The Patient-Centered Outcomes Research Institute (PCORI) has approved $93.5 million to support 29 clinical research data networks that together are forming a new resource known as PCORnet, the National Patient-Centered Clinical Research Network.
One goal of this “learning health system” is to reduce the time and effort needed to launch new studies and focus research on questions and outcomes useful to patients and those who care for them. One of the 29 networks is creating a pediatric learning health system. Called PEDSNet, the network includes eight of the nation's largest children's hospital health systems, and is being coordinated by the Children's Hospital of Philadelphia (CHOP).
Last week I saw a great presentation by Charles Bailey, M.D., a pediatric oncologist at CHOP, about using EHRs as building blocks for a learning health system. Bailey began by describing the goal of a learning health system as fusing clinical care and research into one process, so that every patient encounter becomes an opportunity to take better care of that patient and to learn form the encounter how to take better care of other patients.
Bailey discussed some of the challenges of getting data out of the EHR to do this type of research, as well as how to make the findings relevant to clinicians.
EHRs are proving to be much better as the primary source of research than administrative data, which studies have shown can lose up to 80 percent of diagnoses. In some cases, you find three times as many cases of something like otitis media (middle ear infection) in the notes section of the EHR than you would just looking at diagnosis codes, Bailey added.
There are many reasons for this: some clinicians don’t record chronic conditions; others don't record non-billable diagnoses; still others might use the wrong ICD9 codes.
In terms of data structure in the EHR, Bailey described several challenges. While some data is discrete and unambiguous such as dates, vital signs, some demographics and some lab values, other data is discrete, but not standardized, such as diagnoses or problem lists. Other data may be well defined but not discrete and the location may vary from place to place in the EHR.
There also are issues with harmonizing terminologies. A learning health system across multiple health centers requires agreed-upon ways to name things. “That is something we have not been very good at,” Bailey said, but we are moving toward a world where creating shared terminologies for many clinical uses is becoming more important.
The more data you can make discrete in the EHR, the more you can capture consistently at the point of care, Bailey said. But it takes time to enter discrete data, so you have to make it worthwhile to clinicians, he stressed. “I think it is important to make a social contract with your EHR users. You should make it easy to participate in research via the EHR and the research should make clinical care better,” he added. Don’t make them enter data twice. Tell them why it is important. Make use of the data collected to create tools for them. For instance, in a research network project, you could create for every patient a summary of recent care, detailing preventive care, labs and recommendations. “Return value to them, not just research,” Bailey said. “Then they will work with you.”
In summary, Bailey said that although there are methodological challenges, the EHR provides a unique opportunity to improve effectiveness research. “The challenges are tractable,” Bailey said, “and answers to clinically relevant questions can be obtained.”