December 20: Today on the Conference channel, it’s an Interview in Action live from CHIME with Yauheni Solad, MD, CMIO of Digital Health and VP of Innovation at UC Davis Health. A discussion about the potential and pitfalls of generative AI in healthcare took center stage. While Yauheni acknowledges the excitement around AI, he strongly advocates for a pragmatic approach that recognizes AI's limitations and stresses the importance of understanding what you're dealing with before jumping right in. Is AI the silver bullet for healthcare problems? How can we effectively introduce it into clinical operations? Yauheni poses his framework and deep reflections about AI in healthcare to these questions while highlighting the importance of maintaining realistic expectations.
This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.
Welcome to This Week Health Conference. 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 and events dedicated to leveraging the power of community to propel healthcare forward. Today we have an interview in action from the Fall Conferences on the West Coast.
Here we go.
We are at the Chime Fall Forum. I'm Reid Stefan, VP and CIO of St. Louis Health System, and I'm joined today by Johanni Alad, who is the CMIO of Digital Health and VP of Innovation at uc, Davis Health.
Johanni. Good to see you again, my friend. So you and I are both bullish on generative AI. Like, on a scale of 1 to 10, it's broke the scale rather than 11. But, with pragmatic kind of perspective, so we've talked about this before, but maybe just share like, how you're approaching that at UC Davis Health. Be excited about it, but also to be pragmatic and appreciate that it's not a silver bullet, and there's understandable considerations you have to make going into it.
So how do you, what's the framework and mental model that you're approaching with? So...
Great question. Can it be the similar thing? Yes. Maybe. Right? I think we need to know. Okay. In the same way as when we're looking at nuclear fusion, you cannot just start automatically with nuclear reactors.
You need to ensure you
understand what you're dealing with and that you actually have a proper safety workflow to know when to use it, when not to use it. Right? And we certainly now, it's evident for us how unright we were... Where initially it was radiation, right? But it took some time. We're going in a lot of large language models with a mindset that it can solve a lot of our problem.
And it definitely can. The question is how you properly titrate it into either clinical and operations to both learn from the process as well as start to benefit. Because the last thing you truly want to do is create this unrealistic expectation that tomorrow you'll could be easy. The magic button.
Yeah. Right. And some chat bot will come and solve all of your solutions and you'll be able to automate the full workflows. Right. It's a very nice, mind frame to potentially think about where future may go for human automation. Right. And the support. But before we get there, we need to ensure we establish proper processes on everything.
Yeah. 'cause after all, a lot of the models are sequence to sequence that's predict based on the data. So, we need to ensure that the data is not only FHIR, but we actually, frankly, know what data is being fed. Because you may not
really know a lot of this stuff, right? I like that. So, we
need to, as a healthcare system, be equally excited and realistic.
Okay. So, I like that. So, before we kind of build the manned mission to Mars, let's maybe get good at some, like, low Earth, like, satellites that we launch in the low Earth orbit. So... Are you prioritizing clinical or business opportunities, or both? If you think about that analogy, like, what are the low Earth orbit kind of, wins that you've found so far?
clinical care. Yeah. Use cases possibly are not a good start. Yeah. Right? You're not going to brain surgery immediately.
So you always try
to start with the areas where, A, you have... Of good predictability, right? And you have enough monitoring of the input output as well as the process
so where we're looking at right now is that all the processes that we may have in some of the
or the workflows that can offload monotonous, right, and easy easy to predict things. And we try to. How we build our expertise. Yeah. Or thing that's, it's easy for us to audit.
Yeah. Right. Because a lot of the models that we see right now are relatively black box. Like you think you know what's your inputs and you think you know what's, you get there, but then the model version will change or anything else and all of a sudden be getting a difference. Yeah. Okay. So internally creating not only the processes where we can capture and audit.
The inputs and outputs in the process in the system. But actually learn how to operate with potentially smarter co pilot. Yes. That still require a lot of the alignment. Yeah. So, we actually build workforce that know how to operate in this newer
hybrid world. So as you were talking about that, it made me think of IBM Watson and healthcare.
And like, what if they hadn't started with trying to cure cancer? Like, what if they'd started with something different? True. Like, maybe the outcome would have been a different result as well. And I think we can learn from that. So, I think that there's likely, there's probably no regrets, low risk opportunities in this space.
And not go right to your point, brain surgery or some really advanced clinical care need that we can operate in. Have you found any of those at UC Davis Health? Maybe in non clinical workflows, legal department, you said co pilot. Some of these augmentation tools that are embedded in tools we already own.
Thank you. So, a
lot of the information that's being collected, presented to humans who currently manually have to review things, right? So, your audits, your discoveries, your additional information retrieval, summarization, potential, are the great use cases that we're looking at, right? Your knowledge management, some of the Q& A systems, that's currently, this knowledge lies somewhere on your SharePoints.
