Subscribe to The Podcast by KevinMD. Catch up on old episodes!

Join Shiv Rao, a cardiologist and physician executive, as we delve into the transformative potential of generative AI in health care. We’ll discuss how generative AI can help doctors eliminate distractions, improve patient care, and enhance the human connection in medicine. Explore real-world examples, ethical considerations, and the future of health care with AI.

Shiv Rao is a cardiologist and physician executive.

He discusses the KevinMD article, “I’m tired of being a distracted doctor.”

Careers by KevinMD is your gateway to health care success. We connect you with real-time, exclusive resources like job boards, news updates, and salary insights, all tailored for health care professionals. With expertise in uniting top talent and leading employers across the nation’s largest health care hiring network, we’re your partner in shaping health care’s future. Fulfill your health care journey at KevinMD.com/careers.

VISIT SPONSOR → https://kevinmd.com/careers

Discovering disability insurance? Pattern understands your concerns. Over 20,000 doctors trust us for straightforward, affordable coverage. We handle everything from quotes to paperwork. Say goodbye to insurance stress – visit Pattern today at KevinMD.com/pattern.

VISIT SPONSOR → https://kevinmd.com/pattern

SUBSCRIBE TO THE PODCAST → https://kevinmd.com/podcast

RECOMMENDED BY KEVINMD → https://kevinmd.com/recommended

GET CME FOR THIS EPISODE → https://earnc.me/jtSMmP

Powered by CMEfy.

Transcript

Kevin Pho: Hi, and welcome to the show. Subscribe at KevinMD.com/podcast and get CME for this episode by clicking on the CME link in the show notes. Today we welcome Shiv Rao. He’s a cardiologist and a physician executive. His KevinMD article is titled “I’m tired of being a distracted doctor.” Shiv, welcome to the show.

Shiv Rao: Thank you so much, Kevin. As I was just sharing with you, you’re like a mythological figure for me, and probably for any number of clinicians out there, so it’s such a privilege to be here with you.

Kevin Pho: Oh, thank you so much for those kind words and for sharing your time. So let’s start by briefly sharing your story and journey to where you are today.

Shiv Rao: Absolutely. I’m a cardiologist, and I’m also the CEO and founder of a company called Abridge. A little about me before this: Prior to this, I was an investor at a large health system called UPMC. I sort of violated the Peter principle many times, like 10 times in five years, but I ended up being a leader and check writer for their provider-facing portfolio, which was looking to invest in solutions that could help doctors and nurses in hospitals and clinics. I learned so much in that position. I got to sit on boards and just osmotically absorb as much wisdom as possible from disciplines I hadn’t really been exposed to as a clinician. A lifetime ago, I went to Carnegie Mellon University as an undergrad and studied any number of things I’m not directly doing right now, but in the middle I became a cardiologist, and at the core I really see myself as a clinician who’s now trying to create as much impact as possible using tools like technology.

Kevin Pho: Tell me about that transition from the clinical side into what you’re doing now. I talk to so many doctors about that journey outside of clinical medicine, making a difference outside of clinical medicine and potentially impacting more patients. Tell us a little about that journey.

Shiv Rao: Absolutely. Not to get too deep into a life story that happened too long ago, but I did get to med school in a circuitous way, in the sense that I was subconsciously rebelling in college against all the Indian doctors in my family and in the world. I really wanted to do something different. For a while, all I did was skateboard. I was making music, and my roommate and I were selling records in Japan; he had a record label. I was doing a lot of things that were sort of off the grid.

Then I remember the moment, junior year of college, when I got really inspired by a lecture from an architect, William McDonough. He told the story of an ophthalmologist in India who designed a revolving platform on which he would bring patients around at 12, 3, 6 and 9 o’clock and do cataracts, just 15 minutes each: spin, cataract, spin, cataract, spin. By the time I heard this lecture, he had already given eyesight to over a million people, and his daughter had given eyesight to over 400,000 people. I just remember thinking, “Wow, I want to be part of a team that can create that kind of impact.” Not that it’s any less valuable to create deep impact for one person in your life, but there was something about scale that really resonated with me as a person.

