Keynote: The Future of AI and its Role in Reducing Health Inequities with Ashish Atreja
Episode 17224th November 2023 • This Week Health: Conference • This Week Health
00:00:00 00:29:55

Share Episode

Transcripts

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

Today on This Week Health.

we named our project the Joy of Medicine Project. We did not name it Ambient Intelligence. We named it as a program because Ambient Intelligence may be the first line of bucket that may help it. The second may be the in basket messaging. The third may be the right triage from the ER. So we launched the program of returning the joy of medicine exactly for the same nature

Thanks for joining us on this keynote episode, a this week health conference show. My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of this week Health, A set of channels dedicated to keeping health IT staff current and engaged. For five years, we've been making podcasts that amplify great thinking to propel healthcare forward. Special thanks to our keynote show. CDW, Rubrik, Sectra and Trellix for choosing to invest in our mission to develop the next generation of health leaders. Now onto our show.

  (Main) 📍 📍 All right, here we are from the Health Conference, and we're going to do a keynote episode, a little different from our interviews in action, a little longer, because we have Ashish Atreja, the CIO, CDO, at UC Davis, and big conference for you guys. That session yesterday on AI was was exceptional, and then you had some announcements coming out of it.

I assume we'll talk about that somewhat. We have all these booths behind us, and AI sort of hovers over the whole thing, doesn't it? I mean, everybody you talk to is, AI, we're doing AI. so it makes the valid AI stuff even more important to talk about. And then I just, I want to talk to you about the industry and what's going on.

Let's start with yesterday. A lot of big news around AI. What what's valid AI about and what are we trying to accomplish with it?

Yeah, I think one of the key things we are seeing is, I am a 📍 physician by background as a gastroenterologist. And I was in Manhattan as, in Mount Sinai as Chief Innovation Officer.

And the story starts like, I'm one of the 200 gastroenterologists in Manhattan. But we never felt we were competing, because we were going to the same association and learning the best from each other. And then we used to bring back that best tools or techniques back to our patients. But if you bring to innovation world, that doesn't happen.

As suddenly as it comes to innovation, we put our blinders on, we put everything under IP or potential IP, and we don't share. So what is happening is, in this innovation and transformation space, every healthcare organization has to do... It's own journey completely siloed from what other person is doing.

So repeating the mistakes of each other. We are spending resources which you could have been preciously saved. We're trying to figure out the way integrations and workflows all on our own, alone. And that is an inefficiency that we can't afford anymore. 📍 So, so part of with generative AI, suddenly we are very unique.

I truly think generative AI is a watershed moment. It's, if it was taking six years to get a good quality programmer to program something, it now takes six days to do it, right? Suddenly it's giving the power to the hands of all consumers, and it's creating a problem of plenty, because so many solutions are going to come to us, already started coming, that we can't evaluate at all.

It's also creating a problem of plenty, because they're coming so fast. We do not know how to do security for them, how to do compliance for them, is it really ethical, right? Where is the intellectual property? So, so there are major problems that are happening, but our teams are not increasing in the same size.

No, they can't. As the marketplace is increasing, right? Yeah, we can't afford to. Right? So, our growth is linear, while the solution market is getting exponential. So, and no single organization has bandwidth, resources, or skills to do it alone. So, when I moved from Monsanto to 📍 University of California one of the things that attracted me is all the UC health systems are one legal entity. together in campuses. In that regard, and there's this concept of open innovation by Henry Kempsford in UC Berkeley, which says that if we really want to do transformational stuff, you have to look at all the things that are happening outside and all the things that are happening inside your organization.

Because no single organization controls the knowledge, skills, or resources, and you have to combine the two together and partner as much outside as inside. So the UC Health System started partnering on generative AI, and what we are doing through VALID. AI is opening that up to the entire world. And the VALID.

AI actually stands for Vision, Alignment, Learning, Implementation, and Dissemination. So that's our program which we do internally and now collectively as UC we'll be working with 30 health systems we announced yesterday as founding partners and health plan and technology partners. 📍

30 partners. it's interesting because I'm thinking academic medical centers.

