Oct 24, 2014

Announcer: These are the conversations happening inside health care that are going to transform health care. The Health Care Insider is on the scope.

Wendy: My name is Wendy Chapman and I'm the chair of Biomedical Informatics at the University of Utah. We have with us today Dr. Randall Wetzel. Dr. Wetzel is the chair of anesthesiology, critical care medicine at Children's Hospital Los Angeles. He's also Professor of Anesthesiology and Pediatrics at the University of Southern California. He directs the Virtual Pediatric Intensive Care Unit and he's the CEO of a company called VPS LLC which is an international data analytics company focused on improving the care of critically ill children. Dr. Wetzel joined us as part of our Learning Health Systems seminar series and talked about big data, "What's the Big Deal About Big Data: Adventures in the Intensive Care Unit."

Dr. Wetzel: Well, thank you very much, Wendy, for inviting me. What's the big deal about big data in health care? Almost everyone has heard about big data and it gets talked about an awful lot. Big data is an amount of data, not necessarily information or knowledge, that comes at us in large volumes in an amazing baffling variety and at a very high speed. It has a value and the value I'm interested in is the value of improving the care for critically ill children. It is a truth about what's actually happening to patients. All of that digital data you see being collected around patients is increasingly captured, although I'm still afraid most of it is thrown away and never subsequently analyzed, but the fact that it's now digitally mobile data that we can start collecting, storing, analyzing, and using what actually happens to children to teach us about how best to care for them.

Wendy: With the advent of all the data that we have access to now, how is it that changing the way that we do science?

Dr. Wetzel: It is a radical change, randomized blinded controlled clinical trials. Have an idea, propose a hypothesis, design an experiment, collect the data, analyze the data, get an answer to your first question you hope. That's the old paradigm. The shift is now, with our ability to now capture increasingly large amounts of data, is to now identify data sources, look at the data, let the data teach you about patients, let the data almost pose questions about itself, and then look at the data, mine the data, analyze the data to answer questions that are developed.
The difference between those two is one can take many years and cost billions of dollars. The other one can take a few days and you already have the data, it's already collected. The other difference is in the first example you might 20, 40, 60, 100 patients. I'm interested in gathering information about every heart rate, every blood pressure, every breath, every lab value on every patient that requires pediatric critical care, and be able to analyze what happens to hundreds of thousands of critically ill children. It's an exciting opportunity. I'm not sure if we're ready to bring it to the bedside this week. It's hard to imagine that just 15 years ago hospitals weren't connected to the internet, there weren't electronic health records. Data, although captured and written down on pieces of paper, wasn't really stored or ever analyzed.
So VPS was started to get people used to sending data about patients to a central data repository practice. We now collect data on about 200,000 patients a year.

Wendy: Tell us about some of the things that you've learned from these studies.

Dr. Wetzel: Some simple things. Actually a surprise to me when we first started seeing the data is that 65% of the patients admitted to pediatric ICUs are males and only 45% are girls. Yet girls have a higher mortality rate in the ICU. There was no way to guess that. You expect boys to fall out of buildings, run in front of traffic, be more rambunctious than girls, but it's not a difference in trauma. It's a pretty astounding and a very consistent difference that has sparked whole other areas of research.

Another example is that there's something called septic shock, a severe bloodstream infection. If children get this before puberty, the mortality is the same in boys and girls. If after puberty, then girls survive better than boys. They have a higher mortality from sepsis. Well, that's not just interesting. Obviously there are differences we know about that are gender-based and maybe clues to how we can treat that disease in everyone.

We do a lot of therapies in the ICUs. There's one called high-frequency ventilation. Based on the evidence of maybe 70 patients, none of them studied after 1994, so data that's 20 years old in a handful of patients from three or four different centers, we use high frequency ventilation thinking it's the right thing to do. Nobody has really done a large prospective randomized well-controlled trial because it would cost a fortune. And there's something called clinical equipoise. We all believe that that's the right thing to do. On this database where we've got a million kids, we're able to screen all those kids, find 900 children that had high-frequency oscillatory ventilation, use advanced statistical techniques to match them with patients who only received conventional ventilation, all for the severity of illness, they had the same amount of sickness. We could match them in over 50 different parameters which is usually much better patient matching than you can get in a clinical trial.

Lo and behold, patients who received high-frequency ventilation had twice the mortality of the conventional ventilation, stayed in the ICUs longer, stayed on the ventilator longer, and had a longer hospital stay. If you looked at the data of a very commonly used and quite expensive therapy, you wouldn't use it anymore. It's the sort of thing we can start doing. Nobody would have ever studied this. This data just happened to be available in this large data repository. I would imagine that will save, over the course of a few years, hopefully some lives, but also millions of dollars in therapeutic costs and also spur us on to thinking about improving other forms of ventilation to do a better job.

