Skip to content Skip to navigation

Top Ten Tech Trends: Early Building Blocks of the Learning Health System Falling Into Place

January 22, 2015
by David Raths
| Reprints
Clinical data research networks lay the groundwork for more rapid and cost-effective research

Many informatics experts see the signs of health IT progress taking place today as stepping stones toward the development of a learning health system, in which health systems routinely share data to learn more about the causes of disease and the treatments for disorders.

Although that type of network is years away, in 2015 we already are seeing the foundational elements being put in place and can watch the pioneers doing the early work, especially the 11 clinical data research networks funded by the federal Patient-Centered Outcome Research Institute (PCORI), including the National Pediatric Learning Health System (PEDSNet) and the Chicago Area Patient-Centered Outcomes Research Network (CAPriCorn).

Speaking to the Health Data Consortium on May 28, 2014, PCORI Executive Director Joe Selby, M.D., M.P.H., said the work is “designed to enable practice-changing research by harnessing the vast data locked within health systems and clinical settings, as well as information and experiences reported by patients themselves…Our goal is to use the power of large sets of healthcare data, under policies developed with the help of patients, to enable more rapid and cost-effective clinical research.”

These new networks are federations of entities coming together to learn from sharing data and doing analytics on that data. “They apply learning back to inform practice in ways that improve health,” says Chuck Friedman, Ph.D., chair of the University of Michigan Medical School’s Department of Learning Health Sciences and one of the most vocal advocates of a national-scale learning health system. “Each of these is a valuable experiment when you view the large-scale picture of how we are going to do this for the whole country. We have these experiments busting out all over the place,” he adds. “Complementary to those networks are places like Intermountain, Kaiser and others that have become mature learning systems within the boundaries of their own organizations.”

Chuck Friedman, Ph.D.

Michael Kahn, M.D., professor of epidemiology in the Department of Pediatrics at the University of Colorado Denver, is one of the principal investigators of the PEDSNet project, which is creating a distributed network of standardize data on more than 1 million children to enable data sharing, cohort identification and research. Kahn says the goal is to leverage the data infrastructure being created to establish a learning health system of rapid-cycle improvement, incremental changes, and distributed clinical decision support.

The learning network concepts have been in development for several years, but because of the PCORI funding and attention, it is accelerating, he says. “I think we are learning how to do this better.”

The difficult data and terminology harmonization work is “door-to-door combat,” Kahn says. “But we needed to be sensitive to the fact that different organizations had placed institutional investments in different starting points.”

These networks of large data sets add complexity in terms of governance and consent, he added. How long before PEDSNet starts seeing research results? “I think we are more than a year out and less than two years out,” Kahn says. “You have to have enough data of enough depth to be of interest. Dribs and drabs are not enough. You have to have a fairly heavy base. We are focused on getting to that critical mass of data.”

The American Society of Clinical Oncology (ASCO) is building its own research data network, CancerLinQ, to aggregate and analyze a massive amount of real-world cancer care data. By the end of 2015, it expects to have the platform built and 15 vanguard practices fully engaged in using it, says Robert Hauser, PharmD, senior director of ASCO’s quality department. One goal is to provide real-time clinical decision support to help physicians choose the right therapy for each patient.

“I don’t think it is too early to be thinking about a national learning health system,” Hauser says. “We believe the whole premise of this is to break down silos so you can rapidly learn from the data. So just creating bigger silos wouldn’t help. We are behind the idea of a national learning health system.”

Although a learning health system is a relatively new concept, it is seen as the Holy Grail of healthcare analytics, says Bruce Eckert, national practice director, strategic advisory, for the Weymouth, Mass.-based Beacon Partners consulting firm. “It is great that we are automating processes inside organizations, but ultimately the big win is leveraging that data to improve our clinical performance.”

Eckert says there are many challenges to making this work. “There are all kinds of data mismatch problems. The big one is the patient-matching problem, which has yet to be solved in many settings, including this one,” he notes.

Other problems involve the nuts and bolts of data exchange and analytics, making sure people are using the same ontologies, code sets and definitions. “Then you get into workflow and practices in the organizations. Are their standards of data collection the same? As we get into it, we’ll find there are lots of questions and issues about the ability to compare data across health systems,” Eckert says; but the learning health system movement will continue to drive standardization in these areas, he added, so it is easier to compare data.

Eckert adds that he would like to see the learning network effort linking with the all-payer claims database effort—that might go a long way toward solving some of these issues.

Another concern is the business model around learning health systems. We may see the same sustainability issues we have had with HIEs, he said.