UChicago Medicine Undertaking ‘Data Democratization’ Effort
UChicago Medicine is working to overcome common data-access problems in an effort to achieve “data democratization.” The goal is to allow clinicians, researchers, staff members, and administrators to engage in a dialog with their own data to ask and answer their own questions with the goal of improving care quality. In an interview, Tom Spiegel, M.D., vice president and chief quality officer for UChicago Medicine, explained the impact of the shift to this self-service approach.
Healthcare Innovation: What does the term data democratization mean in the context of hospital research and quality improvement efforts? Could you talk a little bit about that, and maybe some of the challenges that exist in academic medical centers that you have to overcome to get to that point?
Spiegel: Sure. I jokingly wanted to say, “what challenges?” In all seriousness, our approach to data democratization really stemmed from a couple of different areas. One is research. We have three executive sponsors on the team. One is our dean of research. I'm representing the hospital and the quality improvement entities. Our third partner is IT. We said, let’s open up the doors to data. As far as the challenges that we were trying to address, the biggest one is fairly obvious: data accessibility. There are a lot of time delays from data request to fulfillment, and then it's never a one and done. It is very rare that somebody walks in with the perfect request initially. It's an iterative process. After a long delay, you get your data, and then you realize you need six more elements, or the initial data generated new questions and we need to refine and request more data. So it's just delay after delay.
HCI: And that always involved asking someone in IT to deliver the data to you?
Spiegel: Yes. For the most part, if you were going to publish on any of the data, it had to come from the Center for Research informatics.
HCI: So does changing that to self-service involve policy and technology issues?
Spiegel: Absolutely policy and tech. The tech is probably where we spend a lot of our time. From an administrative standpoint, privacy is the big lion's share of what we've had to come to terms with. Policies were already in place that had more to do with ensuring that we're following the policies and not only the letter of the policy, but the spirit of the policy.
We are working with MDClone and it offers a very unique opportunity to create synthetic data sets, which is really a whole new realm in the privacy sector, because it's not real patients. You can't trace any of this back to real patients. They don't exist. The system changes every element just a bit so that you can't trace it back, so at that point, there is no patient to identify. And that data is far more safe and far more usable, and it's still of tremendous value institutionally, because it is based on our patients.
HCI: I read that physicians in your emergency department are starting to use this resource.
Speigel: I’m an ER doc, and I am a big proponent of this entire approach. I’m also a big proponent of opening up our data stores. A lot of the folks in the ED, we talk during shifts and generate a lot of ideas, and there's a lot of enthusiasm around this. So I think they were some of the early adopters of this.
HCI: Does this allow for them to think in terms of a continuous quality improvement approach, by requesting the data and getting it back rapidly, and then making some kind of iterative change in their processes and tracking the results?
Spiegel: We're starting to see that happen. And that's exactly the intent. I’ll give you an example of of one of the early projects, probably the most complete project that we had, that I had nothing to do with, which is exactly the goal. One of our quality chiefs of our Department of Medicine retired last year, and he said, I want to learn MDClone. Basically, he's spending his retirement time in MDClone, and he said he's been able to answer questions that he always wanted to ask.
He's a primary care doc, and he said, How are we doing in controlling patients blood pressure, and could we be doing more? So he looked across all of our thousands of primary care patients and said: who over the last three years had systolic blood pressures over 160? We could be doing a better job of managing long-term blood pressure control. Then he said, based on that population, what medications are they on? And can we create algorithms and have a protocol for our pharmacy team to work with us to get them on the right medication?
There were 586 patients he identified to say that we could be doing a better job, and we should be taking an active role in that. Now he's got teams of folks reaching out to these patients and modifying their blood pressure medications over time, and hopefully improve their health outcomes. So that was something that he just did on his own, because he he wanted to do it.
