Advent of AI: What Will it Take to Drive True Adoption in Clinical Workflows? | LSI USA ‘24

Natasha Allen  0:05  
So my name is Natasha Allen, I'm a partner at the law firm, fully learner. I also have the honor of also being the ai eu Chair of the firm's that sector, which we have a whole 200 Plus attorneys that are really dealing with the intersection of AI in any industry, one of which is healthcare. We're here to talk about today. So, you know, on Friday of last week, the FDA sent out a paper, right, it was CO produced by a bunch of organizations to kind of establish how are we going to regulate AI. And the one thing and the common theme that they started off with was that AI is going to revolutionize the healthcare system, something we can't deny, and that it's something that is going to happen. So we're here today to talk with my colleagues about how can we make that happen, right, how can we implement these changes in the healthcare system? And what are some challenges that are inherent within the established healthcare systems, the processes surrounding health care? So I'm going to ask each of the panelists introduce themselves. So I'm going to start off with Sasha.

Sascha Berger  1:09  
All right, thank you, Sasha. Thanks for having me. Welcome. And I'm pleased to talk about tvm and and how we see the AI landscape evolving. So I'm a partner at TVM Capital, which is a venture firm with a four decade history of investing in innovative solutions in the healthcare space. And in the in the last couple of years, we've been in particularly excited by the developments and the improvements on the software side. And that's what also bring me here today. Rob next to me, we'll we'll certainly also talk about that because we're an investor in vector medical, and I look very much forward to this discussion today.

Rob Krummen  2:01  
Rob. Yeah, thanks so much. Good afternoon, everybody. I'm Rob Krummen with vector medical. And we're an AI based technology company. We leverage the power of AI, computational intelligence and computational modeling to drive better outcomes and healthcare, specifically arrhythmia. And we're delighted that we're seeing so many other advances in this space. We look forward to great conversation. I also want to thank LSI for a great conference also for putting us together. We were at LSI. Eu last year, and we were connected with the TVM team. Fantastic team. They're one of our CO leads in our recent Series A round, and our other co lead is an LSI alum as well soulless bio ventures.

Natasha Allen  2:41  
Great and Daniel.

Daniel Hawkins  2:45  
And when it was Daniel    Hawkins had been in the med tech space about 30 years at this point, starting an angioplasty going into robotics and most recently more on the digital health side. With a company called availment systems. I was recruited by the folks at Khosla ventures to take over a company called Vista AI, we use artificial intelligence and neural networks, to replace what a human has to do in the decision making process to drive higher quality and faster image acquisition for MRIs, we're starting first and cardiac and moving on to other anatomical areas over time.

Natasha Allen  3:18  
Great, so at the beginning, you heard that the title of our panel is advent of AI, what will it take to drive true adoption in clinical workflow? For the purposes of this discussion, we're going to define AI as a computer software, or as computer software used by computers to attempt to mimic aspects of human intelligence. So I think it's always interesting to kind of level set because there's very different different definitions of AI, and the way you define it, and really has a different impact on how it's regulated. So let's get started. So this is a question for all of the panelists. We're going to start off with what are some key value propositions for companies implementing AI in healthcare? I'll start with Daniel.

Daniel Hawkins  4:03  
So I'll say I'll speak in business case, right. So it's probably best a I stick to my own playground. Humans can look at images and see certain things. And the better training you have, the more you'll see. Your judgment comes into play. The less training you have, more will get by you. Artificial intelligence has the benefit of the expertise of the best minds and best experience, but digitized. And we all know that computers can watch a cat and a dog go across the Roman distinct the difference between the two? Well, it also turns out that computers can watch images live and when they watch them live, they see things that the very, very best technicians or clinicians can see in those images, but they do so without fail. That's what this AI does. We do that with him our images. One way of thinking about what this is, is, we're the best photographer for the highest quality camera equipment you can imagine. Whereas if a inexperienced photographer had the same equipment, you'll get a snapshot. One who has really, really good experience, you might get a Pulitzer Prize winning photo. That's the difference, if you will, in our example, humans have certain limitations, and artificial intelligence can overcome those limitations and create consistency in what is produced. And that then generates higher diagnostic yield, which then changes the clinical pathways that that follow from there. So a way of thinking about this is, you're taking the very, very best of what the human experience can be digitizing it and making it operationally repeatable. And that's where we where we come in. Alright.

