Advent of Connected, Sensor-Rich Devices and the Promise of AI for Value Creation | LSI Europe '24

Tonya Dowd  00:04  
All right, hi everyone. Good afternoon, and welcome to today's discussion on the transformative potential of AI technologies in healthcare. It seems to be the hot topic of the event, and as advancements, specifically in sensors and AI intelligence converge, we are seeing a wave of innovation that promises to revolutionize patient care—from real-time monitoring and diagnostics to personalized treatments and more efficient healthcare delivery. Today, we are fortunate to have three key stakeholders who are experts with us, representing the perspective of a leading med tech company at the forefront of developing these technologies, a venture firm, and a strategic investor whose investment expertise is helpful in shaping the future of this rapidly evolving space. So, we'll explore how the clinical applications of AI-based devices are driving value creation in healthcare, the opportunities and challenges faced in scaling these innovations, and the impact they are having on both patients and the broader healthcare ecosystem. So, let's dive into the conversation by understanding how each of the panelists views the potential of AI to reshape healthcare. I'm going to first let each of the panelists introduce themselves. Tal, would you go first?

Tal Wenderow  01:26  
Thanks, Tonya. Hey everyone, Tal Wenderow. I'm a venture partner at Genesis Medtech, which is a medical device strategic company out of Asia, and before that, I operated for most of my life in robotic digital health companies.

Noam Josephy, MD, MSc, MBA  01:40  
Hey everyone, my name is Noam Josephy. I'm an MD by training. I started in the military as a physician and transitioned into med tech 15 years ago at Abiomed. I'm leading all the product development—everything that is bread and butter about pumps and heart recovery. On top of that, we also have a specific focus on a new generation of sensor-based and AI technologies, especially geared towards the ICU, which I think is very relevant for today.

Thom Rasche  02:12  
Tom, yeah, my name is Thom Rasche. I'm a partner with Earlybird Venture Capital. Some of you may know me because I introduced myself before, and I used to work for Johnson & Johnson in the sensor-based business for 16 years before I stumbled into the venture world. I have been doing venture for 20 years, mostly focusing on med tech, at least for me, even though our fund is completely healthcare-focused, going from biotech to diagnostics and medical devices. So, right.

Tonya Dowd  02:38  
Thank you, gentlemen. Great perspectives here. So, AI is the hot buzzword. It's the hot buzzword that has all the hype right now; everybody kind of throws that term "AI" around, but—

Noam Josephy, MD, MSc, MBA  02:51  
It's been around my title, by the way, right? Exactly.

Tonya Dowd  02:55  
It's been around for a long time. Why now? And why in healthcare?

Noam Josephy, MD, MSc, MBA  03:02  
Go ahead. Go ahead.

Tal Wenderow  03:03  
Okay, so I think we're seeing some kind of a similar dynamic that's happening in med tech as we have seen in the buzz in the world of high tech with Nvidia. It doesn't come on its own. There's a set of things that have to happen, and it's the data capture, the ability to get a lot of data at a very fast pace, and then there's the ability to process and host those data at very large quantities and speed. And then you need to start building insights. So that's the word—that's the hot topic of AI and the insights from that. You can go back in time to this evolution all the way from the beginning of Facebook, which, in a way, was a sensor—everyone just putting in their own information. It was a great sensor. And then we've got all the cloud-based housing of all that data, and now the ability to process it is skyrocketing. We have a similar thing in med tech because now we have sensors that are mostly digitized, much less analog. Obviously, the processing power from high tech is coming into that. Everything is going on the cloud. All of our pumps are now housed in the cloud; all the information is housed in the cloud. So now it's really coming to the exciting part of putting it all together. So now you're seeing it here. It's not about hype; it's about clinical outcomes. So that's the next challenge—how does it help patients? It's the next challenge, really, for med tech.

Tonya Dowd  04:40  
I remember the days of MySpace, so I'm going to date myself there. I'm on Facebook.

Tal Wenderow  04:48  
But I think on top of that, healthcare is a regulated environment, so it takes time. To known points, we got to that critical point where there is enough data, enough conviction, and enough small players coming from other industries that are actually trying to enter healthcare. Let's try another industry. Meaning, I think some hospitals still work with faxes even these days, right? I don't know the last time I sent a fax to anyone, but the time is right now because healthcare is a slow follower, especially on technology adoption.

