The relationship between measurement and improvement is a familiar one in our everyday lives. If you wear a fitness tracker to measure your daily step count, you might start changing your habits to walk to more places and get more steps in. Healthcare practitioners know well that the same principle applies in quality-improvement initiatives they undertake on a regular basis. The advent of electronic medical records (EMRs) has made measurement possibilities more powerful and more useful as large-scale research tools.
But there are limits to how large the scale can get in the U.S. while still providing accurate and comprehensive data, according to a team of researchers in Philadelphia who recently authored a viewpoint article in JAMA Pediatrics. Sage Myers, MD, MSCE, an emergency medicine physician at Children’s Hospital of Philadelphia and assistant professor of Pediatrics at the Perelman School of Medicine at the University of Pennsylvania, co-authored the piece with Brendan Carr, MD, associate professor of Emergency Medicine and associate dean at Thomas Jefferson University, and Charles Branas, PhD, professor of Epidemiology at the Perelman School of Medicine at Penn. They point out barriers to developing a national infrastructure for a learning health system and argue that, despite those barriers, there is enormous opportunity to do so with key investments and participation.
Cornerstone recently discussed this viewpoint with Dr. Myers to learn more. The edited conversation follows below.
Before we talk about making a national version of the learning health system model, let’s set the stage. What is a learning health system?
The learning health system is a model where you try to use the data that you’re collecting on an ongoing basis for clinical care to your advantage. The idea is to use existing data, such as data collected in electronic medical records and billing records, to find places where you need to change things to improve care, such as where there are disparities or outcomes that are not ideal. Or you use that data to evaluate changes that you’ve made in healthcare delivery, such as new education offered to your physicians or a change in the EMR display to make them rethink prescribing an antibiotic.
Now that we’ve made this push across the country to have electronic health data everywhere, it’s really harnessing all that work and money to use that data to improve healthcare for patients.
What prompted you and your co-authors to write this viewpoint about taking this concept to a national level?
We’ve been doing work with nationally representative billing datasets that are provided by the Agency for Healthcare Research and Quality (AHRQ) to look at questions about outcomes and care delivery across the nation. We specifically were interested in trauma and were using this to look at trauma outcomes for pediatric patients and how we’ve set up the trauma system across the U.S. and how is that working. Unfortunately, there’s a lag in time between when that data is collected and when it’s publicly available. And then it takes us time to do all the data analysis and put together the manuscripts. Once we had everything together with what, in my view, were interesting results, it was really hard to get them published because by that time the data was so old that journals didn’t want to publish the manuscript. They reviewed it and made great comments about the results themselves, but the sticking point was always that the data is more than five years old, and that is too old.
That experience, combined with some other work that we had been doing with other federal data, made us consider, what are the barriers to being able to use this data on an ongoing fashion? We hit one of those barriers, so then we started thinking about the others.
We also started thinking about how, in an ideal world, you’d be able to use national electronic health data to look at huge systemic questions in healthcare, such as national policy changes in healthcare delivery. Say you’ve decided you want to make stroke centers to specialize in treating stroke patients at those hospitals, and you want to evaluate the outcome. In the past, people have mostly gotten data from the stroke centers themselves because those are large hospitals that can collect that data for research. But that leaves out all the other hospitals that are involved in the larger system, even though they have their own electronic data sitting there. They just don’t get to use it.
Can you review the other barriers and challenges you discussed in the viewpoint, in addition to the timeliness issue you ran into in your own work?
Cost is a big issue. That data is very expensive, in some ways for good reason, because it does take time to clean all that data and put it in a form that is usable. But if that cost could be offset because we see, as a country, that we have already put a lot of money into the upfront cost of creating this data, investing a bit more to make it usable would probably be a much better return on investment for us.
There is also incompleteness in the data now made available. A large percentage of the data that is available is based on billing data. Billing data can give you some information, but it can’t give you a lot of very specific physiologic information. Even some of the care delivery information that’s not billable as separate items is lost. If we could move to a system where we’re able to use the up-front data that we’re collecting in electronic health records, then that would allow us to use information that’s more complete.
The only place where we can get some of that direct information is from Medicare and Medicaid, although still, it’s mostly billing data. It’s much more complete than the others. But sharing data is not a requirement for these agencies. So if an agency [that administers these federal insurance programs at the state or sub-regional level] submits their data, it’s good data, but not all the agencies submit their data. Just making it a requirement of the funding for states to share their data, we could probably get more of it. But even then, including some or all the private payers would be better to complete the picture. There are similar challenges even when data isn’t coming from insurance companies. For example, the data from AHRQ is collected on a state by state basis, and the states themselves sometimes withhold some types of information or information about some subpopulations because they may be concerned about privacy or about how it’s going to make the state look. Not having access to that full data makes it hard to, for example, adjust for race or look at disparities in the data. Hospital information will also sometimes be withheld, or it will be given to a national organization but not be made public. I think that is fine as far as individually identifiable hospitals because we don’t have to know the name of what hospital those patients belong to, but you have to know something about the hospital to be able to know how to adjust for those factors, and sometimes that’s not available.
Aside from the fact that a lot of electronic health data is already being collected, what are the signs of opportunity for a national learning health system, despite those barriers?
One thing is that the federal government has taken on open data as an issue in many other areas. It just hasn’t trickled down into healthcare yet. They’ve created this data.gov portal where there’s public access to tons of data that the federal government collects for different programs that it supports. It is even interactive to the point that end users can create apps that analyze that data, that people can then use. It allows citizens to go in and help to analyze some of the data that’s there. For example, Farm-Plenty is an app that helps farmers better analyze open government data from the U.S. Department of Agriculture on crops grown within a five mile radius of their farms. They can use this data to figure that out, because somebody’s already done that work. Why would we all have to work to find out again?
To have that sort of commitment to health data would make it so much more usable. We could stop recreating the wheels in some way. Things like that would allow the data to be used more efficiently to get us answers more quickly.
Do you see your viewpoint article as a call to action on creating a national learning health system infrastructure? What do you hope the outcome is?
The goal is that it makes people start to think about the investment that we’ve made up front into creating electronic health records across the United States, and to look at how we are harnessing that to get a return on that investment. One benefit people have talked about is being able to share data from place to place [for an individual patient moving from the care of a doctor in one hospital or health network to another], which also hasn’t been realized at this point, but that I think is a small goal of that data. We have the computing technology to be able to handle all of this information to reach for a bigger goal of using that data in near real time to look at big systemic questions.
You wrote this viewpoint piece before the most recent presidential election. Since we’re talking about government policy changes, what are your thoughts about how the new administration and legislature might approach these issues?
We’re definitely stepping into uncharted territory, and I think we all have unanswered questions about the funding environment and priorities.
On this topic specifically, I think it would be foolish for a government to not capitalize on the investments that are already made. The data already exists. It’s not like the data collection part of it is going to be taken away. But what do you do with that data?
There is a good business case not to waste that investment if you can invest a small amount more to have a huge return on that. The case is even stronger if you start to think about how can it feed back into the system to address inefficiencies in healthcare delivery. Where we spend too much on things, this data can be used to try to figure out where those inefficiencies are and what policies do or don’t actually make that better. And you can’t do it all internally in one facility or system. You need the power of the people. You need to make that available so that researchers, with appropriate ethical guidelines, can be working to figure out solutions.