I'm thinking UC. You guys have a lot of patents, a lot of research that goes on and those kind of things. So that roadblock of really wanting to take advantage of your intellectual property, take advantage of those kind of things. How does, Valid. AI must enable that because you have some very strong systems that are going to create IP around AI, but they're still participating in this.

That is correct, and that's what's amazing is Under the Open Innovation, we truly believe we can create more value by collaborating.

But there will be areas where you say, okay, now we're going to take this and we're going to develop a product.

Absolutely. So our first goal was to get the foundation ready of the provider health system.

So we have 30 health systems nearly 📍 across all U. S. We have three global partners. In the next three months, we are opening up technology partners. So we'll be announcing in December, in Node Health Conference, our technology partners. We already have all the major tech vendors signed up. The cloud one.

So we are seeing two, three streams. If it is about organizational structure, it's about the use cases, it's, we can share that openly. bEcause that's a common thing that we can share. When it goes to implementation science or some specific products, we may have limitation because either it's a vendor product from outside, and there may be constraints around that, or the collective health systems may decide to have a new product being launched.

So at least the learning can be common, the organizational structure, and the governance framework. How we want to do and the security and others can be common and shared openly. But when it comes to some specific products, it can be open or it can be more limited based on the intellectual property rights.

So we already 📍 partnered with Berkeley Skydeck, which is eager to actually launch an accelerator within this. Once we are ready in that regard, we are talking with five or six venture groups, which already approached us to say they will love to fund companies which are co incubated. by multiple health system and health plans together.

So we're creating a foundational layer of trust and understanding, and everything that we learn is open, but we want to have certain health system who have some IP, not just UC, but others as well, or a vendor who has IP. They will announce that beforehand. I will carve out a place for them in that place.

Well,

I saw Mayo there yesterday. I mean, some serious players, and I'm not going to ask you to name all 30. Yeah. You probably could, I would imagine. It's

publicly available, but I'm happy to name some of them.

Yeah, yeah, share some of the names. Yeah,

so one of the groups where I was is if the real goal for generative AI is to democratize AI, it is up to us to democratize not just when we look at digital inclusion as an individual, The digital inclusion 📍 also needs to go to organization and states and countries.

What I mean by that is, we didn't want to be limited to an academic centers alone. We know they have bigger research arms, they can do a lot of IP generation. A lot of care is provided by our community partners. A lot of care is provided by FQHCs. We did not want to actually augment the digital divide.

So we've been very intentional in having a very diverse portfolio of the partners. And these are founding partners. We'll be opening up regular membership later on. Anyone can join as an open innovation network. So we have all the five UCs as a collective along with UC Office of President signing on this.

So that includes UC San Diego, UC Irvine, UC Riverside, UCLA and UCSF along with UC Davis. We have leading academic health systems like Vanderbilt. We have Michigan Medicine, we have New York Presbyterian many of those Tauschner leading academic health systems. We have major integrated delivery networks, like Common Spirit Health, which covers a large portion of the hospital chain in that regard.

We 📍 have some very special practices, like FQHC and BLEEC, like, urology practice of Virginia, which solves A lot, one million patients in OB GYN in that area, but it's not like a typical IDN. But that's a patient voice or a physician voice that brings to the

table. one of the presentations yesterday, was it that group, woman who was on the panel?

That is correct. I thought her presentation was compelling because she made the case of We don't have enough people to deliver this care in this market. We have to implement these technologies, but we know there's risks associated with these technologies. We need to move fast, but in order to move fast, we need resources, we need learning, and that shared learning, as you say, democratize the learning across the environment so that we can move forward faster.

And she was painting it as a, this is a critical 📍 moment in healthcare, like, we're going to see solutions that are going to have to step in and fill this gap this void of clinicians clinician shortage.

Right. Very true. And I think, what ChatGPT has shown on the consumer side, this is what is possible.

That's like the peak of the hype cycle of Gartner. But we really have to bring the second peak of Gartner, which is transformation.