Uncompletely lost. Except maybe your admins, no one knows. People who wrote it, people who audited it, right? Possibly the only one who knows the content. Your regular users cannot get it. Unlocking that, very often, is low risk, and that's what we see from OpenAI, Enterprise, DevNet, right? The creation of those targeted bots is now being simplified to the point where you have to start dropping
You still will require a lot of alignment, right? And that's where... The workforce knowledge that is not panacea is important because the answer is likely correct. It's not guaranteed to be correct. Yeah. And it's a very different way to interact with data and the knowledge. But then incrementally, you start building on this information.
So, can I start getting my preliminary report the same way I'm getting in radiology? Right, for a lot of the things in the data. Yes, right, like, your radiology reports are not 100 percent accurate. Right? Sometime you may say preliminary reading was incorrect. Your goal is to be as precise but still you can operate from this.
I think a lot of the generative AI tools may be something similar, right? Into this radiology, early fast reports, right? Where you can get a glimpse of information. You're trying to look through your uh, stored in endless Excel reports, right? That's kind of the story. Clothing and asking some of the questions where precision it may be not as critical as long as you're directionally correct You actually start
getting some of the value.
📍 We want to thank you for a wonderful year. As you know, we have celebrated our five year anniversary at This Week Health, and we are going to enter our sixth year of doing this. And we set out a goal to raise 50, 000 for childhood cancer this year, and you did not disappoint. We have raised close to 60, 000 this year for childhood cancer, and we really appreciate you.
We appreciate you. The community coming together. And we hope to do more of this next year. We hope that you'll join us. 📍 📍 📍 📍
Yeah, I think the point you're making is so important And I worry sometimes that people think well that's so pedestrian like what you're describing That's so kind of you know table stakes kind of an approach But we need to get the reps in because we don't fully know how to use this yet And so I think that is the that's the training ground to get the reps in to get good at this and And then some of the more advanced, more exciting clinical applications will be better prepared for them.
Instead of starting right there. So I would
argue, please, right, that, this Watson movement harmed more than helped. Okay. Why? Because everyone was excited, right? Yes. It's almost a decade ago. Yeah. Right? And then big failures, spectacular failures, and out of a sudden, people are afraid. Yes. People are afraid to talk about it.
People are afraid to dream. People are afraid to. Yeah. And I'm gonna talk about all of it, right? That's what's slow development, right? What's important is the velocity. Right? And you are not getting from 0 to 100 in milliseconds, right? We're still talking about the seconds. You need to properly, kind of get into the right velocity.
And a lot of that depends on the mechanism that you have to build, right? Our innovation engine in artificial intelligence still has missing parts. Right? We're still talking about regularity. Just had executive Yes. Right. Is not, have
you had read the whole thing? Have you concerned Okay, very good. I'm still
reading it, but but it's required, right?
Yeah. And we will have to build a lot of those component while also building. Yeah. So I would say it's not pedestrian, right? Yes. You create the proper way to go fast, but learning the techniques because you don't wanna stumble and fall.
Yeah. That's a great analogy. Okay, we're at Chime. What's a problem that you've brought to.
Chime in your mind that you're trying to look for nuggets or what nuggets have you found one day in that are gonna help you back at uc, Davis Health.
So arguably the chime just started , right? So I'm still looking for a lot of the wisdom. We only had an informatics form so far. But I think the way we think about the clinical products, right, and the way we bring the bedside experience back and the way the organization start to talk about how we can not.
Automate everything, but reasonably start automate things, right, to actually protect our workforms from unnecessary. Questioning is like, yes we can generate a lot of notes, right, to satisfy requirements for billing. Should we? Right? If the information is not easy retrievable based documentation, do you even need to?
Right? Because people are talking about, we can now draft denial error clusters, right, in response to that. Guess what? Insurance can do the same things, right? And it's become net zero, right, of increased computational costs, where we're just competing on a natural language correspondence, right?
That's not what we're striving for, right? Like, we're striving
for clarity. And a lot of the
time when insurance rightfully asks for information, there are gaps. Easy. Is language the best medium to do this? Because Generative AI is not only language model, right? It can do code, it can do interfaces, it can do multi models, right?
So maybe instead of overly focusing on generating yet more bloated nodes with Generative AI, we need to actually use the potential of Generative AI to create better, faster interfaces, now even more interoperable, even more connected. So we actually have less... We have less snail mail. We have less faxes, right?
Fax is still everywhere. It's now digital fax. Does it make it less fax? No. It's still a fax. Right? So a lot of this, in my mind, needs to be questioned. Because now we do have a technology that is, generation defining. A lot of things in 5 10 years will be different. And like with many other things, I think we're overestimating where we're going to be in 5 years.
Very underestimating where
they're gonna be.
Okay. Alright. You heard it here. Community question, everything. That's what I took away from that. So Johanni, thank you for your time and have a great
conference. Thank you, Rita. All right.
Another great interview. I want to thank everybody who spent time with us at the conference. I love hearing from people on the front lines. It is phenomenal that you shared your wisdom and experience with the community and we greatly appreciate it. We also want to thank our channel sponsors who are investing in our mission to develop the next generation of health leaders.
They are CDW, Rubrik, Sectra, and Trellix. Thanks for listening. That's all for now.