Designing for Health: Interview with Justin Schrager, MD and Nick Sterling, MD, PhD [Podcast]

Show Notes:

[00:00] Intros

[01:02] Paths into healthcare and technology

[02:55] The birth of Vital.io

[05:07] From clinician tools to patient empowerment

[06:56] Solving the communication crisis

[14:02] Why EHR portals weren’t enough

[23:50] Designing for accessibility and empathy

[28:54] AI in action

[39:11] Justin and Nick’s favorite well-designed things

[42:37] Outros

Transcript:

Dr. Craig Joseph: Welcome to the pod, Justin, Nick. Tell us a little bit about yourselves. Justin, want to start with you.

Dr. Justin Schrager: Thanks for having us, Craig. I'm Justin Schrager. I'm the chief medical officer and co-founder of a company called Vital.io. We make patient experience/patient communication software for the acute care setting. I'm an ER doctor by training, and I still work in the emergency department in Indianapolis for the IU health system.

Dr. Nick Sterling: Thanks again for having us, Craig. Nick Sterling, also an ER doctor. I'm the chief medical information officer for Vital, focused on the prototypes, safety design, a lot of the research. It's great to be on.

Dr. Craig Joseph: Well, I thank you both for joining. I want to get a little bit more of the background. My understanding is that, when you were five, both of you did not decide that you wanted to be a physician and then work in technology. You kind of had a circuitous route, which I think explains a little bit about how you got to where you are. Justin, tell us a little bit about what you did before you came to this current position.

Dr. Justin Schrager: I guess I'm a non-traditionalist in the medical world. I was an English major then I went and taught high school for a couple of years. Then I went to public health school, worked at the CDC, and found my love for medicine through that work. Frankly, I realized that if I wanted to make change in the world, I would need medical school to have that kind of influence. I worked as a health services researcher and junior faculty for quite a while before a midlife or early career pivot into predictive analytics and software development.

Dr. Craig Joseph: Nick, exactly the same. You were an English major.

Dr. Nick Sterling: I was sort of born for engineering and computer science. Way back, like in the early 2000s, I ended up working professionally as a software developer for some time. But I had this competing passion for health care. At that time, it felt like a total switch. So I decided to go to med school, and I just took on faith that maybe someday I'd be able to merge those worlds back together: computer science and health care. I ended up doing an MD PhD program in a neurodegeneration lab, working on MRIs, shape analysis and human motion, data analysis as well. And then I met Justin, through my residency. He was my faculty adviser at Emory in their residency program. And we did a lot of research together. I make prototypes and do so for Vital.

Dr. Craig Joseph: Justin, why don't you tell us about how Vital got started? My understanding is it involved Thanksgiving?

Dr. Justin Schrager: It's been about eight and a half years since we started working on Vital, which is borderline geriatric for a health tech start-up. It started with my co-founder, Aaron Patzer, he's sort of a genius polymath, but has done a lot of really cool things in technology, including creating Mint.com back in the day and then reinventing into its product platform. Fortuitously, he is my brother-in-law, so he was at my house for Thanksgiving one year. This would have been eight years ago when we started working on this. I had worked a shift overnight and fortunately, I didn't have to cook Thanksgiving dinner, which is something I usually do. I was sleeping all day and I woke up and dinner was not quite ready.

I was in my office signing notes for an hour in an EHR, and he walked in to tell me that dinner was ready, and looked at my screen and said, “What the expletives is that?” I said, “This is the second-best electronic health record, and the other one's not that much better.” And that was how we started talking about doing something. We had talked about things over the years, doing things together just because we're both firstborns and we get along really well and we ideate really well together. But that was the kick in the pants that we needed to start doing something in healthcare. We didn't set out with the intention of radically transforming patient experience. But actually, we set out with the goal of radically transforming how people in general and in this case my poor self, having to use a note writer platform and an archaic EHR transforming how we all experience the digital world of health care. We went into it with open eyes and that's how we started working, approaching problems from that perspective.

Dr. Craig Joseph: And sounds like you were focusing more on the clinician side. Well, tell us that story. How did you pivot to where you are now?

