This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

We want to thank our partner, AvaSure. Over 1, 100 hospitals are using AvaSure's virtual care platform to engage with patients, optimize staffing, and seamlessly blend Remote and in person treatment at scale, drive measurable outcomes and augment clinicians with an AI powered solution that deeply integrates into your clinical workflows.

Aversure offers virtual care solutions supported by a secure, scalable infrastructure that helps you to lead your organization into a future where cutting edge technology is at your fingertips. and Compassionate Care Converge. For more information, check them out at thisweekhealth. com slash AvaSure.

  📍 whatever we do, we need to keep in mind that we want to move technology forward, but not just for the sake of technology, but for the sake of the final goal, to make the lives better. for our caregivers as well as for the patients in the hospitals.   📍   📍 My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of This Week Health. where we are dedicated to transforming healthcare, one connection at a time. Newstay discusses the breaking news in healthcare with industry experts

Now, let's jump right in.

We're thrilled to have you here today, Toby. Thanks for being here. Thanks. It's great to be here.

I love your flagship product, Telesitter. It's a solution that enables trained staff to remotely monitor multiple patients, anticipate needs, and prevent injuries. And if I have my information correct, you guys are in over 1, 100 hospitals now. Is that correct?

So we're going to cover a few newer stories, but more importantly, we're also going to explore how AvaSure's virtual care solutions can align with current trends from AI technology, integration, and healthcare as highlighted.

But also challenges and strategies and implementing such technologies that can provide valuable insights for any of our listeners. So let's jump right in. Are you ready? Ready, let's go. All right DeepSeek's launch shakes markets and threatens U.

S. AI dominance. It's something we've talked about on our today's show as well But the introduction of DeepSeek which is a Chinese AI chatbot has completely disrupted global stock markets And led to over a trillion dollar decline in some of the NASDAQ composites. And it rivals OpenAI's chat GPT in performance, but operates with lower resource consumption.

So Toby, some of your thoughts on DeepSeek and what it means for the industry right now.

DeepSeek is a, I'll back up a little bit and tell you a little bit AvaSure obviously, are a AI player, but mainly in the computer vision space. But one of the challenges that we face is the problem of the amount of processing power that it takes to effectively do AI.

And so deep seek is an exciting interesting case because of the fact that they're able to do what they are with a lower amount of processing power. and so it's really a. It's an exciting situation to see how that is going to play out. I see a lot of challenges around this as well, but the fact that they're able to get higher performance especially in the area of mathematical reasoning and the complexity of the code it's more affordable and it's cost effective.

But challenges in my mind, or at least the thoughts that this brings up, Is really around the fact that it's Chinese. And so when we go to look at this for the health care environment and we run into it with a lot of our customers that they are very specific about the fact that our P. H. I. That we are using through our systems needs to be contained within the United States and not cross the borders.

And so when we look at things like Deep Seek, we're trying to figure out how we can take advantage of it. It's one of the challenges we face is, how can we take advantage of something like this? With the limitations around PHI, which by the way, I think are completely correct and they're the right thing to do is that we do need to make sure that the number one most important thing is that we protect people's information.

So that's why I start to struggle with the deep seek to be like, okay, so how can we take advantage of this? How can we in the medical environment? Take advantage of this with P. H. I. Being the way it is. So we already know that with this being in China that we're going to have issues around, censorship and governmental control.

And so that's part of it. But the other part of it is really around the risks for data privacy.

Can we protect it that way? But the fact that it is open source is also pretty exciting. So there's a lot of opportunities there that we explore to see if there's some way that we can take advantage of those things in everyday products.

And I think that it's this is something that if nothing else will push, all of the AI companies out there to explore and address these competitive challenges to us.

To your point, if we can, or say, when we solve for the privacy aspect of it, China or otherwise, I think about it being more about the fact that it's resource efficient than some of our U.

S. models. What are your thoughts on driving a more sustainable AI solution in healthcare, even if it is with an international footprint?

