Azra Bihorac, MD, professor of medicine, surgery, anesthesiology, and practicing intensivist and researcher in surgical ICU at University of Florida, talks about innovative applications of artificial intelligence for critically ill patients.
Interview conducted by Ivanhoe Broadcast News in November 2021.
What is an intelligent ICU?
PROF BIHORAC: Intelligent ICU is a concept we proposed as a paradigm of how intensive care unit should look in the future. It relies on the use of advanced technology to offset the high human burden for taking care of the critically ill patients. So, if you have ever been in intensive care unit, you would know that usually one nurse take care of one patient. Most of our patients are critically ill; they require a lot of life-sustaining therapies in the form of the machines of different medications, and for that, the intensity of care is really high because of that, taking care of the patients requires a lot of observations, a lot of times human time that at this point we cannot provide both because it’s impractical, and the second we are in a constant state of the workforce shortage, both for nurses and physicians. So, we propose that artificial intelligence and other technologies can be used to our advantage to help us both monitoring patients and, also, intelligently taking care of them. So, what does that mean? So, in our research, we utilize pervasive sensing. Sensors that can be placed on patients in their environment and continuously monitor whatever they are doing. Not just their external movements, but also environmental data, such air quality, light, noise. At the same time, we can also collect physiologic data, so such as heart rate, breathing, oxygen levels, twitching of your muscles and so on. So, with that concept, we have designed a system that can be deployed in intensive care units room, collect this data continuously and then use deep learning and advanced artificial intelligence methodology to process that data in intelligent way. So that can be used for the physician’s advantage. So, for example, what we do is we use physiologic data and environmental data, as well as the sensor data from the patient to develop acuity algorithms. Those are the algorithms that help you predict who is going to get sicker in the next three, four, five hours, right? So, they are basically almost like acute on go real time sensors of the patient’s sickness level.
How does that work? You know, because like, how can you tell someone’s going to get sicker?
PROF BIHORAC: So, what we do, we use our previous data with deidentified information about vital signs and laboratory results to train and develop algorithms. We train AI models to understand which features collected from the patient’s medical records or sensor data are predictable or what is going to happen in the future. So, we are forecasting bad events, if that makes sense to you. It’s very much like a stock market, right? So, we have this score that looks like a line, and it goes from green, that means you are doing great, to yellow, something is not well, to red, meaning you got to take care of these patient. It’s really helping doctor triage their attention to patients who really needs that. And now think about it. Not only we are telling doctors, alarming them which patients needs attention, but also our system and our models allow you to drill into the prediction. You can go inside the prediction and look at the events or data that really causes algorithm thinking that you are in danger. Does it make sense? So, we highlight features that are importance. So, we are bringing what we call interpretability component to the models. For a long time, the problem with the deep learning models was they were considered black box. You would know that something’s going to happen, but you didn’t really understand why. So, our research in the next five years is really launching towards explainability of the models. Understanding not only who is going to get sick, who is at risk for disease, but why and explaining that to physicians and patients, also.
So, you do this all with sensors?
PROF BIHORAC: Sensors and clinical data collected in real time. Everything that is collected by the doctors, by the health care system and is being used for the taking care of you is also utilized by the models in very agnostic fashion. There is no information that are identifiable. It’s all secure in the same security level as the hospital operations. So, this runs together with the operations of the clinical operations
Now, just to talk about the sensors a little bit, how many sensors would be on a patient and what exactly can they tell from these sensors?
PROF BIHORAC: So, the sensors really collect various form of the physiologic data. Some are very small sensors. They collect your continuous EKG, right? Your heartbeats, beat to beat, tracing of your heart. They collect oxygen levels continuously. They collect respiration rate; your heart rate beats. They also can collect twitching of your muscles. We also utilize cameras that using entire body images we can track your movement and activity level in real time. So, we can say when patient goes out of the bed, is he at risk for the fall? We can also measure how active you are. We want you mobilized because that improve your outcomes. So, this system will allow us to independently of any human, observe patients in real time, measure their activity and mobility levels, and then use those visual cues also to predict who is maybe getting better and who is not because it makes sense that patients who are more mobile, who look more agile will do better. We use that in every day in our clinical observation, those visual cues about human behavior. We have not been utilizing them for the prediction of the risk for disease to date, widely. And this is probably the first applied concept that combine all of this type of what we call multimodal data in real time to use that for benefit of physicians and nurses and caregivers.
I mean, before this, luckily, because of the work shortage, luckily, if a nurse can get in every hour to take these measurements that would be a good thing?