So I pivoted. I took a couple of years, did my premed requirements, went to med school, and my soul got crushed; I didn’t see a lot of room for creativity. Then I went to residency at the University of Michigan, and things opened up. I fell in love with patient care, really got my footing, and saw myself as the internist, or the cardiologist, you would really get to know, a generalist who could know all aspects of your care and go deep. But I was also starting to catch this bug around technology. I saw technology as a way to do what I didn’t have the patience to do in basic science research: 13 years from bench to bedside, lab animal experiments. That’s not me. I have a little bit of ADD, and I need tighter feedback loops, and technology checked all the boxes. I built something for the residency program, experienced some level of product-market fit, and along the way to where I am today, I had a failed startup, or a startup experiment, and here I am.

At the end of the day, it’s about impact for me, and, selfishly, the kind of fulfillment you get as a clinician when you feel like you’ve really made a difference, whether it’s in a clinic or in the ICU. I think it’s the same sort of high you can get when you go off the grid a little, or, maybe better to say, when you cross the T and leverage your depth as a clinician to do something different. That different thing could be any number of things, but I think technology is certainly a creative way to flex your expertise.

Kevin Pho: So we’re going to talk about one aspect of that technology that’s really in the news a lot, generative AI. Your KevinMD article is titled “I’m tired of being a distracted doctor.” Tell us how your article came together.

Shiv Rao: I think we all know the statistics about clinician burnout. This is an audience that doesn’t need stats. We all have stories, and I think stories are more powerful for us to share with each other, and that’s what you allow for: storytelling. These are true stories, and that’s where the most impact is for me. There are so many stories. For one, my dad is a retired cardiologist. Why did he retire? He couldn’t type fast enough. At the same time, he’s one of the most inspiring clinicians I know. His patient cohort just adored him, and to this day they see him as something from another era. As much as he was a cardiologist, he was almost like an internist, that primary care clinician who would visit you at home, sit down with your whole family, know your story and know you not just as a patient but as a person. Idealistically, I’ve always wanted to walk in his footsteps and build that kind of rapport with my own patients. And I think what we all realize when we go into practice is that the incentives aren’t aligned for that kind of practice, or at least not naturally, and sometimes you need to do some really creative things to make the economics work so that you can achieve that kind of rapport or get compensated for it.

It’s been a little over two years, probably, since I’ve had a regular weekly clinic, but I still see patients. I’ll take call one weekend a month, and maybe a night a week; I’m on call tonight for cardiology. There’s that pressure of needing to move from one patient to another to another, and there’s always this cognitive overhead. Sometimes I would get over that cognitive overhead by doing a chart biopsy the night before: “OK, I’m going to be present for my patients tomorrow. I’m going to be in the moment. I’m going to look them in the eye. So I’m going to write their notes the day before.” I would basically template everything, predict all the diagnoses I was going to make and what their stories might look like, so that I could feel unburdened to be in the moment with them, and then fill in the gaps and write my notes. That’s the opportunity. That’s really the wedge, but that’s the opportunity generative AI has. I think what it can do for us, and what it’s going to do for us over the next two years, is going to be absolutely paradigm-shifting, and clinicians have the opportunity right now to lead in directing it. They should take advantage of this opportunity to be leaders in this new paradigm.

Kevin Pho: All right, now, just to get everyone on the same page, give us a 30-second synopsis of what generative AI is.

Shiv Rao: It’s a really good question. I think many of us are familiar with AI systems that produce simple outputs, say, a number, for example the predicted length of stay for a patient in the hospital, or a category, like whether a nodule on a CT scan is malignant or benign. Generative AI, though, refers broadly to systems whose outputs take the form of more unstructured, complex media objects: an audio file, an image, a whole document. What’s interesting and fascinating, and what our computer scientists have taught me over the years, is that under the hood, many of those systems are actually built by executing models that serve a more classical purpose. For example, when you’re playing with ChatGPT, you’re creating a whole document, but what’s happening under the covers is that it’s iteratively predicting what word comes next.