We have to get through that disillusionment. That is

correct. And over there, yeah. And the way to do that is creating the implementation science of it, creating the playbook of it, creating how we use technology and what use case in an ethical manner, in an equitable manner to create the biggest impact.

We have to do a lot of learning. And this independent practice will not have resources of a chief compliance officer, chief legal officer, what you see may have. to write a policy, right, to do all those deliberations. But if that is being done and we can share that in an open innovation, they can just adapt that and move faster on that data.

It's interesting, the need 📍 is almost more acute at rural healthcare, FQHCs, and those organizations, and they may or may not have the resources. They are going to rely on this kind of information. They're also going to rely on their vendor partners, right? They're going to lean on their EHR providers and say, Go ahead and implement something and then bring it in.

So, how is Valid AI going to help with that? I mean, there's going to be every one of these partners out here Yes. is implementing something. And we're talking about transparency, we're talking about equity, we're talking about all those things. How are we going to measure all this stuff for those characteristics?

Yeah.

So, we're using the same framework that we have used for evidence based medicine. EDM has been taught to us in our medical school. So through Node Health Association, we expanded that to say EBDM, evidence based digital medicine. Now we're going to be expanding on that, so basically if we put science at the center, then everything works out.

What we have to create is great 📍 operational efficiency of making that happen. Can we have people, when they claim something, put a form, simple form, and say what they're claiming. How are they claiming that? Are the papers they have published, are the data they have shared? So let the owners be a little bit on the vendors, to not just say things, but actually prove with a little bit of value there.

And we'll have an independent group of people, and we'll partner with organizations. So we have already partnered with Coalition of Healthcare AI, we're partnering with NoteHealth, we're partnering with other associations. So we'll have a good body. to look at the literature and say, yes, they meet that threshold which you can trust, or they are not yet ready, but it's great to do a sandbox testing, right?

So, FDA is not going to take all of that's happening in generative AI and have a regulation on that, right? The government is still making some of the rules. Now FDA may put some guardrails around it. Some of the guardrails we have to put in our organization capacity, right? So our goal is to, instead of making it an 📍 inefficient ecosystem, where every vendor is talking to every health system right now, and that's why there's a problem of lending and decision paralysis.

Because if I'm being pitched on Ambient Intelligence by 20 different vendors, I cannot even get time to do a demo with them over the next 20 weeks with my team. So by the time I really implement it, I'm one year behind. And I may still not make the right decision because I'm doing it alone. Versus completely going collective in that approach, if any vendor has already, any health system has already gone through that process, they share their assessment, including the evidence.

So we're creating a much more efficient collective to help in that manner.

It was interesting yesterday, John Hoff goes on one of the panels. He he mentioned, here's a, here's an eloquent way of describing things and what not, but he he was saying, no physician is running up to him and saying, we need AI, we need, and it's the age old, they're not asking for the technology, they're asking for solutions to problems.

What 📍 kind of problem sets are we seeing AI be applied to right now versus what are the kind of things that we're looking at going? We believe that's possible, but that's probably a little ways off.

Great question, and I think what John, great mentor, is alluding to is we really need to anchor ourselves not to market plus first approach.

What's in the marketplace? Oh, I like this goodie, let me get it. To actually a problem centered design thinking approach. What are my biggest challenges from physician's side, patient's side, and others, and then find the right fit there. Right, so he gave some very compelling examples. I did a presentation to...

National Cancer Care Network. Today I'm part of that. There are at least 50 different use cases and some people say there are 100 use cases they're exploring. So part of the deliberation is, so we're working on a model with return on health metrics where we can say is the impact on funds, impact on patient experience, physician experience or patient outcomes and safety.

So we need to have a framework first in place to evaluate the solutions. That's the 📍 one thing, all the use cases. But once you do that, it's a X axis and Y axis how to evaluate the use cases. One is straightforward value, which could be on return on health, five pillars. But it's also about feasibility.

Because you may have something that is amazing. But that may be feasible in five years and may not get immediate value. So if you crawl that, and Gartner has a model called PRISM model, so you have to put value and feasibility, which is technical and operational excellence, inside the organization. Then you can bubble up the right use cases.