Natasha Allen  5:54  
Sasha Rob, do you have anything to add? Yeah,

Rob Krummen  5:55  
I'd love to add to that I, I couldn't agree with Daniel more. At vector medical, we use something very similar. But with ECG, the ECG has been around for over 100 years. And we got really good. And really high level practitioners can read an ECG pretty well. But you also have that level set of folks who don't read an ECG very well. And the neat thing about AI is it can look at those same images and see deflections, modulations, and differences that those beginners can't see, but also a little bit better than the experts can see. And you get to this step change where everyone gets that high level of data, and they can apply it to the patients. And that's the most important part, one of the things I'm seeing from our customers, which are doctors, is they're getting back to their the reason they joined medicine in the first place, which was to help their patients. So they spend a little bit less time looking at the ECG, and a little bit more time looking at the heart and how to treat that patient. So I think that's one of the main advances of AI as we're getting back to the treatment, and a little bit less time on looking at an image or looking at an ECG.

Sascha Berger  7:00  
And actually, I would, I couldn't agree more. Also, that the reason why investors like like tvm, and we are so excited about these developments. Because we're at a stage now where we really see impact, right? It's not just a promise, it's not just something fancy, we're you dream big, you see the true impact of an outcome of these algorithms and software models. Simply because we're now able to demonstrate in clinical trials, right how how valuable and how repeatable and how consistent the outcome of these models is, and that helps to, to drive adoption helps to drive patients to believe in this and doctors to gain trustworthy solutions. And that's why I mean, it's, it's become a buzzword, obviously, AI, but it's not just a buzzword. There is something happening right now, which will change healthcare for the better, I'm fully convinced. And we you see technologies coming out and every angle now, some better or some some less advanced. And I think we we have two very good examples, you're on stage, which can credibly demonstrate that, that this has an impact.

Natasha Allen  8:29  
So Rob, I think you touched on this, and maybe even Daniel as well. But some of the other benefits is also the healthcare professionals, right, and helping to augment their activities and what they have to do on a day to day basis. But I do want to get back to or talk a little about bit about what are some of the limitations surrounding the invitation of AI in the medical system?

Rob Krummen  8:51  
Yeah, it's a great question. So I think it's a really important thing to understand what AI is, what's the quality of the data that's coming in, and what's the training sets. If you haven't spent any time around AI, you should understand that garbage in or results in garbage out. So you have to have real fidelity to the training sets that you're creating, the amount of data that you're getting and how you're using it. And we may touch on this earlier. But that really speaks to the importance of validation and verification of your data. And then also really, really high fidelity studies, clinical results, that you can present a doctors, we are getting more used to the concept of chat GPT I think that's helpful. I think we're getting a little bit less concerned about the black box nature of AI. But at the same time, we have a higher standard of saying, because there's mechanisms going on which a human isn't doing, we have a higher standard to present those results and show that they're repeatable, and that they actually make a difference in the clinical outcome.

Daniel Hawkins  9:51  
And one of the things I think we ought to sort of take a moment and recognize is that AI is scary in healthcare. Just sort of get with that. for just a second, because it's making a diagnosis or suggesting a diagnosis or in the case of ECGs, it's reading that ECG so that as a clinician, yield that control. There's a personal moment in that there's medical legal liability potentially, in some of that. You can clinical, clinical study your way out of a lot of those problems when you approach it properly. And it sounds like that's exactly what you're doing, which makes all the sense in the world. There's another human aspect, though, that's critically important with us. In our case in visit, there's resistance of the staff, in our case, right. So, you know, if you log on to our website, you'll see automated MRI, well, that scares the hell out of it technicians Think about that for a second, except for the fact that they're suffering burnout working 12 hour days, and they're forced to move faster and faster and faster all the time. And they make mistakes. Well, in one of our customer locations at the Brigham and Women's Hospital, they have about a dozen technicians ranging experience from six months to 15 years. And the gentleman who has 15 years experience was wait a minute, I don't need this. I'm good, until he saw those that six months climbing up in the scales of quality and climbing down in the scales of speed. And he said, Well, wait a second, maybe I do need this. But when he got his hands on it, he actually outperformed even better his own results using the AI. So there's a moment that's in there. And to really get it adopted in clinical workflows, there's a release that needs to happen. But prior to a release, anybody worth their salt, and Healthcare's going to resist you, because they're going to ask for the proof. Once the proof is there, and that release happens, it becomes reliance. And that's what we're striving for right is to get something so high fidelity and so useful, that the actual utility of it overcomes everything else. And that's what we're striving. Yeah.