When you think about it, the ability to produce a whole document with desired properties, or a whole image with desired properties, or audio, or video, unlocks a whole host of exciting applications. In health care, you can’t swing a stick without hitting a potentially incredibly impactful opportunity for generative AI, because health care, at the end of the day, is about words: words that are spoken and words that are documented. Over 80 percent of the data in health care is words. That’s the opportunity. That’s why this is going to be so paradigm-shifting, and why, more than in any other industry out there, this is where we’re going to see the most impact.

Kevin Pho: So tell us your vision of how generative AI fits into today’s health care paradigm.

Shiv Rao: I think that what it’s going to do, and what it’s already doing, what we’re seeing in practice, not in a presentation, is actually making health care feel more human again. We clinicians have heard this refrain before: “Technology is going to help you. Just do this extra thing, open this new website, and it’s going to help you in the end.” And it rarely has in the past. It’s usually not the fault of the technology itself. It’s the fault of the incentives guiding or directing that technology to ask us to check off all these new boxes, or do prior authorization for all the medications or imaging studies we need to order for our patients.

What’s interesting and exciting about this technology is that it can take all that clerical work off our plates, or at least set the table. It’s almost like AI should stand for “artificial intern,” that person in the corner of the room setting the table for you. They might not create the perfect set of orders or the perfect note, or do prior auth the way we need it done, but creating that draft allows us to be curators, and I think there’s a creativity in curation that lends itself very well to clinicians. At the end of the day, what do we want to spend our time on? Thinking, listening, looking people in the eye, building rapport and then creatively figuring out the diagnostics and therapeutics. As we all know, however much data there is, and I’m in one of the most data-driven disciplines of all, cardiology, this is alchemy. This is art, and we’re connecting so many dots to make decisions.

So I don’t see AI replacing clinicians, in the sense that we’re going to need creativity and we’re certainly going to need human connection. I do see it augmenting us. There’s a refrain that’s become popular, and I absolutely agree with it, that doctors who use AI are going to replace doctors who don’t. I think that over the next two years, 80 percent of clinicians are going to be using generative AI in some way, shape or form so that they can be more present with their patients.

Kevin Pho: So tell us about a scenario where you use generative AI, say, in an outpatient cardiology clinic. Walk us through a hypothetical scenario so listeners can really visualize the practical application.

Shiv Rao: Absolutely. There are any number of applications, and I’ll go through the one my team focuses on building. Two weekends ago, I was in clinic, walking from room to room. I’d walk into a room with this technology on my phone, open up Abridge and have a normal conversation with my patient. At the end, I’d hit stop, swivel my chair, and my note is there, written the way I like it. It reads like a clinical note, with all the style and formatting. It has structured the assessment and plan in a problem-based way. It has also helped me understand which diagnoses I spoke to and mapped those diagnoses to codes, so I can create the most complete representation of the conversation I had. It has captured the patient’s history in a colorful, dense and rich way, per my preference, which I can set, so it’s capturing some social determinants of health information and social history that I’ll be able to leverage to hit the ground running with rapport when I see the patient again in six months. I’m just trusting and verifying: maybe making some edits and then finalizing.

I’m also able to send a version of that note to my patient, which they can get in the portal or somewhere else. It represents the same conversation, but written at a level that will hopefully help them improve their own health literacy, and also improve their ability to follow through and be the healthiest versions of themselves. That’s what we’re building, that’s what’s live, and that’s what thousands of clinicians are using, sometimes saving three hours a day in primary care. And that’s just the tip of the spear in terms of what generative AI could do. Imagine multimodal generative AI that can parse through the electronic medical record, look at imaging, labs and all the notes, and summarize that whole history into a pithy paragraph. That’s where this is going to go very quickly, and there are any number of use cases beyond that.

Kevin Pho: So just to be clear: You go into an exam room, you bring out your phone and turn on this app, it records your interaction with the patient, and shortly afterward it has a whole clinical note documenting that interaction. When you say you have to do some editing afterward, how long does that normally take?