Thom Rasche  05:19  
Yeah, I think "why now" is an interesting question. Because I think, one, the devices are now at a stage where they actually transmit, which they haven't done for a long time. And to me, it's out of the question that the granularity you can get with AI analysis will lead to better outcomes. That leads to better information. Let's put it this way: a physician would never be able to detect when you take it at a very low level in image recognition and radiology or anything like this. I think to me, it's clear that a computer can do this much better than a human being. I think there is no question about it. What I, as a venture capital investor, always question is whether that is actually a self-standing business. Is that truly a service business you can run on your own in parallel to the devices or the drugs, or in terms of the providers? Or is that actually an integral part of either a radiology company or a med tech company, or anything like this, where you say, "This is a service I will have to provide as an analytical service." Now you can argue that, let's take Abiomed as a good example. They are covering a certain segment of the treatment path, naturally so. But what about all of the other infusion pumps, the drugs, the clinical information—all of that—and how do you get all that together? And who's going to do this? This is Abiomed—very unlikely. Why? So who's going to do this, and who's going to pay for it?

Tonya Dowd  06:54  
That’s a great segue into now. Can you explain what specifically Abiomed is doing in this space and why it's so unique? And I know you brought some toys with you, so—

Noam Josephy, MD, MSc, MBA  07:10  
Well, I learned very quickly that you always sell, right?

Tal Wenderow  07:18  
Always be selling, right? Always be selling.

Noam Josephy, MD, MSc, MBA  07:21  
So let me just quickly explain what we do. This is the Impella heart pump. It gets implanted; it goes into the heart. This part sits in the left ventricle, and the other part, this would be the aortic valve, and this would sit in the aorta. It's a very simple conceptually—there's a cannula, and there's a pump here that pulls blood from the ventricle, and it does it in a physiological way. So it doesn't bypass the heart; the blood will flow in a physiological way. So that's been our journey, and that's getting to a $1.5 billion business as of coming out of 2024. The challenge that we had was about making it smaller, with more flow, more durability, and more blood compatibility. Those things would capture about 90% of all of our product development. But we've broken a ceiling because now this pump, for example, can sit in patients for months. Patients can get out of bed; patients can go to the restroom. They can walk around the hospital with this type of cardiac support, which is like walking with a crutch for someone who has a muscle injury. But now the next challenge is: what do we do with this old information? So more time on the pump means a lot of variability in care, nursing staff, physicians, residents, and fellows, and all that. Our approach to that, to your point, is that we want to standardize that care because that will determine the outcome of the patients. How do we standardize the care? We can't train everyone because that's literally boiling the ocean. Our approach would be through data. So we want to start using that pump as real estate. We want to start putting more and more sensors on the pump. And that gets me to what I started with, which is that we have more sensors. We need a place to receive the sensors; we need a place to integrate the data, to show the data, and then get insights from the data. So that's, in a nutshell, our vision for doing that. We are not a software company; we know how to do coils and motors. We would love to partner with those companies that don't have the real estate in the ICU so we can partner around that and create that umbrella of patient care in the ICU.

Tonya Dowd  09:40  
Where do you see the data going, and who is it being used by?

Noam Josephy, MD, MSc, MBA  09:46  
Should I take this? Well, perfect. We'll start. You always have to start with the bedside. That's our view. It's easily challenged, but in my view—and that's the conviction we have at Abiomed as well with our Impella Connect, which is on the cloud—it comes down to the bedside. The bedside has to run the management of the patient, so the first information has to come there. You can have a reflection of that information on the cloud, and you can use the cloud sometimes for computing or storage, but it has to be bedside. There's enough to solve at the bedside, with all the variability of care and different providers that there's room for a lot of innovation.

Tal Wenderow  10:33  
I assume we send anonymous data to the cloud, right? And you use it for your own purposes. Did you encounter hospitals that will tell you, "Data is mine; you can't touch it for your processing," or are they more open right now to do whatever you want, as long as it's anonymous?