Dr. Justin Schrager: When we started, we wanted to do operational and clinical decision support, primarily operational. In Aaron's words, “I want to fix the part of health care that has the most entropy.” I said,” I know where the entropy is. It's in the ER, specifically the waiting room in the ER.” So what we tried to do is apply machine learning. And Nick was working on this with me at the time, applying machine learning to help with throughput, prognosticating bottlenecks, figuring out how long people would be there overall so we could predict who is going to be admitted, where they were going to be admitted, and the bed type they needed.

It went really well so we formed a development partnership with Emory, where I was working at the time. We were able to prototype a model and get it running in a software platform, a SAS platform, which was helping effectively with the bed control of the hospital, figuring out who needed beds, where they needed to go, if they needed to call extra nurses and do things like that. And it was working really well. And then we ran into a health/life changing event in early 2020 during our pilot period and everything changed. It didn't just change for tech start-ups. Obviously, life changed for everyone. I realized during the height of the pandemic that we were having trouble keeping up with updating patients and telling them what's going on, educating them, talking with them about things, figuring out what matters to them, talking to their family members.

That's the part of emergency medicine that is super important. And it's also the part of emergency medicine that goes out the window first, when things get hectic and busy. You can't think of anything more hectic and busy than the middle of the pandemic when we had no PPE and, you know, the whole story. What I realized, though, was that we had this amazing machine learning work that we had done to predict how long people had been in the E.R., what types of beds they needed, how backed up the CT scanner was, etc., etc. We had it at a really granular level. Plus, we were running a pilot, so we had all the data flowing through our system which was already EHR connected. Aaron and I were sitting on opposite ends of the earth talking about this. He was living in New Zealand at the time, and still does live there. But we were commiserating about what it was like to be an ER doctor during the pandemic.

And I was like, “You know what, Aaron? We just need to take all this data, all this stuff we're trying to do to improve clinical throughput and decision support and all the other stuff we have been working on. We just need to give it to patients because I don't have the time to be on the phone for half of my shift, calling everybody's relative to let them know that they still haven't gotten a CT scan, or they're still in the waiting room or still in the hallway waiting to go upstairs. I don't have time. It's not safe. I don't have enough people to be doing that kind of customer service.” And I said, “We have the software, let's just go right to the patients and just tell them what we know.” Fortunately, we had an innovative, creative partner in Emory and in the Department of Emergency Medicine that allowed us to do this even when all the other COVID stuff was happening and it was extremely successful. The patients loved it. That year, we were able to get upwards of two-thirds of every patient who comes through the door in the E.R. using Vital, seeing their wait times, learning about labs, education, all kinds of things like that, which I considered to be a great success.

And that was our first great success: getting a plurality of all patients using our platform. So that's the story. We pivoted during the pandemic. I won't say that it was financially existential at the time. It was more about what the key driver was. Or what was the key pain point for the department? At that time, we were just doing emergency room software, not inpatient like we do now as well. But what was the big communication hurdle that we had was the big problem in health care that I wanted to help solve. Fortunately, we were in a good place to do it, and that's kind of where we've been ever since.

Dr. Craig Joseph: That's interesting. It wasn't an existential threat, but it was the problem of the day. And you went about solving that or at least making it a lot better. It's like solving cancer.

Dr. Justin Schrager: I love how you've equated cancer and communicating with patients in terms of…they're like the Mount Rushmore of problems in health care!

Dr. Craig Joseph: Well, I think one could make an argument that a lot of the problems that we have in health care in the U.S. especially is in communication, getting the right message across to the right person at the right time and in the right way. So, yeah, I'm not backing down from that.

Dr. Justin Schrager: I agree with you. It's a huge problem. And at the fundamental level getting patients to follow our recommendations and give them the plan comes down to communication skill. But there's so much stuff that happens around doctors, outside of the nurses, around the nurses, that you could bucket as communication and all of that goes part and parcel with, you know, everybody needs to be on the same page and the patient has to want to read that page. They have to be in the right place to read it.

Dr. Craig Joseph: Yeah, for sure. Nick, you had mentioned you started your career getting a PhD and an MD at the same time. And you were working on MRIs. I'm curious, how different was it moving from looking at the tiny little pixels on a screen (or whatever MRIs produce) to looking at hospital data to predict, hey, is this patient going to be admitted? Where is this patient going to be admitted to? Is it similar or is it completely different?