Shiv Rao: On average, it takes people 28 seconds. That’s what we see with some of the clients we’re serving right now, so very, very little time. And there’s something more profound here, too. There’s ROI in the time saved; it’s hours. There’s ROI in a complete note. But the ROI you get from unburdening clinicians from the psychological overhead of having to shift gears, thinking not just from the clinical perspective but from the billing perspective and then the patient perspective, is really hard to put a number on. I’d love for all of us, as a clinician community, to collaborate and put a number on burnout. It’s invaluable; you can’t put a number on it.

You see a patient and leave, and in my case I’m always hyperconscious that I’m writing a note for other clinicians to read and grok my thinking. Then I’m sorting through all these drop-down menus to get to a specific diagnosis and ICD code, which crushes my soul. It’s the last thing I want to do. I’m always picking the lowest-hanging diagnosis, which is nonspecific, so the coders chart-chase me for the next few weeks. And then I’m thinking, “Well, wait, my patient’s going to read this note. They’re going to go to the portal, and in my case, they’re going to see a term like ‘transcatheter aortic valvuloplasty,'” and understandably, reasonably, they’re going to have questions. I’m going to get inbox messages and phone calls asking, “What’s that word? I Googled it, and it sounds scary.” That’s what we’re doing about everything I just described. I really do believe this is one of any number of use cases that companies can build now, and that physicians can help build, but the use case we’re building is ultimately about helping clinicians be less distracted and more present, feel less burned out and get more energy from the thing that drew them to this profession in the first place, which is human service.

Kevin Pho: Now, do you have to train the tool you’re talking about? I know with ChatGPT you have to play around with prompts to make sure you get the right output, and even with dictation, you have to train Dragon and some other systems to make sure they produce the output you want. For these specific tools geared toward writing notes, is there any training or buildup time before they’re functional?

Shiv Rao: No, you go with it immediately. In fact, any clinician could ping me right now and just start going with our solution. That’s what’s elegant about certain applications of generative AI: It’s not yesterday’s technology. This is technology that will immediately create value. Will it get better over time? Absolutely. It’s going to get further tuned, and you’re going to be able to further train models. I think that’s part of the magic of this moment right now, in 2023. It feels historic. It feels like these aren’t tailwinds for what can happen to our profession right now; it feels like a tornado. I think this is our moment as clinicians to direct the tornado in the direction we need it to go, and make sure it doesn’t go somewhere else, because technology is certainly a tool; you can use it for whatever purpose. In this case, we have an opportunity to make sure it does the thing we’ve always needed: help us return to our roots as the original care advocates, as creative thinkers working through differentials, and, at the end of the day, deliver care that improves outcomes.

Kevin Pho: One of the critiques I hear about generative AI in health care concerns privacy, because it’s pretty much going to be using a lot of that patient information for future iterations. Talk about some of the privacy issues that people sometimes raise about these tools.

Shiv Rao: Privacy is paramount; we all know that. To be quite honest with you, I believe that privacy, and the bigger umbrella of trust, is actually going to be the big differentiator for generative AI-centered applications in health care. If you’re one of us as consumers and you start to use ChatGPT, or, in the case of my 13-year-old, use ChatGPT to write your papers and do all your homework, which she clearly is doing, that’s a different use case. I don’t mind so much when she’s feeding public models a lot of data. It’s an amazing service that we all have access to now, and I’m incredibly grateful for what companies like OpenAI are doing for all of us. At the same time, when you apply that same technology to an industry like health care, where trust is table stakes and privacy is paramount, you have to put a lot of rigor into what you build beside, underneath and on top of those models. You might even have to fine-tune those models, and you certainly need to build infrastructure that allows the data to do what needs to be done to deliver the value, but nothing more. You need to create a covenant with your users, the clinicians, and with those patients that you’re going to treat their data with the utmost care. That’s going to be the difference, and that’s easier said than done. I think there will probably be a flood of tech solutions over these coming months and years that are generative AI-focused but don’t have the hygiene health care should demand. But I do think the dust will settle over the next year, and it’ll be clear who is creating solutions that carefully manage data, and who is building auditability into their models and transparency into how things are working, so that we’re comfortable using them with our patients.

