In a recent post by Andy Oram on EMR and EHR titles “Exploring the Role of Clinical Documentation: a Step Toward EHRs for Learning” he introduced me to the idea of what he called a Learning Health Care System. Here’s his description:
Currently a popular buzzword, a learning health care system collects data from clinicians, patients, and the general population to look for evidence and correlations that can improve the delivery of health care. The learning system can determine the prevalence of health disorders in an area, pick out which people are most at risk, find out how well treatments work, etc. It is often called a “closed loop system” because it can draw on information generated from within the system to change course quickly.
I really love the concept and description of a learning healthcare system. Unfortunately, I see so very little of this in our current EHR technology and that’s a travesty. However, it’s absolutely the way we need to head. Andy add this insight into why we don’t yet have a learning health care system:
“Vendors need to improve the ability of systems to capture and manage structured data.” We need structured data for our learning health care system, and we can’t wait for natural language processing to evolve to the point where it can reliably extract the necessary elements of a document.
While I agree that managed structured data would be helpful in reaching the vision of a learning healthcare system, I don’t think we have to wait for that to happen. We can already use the data that’s available to make our EHRs smarter than they are today. Certainly we can’t do everything that we’d like to do with them, but we can do something. We shouldn’t do nothing just because we can’t do everything.
Plus, I’ve written about this a number of times before, but we need to create a means for the healthcare system to learn and for healthcare systems to be able to easily share that learning. This might be a different definition of leaning than what Andy described. I think he was referencing a learning system that learns about the patient. I’m taking it one step further and we need a healthcare system that learns something about technology or data to be able to easily share that learning with other outside healthcare systems. That would be powerful.
What are your thoughts on what Andy calls a popular buzzword: A Learning Health Care System? Are we heading that direction? What’s holding us back?
Hi John, This topic hits my greatest interest. I strongly believe that when we gather enough data across a broad spectrum of cohorts that represent all populations (not just those that are easy to capture or are laser focused on one part of the patient population that shows up for care) we will have the right “big data” to start to ask good questions and get better answers, both for individual patients and to treat populations. We aren’t there yet, but we have most of the pieces in place to get there. We just need to organize and connect them. And yes to the fact we have to begin to imagine the potential of technology to catapult us into a true knowledge-based system that goes beyond what we currently think of as evidence-based medicine. As you probably know, I’ve been blogging in obscurity about this for about five years over at http://www.healthsystemed.com – the “ed” is for education and the icon is a little blue androgynous doc “Ed” holding a tablet instead of a clipboard. I harp on this issue for Health Wonk Review where some policy wonks chime in on their different areas of interest. If you want to run this comment, maybe some people will head over to join the conversation on a learning health care system at Health System Ed. Thanks again for all the great work you are doing out there in the trenches! Keep it up! Best, Peggy
Thanks for taking off from my article, John. I think, in response to Peggy, that one tenet of big data miners is that you need to sweep in a lot of information. Making judgments based on 1,000 patients can be risky. Making the judgments on 1,000,000 patients is more likely to shake out oddities in the data and find real trends. This is why the VA or Kaiser are better at finding trends than a typical stand-alone hospital or clinic. But we must learn how to safely deidentify data and license it in ways that prevent abuse.
As to your point, John, I think it’s a neat idea that we learn not only what treatments work well (and on whom) but what technological interventions help.
Peggy,
I’ve been blogging in obscurity here for 10 years, so I know the feeling. Thanks for joining the conversation. Glad to see our interests align.
Andy,
I think treatment is treatment. I don’t care if it’s a drug, an exercise plan, a tech solution, etc. We need a platform that covers all of those things and shares insights that people find.