Speaker 2: It's the science, it's the engineering, and it's the evals.
• 28:42
  Speaker 2: So are you gonna point the biggest brain at the smallest, tiniest sort of challenge about, you know, prioritizing which medication to put in the med list and the note?
• 28:52
  Speaker 2: Or are you gonna use that-- Are you, are you gonna create your own in-house model?
• 28:55
  Speaker 2: And there are any number of tasks just in notes, much less clinical decision support, that you can handle.
• 29:02
  Speaker 2: You can actually handle better than the frontier labs by mid-training, by post-training your own models.
• 29:07
  Speaker 2: And those aren't cheap.
• 29:09
  Speaker 2: It still takes a lot of resources, and most importantly, it
• 29:12
  Speaker: takes- But you seem to be very good at raising capital.
• 29:15
  Speaker: Uh, uh, uh, congratulations on-
• 29:17
  Speaker 2: Thank you… Speaker: you know, rai- that, that, It's, it's, it is, it is a hard game.
• 29:22
  Speaker 2: I'd like to prep, probably close with a, a little bit of a- Yeah … a lightning round.
• 29:27
  Speaker 2: Uh, I keep calling you Shiva as opposed to Dr. Rao.
• 29:29
  Speaker 2: I hope that's okay.
• 29:30
  Speaker 2: Please.
• 29:30
  Speaker 2: It would probably insult your father, but hopefully you- No you can, as a friend, I can do that.
• 29:35
  Speaker 2: Um.
• 29:35
  Speaker 2: Yeah.
• 29:35
  Speaker 2: So just quickly.
• 29:37
  Speaker 2: Yeah.
• 29:37
  Speaker: Most, uh, overhyped aspect of healthcare and AI.
• 29:42
  Speaker 2: Overhyped, uh, uh, I-- Right now, there's, there's the big conversation around being able to fully automate doctors and-
• 29:49
  Speaker: Our friend Zeke, yes.
• 29:51
  Speaker: He's pissed off many, many people.
• 29:52
  Speaker: And,
• 29:53
  Speaker 2: uh-
• 29:54
  Speaker: And that's, that's basically does-- Will the AI model replace doctors?
• 29:58
  Speaker: What-- Where, where do you stand, Dr. Rao?
• 30:01
  Speaker 2: High frequency, low acuity workflows, clinical workflows are absolutely going to get automated.
• 30:06
  Speaker 2: And if we do this right, that will simply give PCPs leverage to do more, um, to operate at an even higher level of the stack.
• 30:14
  Speaker 2: And I, I, I think people are maybe purposefully conflating the argument right now to make it seem more provocative, um, and to get more headlines.
• 30:24
  Speaker: Provocation does drive clicks.
• 30:26
  Speaker: Next question.
• 30:27
  Speaker: Absolutely.
• 30:27
  Speaker: What, what, what's the most underhyped aspect of AI in healthcare?
• 30:33
  Speaker 2: Well, uh, there is that refrain, as cheesy as it is, "The future is here, it's just not evenly distributed." That's really, like, the world that we're living in right now.
• 30:41
  Speaker 2: Um, we're, we're living in science fiction.
• 30:44
  Speaker 2: And I think the ability to help clinicians make better decisions, that's the world that I'm l- that I'm super excited about with Abridge right now.
• 30:51
  Speaker 2: Just the, the-- what we're doing right now is feeding the story of the patient into these models, and so when you ask it a question, you're gonna get a way better answer than you could before.
• 31:02
  Speaker 2: And, uh, you know, speaking of superhero powers, it's one thing to do revenue cycle, it's another thing, you know, to actually do decision support.
• 31:09
  Speaker 2: So that's where I think we are n- we're not hyping that up enough.
• 31:12
  Speaker 2: I think that quality measures, Leapfrog scores, U.S. News & World Report rankings, there's gonna be an incredible arbitrage opportunity between the systems who adopt this kind of technology into their workflows and those that, that wait too long.
• 31:25
  Speaker: So fin-final question, Shiv.
• 31:27
  Speaker: For the patient, what should the patient be asking the doctor about AI in healthcare when they're visiting either an emergency room or a, or a specialist or s- or, or a consult with a cardiologist?
• 31:40
  Speaker: What's the-- what should be in the patient's mind?
• 31:44
  Speaker 2: Yeah.
• 31:44
  Speaker 2: Well, th-there are certain parts of this technology and AI in general and in care that, um… or all of this is based on trust, and so consent is a really big thing.
• 31:55
  Speaker 2: Transparency is a real thing.
• 31:57
  Speaker 2: Making sure that the patient is a part, in the loop.
• 31:59
  Speaker 2: It's hard to not be in the loop with our technology 'cause you're putting this phone between you and the patient, and it's it's red, and it says record.
• 32:06
  Speaker 2: And at the s- at the same time, you wanna be able to h- broach that conversation, you know, properly.
• 32:11
  Speaker 2: Increasingly, I think patients are gonna ask, you know, what else is in it for them in addition to the clinician being more present, making eye contact, making better decisions.
• 32:20
  Speaker 2: As profound as that might be, what else is in it for them?
• 32:22
  Speaker 2: And that's where today we create summaries that are, can be written at the right reading level that can help the patient better understand and follow through later.
• 32:29
  Speaker 2: But that's certainly more surface area for us to extend into.
• 32:33
  Speaker 2: Can these summaries… At one point in time, we would nudge the patient in a week or two and say, "Did you follow up and get that echocardiogram?" You know, "How are you doing with those diet and lifestyle recommendations?" And there's a real opportunity to create an extension of every clinician's best intention to help their patient even when they're not in front of them.
• 32:50
  Speaker: Awesome.
• 32:51
  Speaker: Well, Shiv, um, that's a, that's a great answer.
• 32:54
  Speaker: I think that much of this, whether it's on the physician or the t- or the patient side, will move as much at the speed of trust as, as, as the speed of technology.
• 33:04
  Speaker: So that's it for this episode of"Care Talk." If you like what you heard or you didn't, we'd love you to subscribe on your favorite service.
• 33:11
  Speaker: And thank you, Dr. Shiv Rao, skateboarder, film theory, AI-informed cardiologist turned into gr- successful CEO.
• 33:22
  Speaker: Thanks for joining us today.
• 33:24
  Speaker 2: Thanks so much, John.
• 33:24
  Speaker 2: Take care.