Kevin Pho: As you know, there are several companies with generative AI solutions similar to what your company is doing. From a physician’s standpoint, what should we be looking for, specifically as it relates to patient privacy and security, that would separate these companies when we’re making a choice?

Shiv Rao: Absolutely. I think there’s a high-level phenotype for a company that makes sense to me and is magical in health care, and that high-level phenotype translates into the more granular things they might implement inside the company: the processes, and the sorts of professionals in place. At the highest level, I think these companies should be health care native as much as they are AI native. If you have one without the other, it’s interesting. If you’re AI native but don’t have clinicians on the team, I think the risk is that it’s an opportunistic solution that’s going to try to make a quick buck but not really have a long-term vision or road map to create the kind of value that can truly transform the profession. Similarly, if you’re health care native without the computer scientists, the technologists, the engineers and the designers, who should have a level of pedigree, then you’ll have a lot of vision and a lot of road map, but an inability to scale to it or build to it, so your ceiling is a little lower in terms of what you can achieve. You need both ingredients to really create magic.

That magic manifests in many ways. In terms of privacy policy, data should be encrypted at rest and in transit. Maybe that sounds like table stakes, but sometimes it’s not there. Certain data use agreements should be baked into the policies you sign up for as an individual clinician, making clear that the company is not going to sell your data and is going to use it simply to improve the service for you as a clinician. So maybe on the first day it doesn’t recognize the name of a new oral anticoagulant, but on the third or fourth day it does. That’s the sort of magic AI can deliver, and certainly that we aspire to deliver faster than anyone else out there.

Beyond that, one piece of this space that I think will prove to be table stakes, and could potentially be a subject for regulation, is auditability and transparency. Many of us who’ve played with ChatGPT, or with generative AI technology in general, have probably heard the word “hallucination.” It basically means that a generative model is creating output that was not reflected in any way in the input. Honestly, in the arts that could be a happy accident. It could be great; sometimes those hallucinations should be embraced. What I’m particularly interested in and passionate about outside of health care is what generative AI could do for music, and it’s really cool to see new things come out and figure out how to leverage that stuff in your own compositions. However, in health care it’s dangerous. If, for example in our space, you have a conversation with a patient and then get a note that contains generated summary that doesn’t reflect the conversation you had, it doesn’t take long for your trust to erode. So that’s a lot of where we’ve invested. It’s a long game, and I feel strongly about it because I’m a doctor first and foremost. It’s about trust, and trust is some combination of transparency, reliability and credibility. With auditable models, in our case the ability to highlight a word, or a sentence generated by the machine learning system, and see where all the evidence came from in the conversation, and even listen to it, that’s the sort of auditability that I think will start to differentiate companies, not just in this space. Think about what could happen to radiology in the coming years as well.

Kevin Pho: And my final question: Let’s end with some take-home messages that you want to leave with the KevinMD audience.

Shiv Rao: I think the biggest message is that this moment right now, in 2023, is historic, and clinicians should take leadership roles and make sure we direct this technology where we need it to be, in order to create the value we’ve been waiting for, because ultimately this is technology that can make health care feel more human again.

One last message I’ll relay is about clinicians. In my interactions with clinicians, so many of them are so much smarter than me in so many different ways, but so many of them underestimate their ability to transpose the skills they’ve collected in their medical training to a different sphere, whether that’s business or technology. What’s so fascinating is that so many of the muscles we built in residency, rounding in the ICU, doing psychiatry or listening to a patient’s story as an internist, translate so well to leadership roles in any number of industries, but certainly in technology. So I don’t think we should limit our ambitions when it comes to playing key roles in this new paradigm.

Kevin Pho: Shiv, thank you so much for sharing your perspective and insight, and thanks again for coming on the show.

Shiv Rao: Thanks so much.