We have another group in GI who noticed a lot of our patients are on medications that they might not need. There are a lot of acid-reducing medications. There are a couple different varieties, but patients stay on these things for years, thinking they don't want to get heartburn again, so they just keep taking that medicine. Well, the evidence shows many of those patients don't need it. And if we do an upper GI scope and take a look to see what's going on in the stomach, it may be that they don't need those. So that's another project that he's involved in. He and I were just talking about this the other night to say: can we through better screening eliminate some of these medications that patients don't need? I think these are just the types of questions that, as a primary care doc, he's wondered about, and now he could sit down at his computer and get the data to take an action.
HCI: Can you think of an example from the ED?
Spiegel: Just a few weeks ago I was talking to one of the other docs and saying, hey, we we use this medication for anti-nausea that costs literally 100 times more than the other seven alternatives we have now. The seven alternatives cost like 44 cents, and this one that's so expensive costs $44. My point is, if we use this three or four times a day over the course of a year, that's over $50,000 that we don't need to be spending. So my question was, how often do we use that? Previously it would have taken quite a bit of effort and time to even find that answer. We had that answer in minutes. We just pulled up MDClone. We looked at all the ED patients from this last calendar year, pulled up this medications, and how many times was this administered. Then we can look across the healthcare system to find out which hospitals are doing it, and then what departments are using it to find out the ER is the second most, but our post-operative unit was the most active user of this medication. There's a clinical indication for it, but I think it's being vastly overused. So talking to our post-op team, they took it off their order list, and we've seen a dramatic decline of this already within the first few weeks.
From a quality perspective across the institution, it starts in the ER, but it continues out from there. We could put our finger on the pulse of any of these care issues and assess them ourselves, without any other data teams or any delays or anything. It's really remarkable.
HCI: So is the data in MDClone organized differently than it is in a data warehouse or in in the EHR system? It sounds like you're able to do these searches and ask these specific questions and get answers in a way that I didn't think people could do that readily.
Spiegel: Yeah, it's really amazing. They base everything on events. So you have to declare an event, and the event could be showing up at the ER. At that event, you could then have any type of factual relations, like, what was your white blood cell count on that visit? Or you could do time-related events. So I could say, within one hour of arriving at the emergency department with chest pain, did you get an aspirin? Or within 24 hours, did you get a beta blocker? You could start with an event, and then you could either add data, or you could add time-related information that would have been very cumbersome in a normal relational database.
HCI: One thing that people often tell us is that there is a difference between accessing structured and unstructured data, and that in the past, the unstructured data in notes was hard to get at. But is that being addressed by large language models?
Spiegel: Yes, it's being addressed by LLMs. This system does have a natural natural language processing application with it. So we could do notes queries. For instance, our radiology reports have very few discrete data elements, so it has to be done through an NLP.
HCI: Is there an element of this that involves training clinicians to start thinking in this way and asking more of these questions?
Spiegel: Yes, that’s exactly where we're going with this. What we're realizing is that as people get the data, now they want to take action on it. So we're having a parallel work stream that's basically creating a self-service toolkit to enact change. How do you change order sets? How do you change a care pathway? How do you request EHR changes? So it's really become a fascinating thing where we're enabling the front-line providers to not only ask questions, but then, how do we enable them to take action?
HCI: This seems like a quite a significant shift in how you do research and quality improvement in the hospital. Anything else that I haven't asked about that you'd want to mention?
Spiegel: There are so many different opportunities to ensure we're providing the right care. We started off with a small cohort of our champions, who are now out talking about it with their colleagues. We’re going to be doing this campus-wide. And by campus I mean not just the medical center, but also the business school, the public policy school — all across the campus of University of Chicago, to say, hey, the healthcare system has data that you can use in a de-identified, synthetic way to ask and answer your questions and really open up the doors to research.
HCI: So people doing economic research or policy research could access the data?
Spiegel: Yes, we want to form those collaborations and also to make this data available. This was the enactment of what our dean called the One UChicago Vision. It wasn’t just the healthcare system but University of Chicago as an institution having access to this, so that that's where we're going, and that's why synthetic data is so important to us.