Noam Josephy, MD, MSc, MBA  10:51  
In the US, it varies a lot, so there are probably as many opinions or more than the number of hospitals, or more so the number of chief information officers or IT personnel. Also, post-COVID, the IT personnel in hospitals are drained—there's no one there, so you're trying to find people to talk to, and they're overwhelmed and doing the mundane things. I think that's an opportunity. It's an opportunity for companies to come and take this over, make it very easy and streamlined. To put it specifically to your question, we started our Impella Connect, which is our cloud infrastructure, to see where all the pumps are running, and we have it for the entire US now. We have 1,500 hospitals with the pump data streaming online. We started this in COVID. So that's one of those examples where COVID had some degree of blessing to facilitate and accelerate certain things. We went into COVID with 50 accounts online; we came out of COVID with 1,000 accounts online. Wow. And it was mostly because the hospitals needed the help, and they said, "Let's get through this because we want your people bedside to help with the implants." I think now hospitals are having to pull back from being so permissive with certain things and being more risk-tolerant. So now you have to get into the rigor of business associate agreements and being part of that, or have partners to do the BA for you.

Tonya Dowd  12:12  
What about outside of the acute care setting? So, you know, the watches and the other sensor types of technologies that we might be wearing or using in the home. Where do you see that data being useful?

Tal Wenderow  12:25  
I'm with you guys. We can help you a little bit. You want some water? But we can help. It's all about who uses the data. What do you do with the data, and is it a standalone? As an example, I ran a vocal biomarker company, and I can tell you there's a lot of startups doing things in silos, and it's not good enough to move the needle, whether it's outcome, cost, or patient care, and you kind of need to combine them. But a startup cannot do that; they cannot start three companies together, meaning it happens with the big tower that everyone is connected. And I think Noam is trying to do the same with Impella. But the Apple Watch, I think it's a great tool. It'll take time, and the question is: what do you do with that? And if they, for example, view, if you send me some information, that's a risk because what happens if you miss it, and there's information that you actually missed? There's a lot of risk. And then who do you send it to—the patient, the physician? How fast does the physician think? We're not there yet. We're not even close to being there.

Thom Rasche  13:26  
I think that's one of the biggest questions. I think the Apple Watch is detecting an AFib in the middle of the night. Is Apple going to call the physician? And by the way, talking is like, there are liability issues involved with Apple. So are they saying, "It's like you call 911 to actually get an ambulance sent out," or stuff like that? So how the hell are you going to solve that? Now that should hopefully not prevent us from collecting the data, but I think the whole data integration part is actually really a catch-22. I really don't have an answer to this, but if you look at it, as much as I understand any company, be it Abiomed or anybody else, to say I'm interested in my analytics or for my pump in order to get better treatment for the patient or better adjustment for my device, I think this is perfectly fine and understood, but that's only such a small piece because margin comes to the infusion pump, comes to this old diagnostics labs. Am I shredding the blood? Yes or no? So there's so much information which is currently available. It's already put in a format which is standardized. So that was one of the problems we faced, but I think that's now solved. But who is going to integrate all of that? Is this the hospital information services companies? Is it Apple, Google? I don't know. Who is it? So who brings all of that wealth of information together? And if they do so, somebody's got to pay for it.

Tonya Dowd  15:45  
How? Yeah, these are great points. What about the business model and the reimbursement? Who's going to pay for it? Maybe Noam, you can speak to what is the business model for your sensor technology currently?

Noam Josephy, MD, MSc, MBA  15:59  
We are fortunate, slash spoiled, in a way, because we in our business, it's the sickest patients in the hospital. So one in two patients that get our pump is statistically likely to die. So we know our outcome very quickly, and what we've learned over the years is that better outcomes drive better business. It seems very trivial, but we see this very timely. So we have the ability to step into a hospital, look through their patients from the last 10 months or 12 months. We actually do this regularly, and we do our outcome review, focusing on the clinical and trying to understand what the variants were and where the variability came in. So we have an incentive to keep on doing this, and we have a large margin of improvement still. I mean, we are moving the needle. We just had an RCT that was published; we showed a 12% improvement in absolute terms, but it's still 60% of patients that are surviving and 40% that don't, so there's a lot of room to improve that. So we have the incentive, and we know that it drives business, or the economics for us is one that is mostly internally faced. We don't necessarily need to move prices in order to justify this because we know that it moves the business. I think a lot of the things for smaller companies will be to find us or some others that have this thing and are missing that expertise, so we will partner to be able to do that.