It's exactly the challenge, is how do we take advantage of this? Because that international footprint is challenging. And so when we look at the ways that we could take advantage of this, I think there has to be a lot of precautions, and we're going to have a lot of oversight. And openness, if we're going to be able to take advantage of this, I think there may need to be some more work around being able to have the open source models be somewhere that is a neutral territory, perhaps that can be opened up better and that can be controlled by the parties involved.

But there's a lot of challenges that I don't even know the answer to, but there's some great opportunities out there. That if we can find a way around it, it will absolutely give us some additional improvements to our performance in the AI area.

When you think about that overall performance curve and how the competition like this can shape the next generation of, say, clinical decision support tools, what does it look like?

The assumption of, hey, AI competition is going to be good. We've built it on these early models. What does that mean for next gen in these spaces?

I think you're hitting on exactly the point DeepSeek really hits here is there's this point of diminishing returns where amount of power it's going to take for us to be able to continue to advance the AI models, one of two things is going to have to happen either.

We're going to have to continue to. Come up with more and more power, or we're gonna have to come up with Deep Seek has more efficient models and there's gonna be some sacrifices that you have to make in order to be able to do that. Such as, it can't be necessarily addressing as many things all as well at launch or we're just much more targeted.

On what we're going to do with our A. I. Models as opposed to having a very broad model. That's just thinking about everything. This would be something where really have to focus it in on different areas, which would allow us to have smaller models and less processing power. Therefore, and that's my understanding is that's a big way that Deep Seek goes about.

This is that they've stripped out pieces. And really tried to focus in the areas that are most important to these models. So that's where I think it'll grow. I think we're going to see that it's a problem that we faced early on If we put processing at the edge, put our GPUs right there at the edge and the cameras, the problem we'll have is over time that hardware is going to have to be replaced more, more quickly as technology advances, because As we want to add more classifiers, detect more things in the hospital rooms, we're gonna have a limitation to the amount of processing power at that edge.

Whereas if we do it in the cloud, there's a lot more power there that can grow over time, that we can continue to add more GPUs without having to replace the edge processing. And so that's where this type of efficiency really comes into play because although you're not replacing the hardware, the cost will continue to go up as you add more and more classifiers and you use more and more GPUs.

Or more processing power in addition to the whole utility aspect

But that's where this type of efficiency, it's something that we need to see the industry really grow in and that, that level of focus I think is going to be really something that will be helpful to the entire industry.

I think a big part of it is Really having some partnership mentality that is going to be key to the A. I advancement because the reality is that everything changes and the A.

I. Models will need to change with it. So let me give you an example. So today, if you were to go into a hospital, fashions are a particular way. Gowns look a particular way. 10 years from now, we don't know what people are going to be wearing, and we don't know what the gowns will look like. Maybe people will have gotten rid of gowns and they'll be using, I don't know, sleeping bags.

But effectively, there's going to be something different. And so having a partnership mentality with your AI partners is really going to be what's going to make your AI programs really successful. Because if you go into it thinking, I'm going to buy this off the shelf model and it's going to work, It's gonna work today.

And the fact that we're using it in the cloud is we're able to take advantage of all of the video From the different hospitals, and we use we call it a anonymization model where we're able to anonymize the video and then feed it in that train our models accordingly with that. By having this type of a partnership mentality with our customers, we can work together with them to make their models better and better, and that's the way it really is with most AI models is you have to have

that partnership mentality and you have to pick your partners. Who is it that, can work well with you and that can, have the resources and the willingness to, make that a long term relationship.

Absolutely. And you made a key point about using your information to be training yours.

And then if you do source it with a partner, it's a trusted. That's the one that source of data, whether that's an outside. Although something that's been de identified to still make a scenario like an individual systems. That much better having that kind of rapport. The partner like AvaSure is going to be a definite value prop in game changer in terms of the safety and the quality and the information and the outcomes it produces,

which

is why our next article we see Baycare.

Piloting AI assisted voice technology for nurses. Three big things came up for me on this one. Workflow optimization, the ability to integrate AI voice assistance, which can completely streamline your documentation. The burnout reduction, very real. We hear about it at every summit, every dinner, every meetup that we have.