Dr. Nick Sterling: The closest comparison I would say, I mean you always have the same research design problems, right? Like causality versus association-correlation. That's always going to be, in the scientific field, a consideration. In terms of the data itself, MRIs are all voxels, you know, motion data, sub-millisecond information captured in three dimensions across extremities with limited subjects. So, you’re still in a language at the same time, like the lexicon is huge. You know, the corpus of language is potentially enormous. And then you end up with these long vectors. So basically what I'm getting to is that you still end up with this problem of, potentially, a limited number of subjects to fit a model to with a gigantic number of variables.

So you have to be able to collapse those variables down into a more meaningful set, like whether it's through, you know, PCA or something like that or, or just selecting the most important features. I would say that's probably the biggest similarity and we've been really fortunate to have great datasets for that so it hasn't been really an issue. But certainly, for any start up getting started in health care, finding a suitable data set for training is pretty important.

Dr. Craig Joseph: This is what I keep hearing. It all comes down to the data. Apparently, that's still true.

Dr. Nick Sterling: Data quality as well.

Dr. Craig Joseph: Okay, I'll buy it.

Dr. Nick Sterling: I would say though, on your theme of communication, I would notice that in the lab, we were studying Parkinson’s disease and neurodegeneration and it seemed like there were still communication problems between neurology and getting patients what they need. Early diagnosis was actually a pretty common theme, too, steering the course early in the disease to prevent further progression. So that's been a constant theme probably since 2011. Must have early communication to facilitate intervention for me is something I'm really interested in.

Dr. Craig Joseph: Let's continue that communication theme. Justin, apparently you're aware that there are electronic health records out there that have things called patient portals where upon patients can get lab information and see prescriptions and do all of that stuff. It sounds like you're playing in the same area, but not, so tell us why did you not think that the EHRs were solving that particular problem for patients waiting in the emergency department?

Dr. Justin Schrager: You know, we have a problem in health care where our technology, or at least the technology that we use and then the technology that we ask our patients to use is really pretty outdated, to be perfectly honest. You know, we're decades out, compared to the rest of the service industry. Really, my philosophy is you fly on an airplane, you're using an app, you're getting real-time updates, you're getting personalized communication. You order pizza online, you can select what toppings you want. And next time you check into a hotel, they know what you ordered on room service the last three times by your name. This is like table stakes for any service industry, and I hesitate to call medicine a service industry, but let's be honest, it is literally the most important and most expensive service that you will ever pay for.

And so, for us to be resting on our laurels, sitting back on the cart, being pulled by a donkey through medieval Europe using the roadmap of the electronic health records, I think it’s a bit passive. And I would like medicine to be a little more active. We need to be active and demanding when it comes to what we use to communicate with our patients. And the reason that we were able to be successful with the acute care transformation of communication, which is what we do — we don't do outside the hospital, we don't do quarterly. We don't want to change portals. We don't want to build a portal. We don’t want to replace portals. We just want to help with the acute care setting ERs, inpatient, and shortly after inpatient. The reason we want to do that is there's just so much communication that can be conveyed electronically using new techniques or simple algorithms that just wasn't happening. And very few patients, even patients that have portal accounts, were using it when they were in the E.R., the inpatient space, and those that were using it were having negative experiences.

You know, they're getting pinged with lab results, whether they're ready to receive them or not, whether or not they're being treated for a devastating illness or something like that. It's just not a configurable, personalized experience. And very few patients are using it. I mean, 15% usage rate, that's not good. We need to reach everyone. We need to make it so that everyone can use it. It needs to be multilingual. It needs to be configurable. It needs to be something that technically anyone could use. And so I didn't really deem our decision to move into the patient communication space to be successful until we had reached, we call them “vertical patients” in emergency medicine. Basically, anyone that's awake and alert and can walk. Anyone who is not clinically ill or acutely psychic or suffering from major trauma should be using our platform. And we were able to achieve that. It’s a radically different way of looking at it. It's like the difference between what we do and what patient portals do. Remember when you used to call a travel agent and request to buy a plane ticket to Phoenix or something, whereas now you're on your phone doing it while you're flying somewhere else. It's just a new generation of doing things. It's modernization. And it is something, frankly, nice that we can have in healthcare, in a world where we don't get to have a lot of nice things in the digital space.

