Health data analytics is shaping up to become a global issue. But for that day to come, patients have to consent to sharing their data with providers and, likewise, providers must agree to do the same with their competitors.
While the politics are already proving difficult, the technology is stepping forward. IBM distinguished engineer Scott Schumacher and global healthcare ambassador Lorraine Fernandes spoke with Government Health IT Editor Tom Sullivan about the challenges of health data analytics, what it will take for physicians to actually be able to use IT systems as robust as Watson, and how the days of nations functioning as islands are disappearing.
Q: Reports circulated last week about the University of Tennessee’s Nautilus supercomputer, which a researcher claims applied sentiment analysis algorithms to determine the hit on Osama bin Laden and the conflict in Egypt, in arrears. Is that sort of predictive analytics happening in healthcare?
Schumacher: They’re not there yet. The underlying infrastructure is not there to do that. I think they would like to do that but the only industry where we’re hearing that is law enforcement and policing. It doesn’t seem that healthcare, from what I’ve seen, has been jumping on that bandwagon.
Fernandes: The point I’d add to that is that traditionally healthcare has underinvested in IT. Now finally we’re seeing providers, payers and the other components investing more, so there’s still a lot of putting their house in order, saying ‘Let’s make sure we have the pieces we need in order to do predictive analytics.’ There’s still a big segment of healthcare providers, the bricks-and-mortar set that are just starting to work toward attaining meaningful use, ensuring their patient and provider identities, get the clinical data out there to demonstrate they’re delivering better care for ACOs and other incentive programs. But there is a segment of the provider community that’s beyond that foundation and beginning to embrace predictive analytics and other heavy lifting.
Q: What sort of predictive analytics are they running?
Fernandes: When they decide to do that, they look at where the cost and quality issues are in healthcare. It’s the old 80/20 rule, so it’s your chronic illness patients, diabetics, congestive heart failure patients, hypertension patients. It’s where you can really do some analytics based on the data available today and impact care models, care delivery processes, drugs coming to market, look at the effectiveness of both quality and costs.
Q: That’s timely considering the study published this week estimating some 366 million diabetics globally and others saying that by 2020 more than half of Americans will be diabetic or pre-diabetic, though I have to be somewhat skeptical …
Schumacher: I do share skepticism with you on the predictive analytics stories. It’s hard to do predictive analytics right. They often report things they could have found but don't always include the number of false reports. We need to be able to know the acuracy of that prediction and the type of errors that come out of that. Still, it's hard to predict future trends but even harder without good analytic techniques.
Q: There’s a lot of buzz about Watson in healthcare but we don’t actually hear all that much in the way of detail, so what are some practical real-world examples?
Schumacher: Watson is sort of the showcase and it’s interesting to see how to bring this query capability to healthcare providers to decide if I have somebody in front of me with a particular set of diseases, what’s the literature on that set? And that’s where Watson works. There are a couple of plots that are going to go around Watson. The main practical use of Watson within healthcare is as a way to make that vast literature available in a usable way to the physician. That’s one of the key things. Healthcare providers don’t have time to keep up on all of the literature out there and having a way, not just Google where you type something in and what you want is on page 10, but here you could put in a complex query and get targeted information back. So the key to Watson in healthcare is the ability to ingest a large corpus of medical information, then it does the parsing, the natural language processing, and then respond at the point of care.
Q: Now, do you physicians at the point of care really know how to conduct complex queries?
Schumacher: That was one of the things if you watched Watson on "Jeopardy," it was being fed human questions. Who’s the president of Mexico? That’s one problem. But when a doctor has a patient with heart failure plus diabetes, plus this and plus that, what is out there about him? There’s a big difference between those two. There’s going to need to be a UI developed that will help the physician go through that. But the basic ability to do a snippet of human language is one of the key components.
Q: How would the average health system pay for, implement and ultimately reap the rewards of these powerful analytic systems at a time when technology is just starting to come into its own in the realm long known for being behind the cutting edge?
Schumacher: That is a real good question. Most of the providers we talk to are biting the bullet and investing in clinical warehousing capabilities. They see that to stay in business, with ACOs and changing payment models, they have to get ahead of what things cost and how patients move through. There’s an awareness that getting up to speed with basic analytics is just something that they have to do. Like the ICD-10 conversion, you’ve got to do it. When we talk to our customers they’re putting those things at the top of their lists. I think it goes back to your point about things changing, if I don’t understand my data, I don’t understand how patients move through, I don’t understand what it costs, I’m going to get killed on an ACO if I don’t bid it right.
If you look at the data that all these people have right now, a lot of it is free text, stuff that they need to get a handle on, transcription notes, lab results. So that brings the clinical health a little bit more in focus if they can unlock this information and get into some sort of a structured form that would support analytics. That’s one where there’s a return for providers that doesn’t require manual coding. If I switch to ICD-10, it might be a year or two years before I have sufficient information generated with that structure to use. But if you can start putting some information that you have in a longitudinal patient record, that’s some analysis I can start doing today really ahead of the new coding or other components.
Q: Health reform is essentially predicated on EHR adoption, but the underlying problem today is that health records exchange happens via print/fax/scan. How can we get over that hump?
Fernandes: I don’t think there is any singular solution. I think some of what ONC is doing in their various programs is going to begin to address the challenges. But it’s going to be a long, tough journey because you have the proprietary interests. If I share data am I potentially losing market share, losing patients? We hope that doesn’t inhibit people, but it is human behavior. We have the technological challenges, which we are chipping away, certainly. And then EHR adoption will hopefully do away with paper silos, but you don’t want to create electronic silos where the paper silos were in the past. I put some faith in consumers. I think we are going to have a population that is going to expect electronic records, expect that information is available for sharing, and that when I go from my physician to my dietician because I’ve been told I need to lose weight or manage my diabetes, that my information is appropriately shared. I put some faith in consumers building a groundswell along with what’s happening at a federal level with reform as a generic term.
Schumacher: The amusing part is that the younger generation assumes their doctor in Santa Monica has access to whatever happened to them when they went to Pittsburgh. As they realize that’s not the case, there will be pressure to make that the case. One of the real issues, though, is the dichotomy of sharing and privacy. As long as there are stiff penalties for inadvertent disclosure, I think people are going to be pretty reluctant to share data. Any sharing mechanism is going to have a certain amount of error rate and I don’t believe there’s a way out of that. So we have to be able to minimize that, obviously, but live with it to a certain extent.
Q: So when we look at the changes coming, there’s more granular health data, with ICD-10, more fluid health information exchange, we’re on the verge of a veritable mountain range of data – ideally to be followed by more robust analytic technologies that can produce better trends that improve outcomes. What does that world look like?
Schumacher: If we get stuck in this mentality that we’re going to create distributed electronic silos of information, then we always come back to ‘that’s a piece of the solution.’ I think if we can get to the point where there are places that data flow freely and can be aggregated in a de-identified way and can be used in an analytic way then I think the world becomes a place where we can start to take actions earlier, understand those actions and account for a better population. But there’s a lot of flux in the current design that makes it perhaps not as certain as we might want.
Fernandes: I would add to that, particularly because my role is global and I spend a lot of time outside the U.S. visiting customers and government officials around the world, that those mountain ranges are going to have to embrace the fact that it’s a global economy and a global society and that what might sit out on one of those mountain peaks in a de-identified fashion supports things like biosurveillance, and having data available for research around the world. It’s going to be really interesting how we embrace that change as a society and the economics associated with healthcare, and where that leads countries like the U.S. and Brazil and Australia and Hong Kong, the UK, everybody else out there, too. The days of functioning as an island are gone – it’s going to be a global issue as much as a local issue.