If you can lessen those admin tasks, then the tools can help mitigate the burnout, which is a critical issue for all healthcare systems. And the most important piece, which I'd love to hear you touch on, data accuracy. This improved documentation ensures better patient records. Which leads to enhanced care quality,

so what were your summer thoughts on this one?

When I read this article, this was very close to my heart. My wife was a nurse and is now a nurse practitioner, and my daughter in law is a nurse as well. And so I've heard for a long time about how charting is a big task

these nurses spend a lot of time on and E.

M. R. S. Although it's been great for being able to have standardized, form of charting and being able to provide that information more easily. to different hospitals. It is a very time intensive process for the nurses to spend on charting. And so it's different, honestly, for nurses than it is for doctors the way that they chart and the things that they have to chart for.

And so this was a very interesting article to me from that perspective, because I do think that they are two very different, important

of charting that need to be explored. Now for AvaSure, we don't do, this type of a product. But we do recognize how important this is.

And so what we've done with our product is being that we have a microphone and the camera in the rooms. Already, we've created an open architecture for our product. So that way, if you have a partner like this that can take in the ambient listening, the audio from the room and be able to take that and put it into the EMR systems, we're providing the microphone and that availability.

For any partner that you want to work with. So we have the ability to provide that for our customers to use with other partners. And so we're always looking for opportunities to make, the caregiver's lives better. On accuracy. This is an interesting problem and this comes back almost to my previous life before medical when I used to work in automotive.

The products that I worked on were around what we called connected center stack. And that was where you had voice recognition in the vehicles. And so I've worked with voice recognition before we had the natural language models really at a point where they were very useful. And so I've had a lot of experience through the years here.

And so some of the challenges that you get into, and especially I'm very impressed with the things that the ambient listening products have done today around the accuracy with the medical records. Is some of the challenges though that I see that are interesting is that you have a lot of discussions that happen in the room that are not related to or should not be in the medical record.

And so this comes back to, how does the application work? And so you're looking at it from the perspective of, when we were in cars, we used to press a button and then you give your voice command, dial so and things like that. They're going to have to do something similar to that, similar to, how we, you might give a, verbal command we have a virtual companion or virtual assistant.

That we have been developing, and it's that same kind of idea where you would say, Hey, Vicki, and you could start, telling you what you want. Hey, I need a drink of water. I'm in pain, whatever need is that the patient has. And so we're picturing something similar with these applications where you have to give some type of a verbal command.

That would say, start transcription or whatever the command is for these different partners where you would be able to, give the information as you talk to the patient and then be able to end it So you don't end up with extraneous information being entered into the E.

M. R. But the accuracy of these language models just gotten very good over the years. And yes, you still need to read it. It's not going to be perfect, but. It does really well. And so that's where we feel like there's a lot of opportunities here for making the lives of the clinicians better.

How scalable are these types of solutions that are AI based? Do you see that they have any barriers to widespread adoption? Or is it really about the planning and the understanding organizationally that needs to happen for not only a smooth implementation, but also making sure it gets ingrained in the hospital as much as feasible.

So scalability is interesting question because what this really comes down to is, and I can't speak to any, the specific one being used there at Baycare or anything like that. But when you think about how these systems will work, Is, you don't have to literally be recording in every room all the time, right?

You only have to be listening for just the, initial commands into those rooms. And from that point, then you need to bring in the additional processing to be able to do the higher, natural language models. So really this comes down to more how many of the concurrent calls will be processed at any given time.

When I say concurrent calls, I'm referring to concurrent voice recognition sessions. Or natural language sessions that are all processing at the same time. So these things could be auto scaled in the cloud. If they're at the edge, may end up with more limitations as you add additional languages and things like that.

But in the cloud, you can always do additional auto scaling. But if the models aren't efficient, obviously costs will continue to increase as we do more models. However, generally speaking, audio is a lot smaller, a lot easier to process than video is, and so more of a lightweight processing that you would do with audio than with the video.

So it really comes down to how many different sessions would be entering into the EMR at any given time. And so when you look at it from the perspective of how many nurses you have on a given shift and how many of them are actually doing anything that needs to be entered into the medical record, that's how you determine how much load you will need for each of those, because you don't want to end up with a situation where you have to tell it to wait.