So if you use that framework, which we are using at UC Davis, the things that are coming, I will say no brainer, which I would say, within two years, majority of the health system should be getting, or would be there. AI scribe, ambient intelligence. That's a great example of generative AI. So it's not just that I'm talking to a patient and I'm doing a dragon later on, and then editing that.

I'm talking to the patient like I'm talking to you, though. 📍 Like a patient wants a physician to be talking to them. We're bringing back what EHR took us away from. There are

so many solutions in that space right now. It used to be Nuance and M Modal, essentially, were the two solutions. And now you have, there there's ambience, there's a bridge, they see me popping up all over.

Now what I'm finding with those solutions is they're ready for the outpatient setting, but the inpatient setting is still going to take a little bit of time to get right. So we're still tied to the old solutions. Because the workflow is different, right? The specialties are different, and those kinds of things.

And that's an example of outpatient. Hey, we're good to go. Let's go to town. And the beautiful thing of that is, the price point has come way down. when I heard that the biggest problem with scaling, that an AI scribe kind of solution was, it wasn't AI before. It was essentially outsourcing scribe Yes.

Somewhere else. Now it's true. That's 📍 right. AI technology. And so we're seeing the price come down

half

third almost. Yeah. Yeah. What it used to be, and that's the promise. The promise is, Hey, we couldn't, when I talked to CIOs, they said, Hey I can't scale this across my health system. I can only afford it, afford this third of our positions.

Now, in theory, if it's third in price, we're gonna

get it across the board. Yeah. And the technical feasibility, that's exactly a great example of value was always there. But the feasibility did not move higher on that regard. Now, from a financial and technical feasibility, with generative AI, it's moving right into that sweet spot.

It becomes the top. That's the number one use case, I believe, for majority. Because we cannot impact patient experience, patient access. If we cannot take care of the workforce, which is suffering today. We've got to get rid of pajama time. That's correct. That's correct. So on a similar note, other technologies are screening in basket messages from because once we open a patient portal, a patient can ask any question or...

And that comes to the physician in uncompensated time, the pajama time after they're done with the 📍 work. a major barrier or a major factor in physician fatigue and burnout as well. So you can automate some of that by putting a generative AI, which can say this question is simple. If you need to make an appointment, this is the link to make an appointment, right?

So, and you can automatically triage to nurses based on the question that's asked. So, so you can actually take care of a lot of the burden from that part. So, again, clinical exchange on an ambient site doing the workflow integration from the in basket messaging, you're decreasing the burden. And Epic already has launched this as a module.

UC San Diego, our partner site, was one of the first ones at Stanford to actually launch that. So now... We take that evidence of how it is working. Now the evidence for ambient intelligence is very clear. Around 25 percent productivity output. Around 2 hours saved. Now you can take some of those 2 hours saved in a day in an outpatient setting to really have the physicians being relaxed.

And some of that can increase the output as well. Well, I'm

encouraged by the advancement 📍 in the ambient. We were trying to create rooms, these specialized rooms with cameras. Outfit them with all sorts of stuff. And now, almost all of it is based on your phone. You come in, you get the consent.

Can I? Yeah, feel free to record. And then the phone goes down, and you're essentially... It's an interact... Our phone is recording this right now. It's this same interaction. It's a lot of eye contact. It's not the old thing. I've heard people talk about returning the joy to medicine, and this is why people got into medicine.

Look at somebody, help them, serve them, and help them get healthy. Yeah. It's a really exciting time.

In fact, we named our project the Joy of Medicine Project. We did not name it Ambient Intelligence. Oh, interesting. We named it as a program because Ambient Intelligence may be the first line of bucket that may help it.

The second may be the in basket messaging. The 📍 third may be the right triage from the ER. So we launched the program of returning the joy of medicine exactly for the same nature.

So we had Chris Longhurst on our recent webinar, and Mike Pfeffer was on as well. Yes. And they were talking about the inbox.