I think that's the core difference. I don't like to think of our product as competing with the electronic health records or the EHR’s patient portals. In fact, we drive adoption of patient portals up very substantially to the tune of 20 to 30% in every health system where we work. And that helps with all their business KPIs and all the things that patient portal vendors tell hospitals that they do for them. We’re actually putting people back into the ecosystem of the health system. What we're doing is we're providing an extra level of service during the acute care phase when it's extremely, extremely needed.

Dr. Craig Joseph: What is the difference? So, if I'm a patient in your ER, what am I signing up for? Do I need a PhD to use your product? If so, do you grant that? Like, how do I do? I have to write a thesis? Give me all the details, Nick.

Dr. Nick Sterling: Going with what Justin was saying, I find that when things get busy in the E.R., I would say the overall sentiment of certain patients are that they feel forgotten, like in the waiting room. They're just kind of waiting out there. Nobody's updating them about what's going on. They know they got some tests, but they don't know what their results are. They don't know how to check on it. And in reality, that's not necessarily true. You know, there's a charge nurse going through and cycling through patients on the board constantly. The doctors are looking at patients on the board, too. But the patient perception is that nobody really understands why they're there. That's kind of evasive, I think. What we try to do is essentially communicate to the patient that they're more than just a number in the E.R. When someone comes in, they'll register or confirm their information, like a cell phone number, and get a text link. It's a secure link. Almost like a forgot your password link, but a unique link just for them to log in.

It's a web-based app and patients don't have to download any additional applications to their phone; they'll be greeted by a secure login. They proceed through that, and then they can get information about why they're there. So, it's pretty seamless. It's immediate. After registration, we have a high adoption rate, like Justin was saying. And the idea is to walk through care. So, you know, in the E.R., there would be things like wait times, what to expect at different stages of the visit, how to get around the facility, labs that are ordered and what they might mean, we also provide interpretations.

We provide plain language summaries of more complicated tests like radiology images. Say, someone has a trifling fracture of their ankle. We’ll explain to them what that means through large language models. And then if they move into the inpatient space, we have some more features. So, we'll walk them through what to expect there. And we'll give them the care summary, things that they're going to be expecting to get throughout the day, like procedures and tests, like maybe they have a stress test or a radiogram, that kind of thing. We stay with them all the way through discharge. Where are they supposed to go? The kind of doctor they're supposed to see? And they get to that kind of doctor, now are they going to do that with their insurance? Insurance matching, that sort of thing. We're essentially guiding them through the whole process, making sure that they get the care that they need.

Dr. Craig Joseph: So it's a one-way communication? Am I talking to nurses or am I just getting information?

Dr. Nick Sterling: As a patient, when you're using the application, it's currently entirely one way. We've looked into bidirectional communication but at this moment, basically, we're telling the patient what's happening. The good thing, though, is it doesn't require any intervention from nursing staff. So, nurses don't have to do anything extra, registration staff doesn't really have to do anything extra either, doctors don't have to do anything. In addition, it's configurable. We can control what sort of messaging goes out to the patient and what doesn't. For example, one of the cool things that we've implemented into our language models, for example, is the detection of sensitive topics. These would be things like, is this an acute psychiatric-related complaint or something having to do with a test for an unexpected event during pregnancy? Is this accidental trauma in a pediatric or elderly case? If it's neglect, abuse, all those things that you wouldn't want a patient finding out via an app. Really, a doctor needs to come down and sit in front of them to have that talk. He can control that as well, no additional work for hospital staff.

Dr. Craig Joseph: Nick, as you may know, Justin has this concept of the big six problems that he's trying to solve. And, apparently one of them, number three, is depersonalization, which you just hit on. I'm seeing a bonus in your features, is what I'm saying here.

Dr. Nick Sterling: It's a huge problem, like Justin said. It still happens to me, you know, on shifts. I work at a few hospitals, but I'll go into the hospital if it's an ordinary shift. I often feel like I can spend the time that I need to with the patients, but there’s a constant tension between having to move and get patients seen and get them where they need to go. And mathematically, you come out with a limited time with each patient, so the idea that there's an AI-based product that's facilitating and enforcing the kinds of stuff that I'm putting in the plan and in my notes for a patient would be great.