Because it's busy with somebody else. So you want to make sure that you can scale to that size. So those are really just, there's a lot of things to think through around that scalability. But it's definitely, things in the cloud like that can be scaled. And so I think that there's a lot of opportunity there.

I think some of the, advantages around concepts with Deep Seek. May be able to provide some good opportunities for improvements in that efficiency and be able to increase the parallel processing aspect

In addition to the privacy and quality component we talked about, and you said, maybe it's the ability to recognize a certain person's voice, what does and doesn't get into the record.

What are some of the safeguards you've put in place to ensure you're up with accuracy of information, but also meeting those compliance requirements?

So for us, we do not cross any borders with any of our data. We're very careful about that to make sure that it is. It stays within the country that is the host or the customers country.

So we're very careful about that in order to make sure that we don't cross any boundaries there. In addition to that, when you have data entry like that, already integrate with EMRs for our system where we'll provide additional information into medical records. Specifically around like the telecenter and things like that, as it can keep track of how many stat alarms have been set off and then information along those lines can be entered into the system.

There's also a note section, though, so you always have this question of how can you be sure that the notes that are being entered are. Worthy to be in the medical realm. And so what we have done is we've added a human verification to that. So when the system does something automatic and it's going to go into the medical record, and it's not a simple of items of which they selected from, but if it's truly going to be, they're text before entering it into the medical record. It gets validated to make sure by a human to make sure that it's correct and the advantage of that as well for the models is that you get the correction back to the models to let us know what's good and what's not. And so that allows us to continue to train and make the systems better

And yet all of this takes quite a bit of power consumption, which we talked about in the first perspective with DeepSeek.

So our third article is about tech giants pursuing direct power plant access amid. Grid congestion concerns to major technology companies looking for direct connections to power plants to meet their substantial electricity needs. They're bypassing traditional power grids. Some of the key things for me here, the data center reliability, you've got to have that consistent power supply to manage sensitive patient information, your sustainability initiatives.

You've got to have opportunities for healthcare organizations to engage in sustainable energy practices, and then the regulatory navigation, understanding energy regulations and the essential perspectives that healthcare. IT infrastructure as it grows will be one of the key factors

as IT relies more on cloud based solutions, what risks do power grid disruptions pose to health system operations, and what are some of the things you've done to protect your partners and clients?

Great question. First of all, I was intrigued by this article when I read it as well, because of the fact that this is a very hot topic for us is with our own systems.

We are very careful to make sure that we have. As much of a continuous power supply as possible, whether that is a generator or will often do a generator plus two different directions of power from two different companies that both have the option of being turned on for data centers that we've worked in.

And so a lot of these challenges and we do the same thing with our fiber Internet connections is you've got to have multiple options. And so what's interesting to me about this one is they're connecting to a nuclear power plant, which is You got straight power. You're right there. And so that's pretty cool.

And I thought that was a pretty, ingenious idea was to go straight to the power plant. It did make me wonder, though, how often would you have a situation where that power plant has to have maintenance or other things where you would need to pick up power from somewhere else on the grid?

Or is that power plant truly generating all the time? I don't, I'm not a power guy, so I don't know that much about nuclear power, but it was an interesting thought that just went through my head about Is this like a great solution to make sure that we maintain uptime? The way we deal with it today is we try to have, implementations in different regions as well as different data centers when we're in the cloud so that we have the ability to switch over in the scenario that something were to go down in a particular data center.

Honestly, we've never had to use it. But I should say we've had to use it once around the crowd strike time, but other than that, it's really worked quite well. But it's an interesting problem because the amount of power that we're drawing is constantly increasing. Every customer that we add to AI is more power that we need in order to continue to run more and more AI.

And the more classifiers that we add, the more power we need. So this again comes back to then we got to be more efficient with our AI in order to make sure that we , are being good stewards. With the amount of power that we're using. And I think they use the example there of half a million homes.

I think they said that 960 megawatts would be equivalent to, or, and so I was like that's huge, and so there's a lot of power that we're looking using for GPUs and being able to do, the AI processing that we're doing to be able to train these models and to be able to use these models on a regular basis.