Tal Wenderow  17:34  
That's also the challenge to your point of startups. They cannot just wait for the strategic to come and partner, and you can do 124 deals a year if you're really aggressive. Yeah, there are hundreds of startups out there, and startups have the luxury, I think, in my opinion, not just that, but you started with the therapy, highly reimbursed, and then you add the sensors in the eye, which is a much easier path. If you're only a software-based company, that's exactly the challenge. How do you show that, and you don't what I call "death by pilot"? When you go to all these new players and they say, "All right, let's do it for three months." You're all excited; you come to the board, "Amazing! United wants to work with me!" And that three months turns into six months, and the results are not conclusive, right? It's not black and white because it's never black and white. So unfortunately, we don't have the recipe or the answer. Otherwise, we'd be doing it right now.

Tonya Dowd  18:27  
I think, Thom, you had mentioned the key with partnerships, right? So med tech or AI partnering with pharma companies to have more precision medicine. Can you speak to that a little bit more and the business model for that?

Thom Rasche  18:40  
Yeah. I mean, at the end of the day, I think if we collect enough data, be it in the intensive care unit or even in any hospital, you will be able to determine, for instance, any of those new antibody therapies which are looking at AI models to determine which patients will react to a therapy. You can do this in pathology labs, where you actually do AI analysis of the slides in order to help pharma research understand what drugs react to what patient population so that you can segregate patients much better. So all of that, I think, is actually very feasible. And of course, pharma has a high interest as well. You could argue you limit your market, but on the other hand, you will say that those where the therapy is actually active, and you know those antibody therapies are extremely expensive—they go for $300,000 to $500,000 a year—to say, "Well, you can go to the payer and say, at least we know it works for this patient population and not for that patient population." So those are, for me, kind of like the initial steps to segregate patient populations. And there it's actually rather simple because, well, it's not simple, but at least you have a target. So it's not all data together. And when we talk about the whole hospital process and how do you gather the data there and actually make it meaningful for outcomes, then I wonder—and again, I have no idea—how do you define a stepwise approach to get to ultimately the whole thing? I mean, how do you go? I mean, any company will do something. I know Braun is doing something on the perfusion machines, and I know Siemens and GE or Philips are doing something on the monitoring, which is great, but there is a piece missing, and I don't know what the step approach would be.

Tonya Dowd  20:37  
What about just shifting gears a little bit? What about patient privacy from all of the data going around everywhere, and we're not sure, as a patient, right, where it's going? What can you speak to that a little bit and the challenges there?

Tal Wenderow  20:51  
Meaning, first, obviously, individually, I'm not concerned about that. I remember my wife used to work for early detection, you know that, so the patient monitoring underneath the bed, and we put it under our bed at our house, which was listening to that. But I'm personally not concerned about that. I think obviously the regulators are concerned. And you kind of raise a bigger question than that: who owns the data? Is it the patient? Is it the hospital? Is it the companies? And that's a huge one. Meaning, officially, it's the patient, but hospitals own all this data. I think you have to trust the companies that are doing a good job protecting the data. No one asks, "Is the data secure?" It's obvious. And I think we, as a community of startups and strategics, have to do a good job in protecting the data and not take shortcuts because one loss of trust or a breach can be not only deadly for the patient but also for the brand going forward.

Noam Josephy, MD, MSc, MBA  21:46  
Without your loan, I would second Tal's point in terms of the personal aspect, especially if I, you know, each one of us relates to their own therapeutic area. So if I, when I think about patient privacy, and I think, God forbid, myself in an ICU, I don't care about my privacy. I need to get stuff done at this point. However, it is something to think about in a broader sense, both because not everyone is in the ICU and there are data that are private and can be used in a malicious way. The other part is also maybe hand in hand, the regulators are coming in much stronger on cybersecurity on every level. We're seeing that there's a new regulation that came out last March, and it's not only elevating the level; it sends a direction which then your risk-averse colleagues can easily overly interpret that. So it creates a higher level of scrutiny even within the organization because you say, "Well, I'm going to do only this," but a year from now, something else will come. We have to actually plan for that, and that starts delaying things, and I'm sure for smaller companies, it delays their timeline or increases their cash needs and all those things.

Thom Rasche  23:07  
Yeah, I think I would second that. I think only the health—

Noam Josephy, MD, MSc, MBA  23:09  
You worry about data security, yeah.

Thom Rasche  23:11  
So once you are sick, I think this is the last thing on your list you worry about. And I always make, and I think we talked about this, the comparison to banking. We do electronic banking, I don't know, for what, 30 years or something like this. And is anybody worried about the data security of my banking system? I think everybody uses this, everybody that uses the app and everything. And with banking, AI with money, you can do something immediately with your health data. That's definitely not a direct connection in order to make something with it financially or something like this. This is actually not that easy, not at least not that straightforward. So I would actually be much less worried about the data misuse and all of that compared to my financial data, which I would like to be protected.