I want to direct this conversation to physician adoption. Because it was interesting, the conversations, as Chris is doing a lot of research. He got some funding to do the research. Published papers and peer reviewed studies, which is needed, much necessary. I forget which one talked about it, but they said.

25 percent of the physicians are like, this is great. Like I'm not even touching the note. It's fantastic. And the way it goes, but they still have , 50 percent that come in there and their immediate reaction is deleted and sort of,

is it just going to take time 📍 for physicians to get comfortable with this and get comfortable and just more swings to the bat, more uses of it. And then they go, yeah, this is pretty accurate. I'm going to check it, but it's pretty accurate.

Yeah. So we call this a great example of fit journey from innovation to transformation.

Innovation happens, we say this is possible, but the transformation really requires implementation science. So we are now getting deeper into implementation science of creating maximum value from ambient intelligence. And to be honest, Bill, I don't know the perfect answer we're to lab. Is 90 percent ambient intelligence is perfect or maybe 50 percent people who need it, get it.

And the reason for that is you have many specialties who have got so much better in asking pointed questions because it's an orthopedics or whatever those specialties are, and they have their smart phrases and a smart template. And for them, the improvement may be a fraction, 5 10 percent improvement.

And some practices like 📍 internal medicine, for example, which are mostly on multiple symptoms, social demographics and all that kind of stuff, may get much more benefit because they don't have that much structure or smart phrases in that regard. So I do feel based on the practice you are doing, and what kind of position, maybe if you're very fast in typing, and it's easy for you just to type and smart phrase, you will be done with the note in 10 minutes.

And you are so comfortable because you're doing it so many times. Initially, it may be inefficient for you even to change it because you are just comfortable doing it, right? So, so one of the things that I've learned from this implementation science, which is great to share across this network like Valid AI is, and these implementation science are not published on the paper.

People publish on the impact, but not nitty gritty details, is which speciality is the best one to go live with? Do we buy the license for entire physician group? Or we buy the license for FTE, for physicians who want to take it. And, if we really look further, maybe we are leaning 📍 towards having, if physicians want to use it, great.

If they don't, that's fine too. If the technology is really of value to them, we'll share enough details for them. They can decide now or next year to buy, right? So, let it be a pull. This is a fascinating technology.

It's amazing to me how many use cases I've already seen. I was talking to some health systems.

I interviewed a smaller health system last week and by layering AI technologies, and we forget we've been using AI technologies for quite some time, but you layer OCR, NLP you layer RPA and then you, the natural language generative AI on top of it. One of the things that was fascinating to me is they kept talking about, we've unlocked the 80 percent of the data that's unstructured, like we're making sense of it.

When a fax comes in, it automatically goes through OCR, automatically goes through. Various layers, and then it says, now this information needs to be put in these places. And and those are deterministic AI models. I'd say people, we trust them more 📍 because we can validate them more. But I'm getting off track here. Unlocking that 80 percent of the data. As a clinician, that's going to excite you in terms of the ability to get a complete picture of the patient you're sitting across. And that's really made possible. We would have to put an army of people on that versus technology stepping in and doing that.

It has to be an exciting moment.

It is. And I think part of it is just to explain, and there was a great discussion in the panel yesterday. This concept of hybrid AI. Many times, and Ken Harris was mentioning from AWS, when people talk about generative AI, sometimes the use case actually can be solved by the already existing traditional AI.

And many use cases will actually take combination. To really make things happen, you need a combination of 📍 OCR with NLP and generative AI. So, many of the medical decision making, I'm a gastroenterologist by background, the very good guidelines and rules are written. That's what is our thinking and that's why in many things it is.

So we still need a way to, a better way, to have the rule based system, like clinical decision supports in some way, combined with the generative AI. And I think it's coming in that regard. You can have a structured thing, you trust that 100%, but then you have generative AI. What generative AI is great at is summarizing from this multi dimensional data, whether it's text, PDFs, CAD, because even if a patient comes to me, I have 20 minutes, I may not have time within 20 minutes.