Dr. Craig Joseph: Justin, getting back to your big six problems, problem number one, I'm not sure if it's the most important or just listed first, but reach as many patients as possible, which, Hallelujah brother, how do you design? You clearly spent a lot of time, and we like to talk about design here on this podcast, how do you design your tools to reach as many patients as possible? I will start and answer the question that I just asked you, which is, hey, sounds like I don't need an app, I just get a text message. And I probably put in my birthday to make sure it's me or some information that I know. And then I'm in. I think that's how you said it works. Are there other design techniques that you've utilized to reach as many patients as possible?

Dr. Justin Schrager: Oh, yes, there are about 100 of them. I'll give you a few examples other than the login and provisioning process, which needs to be extremely secure in health care for a multitude of different reasons, not just because patients expect it. Once you move past the security, provisioning, and all the other things that we do really well at Vital, you have to get into the world of the patient, get in the head of the user of your product, and that's what we do. So, I sit down in the waiting room, I say, tell me what it's like for you to try to get your health data while you're here during the E.R. visit. And we all know what it's like. We've all worked in health care and watched patients struggle to remember their login to their portal. For a long time and still, nurses are having to lug around iPads to bring them to the bedside and try and help people log in.

That's old school. I mean, downloading these apps and doing things like that is not necessary. The way our product is designed is such that we want to decrease friction absolutely. And you do that through thoughtful design, talking to patients and finding out what their problems are and barriers to use. And at the same time, you make AB testing and iterative improvement part of your product development. We do all of that. For example, if you went to one of the ERs that was using our platform or one of the inpatient settings that uses our solution, you would be invited to use the application with your first name. It would say, “Craig, welcome to such and such hospital. If you'd like to learn about your care plan for the day etc. or whatever the hospital decides from the invite, you've reached out to someone in a way that is personal with your name. You're offering them something in exchange for their time in your app and all the other good things that come along with it for patient success, outcomes, loyalty, whatever you want from a business standpoint. You're giving them something that's commensurate with their time so it's an exchange. It's an exchange that I think is fair. And then you're also inferring based on where they are clinically, what it is that's most important in the E.R. waiting room. What do you care about? You're in the room. You've seen the doctor. What do you care about? You're in an inpatient setting. You've been through morning rounds. The cardiologist has stopped by. The PT has already come by. What do you care about? And then even more important than that, probably for patients, is what does your family who's not there care about? Whether they missed morning rounds or they're in New York, across the country, they want to know what the care plan is. It's a huge problem in health care, this lack of communication. It's experientally difficult for the doctors and nurses to call a bunch of family members and explain themselves over and over again, answer questions at different education levels. So, if you can make that information which is already present in the EHR available to patients and then synthesize it for what they want, tell them what questions to ask the next time the doctor comes into the room. Learn more. Their family members follow along with the same information. Those are iterative things that we've done to increase utilization. So, I wouldn't say there's one thing that we've done to increase utilization, but what I think we've done is iteratively improve the product over and over and over, and we are maniacally focused on user time and on stability, all the other things that normal software companies do.

Dr. Craig Joseph: It's hard. Health care's hard. I don't think anyone who's listening to this podcast doesn't agree with that statement, because we're generally all in the industry and we get it. And to your point about logging in or using an app in the emergency department, no one is thinking clearly at their best when they're a patient or the family member or friends of someone in the ED, and probably in the hospital as well but for sure, in the emergency department. If you don't work there, chances are you're not at the top of your game and so you have to take that into consideration as well, as you do.

Nick, there are many kinds of artificial intelligence that's used. Can you describe some of it to us? I get, okay, I'm a patient. I'm in the back. I've got blood drawn. I get my SED rate and my electrolytes back. Not a lot of AI there or is there when you’re giving a high-level interpretation of test results?

