Loading Results
We have updated our Online Services Terms of Use and Privacy Policy. See our Cookies Notice for information concerning our use of cookies and similar technologies. By using this website or clicking “I ACCEPT”, you consent to our Online Services Terms of Use.
Research & Education Institute
Science@UH Podcast

The Integrated Medical Environment: Using AI to Translate Big Data to Precision Medicine


Click the play button to listen now:

Subscribe: Apple Podcasts | Google Podcasts | Amazon Music | Spotify | iHeart Radio


Dan Simon, MD: Hello everyone. My name is Dr. Daniel Simon, and I'm your host of the Science at UH podcast sponsored by the University Hospital's Research and Education Institute. This podcast features University Hospital's cutting edge research and innovations. Thank you for listening to another episode.

Today, I am happy to be joined by two guests, Dr. Michael DeGeorgia and Dr. Kenneth Loparo. Michael is the Director of Neurocritical Care Center and Director of the Center for Music and Medicine, UH Cleveland Medical Center and Chief Neurologic Services and Director of the Primary Stroke Center at UH Ahuja Medical Center. Ken is the Arthur L. Parker Professor of Engineering and Founder and Faculty Director of the Institute for Smart, Secure, and Connected Systems at Case Western Reserve University.

Together, Michael and Ken are building the Center for Connected Health Innovation through a $1.2 million grant from JobsOhio. Welcome, Michael and Ken.

Michael DeGeorgia, MD: Thanks for having us, Dan.

Ken Loparo, MD: Thank you, Dan

Dan Simon, MD: Michael, you are an internationally recognized expert in neurocritical care and cardiovascular disease. And Ken, your research interests focus on stability and control of non-linear and stochastic systems with applications to large scale electricity systems. How did the two of you begin to work together?

Michael DeGeorgia, MD: Well, I first met Ken shortly after journeying to University Hospital in 2007. So, I had been doing a research in the neuro ICU on multimodal monitoring using medical devices to measure things like cerebral blood flow and brain tissue oxygenation and I was frustrated that none of the clocks of these devices were synchronized, which made assessing correlations very difficult. So, in December of 2007, I set up a meeting with several case engineers and presented my dilemma, naively thinking that it would be a relatively simple fix. And it was Ken, who explained that the problem is actually much bigger than just limited interoperability of medical devices in that none of the monitors in the ICU, in fact, none of the data collected in the ICU, is integrated or synchronized or stored. And that is what set us down this pathway to try to solve this problem. Ken, do you remember that meeting?

Ken Loparo, MD: I remember that meeting and I remember the room was full.

Michael DeGeorgia, MD: Yeah.

Ken Loparo, MD: And, I also remember about a year and a half later, it was like three or four of us that all remained together and were charting the course... and my experience, of course, has been not only on the theoretical side of sort of systems control and dynamics, but a lot of industry applications and most of the problems that we deal with are highly complex, involve lots of non-linearities, noise and non-stationary data. And when Michael talked about what was happening in the ICU and what he wanted to have happen in the ICU, it was very reminiscent of many things that I had done in industry practice. And that when you look at automation, automation about 25 or 30 years ago is where sort of the connectivity and interoperability that Michael is talking about now, that's where they were. And so, I understood the path of what it would take to move from a proprietary infrastructure of devices that are fairly autonomous and not collaborating, to what it would mean to have a truly orchestrated set of devices within a data-intensive place like the ICU.

Dan Simon, MD: Well, that's really cool. And I would say, how great is it that you can walk out of the hospital a block away, knock on the door of BME and engineering, and have all these great people who can really help. And so, that is the beauty of our university circle community. So, the focus of your research is big data and precision medicine and critical care. Why focus on the ICU? Why is it so important?

Michael DeGeorgia, MD: Well, more than 5 million patients with life-threatening conditions are admitted to the intensive care unit each year in America. And it's a privilege to be able to work in this area and help so many people when they're most vulnerable, but as Ken pointed out, critical care involves highly complex decision-making with staggering amounts of data… beyond the capability of any person to absorb, integrate or act upon reliably. So, there's an urgent need to translate this raw data into actionable information at the bedside similar to the data-integrated glass cockpit that provides pilots with situational awareness in today's modern aircraft.

And in addition, critical care is also very expensive, right? It's at 13% of hospital costs. It's almost 1% of the gross domestic product in the United States. And so, this is really ground zero of data in hospitals. And so, that is the reason that we feel this is where we should start our work on.

Ken Loparo, MD: And I think too, Michael, as we've talked about, if you can solve these problems in the ICU, then it can move to every operational setting within the hospital.

Michael DeGeorgia, MD: Right. If you can make it there, you can make it anywhere.

Dan Simon, MD: Well, that's I think a really good point, but let's try to get our listeners onto the same page and just ask a really basic question. So, tell me, what do you mean by big data and precision medicine? Define those two things for us.