Tonya Dowd  24:00  
So from a regulatory perspective, the FDA just came out with a kind of a joint statement with four different centers from the FDA in terms of what they promised to do regulating AI technologies—how they're going to put more scrutiny around this from the concept all the way through utilization. What are your thoughts on this, and is there too much regulation now? Not enough? Where does it need to go?

Tal Wenderow  24:30  
You want me to start? Go ahead. I don't think there's too much regulation. I do agree with what Noam mentioned: that each one reads the regulation differently, and startups look at it like, "Here, big companies are there," and there's some place in the middle. I really believe that when I interview a quality guy or regulatory guy, they need to provide the boundary lines, right? You cannot do tests and this in between; there's a lot of gray zone, and you need to be comfortable and honest with yourself that you're doing the right thing for the industry, for the patients, for your product. I don't want too much regulation, but I do think there need to be some guardrails from a risk analysis and all that, that you actually walk through them, but not overly because it changes, as mentioned, every day.

Noam Josephy, MD, MSc, MBA  25:14  
Yeah, it's evolving.

Noam Josephy, MD, MSc, MBA  25:18  
I don't know actually if there's too much because it's relatively new, and it's like with the law system—until there's a ruling by a judge, you don't really know where the limit is. So we're testing it, and it's hard because many times when you test, you want to test to see if you succeed. So you put out a device, you put a submission, and it goes through. You actually don't know what the limit is because if the boundary was here and you gave yourself so much room, then you succeeded, but maybe you had some more flexibility that you didn't allow yourself. But companies are afraid of getting a "no" from the FDA, so I think it's not until we get a few "no's" from the FDA and then negotiate that we'll know exactly what the boundaries are. So I'm at the position now with the new guidance that I'm actually concerned with over-interpretation rather than over-regulation at this particular point in time.

Tonya Dowd  26:23  
Yeah, and the AI technology that's been cleared to date has been through the 510(k) predicate process. Do you think that process needs to change and be specific to AI technologies because they're kind of, you know, applying an old process to new innovation?

Noam Josephy, MD, MSc, MBA  26:45  
I live in a PMA world, and it's a nightmare. It's the right thing for patients, but it's really hard. I hope they find a way that is not PMA for PMA-like because the PMA is not just the burden of proof that you have; it's also the systems that you need to have behind it to sustain it. And if you apply this to software, then you miss the whole edge of software, which is agile and quick learning and the ability to adapt quickly. So I would love the regulators to be able to adapt to that versus force AI into a PMA universe, which I think will be a nightmare. Meaning, I think—

Tal Wenderow  27:30  
The FDA is most concerned about software that actually adapts and learns on the fly. That's what they're really concerned about. Because if you do design phase, this is the version that goes out, they're typically okay with that. If you start saying, "My AI engine, I'm going to learn from your patient," and I would change the Impella product on the fly based on that, they're not comfortable. And I don't know how to solve for that right now, and I think that's a big challenge for AI—the self-learning and iterating without your interruption. That has to be the biggest challenge for regulators to solve.

Thom Rasche  28:01  
Yeah, I think the durations—it's really difficult to define what I would call the borderline. How big can the duration be before it has at least perceived or even after the fact? Mostly no, it actually had an impact on patient treatment. And so this is the hard piece to actually say, "Okay, how much do you allow software as a service to actually get iterated without maybe just with a notification to the regulators or something like this?" So this is the most difficult thing. And by the way, don't forget, this is all global. All of a sudden, this is not FDA-based anymore because any company would love and mostly actually have global products rather than one specifically designed for, let's say, the FDA as a regulator. And I mean, it's a good example because at the end of the day, you would like to actually have tech companies and med tech companies as innovators act in the market initially fairly freely, but when you look at social media and stuff like that, you can say maybe, "Well, that was a little bit too freely." Nobody knew, I mean, only after the fact. But so it's really a very difficult answer to give. And here we talk about people's lives, potentially at least. So where that boundary is and where software can be iterated with a notification versus where you need to go through a potentially amended approval process is hard to say, really, because you only know after the fact.

Tonya Dowd  29:38  
Kind of on that note, from an investment standpoint, what key performance indicators are you looking for in these technologies before you invest or consider investing?