Dr. Nick Sterling: I would say we have AI in two contexts: one is in the foreground where the patient actually can see the test results, and then more in the background. In the background, we have things like the sensitivity filter, filtering on sensitive topics. As I mentioned, we use a more basic version to decide who can be appropriate to invite in to use the app. To filter out, make sure as Justin was talking about with vertical patients also, identified the problem that, as you know, there are many, many, many different kinds of note types. And they're all labelled different things in all EHRs. And they are also all different across specialties, and they can be different by the author type, by user preference. And so, there's this massive problem in healthcare. Now, do we standardize notes so that we can actually use the information that are within notes, like, how do you know how to just immediately go in and pull out? But generally, what you would call an assessment and plan section or an API, which those can be called different things, as you know, across different specialties and notes, and they can be located in places. We have a very complex and comprehensive note taxonomies system that allows us to ingest a gigantically heterogenous data set of notes and standardize them down so that we can pull out information like follow-up information, like where is the patient supposed to go? And, yes, the patient will see that in the foreground too, like when we're telling patients what they're supposed to do after they leave the hospital, that is fed by a lot of that taxonomic data. I mentioned radiology summaries is one that we do. We published a study in Journal of Patient Experience a couple of years ago. We had a panel of eight doctors review our language model summaries to make sure that they're pristinely accurate for the patient before they reach the patient. We also have an AI safety architecture, which I'm really excited about. There are basically 3-5 core components to it. And it basically involves making sure like, are you feeding the AI models what you believe that you're feeding them because if you're taking garbage input, like let's say a nursing triage note, you believe you're feeding a nursing triage note to a core model, but it turns out you're actually feeding like a set of vital signs with no human written language.

That's an opportunity for a poor-quality result. And then on the back end after the results are produced, we also have a judgment layer as well. We're very comprehensive about the way that we approach AI. And that's a system in itself on the back end. Probably another exciting component in the foreground is inpatient care summaries. So like I was saying, how we tell patients basically what they're there for, how they're revolving if that's specified in the notes and things to expect throughout the day, we just try to make it almost like a GPS for the patient to kind of know where they are through their hospital stay. So there's probably ten different instances of AI throughout what we do. And I would say maybe 60/40 background/foreground or something like that.

Dr. Craig Joseph: I was going to ask you about safety, because again, with health care, it's one thing if ChatGPT messes up my answer to what restaurant I should go to tonight but it's a completely different thing when you're giving information that the patients are trusting is right.

Dr. Nick Sterling: It's so important. There are challenges that you wouldn't think of immediately, which is why you need a super well-designed system. Let's just take, for example, I mentioned garbage input, right? Well, content is one thing — are you expecting a triage and you get a set of vital signs. What about errors that are hard for even a human to understand? I'm sure you've seen plenty of these. Justin has seen a lot of these, where you get a radiology report and often the finding is there is no diverticulitis, for example. The large intestine appears normal and then down in the impression, they have sigmoid diverticulitis. And so you have those two contradictory pieces of information.

And when you feed that into a language model, again, that's an opportunity for a poor result. We do a lot of work to filter out those edge cases to identify them in a robust manner. We have multiple layers of safety that things have to go through. It's easy to just throw stuff at a large language model. I use ChatGPT all the time and there are low consequences for daily use. But when you put the healthcare data through there, especially at scale processing so many encounters, it's important.

Dr. Justin Schrager: One of the things that Nick is not telling you is that Nick is an absolute wizard with this stuff. And one of the things that we feel is vital philosophically, is that we should be using modern AI for the democratization of medical information and access, communication, education, and all the things that it's doing for the rest of society. One-on-one, you use your computer to get a question answered, whether it's evidence or whatever. Patients are doing the same thing on the internet when they receive health information and, whether doctors like it or not, from a paternalistic standpoint, patients are going to be doing this, and we need to get ahead of it. We need to learn how to keep it safe. And to be perfectly honest with you, if you want to know where the stuff breaks and where the hallucinations start, where the security risks exist, you have to be in the world of using it and making it and adding it to your product, because that's how you learn this stuff. You want to know what the difference is between GPT 3.5 and 4.0, or when you're trying to do certain things that go directly to patients in a semi-ungoverned way, you kind of have to get in the weeds.

We're very privileged to be able to do that kind of thing where we're doing something that is a little outside of the norm in healthcare, especially with AI. We do a lot of other machine learning things, too in our app, predicting wait times. When you're talking about, oh, I'm generating a text for patients to help them understand what's going on, goes back to point number one: reach as many patients as possible. You're never going to reach as many patients as possible if you're making the doctors review every little piece of information generated by an LLM before it goes out to a patient, it's just one more task, right? Who's going to do that task? Is it the nurses, the radiologist who's responsible for reading through these things that frankly, any patient copy and paste their note from patient portal, plug it in Gemini or ChatGPT. We're putting guardrails around it. We’re putting in safety strategies. We have an entire schema frankly, multiple models just watching the output of the AI model for quality and safety. It's a philosophical decision that we make at our company to keep patients safe. I'm assuming other companies do the same thing, but we're in a unique position of being the predominant place where patients are getting their information.