To actually look in a document repository, which is in PDF for the outside labs. And that may be a key piece of information that may change my decision making. Right? So many times the unstructured data is there, but it is logged in a certain way that we are force feeding ourselves to make decisions based on structured data.

Well, it is social determinants of health, many 📍 times. Right? A case manager may write something about it, but I have zero visibility on that or not time enough to do that. So Generative AI can create a summary of all those things right in front of me. And make me to allow their decisions, which is amazing, but think of it from a research perspective.

I was just in a cancer world just talking about the genetic AI, and they talk about clinical trial matching. The clinical trial matching in my world in inflammatory bowel disease and in cancer is limited by structured data. Because a lot of the things, how to recruit a patient in a clinical trial, is not in medication or labs, but in free text note or in genomics labs.

Now you can completely unlock them. And you can have a patient, you can actually have a very good Not 100%, but better than 50 percent accuracy to say this person likely will be a good fit for this trial. We can reach out. I think it's just uncovering many things we've wanted to do for many years, but never know how to close that gap.

Now it's coming there.

📍 Well, I don't usually do this, but what's top of mind for you? What do you want to talk about? I'm curious.

So I think we launched ValidAI yesterday. We got coverage. We have 30 health systems. My first goal is to get every state in U. S. covered.

I saw the map, a lot of it is covered.

What is so interesting is Maine was not covered, we just didn't get a chance. As soon as we launched it, we gotta call Daniel. You want me to call Daniel? Yeah, no, he says, hey, I want to be from MaineHealth, we want to be included. So that's great, it's a cool thing.

I Think part of it is, we have such a unique opportunity to really transform care that we could not have ever imagined in medical school. And care goes beyond just the medical care, it's at home care as well. The need for health, truly. And we have a unique opportunity to do it in the most equitable 📍 manner.

And do the science of generative AI. So while there is so much excitement, just do it. Just implement it and we'll learn from it. If we do it together like a science, we'll create a structure which will be a legacy. Like which will stay the test of time. Each one of us can build on each other. And through that, everyone will get

What you're describing is, what I think everybody's feeling is, this is a moment. At the 229 Projects event, I will ask, scale 1 to 10. 1 meaning, why are we talking about this? 10 meaning, this is going to change things. I asked CISOs the last time we were together, but CIOs I'll ask.

And 90 percent will say 10. And if you could give me a higher number, I'd give you a higher number. We're seeing, we're able to do things. This is the phrase I've been saying to people, the promise of meaningful use is finally being realized. And it's been a long time, but we've digitized all this stuff and we weren't able to get value out of it.

We got value out of it, but it was locked in a lot of different areas. Now we're able to make use of that.

And my central message would be, don't be dependent on any one technologist or engineer or one company or something. My urge is, unlike EHRs, when technology... We went far ahead and the clinicians were lagging behind, we were force feeding because we couldn't take part in designing that solutions for us.

This is our opportunity. Nurses, physicians, everyone in healthcare, startups, to actually become co creators. We need to be co creating solutions because we know the implementation science and what's better for us. We should not be just sitting and saying, let the best solution come and that should take us from zero to 100 percent right there.

We should be sitting with them, co creating, co designing, creating value and then sharing that knowledge. It's a moment for us.

Ashish, always great to catch up. Thank you, Bill. Thank you.   📍 I love the chance to have these conversations. I think If I were a CIO today, I would have every team member listen to a show like this one. I believe it's conference level value every week. If you wanna support this week health, tell someone about our channels that would really benefit us. We have a mission of getting our content into as many hands as possible, and if you're listening to it, hopefully you find value and if you could tell somebody else about it, it helps us to achieve our mission. We have two channels. We have the conference channel, which you're listening. And this week, health Newsroom. Check them out today. You can find them wherever you listen to podcasts. Apple, Google, overcast. You get the picture. We are everywhere. We wanna thank our keynote partners, CDW, Rubrik, Sectra and Trellix, who invest in 📍 our mission to develop the next generation of health leaders. Thanks for listening. That's all for now.

Chapters

Video

More from YouTube