Thom Rasche  29:51  
I can answer this very easily. I mean, it's outcome, outcome, outcome—show me. I mean, most of the AI and the lit six we see. And, I mean, we have a lot of business plans. I mean, actually, I think it seems to be fashionable. If you don't have AI on your business plan written, nobody is going to look at it, which is not right, but right now, that's kind of like the impression I have. But the question I have is always the "so what" question. So you do all of this data analytics, you do all of the analysis, you have coming by, and I have charity data together with the LMU, and then the next one comes by and says, "I have the Mayo data, and mine is better than yours." So protection, uniqueness, outcome—those are the things which I would actually look at to say, "Let me understand why Abiomed made at some point in time would want to acquire you, and why you believe you are unique enough that they say, 'This is a cool technology. We really need this or want this, and it helps people.'" And I think that's kind of like the measure we would always apply, which I know is frustrating for most of the tech companies because the tech companies say we can't really protect our algorithms. I get that when that goes back to my initial comment to say maybe it's not a self-standing business, which I really—

Tal Wenderow  31:17  
To add to that, when I look at things like that, I look at the standard of care in healthcare, which is very low. There's, as Noam mentioned, there's a lot of improvement. So you see a lot of AI companies saying, "I'm better than the standard of care." But when you look into that, the bar is much higher for AI. So the standard of care is 60%, and you come with 75%. Is it good enough? Right? Does it move the needle? And so that's one thing. The second thing is, I'm not a data engineer or studied any CS, but it's the strength of the algorithm because you can find a point in the use case that you have high accuracy and high specificity. I always ask, especially AI companies, "What's your AUC?" and see what's the strength of the algorithm, and that's built on data. And how do they train the data? And is it applicable? And we discussed global: I want to ask Noam, is it the same in the US, Germany, Israel, and China? Is it the same algorithm? Do you need to adapt it, or is it globally applicable?

Noam Josephy, MD, MSc, MBA  32:15  
I agree with that. Let me go back to from an investment standpoint. I'll say for us as a strategic—and now I'm becoming much more comfortable with the J&J environment—there are two avenues for investment. I think one is the traditional, which I think the same way that you guys are looking at it, and the other one is what we call "build to buy," which I think David Voortman from Ultragenyx has mentioned. I think that's public information now that they have a build to buy with a beam, and this is an opportunity for them to come and solve something very specific that we need. And there are a lot of things that are pre-negotiated. There's de-risking from our end, which comes mostly as non-dilutive to get like in our A, so to get the engineering and the R&D going, and they need to meet specific milestones, and they come with a commitment for a long-term commercial agreement. It can take different forms. So that's our investment. And they have a heavy AI component. So for us, if they come and they solve a specific problem, to us, obviously it links to patient care and outcomes eventually, but we can reduce it to something that is much more defined. So if they come and they solve that, this is attractive for us. I think on the other part, in terms of improving outcomes, it is a challenge. It is a challenge. But I would come back to the fundamental: just like with devices, you don't want a device that's looking for a problem to solve, right? You have to find the clinical problem or whatever problem it is. It may be an adverse event, or maybe it's an efficiency or standardization that you're able to quantify and solve, yeah. So get in the marry the two—the problem and the solution. And I see a lot that are enamored by the AI. And definitely, when I talk about AI, the hands start waving, and there's a lot of things like, but eventually you have to marry the problem with the solution. AI or not, yeah?

Tonya Dowd  34:18  
And it's like with any technology, it's what is the unmet need, right? And I think, you know, Tal, you talked about specificity and sensitivity in terms of does it work? But what about clinical utility data to show that it's actually making a difference in patient care and changing decisions, right, of the provider, of how they care for that patient?

Tal Wenderow  34:38  
Of course. And so far, I think we're focused on this discussion more into the in-hospital setting to help surgical procedures and others. There's a big area of hospital-at-home, skilled nursing facilities. The bar is lower because they have stuff there, but I agree, meaning it's all like everyone mentioned. There's a lot of amazing technologies. I'm a startup guy by heart, so I see that. It's like, "Wow, unbelievable!" Without sending any bad words, here I go back home, and I was like, "That doesn't make sense. How do you adapt that? What's happened there?" So without being too generalized, it's all about the outcomes, like Tom mentioned.

Tonya Dowd  35:17  
And the outcomes being proven in what way?