Natasha Allen  12:02  
So I'm a lawyer, and everyone asked me if AI is going to take my job, so. So I do want to talk a bit both about your companies, right. So tell us a little bit about what you're doing. You know, how is you know, for you, Daniel, how is one click, you know, reshaping how you're doing diagnostics and patient care. So,

Daniel Hawkins  12:21  
I joined the company after about 10 years where the really good work had been done on the engineering side, right. I mean, I frankly joined when there were a dozen engineers, a clinical founder and vice president of sales, and that's kind of it. Through the experiences of the clinical founder, they were able to get into an MRI lab at the Brigham and Women's Hospital. They were able to change the speed in which cardiac MRIs were being done. They studied the fidelity of the images that came from it with an independent review of pre Vista post Vista images. In other words, 4100 images that were done without Vista, and about 1100 images that were done with VISTA, and they were put in front of an independent group to grade one to five, the quality went up, and the time went down. What does that actually mean? Well, for a hospital, that means throughput. Throughput matters, because you know, an MRI machine cost a few dollars. So when you can put more patients through more reimbursement now you have the attention of the Chief Operating Officer, and that set of economics feels good for everybody. But then let's talk about the clinicians for a second. Prior to VISTA, they had a six week waitlist to get a CMR cardiac MRI six weeks. Now it's one day. So they were they completely changed their clinical pathways based on that. So what are we going out with commercially, those kinds of messages around changing the wait times expanding capability CMR happens to be one of those categories of Mr. Where most centers don't do it because it's too intimidating. It's too difficult. And the training necessary for technicians is so high that they just simply can't get the people. There's a 30% shortage across all of MRI technologists across all specialties. But within CMR, it's far higher than that. We're actually enabling it to happen in facilities where it can't otherwise be. So that's yet another benefit. Economics, shortening the time and pure enablement to even allow somebody to get the benefits of a CMR. So that's that's really where we're approaching the market. Right.

Natasha Allen  14:40  
Rob? Can you talk a little bit about vector? Yeah,

Rob Krummen  14:43  
I'd be delighted to so we're in a motion where we're building out to more hospitals around the country. And part of that is interacting, doing a lot of the same things that Daniel just spoke about dealing with the skeptics building out among the the staff and the technicians and the physicians. One of the key things things that I think people need to understand about AI is particularly in healthcare is that our healthcare system is under continual strain. The wait time is to see a doctor are getting worse, and AI is a real opportunity to help fix that problem. Because of the efficiencies that that take place, we did a study of our first 30 commercial cases, we ever use the products in and compare them like case like Dr. Seven different electrophysiologist. This was just peer reviewed published last month, and we found is 25 Productions 25% reduction in procedure time. Why is that important? Well, in an eight hour procedure, if you can cut 25% Out of that, two benefits, one, that Cath Lab cost $2,500 an hour to run. So massive savings for the hospital system. But more importantly, like you say, we can have another procedure at the end of the day, we can treat someone else who needs treatment at the end of the day. And that's really, really important. If we're going to deal with our problems and numbers. You can't train enough physicians to catch up to our numbers problem. It's just the fact. And it's worse, if you go to other countries, if you go to the developing world, the numbers are worse. We need these technological solutions. So we've been delighted to work with those. One other thing I want to say on the staffing side. One of my proudest feedbacks that we weren't measuring, but I recently heard is staff, particularly nurses that are under strain. They're incredibly stressed out by the amount of demand they have, they miss family meals, because the procedure, it goes along at one VA that we happen to be affiliated with, they haven't had overtime, since we were introduced. That's a cost savings. But the nurses have not left because they like that's a big deal. Yeah, that's a really big deal. So anyway, we're excited to be able to help the physician be partner with the hospital system. And we're excited about that.

Natasha Allen  16:49  
Okay, let's talk dollars and cents. Sacha. Why, why are investors doubling down now in AI? I think you talked a little bit about adoption, right? And people recognizing a little more that AI is out there. It's not new, but more people having an understanding of kind of what it does. But why do you think investors are doubling down now?