Michael DeGeorgia, MD: So, if there's any phrase that's captured the public's imagination, it's big data. The idea of harnessing large data sets using sophisticated processing tools to extract hidden information is very powerful… And in medicine, no area holds more promise for the use of big data than in critical care again, by nature, it's a data-intense environment. But despite the growth of critical care, as Ken mentioned, the basic approach to information management in the ICU has remained largely unchanged over the past 50 years. If you had walked into an ICU in the 1970s, it would look pretty much the same as it is today. And in terms of how we manage the data and how we collect the data and how physicians make decisions, it has not really changed nor evolved. And so, what we're developing is a new integrated architecture that will enable clinicians to provide precision medicine or the right treatment for the right patient at the right time. That's really what the goal is.

Dan Simon, MD: So, when you say, obviously, collecting and harnessing all this data…clearly one person there's no way could integrate all of this and see patterns. So, when do we start getting into the notion of what does it mean to say machine learning and AI? What's happening? Are you teaching a computer what to look for, patterns to look for of, let's say, who's going to get sick? Who's going to get better? Tell me a little bit more about how collecting this data is going to help a patient.

Michael DeGeorgia, MD: I will let Ken take this. But let me just say that, yes, the future is artificial intelligence, machine learning, deep learning, big data. The problem really is, the sort of upfront problem, is getting the data into a surgical database. And so, once you get the data into a database, yes, there's a whole world of mathematical tools, including AI and machine learning that can translate this data. But right now, we can't even get any of the data. Being physicians, you can't even get an average in the ICU…right? If a patient's tachycardic and I say, "Geez, what's the average heart rate in the last four hours? Is that statistically faster or slower than it was four hours before? But what's the P value?" There's no way you can do that. That's eighth grade math, right? You can't do that. That's eighth grade math on one parameter. And so, we need to get the data into a data set that's acquired synchronized, time synchronized, into the same data format to be able to start to run these algorithms.

And then Ken, you can take it from there in terms of the AI and machine learning.

Ken Loparo, MD: Sure… Well, I think that the operating words these days are all about big data and the kind of data that's necessary to empower artificial intelligence or machine learning algorithms. But we need to remember that what's really, really important is the context needs to guide how you extract features and pull out relevant information from the data. And very often, there are data sets that are out there that include numeric data, that would be sort of the information maybe you would move up from an ICU to the EMR, and also, time series data, the waveform data that Michael can look at on a monitor. But then, they trickle off the screen and they're kind of gone because nobody's capturing them. And that data is what is necessary to begin the process of what we call feature extraction. That is looking for meaningful patterns and relationships in the data that have some correlation, but maybe more important, have some causal mechanistic relationship to what's happening clinically. Without the clinical data in the insights of clinicians and clinical personnel on what's actually happening with not just data that is coming off of a particular monitor, but off of all of the monitors, lab tests, any kind of imaging analysis, CAT scans, MRIs, also, what's being infused into the patient, both automatically and physically by personnel in the ICU. All of these affect the complex physiological dynamics of the individual and without understanding that collectively and completely, AI and ML and any other algorithms that we handcraft and develop are simply irrelevant. So, that's why this can't be done by simply individuals pulling data off of the internet and trying to do something. It has to be a collaborative, integrated operation involving engineers and scientists and clinicians. All working together with the same mindset, the same comprehensive data sets, and developing a language that we can speak to each other so that when we talk, we can not only listen to what they're saying, but we can comprehend.

Dan Simon, MD: You know, it strikes me that the obstacles are really significant. I mean, I remember when I was a resident, we had these massive paper spreadsheets on these big boards, and everybody was writing and filling in the boxes. And actually, although it looks fancier now because it's done on a spreadsheet on the computer, there's still someone entering the data. And when you walk into an ICU bed, you know, Mike and I will tell you, you got eight IV pumps going with different medications and fluids, you've got tube feeds going in, you could have a balloon pump, you got a ventilator. I mean, you got all this stuff going, everything's got electronics and people are still entering it by hand. I mean, it's unbelievable. And there doesn't seem to be smart Bluetooth technology that's dumping this information to allow doctors and nurses to actually take care of patients as opposed to recording. Is this some of this problem that we have, which is that the technology's out there, but for some reason it hasn't been applied in this medical ICU or other medical settings?

Michael DeGeorgia, MD: I mean, part of the challenge is-- I'll let Ken talk about it-- but part of the challenge is that all of these devices have proprietary limited data formats because they all sort of grew up organically and they were never designed to be integrated in a cohesive, coherent way. And so, as you say, there's a balloon pump here in the Neuro ICU, we've got continuous EEG and blood flow and percent of oxygenation, ICP, and these are all separate boxes essentially that were never really designed to be integrated and processed together. And then, coming back to where I started, that none of these are time synchronized. So, these infusion pumps, when you open them out of the box, you set the clock like your VCR, right? You just look at your watch and you set the clock. So, all of these clocks are not synced to another. So, if you want to sort out whether the medication that infused into the patient's arm came before or after the cardiac arrest, which is kind of an important issue, you can't tell, because there's just no way of networking them.

Ken Loparo, MD: Right. And so, that I think is the challenge, and that's what Michael and I over the years have been looking at building and demonstrating in his ICU. That is the ability to not only interrogate a single device, but to interrogate multiple devices across the ICU field, and then bring those together at a central point of connectivity where the clocks then are automatically in sync, and now the data can be correlated, but also can now be analyzed for trying to uncover causal mechanisms to really understand the interrelationships of all of these various organ systems that are responding both autonomously and in some sort of controlled manner and how they're responding to external influences that are occurring in the ICU.