So if we don't come up with better, more efficient models, and if we don't come up with better, more efficient power that we can use for these, we're going to end up hitting a wall at some point. So by the fact that this discussion happened, got, I think it opened a lot of people's eyes to how big the draw is, because we all say the numbers.

Or, we say it, but it's like this one really is in your face of this is the kind of power that we need. And this is one data center right from AWS, as opposed to. All of the, different companies out there. So it was really surprising, how much that got me thinking about, this really does make mean that we need to be more efficient and we've got to take advantage of all of the tools in the toolbox to try to find the right things to be classifying and to make sure that we're focusing our power and our energy on the most important things first.

It was a conversation we covered about a month ago when we first heard rumblings about this, and we covered it in our today show because it put us down the research path of so how much nuclear energy is common in the U. S. and in the world, and actually it's still a really popular form of energy production, especially on the east coast,

so it wasn't far fetched, actually. brought me to the perspective of, do we either see health systems potentially following tech giants and seeking direct energy sourcing, or how can a CIO ensure redundancy and power for their patient critical systems?

What's a story or conversation you can share in that aspect of what you've experienced in the success of your implementations?

Great question. And that's something that was, it's been on my mind ever since I started with AvaSure.

We started out in tele sitting and we're much bigger than that now. We're, we also have what we call our episodic product, which can be used with tele health and we have a lot of virtual nursing products that we use there. But our first product was really tele sitting.

And what we did with tele sitting was we made it so that we had a more efficient and effective method. For sitting with patients that got rid of the need or reduce the need, there's still a need for certain acuity of patients where you would need sitters still, but we really reduce the need for it and we were very effective at it.

And as time went on, we started to realize we're so good at this that people were not hiring as many sitters and they were reallocating these employees to other areas. , that opens up a problem to what happens if we go down. If we go down, then they don't have the people to be able to cover.

And this was probably 10 years ago or maybe even longer than that. We went down the path of saying we have to make sure that our observer's eyes are always on the patient. We don't want to have any outages.

We don't want to have, the monthly patch updates that have to be installed. We don't want any of those things to make it so that people just, that the observers cannot be keeping the patient safe. And we actually went about designing our system in such a way that the video stays live throughout as many different use cases as we could.

And so we try to minimize those outages so that as if, for instance, a particular node of one of our clusters goes down in a particular fault domain. The other ones pick up the load, and you don't lose patient visibility during that time, and so hopefully the observers won't even notice that it's happened

Final thoughts, Toby.

So as we look forward from this point and we take a look at the challenges that DeepSeek have put out there for all of us, AI companies and the challenges that this has put forth to us of really being more efficient with the resources that we have, I think that this really is a great opportunity for us to work together with our acute care and hospital partners in IT to really advance AI to the next level and to be able to take advantage of these opportunities to focus on what is most important with our models.

And really, whatever we do, we need to keep in mind that we want to move technology forward, but not just for the sake of technology, but for the sake of the final goal, which is of course, to make the lives better. for our caregivers as well as for the patients in the hospitals. And I think this is a key opportunity we all are standing at as we look at how will we power it?

Will it, will we put it on site? Will these be servers that are on prem? Will they be in the cloud? Is there a hybrid solution for how we can power this? Or should we just go about it from the perspective of what is most efficient for each use case? So there's a lot of opportunities and I look forward to the future.

I do too, because we're talking about it and we're thinking about it and it's not happening just to us. It's this real time perspective and with the base that you have, those ideas and innovation with something that's already true for so many of these healthcare systems. So thank you for being a partner with This Week Health and the 229 communities.

Thank you for AvaSure bringing so many amazing ideas forward. Thank you for the conversation today because I know for sure it will not be our last.

Sounds great. Thank you for having me here.

Of course. Thanks for tuning in to Newsday. That's all for now.

  📍 Thanks for listening to Newstay. There's a lot happening in our industry and while Newstay covers interesting stuff, another way to stay informed is by subscribing to our daily insights email, which delivers Expertly curated health IT news straight to your inbox. Sign up at thisweekealth. com slash news.

Thanks for listening. That's all for now