Sascha Berger  17:08  
Well, because I think we have reached the stage of maturity of those technologies where where you see tangible benefits. And we've just talked about real tangible benefits on the stage. So I mean, everyone knows, chat GPT, everyone's speaking about that. But that also helps, because people now get into a mode where they try it out themselves where they see it's not not just a black box, and I cannot trust it, they see they see the results. And that's the same what we're seeing in healthcare. The software has been around for a while, and machine learning tools have been around for ages. So that's not new computer power, it has increased, obviously. So that's all good. That's all in the back. But the trust piece was what was preventing a broader adoption of it. And that is what's changing. Now. That's what we see when we talk at talk with potential clients. And that's always what we do, right? We learn not just excited by a technology. But we want to understand, does that really make an impact does that change practice? So that the implementation later on is not just driven by some few early adopters, which you always have, there are always those who are excited by by new technologies, but that you get broader, broader coverage in actual workflows. And that is what what we see today and by with many companies, and that's why we're doubling down. And we see a lot of also investors who have been reluctant to go into that field now being open. And actually the same within tvm. I mean, we're, we're around for, as I said, initially for four decades. So we have seen a lot of waves of different technologies. And when I joined tvm, roughly 10 years ago, I was excited by the first models of machine learning out there, but I had a hard time internally selling it right. So everybody had this fancy stuff, right? The young kid on that, that he's excited by these technologies. And now even my partners realize that this is a real thing. And well, guess what? We invest it so great.

Natasha Allen  19:31  
Thanks. Exactly. So maybe could you go into some of the considerations you take into because as you said, this is new adoption. So it may be a little more riskier than something I don't know, like, I'm from Silicon Valley, so like an app, but you know, there is more risk or maybe a little bit more heightened risk. And you said your partners were a little hesitant to begin with but now they're turning the corner. So are there specific things that you look at when you're looking at a company to try to assess okay, this is, you know, something that we think will be accessible?

Sascha Berger  20:00  
Absolutely. So initially, we talked to you also about the fact of the importance of good data. garbage in garbage out was the what the term and and that's an important element. If we're looking at software, what's what's the background? How do these companies get their models trained? What's the source of data? And how unique is that? Because it's great if you have a team in imaging, AI, and they say they have this proprietary data set from Stanford, it's wonderful. So we build a proprietary algorithm based on this data set. Well, but who has another data set with the Harvard data or Mayo Clinic data? Right? So that's challenging from a business perspective, how, how unique is that and how defendable is that advantage, you may gain? So that's one element. And I think right now we're even past that. Because we see clinical studies, we see clients using it, we see, we see more than just a software code. And in the end, it's the same across all the different investments, whether we look at software, or hardware, or service or theoretically, it's, it's making a business out of it. So is this a business model which will fly? Is this defendable? And is there in the end also an exit route, because for us, clearly, we're in for a certain period of time. And it will only gain momentum? If there is if there's a true solution to a true need. And not just something nice to have, right? It has to be, it has to be truly differentiated.

Natasha Allen  21:51  
Okay, let's turn to how do you actually achieve true adoption? So my question is for everybody, but can you discuss specific instances when AI solutions were involved into a clinical workflow? And what were some of the main challenges or successes? I know we had a discussion on the call. But I'll let the panelists talk. Rob, do you want to take this one? Yeah, I'd be

Rob Krummen  22:14  
I'd be delighted. So Daniel, hit on this earlier, but I do think there's a higher bar to adoption with an AI technology. And in our case, and we still see it today, doctors want to use the technology, and we're delighted to have them use it. So we often allow them to use it 10 or 20 times until they're comfortable. And and then once they see the results, we're happy to let them continue using it because they asked for it again, and they become reliant on it. There's another step in healthcare though, because doctors are just one vase, there's also an admin. And so one of the big buy in moments of step change for us is when the doctors started helping us make the case to their admins that they needed it. That was a moment that we knew we'd done something really special, because it wasn't just, this is a cool thing to have. It's I'm using this, it's helping my patients and we have to pay for it. And that's the moment you have to have,

Daniel Hawkins  23:05  
you know, to drive adoption. One of the things companies hate somebody very spark told me once the trials are for attorneys and pilots or for airplanes, none of those should be in commercial processes. The reality is AI requires some of that, right? Because they need to release you know, former company in mind, Shockwave medical, we didn't give away the first few devices, but we allowed folks to use them, right, three or four cases. And then you need to be on the shelf. And that was it. That was a paid trial, if you will, but in something that is driven by software, something that is changing practice pattern. For five days, that's not enough, you need some time, right. And in our case, one of the things that we need with VISTA is time in multiple different uses of the software by multiple different technicians. And now that starts to look like a pilot or a trial. And that looks like a long selling cycle. And so we're actually pre setting the success criteria, having an agreement that should we achieve these success criteria objective, not subjective. We achieve these, we've pre negotiated the agreement and flip immediately into a commercial agreement. In some respects, think of it in the app world as freemium, but it's not really free, right. It's some version of freemium where after a certain time period, it automatically goes, will folks resist that at the administration level, right. But that's our challenge. We're market creating to be very clear, this is market creation. Whereas AI was not used. We're trying to make it a standard. So we're fighting the established workflows. And yes, there'll be early adopters, and then you'll get to 16 18% adoption and crossing the chasm is your challenge. It always is. And that is both ultimately driven by studies by economic pressure by a number of things that will find its way through the network inside of a hospital to be a decision making process. And that's where good investment capital and patient capital, with the right vision allows you to cross that, that Rubicon, if you will.