Tal Wenderow  35:26  
This one is critical benefit. That says this one, there's a lot of companies trying to solve cost. And I think on paper, it sounds really good. The problem with solving cost is demonstrating that when you're that small of the therapy solution, it's really hard when you come, "Okay, I can reduce readmission by 20%."

Thom Rasche  35:44  
There will always be somebody who can do it cheaper. So cost is usually, most of the time, not really an argument to introduce a new technology because unless it really dramatically changes the process. I mean, and you talk about cost savings in the 30%, 40%, 50% range rather than 10% or 5%, something like this. I think at the end of the day, it's mostly about clinical outcomes. It's all about show me patient outcomes. Does it improve the outcome? And what is the measure—side effects? We may change side of care or anything. What is the measure you would like to show, and how big is the trial you need to perform in order to show that, which is basically the question we ask all the time, be it in drugs or med tech.

Tonya Dowd  36:29  
Yeah. Can you speak to the trust factor from the physician side and perspective? Do you think the majority are there in terms of trusting AI to augment their decisions? I dare I say, replace their decisions, but augment.

Noam Josephy, MD, MSc, MBA  36:46  
I'll jump into that. AI was disserviced several years ago at the outset of that—a lot of promise—and there, I don't know if people are familiar, there was a sepsis algorithm that was developed by Epic, and there were a few that came up with the hype. And also, to Tal's point, they assumed it's okay for everyone. It applies to everyone, but it actually applies to the population that you trained it on. So it's as good as the people that you train on. People in South America are different than people in the US, and they're different than those of European descent, and when you go to Japan, we need to do sizing because the population as a whole is smaller than the population in the US. So we need to show, and we need to give guidance about that specifically. So there's a little bit of a reckoning in that. And the other part, from the physician standpoint, they are overwhelmed with notifications. Yep. They open up Epic, and then click, click, click, click, click, click, click, click—okay, now I can start working—which is dismiss, dismiss, dismiss, dismiss, dismiss, dismiss. So you're up against that. It's not a naive space anymore; it's something that you work against an exhaustion of alarms and alerts.

Thom Rasche  38:01  
But then actually, I think that comes down to the user interface. I completely agree. Most of them are either deleted or dismissed. I think it's kind of like introducing the electronic calculator where you say you still need to use your brain when you actually do what you do, rather than just trusting the system. So I think at the end of the day, I see any tools—and I call them tools—as support tools, but that is not to say that you should be shutting off your brain. So I think that's the key thing here, which I think any software tool will provide. So I'm not really scared. I think it's actually, if it's utilized right, it should be supportive.

Tonya Dowd  38:49  
So we're out of time. I can't believe this. I could sit here and talk all day with you gentlemen. Parting thoughts on the opportunities that AI brings to healthcare delivery? Tal, first.

Tal Wenderow  39:01  
First, I just noticed that we did not even mention ChatGPT on this panel, which is really impressive. Well, I used—

Tonya Dowd  39:06  
It today at least a few times.

Tal Wenderow  39:14  
That point. But I will say for, you know, it's very easy to sit here and say, "What are the challenges and how hard it is?" In the end of the day, it will come. So for all the innovators, all the startups, all the companies, just find that really where you feel comfortable, that it helps work with the clinician and make it happen, and then with the strategic or without, or with the investor or without, just make it happen, and it will come. It's just a matter of, you know, the transition from manual shift to automatic shift—from BlackBerry to iPhone. It's not an "if," it's a "when."

Noam Josephy, MD, MSc, MBA  39:48  
I will just second that. We call the environment—and going back to the ICU—but I think it applies to everywhere. We live in a data-rich environment, which is also relevant for the ICU. It's a data-rich insight-driven environment, and you can apply this to any system, and the computing power is just so much better than what we can do in our own head, which allows you to do the classic elements of disruption, right? You can reduce the level of training needed to do certain tasks and certain things if you have a good enough supporting system. I think this is the future, but I'll just echo Tal: you have to know what you're solving.

Thom Rasche  40:27  
It's a great opportunity for patients. I am 100% convinced. And any startup, I can only recommend: find the boundaries on where you can operate in because I think the biggest problem is if you try to solve it all, and they gotta work. So you have to define the boundaries in which you operate and what problem you can actually address and solve. But there is a great opportunity, I'm honest and convinced.

Tonya Dowd  40:54  
Tal, Noam, Thom, thank you so much for your time today. Excellent conversation. Thank you everyone for attending.