Wendy: Yeah, so it's clear how the large data sets really give us access to asking new types of scientific questions. What about the thought of bringing that data to the bedside? How do we better support clinicians at the point of care make the right decisions? How do we present all that data to them in a way that can support their decision making?

Dr. Wetzel: There certainly is a lot of data in the ICU. Any day in the ICU I may be responsible for caring for 20 children. Streaming from their bedside are over a 100 data streams per child. And it's my responsibility to look after all those kids and be aware of what's happening to all those kids. Well, anybody knows you can't look after 2000 data streams as a human being. Forget it. Impossible. So I edit the data. I look for patterns in the data. I rely on intuition and I occasionally actually rely on knowledge. Managing that data is a communications challenge, it's a learning challenge, it's a knowledge discovery. That data is just data.
As a clinician, I have to find patterns in that data that tells me about the physiology that's happening to the child. What I would like to do is be able to look at a patient, have the data from that patient compared to the data on a million critically-ill children for whom the outcome is known. I know what course they took, I know what treatments they received, I know what their outcomes were, I knew what their prognosis, and, by using advanced machine learning techniques, find a population of kids just like the one in front of me now and have suggestions, advice that are accurate, reliable, what actually happened guide the therapy of the next child.

Wendy: Now that we have electronic health records we have access to all this data, whereas 20 years ago we didn't really have that data in digital form. So we've overcome that barrier. But it's been 20 years. Why are we not taking advantage of this data right now at the patient's bedside? What are the barriers that are keeping us from learning from this data?

Dr. Wetzel: Well, progress often occurs slowly and radical change doesn't occur very fast. When the new electronic health records came in, they were not designed to do what we're talking about. They were designed to take the old paper record and put it on a screen. That was it. It was for data presentation about a single patient in detail near real-time at the bedside. The design of the databases behind those was totally incompatible with the way you would design a database to do this kind of research and provide this kind of information.
I can find detailed data about Jimmy in an electronic health record, but I can't find easily data about all the people who had the same temperature and blood pressure as Jimmy. That has required years of learning how to extract the data, transform the data, and put it in a fashion that can be analyzed. Those barriers are still there and they are time and research. The barriers are less and less because the quality of the technology has improved.

The other biggest barrier is to realize the value of doing this, although every physician knows that they are to learn from their patients and we practice medicine, and in fact every interaction between a patient and health care is in some aspects an experiment. The outcomes are never 100% certain, odd things happen, people respond differently, and if you miss those you may not learn something important. And if we are interacting with patients in an experimental fashion, I know that sounds a little scary, but that's the reality, we are compelled to observe closely and to learn from it, and to build learning health systems that many of you have heard about recently you've got to have the data to learn from. And that means every one of those transactions should be captured, stored, studied, and there are ways of doing that.

So I guess I'd say that in 20 years we've moved from paper records and static data that's un-analyzable to beginning to free the data up from paper, make it digital, begin to move it, and make it comparable, and learn new databasing and new analytic techniques, and obviously vastly increased computational power and storage capacity, it really makes this no longer a technological challenge. It's a political challenge. It's a matter of not spending much money actually to make this happen. The technology is there. The cutting edge stuff is on how to analyze large amounts of data, but we've been making pretty pleasing progress on that. It's now possible for our computer-driven machine learning algorithms to perform better than groups of clinicians.
I'm not worried about or thinking about replacing the doctor at the bedside. I'm working at taking the doctor with computational science and improving the care the doctor can deliver at the bedside and the time that doctor can actually spend interacting with the parents and the child, also knowing that what I'm proposing or suggesting for care is based on hard evidence and comparison to lots and lots of previous experience.

Wendy: The data that we collect at the hospital and when we go to the doctor is largely recorded by the clinician. Could we take better advantage of the patients and the patient's family members to able to capture broader and deeper data about our health?

Dr. Wetzel: As everybody knows, who has a Fitbit have increasing ability to capture physiologic data about themselves all the time. There's a burgeoning great potential of being able to monitor patients outside of the hospital and it's exactly the same problem. I like the ICU because things happen in a short period of time, so we can see if we're doing a good job at where we're going. It's measured in days rather than most people's illness which is measured in lifetime.
As people become more aware of wearable technologies for monitoring, there's a great opportunity for us to have a lot more of that information constantly analyzed. It's the same mathematics and informatics problem. Then to turn that into an alert that says when they pick up their telephone says, "Time to see the doctor," and, "Would you like the doctor to have your data? Push button." Obviously that will lead to much better communication between the patients and the health care providers. I can't imagine it can do anything other than improve the process.

Announcer: TheScopeRadio.com is University of Utah Health Science's radio. If you like what you heard be sure to get our latest content by following us on Facebook. Just click on the Facebook icon at TheScopeRadio.com.

For Patients