Most patients, we’re talking about two-thirds of patients, are using our app for 30 to 45 minutes a day, and they're going there for everything from LLM care summaries for what's going on in the inpatient setting to all the other things. So it’s really important that we do it in such a way that, frankly, as doctors, we could look at this and be like, you know what? This is great. This is about as good as I would do if I had five minutes to sit at the bedside and help a patient understand the radiology report. That's what I would consider high quality.

Dr. Craig Joseph: Yeah. When you compare to perfection, it's hard. But when you compare it to the real world of what would you get? Otherwise, what are you going to get from another human? And then it seems like a fair fight. So, yeah, I hear you.

Dr. Justin Schrager: In medicine we have a perfection problem. Perfection is the enemy of the good. And I think that the rest of the world with AI, specifically LLMs, is just way down the road. I mean, it's something that we need to come to terms with in the medical world. And we need to find ways to be okay with it.

And what I mean by that is research, understanding how these things work, understand how they're changing. It's easy for doctors to learn how they work and to get in the weeds and learn about prompt engineering and all this stuff. I would encourage anyone listening that's a doctor, nurse or anybody else in field, get your hands dirty. It's fun and interesting and for those of you that are on the side of the fence, that thinks, “This is too scary. Don't let patients use it, etc. etc.” I challenge you to try it out and see what happens. I think you'll be pleasantly surprised.

Dr. Craig Joseph: I agree playing around is the best way, and in a low stress environment you get some ideas. You don't have to understand everything but understanding at a high level how it works, is super helpful.

Well, we could go on and on, but we have reached the end of our allotted time. However, don't get too depressed yet. I always like to end on the same question. And this is for both of you so each of you get to answer. Is there something that's so well designed in your life that it brings you joy whenever you interact with it? Why don't we start with Nick?

Dr. Nick Sterling: In general, things that confer certain leverage are really fun to use. Back in 2005, I got into biking and I got my first bike after, but when I started working as a software developer and, carried that bike with me through 2021. And then I was training for a longer triathlon and it just wasn't reliable enough.

So I went out and I bought a triathlon bike, a Cervelo. And the difference was just shocking, it was like I was hopping on to a Formula One car. I think it's something having to do with precision engineering and that every single movement that you make on that bike translates into an outside effect. Like, you put your foot on the pedal, and you just go, or you steer half a centimeter to the left and you go, you know, two, three meters to the left. It's super fun to ride. I love how precision engineered it is.

Dr. Craig Joseph: That's great. Well, Justin, what kind of bike do you have to talk about?

Dr. Justin Schrager: Contrary to popular belief, I do not have a large bike collection even though I am an ER doctor. The thing that I enjoy using is a little different. I have a Rancilio Silvia espresso machine. It's like an old school manual machine, no buttons, no switches. Everything is completely tactile. It reminds me of a lot of different things in my life that I enjoy. It's just nice when you're in the middle of the day, you know, you go down to the kitchen, you grind some beans. It smells nice. You get to use the machine. It's loud and chaotic and there's steam shooting out of it. And you know, the kids are interested by it. To me, it's just like this heavy metal machinery of an espresso machine. It is timeless. I feel like it's going to keep working until long after I stop drinking coffee and leave this world. And it's one of those things that you just look at it and you're like, all right, this is a simple design that has not changed in, frankly, 100 years. And it's just very pleasurable, too.

Dr. Craig Joseph: At some point we'll have enough of these responses that I can feed it into an LLM and, Nick, I'll call on you to help me design that. We've had a bunch of people say something along the lines of what you're talking about, like, hey, it's real basic. I have to do all the work; it doesn't take any of the work from me. There's no AI. There's no LCD panel. And I love it for that reason. And I find it's often technologists such as you two who pine for these simple things that just work and will continue to work. So I thank both of you.

This was a great conversation and I'm excited to see what Vital will do as both of you contribute to its success. At some point, we might have you on to talk about the other four big problems in health care that you're working on, Justin. I appreciate you giving us all your insights and thank you again.

Dr. Justin Schrager: And thank you. This has been awesome. Very fun conversation.

Dr. Nick Sterling: Thanks again.