PROF BIHORAC: That’s exactly right, because the nurses, they are overwhelmed with more and more patients in ICU. Even in the regular time when we are not overwhelmed with the ongoing pandemic, nurse could attend perhaps every hour to the patients. With this system, we have second-to-second observation, almost continuous. Those observations are not only made, but they are also brought in coherent form so the nurse can process them in intelligent way and get information that is ready for them to use.
I have to think it’s really good for the doctor, as well, because they get a more constant read of what’s happening and kind of, not more reliable, but almost more reliable, in a way, because it’s real time all the time.
PROF BIHORAC: It is, and it is more comprehensive than what human can observe. It’s not like multiple humans observing the patients at the same time and then bringing that information in summarized form to the doctors. This is not only for health care workers. This is also for the patients and their caregivers. Think about that when you are in ICU alone, you have all of these machines without any information being translated to you as a patient, how alienated that environment is. So, we are developing an interface that can show to the patients and their caregivers what’s going on. Right? Hello, everything looks good. You have maybe exercised this much yesterday. You eat this much. You are doing better and so on. We want really to make this as more friendly environment for everybody in the ICU. We called it human centered intensive care unit. That humans are patients and caregivers and doctors and nurses.
How many studies on this have you done so far?
PROF BIHORAC: Well, we have been working on this work since 2010. Our initial study that was funded by NIH was about surgical risk assessment, using this type of data for predicting who is going to have complications after surgery. That study was funded now in the fifth year. We have validated algorithms that are being used by the surgeons in real time here to assess this surgical risk. This study that extended to this one for intelligent ICU from surgical environment to ICU environment and then we have the broadening this to apply now for what we call delirium, for agitation and brain dysfunction in intensive care unit that happens to a lot of our patients, and you heard about it, COVID brain, right? So, this system now not only collect information about patients, but also collect information about environment. Often patients are exposed to not usual amount of noise in ICU. They also are not exposed to enough light during the daytime and too much light in the nighttime. So, in our next study call Adapt, we use information not only to inform the doctors, but also to adjust environmental level of noise and light so it actually helps us treat delirium. So, we are moving from observing, predicting to intervention using A.I. in real time.
So, I can understand the oxygen levels, the heartbeat, the blood, everything like that, but getting into delirium, it seems like that’s a whole new ballgame because it’s not something that can be like you can check your pulse.
PROF BIHORAC: No, you can’t. It is really interesting work, and it’s intriguing because for the first time, we are using environmental cues to understand what’s going on with the patients. We know from the data that simple interventions like making sure that patient sleeps, make sure that the daytime is bright, and nighttime is dark, are very important for what we call circadian rhythm. Our body operates in the 12-hour rhythm in the day and night shifts that are completely lost in intensive care unit because of the nature of the care we provide. So, I think this is a really good example how a technology can be used to our advantage, both by collecting data in a fashion that no human can do, and the second, implementing that in a fashion that is systematic and automated. Again, in a way that we cannot do very well as only by ourselves.
Well, I think any person that’s ever-spent time in the hospital, I think they come out more tired than what they went in because every hour a nurse is coming in, waking them up, checking their pulse. So, this can change the whole dynamic of it.
PROF BIHORAC: Absolutely. Exactly. If you think about how actually processes of care right? So, that’s why what we call this intelligent care. Processes of care are not very much standardized and automated. Right? You know, today one nurse made that do this this way. Next day, it’s going to be done this way. Often, it’s hard to apply or be compliant with these protocols. That’s a big challenge in health care. This system allows us to precisely in a standardized way, always apply same criteria and treat patients in a similar way. That is very important for certain conditions like delirium.
So, do you think this is the future, not only a surgery, of ICU, but of just hospitals in general?
PROF BIHORAC: Absolutely. I think this is the future of health care. If you think that in that ICU environment is not observing patients enough, just think about floors where one nurse takes care of eight, 10 patients sometimes. So how much time they can really spend in each patient’s room. Think about patients in nursing homes or even patients at home or you are being at home. The concept of the health care coming to the patient, like human centered health care, goes to you. You don’t come to them. So that is the future. I think technology is going to play a very big role in this. I think we are very privileged at University of Florida that we have access to this data in a secure fashion that we are able to actually implement all of this in real time setting. Is a huge advantage we have here.
Are you using it for all your ICU patients right now?
PROF BIHORAC: As of now, this is the research study, so we are not using as a part of routine care, but we are using as a part of three NIH-funded large studies ongoing at the University of Florida.
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
If you would like more information, please contact:
Doug Bennett
352-265-9400
Sign up for a free weekly e-mail on Medical Breakthroughs called First to Know by clicking here