Dan Simon, MD: So, what do we need to get to your vision of the ICU of the future? How far are you in developing this technology, and what will this data ultimately allow our caregivers to do?

Michael DeGeorgia, MD: Well, yeah… So, we need first an integrated informatics architecture that facilitates data acquisition, consumerization integration, storage of all relevant patient data from all the devices, from the EMR, the phenotypic and physiologic data into a searchable database that stores the context of when the data was collected. And then, we need the high-level data processing… algorithms to extract the relevant features and that's what translates the raw data into something that is actionable for the clinician. And then in the end is really the cool part. You need this intuitive data visualization to present it all back to the clinician in some sort of user-friendly, intuitive way. Right now, there's really no data visualization, really. But you need some way of presenting this back to the clinician and some type of clinical decision support that is very patient-specific. And that then changes the way physicians and providers can process all this complex data and then act upon it. And you really need those three things and there's no shortcut. And so, that's what we've been working on. And we've developed a system here called TIME, which is The Integrated Medical Environment…that does, in the large part, the first two parts of that. We focus mainly on the data acquisition and the processing, less so yet on the data visualization, but that's kind of where we are. Ken, you want to take it from there?

Ken Loparo, MD: Yeah, and as Michael pointed out, once you have the data and you have it in a format that makes it accessible and can be linked with all the relevant clinical data so that this is now a continuum from raw data all the way through clinical interpretations and understanding, then I think it's a matter of not making decisions, providing decision support in a way that is interactive at the bedside. This means that when a respiratory therapist comes to the bedside, they have a visualization that's intuitive to what they need to do and what they need to focus on. When Michael comes to the bedside, he has the view and the ability to interact with those tools based on what he needs and the information that he needs in order to support the decisions that he needs to make.

And so, this flexible user interface that now not only facilitates an individual to work within their domain or sphere of influence for the patient, but I think also works logistically across all of the workflows of all of the clinical personnel that are treating that patient in the ICU. So, one of the things about this ICU or the future laboratory that Michael and I are trying to put together, is we want engineers, nurses, physicians, all coexist and collaborate within the same physical environment, so that technology is developed, not based on me pushing something out of my lab, which shouldn't be a technology push, it should be a clinical pull. It has to be the rationale and the needs have to come from the clinical side and the technology has to meet those needs. And so it's innovation, not invention.

Dan Simon, MD: Wow! I mean, that's just really cool. And, I have to say, I would really look forward to that because right now we just have a fancy spreadsheet on a computer screen instead of on a roll-able table. So, congratulations on your grant to establish the Center for Connected Health Innovation. Can you tell us a little bit about what its purpose, and how that JobsOhio Grant is going to help?

Michael DeGeorgia, MD: Right. So, the Center for Connected Health Innovation was launched this summer. It's a multidisciplinary center, designed really to catalyze this digital health research, promote interdisciplinary education and really start to harness this technology to improve patient outcomes, not just in the ICU, but across the whole continuum.

Clearly, what we've been talking about is the future of medicine there's no question this is the future of medicine, it's just to get there is hard, which is why it hasn't really come to fruition yet. But this is clearly the future of medicine. So, we applied for and received this 1.2 million dollar grant to establish this center to work on not only the ICU, but even beyond the ICU. And the Institute for Smart, Secure and Connected Systems, ISSACS, which Ken was the faculty director, is really the administrative home of the center. And one of the goals… was to establish what Ken mentioned, essentially the ICU of the future, but a digital health test bed on campus, which would be a center for research and development of testing all of these digital hardware and devices and ICU data acquisition software and equipment and emerging technologies in some type of space that would be available to multiple disciplines, nurses, physicians, respiratory therapists, to get their input into how best to utilize this data.

Ken Loparo, MD: And I think education is also important part of this picture in a sense that, now, when we think about educating the engineers of the future, the nurses of the future, the physicians of the future, this kind of technology is going to be part of what is going to be routine for them, we're hoping it'll be routine for them. And therefore, it'd be good to have them together at the ground floor as they're working forward in their own educational curriculum and professional development.

And we feel that the environment of such a digital health test bed, could be something that motivates engineers and nurses and physicians of the future to begin to think about how they collaborate across these boundaries, so that they're not isolated domains, but they're a continuum of education and research and technology development that meets the future healthcare needs of this country and of the world.

Dan Simon, MD: Wow, that's incredibly inspirational to hear and as a practicing cardiologist, we face these difficulties every single day. So, you're going after something big and that's really, really exciting. I want to congratulate both of you on your dedication and persistence because you've been at this for a while, but it sounds like you're making great progress. I want to thank both of you for taking time to speak with us today, Dr. DeGeorgia and Dr. Loparo.

For our listeners interested in learning more about research at University Hospitals, please visit UHhospitals.org. Thank you so much for joining us today.

Ken Loparo, MD: Thank you, Dan. My pleasure.

Michael DeGeorgia, MD: Thank you.