Natasha Allen  25:23  
So I had a question. So we spoke about this a bit. But what about in terms of just integrating into the existing systems? Is that a challenge?

Daniel Hawkins  25:32  
For us, it's really not at Vista in as much as we don't have our own hardware, we just spec a piece of hardware from HP, and it has an NVIDIA chip, and here's what you need, and etc. We install our software on that. And then it is connected to the MRI machine, and it literally controls it. So we don't do post processing, we do image acquisition. So the actual integration of that takes about four hours. Perfect. That's it. And we're all 100% on premises. So we don't need to worry about the cloud access and all the security things that come with that, when you do introduce those cloud access, different game. And I'll I'll let some comments here. I'm seeing some nods here.

Sascha Berger  26:13  
Yeah, absolutely. So it's not necessarily related to AI, per se, but to two different software, software technologies. And Cloud is one of those major steps, which is required for some of the AI applications. And that is a challenge, obviously, going from an on premise to a cloud solution from a data protection point of view from, from a authorization point of view, certification, internal workflow. So if you have to turn that on it, it is a harder sell, and it will take time. But that's as as Daniel said, that's for those who, who basically create that market. Once you're beyond that hurdle, and we all know that we'll get there, right? I mean, this is digitization will not stop. And better IT infrastructure is required for all different sorts of reasons. So if you're a fast follower in that market, then it's much easier. But you have to fight that fight for the first time, then then of course, it will take time. And also, for example, one one element, which we saw a couple of years ago, which is now becoming much easier is the regulatory piece of it. So the FDA was confronted with these new technologies and the idea of having a continuous improvement of those software algorithms. That has been a challenge. Well, how do you deal with that, but by now, in 2024, there are established processes, it's still early days. But when I speak to companies who are in frequent exchange with the FDA, it's becoming much smoother these days. I don't know what what's your experience here. But I mean, there are established workflows established ways of pre approval of future improvements. And the FDA is just much, much more receptive to these, these changes. And that helps also for the next couple of years tremendously.

Natasha Allen  28:30  
Yeah, I would be interested if you'd want to share maybe your experiences with the FDA, but they FDA has been on the forefront in terms of trying to get in front of AI AI technologies and processes and how to assess them. But do you have any stories or

Rob Krummen  28:43  
why I'd like to actually, thank them, I think that they're the FDA is in the forefront of trying to fix these issues and confront them and be ahead of it. It's controversial, but I've told everyone who will listen, we over disclose with the FDA, we want to be a trusted partner with them. And I try to communicate with them on a regular basis, it builds the trust that they know my data is going to be as good as I say it is. And that allows for that process to move forward. There. There is a reluctance in this industry to work hand in hand with the regulators. But I think we have that onus to do that. And I again, I would thank the FDA for advancing this issue as much as they have. You know, it's a step change, and it's not always that way around the rest of the world.

Daniel Hawkins  29:25  
So, I would agree in terms of being close with FDA and providing data early. That's what I did at Shockwave. In fact, before I even went into the first animal, I was talking with the FDA, super aggressive. My board kind of looked at me side eyed, but it turned out to be the right thing to do. You know, for VISTA I walked in it was already FDA approved. So I have absolutely no idea. We'll see what that's like. But when I walked in what's FDA approved, right?

Natasha Allen  29:52  
Is that the way to do it, then there's a way to do it walk in after all that

Daniel Hawkins  29:56  
work stuff. So

Natasha Allen  29:57  
we have a little bit of time. I don't know if there are any questions before we wrap up? No. All right. Well, you know, I want to take this time to thank our panelists. This was, you know, a very informative, honestly, we're at the forefront, right. This is not the end of the story. There'll be many chapters. But I do think that the FDA, everyone somewhat recognizes that this is going to change the way we interact with patients, the way we interact with our healthcare system, even the way that we help our health professionals right to deal with shortages of work. But, you know, I want to thank everyone, for attending. And if you can, please give everyone all of our panelists a white round of applause. That'd be great. Thank you. Thanks so much.