August 7, 2014

Group Health & UW showcase analytic methods built for the real world

As our nation tries to revamp an ailing health care system, many are looking to a new kind of research to provide solutions: pragmatic clinical trials. Unlike traditional randomized controlled trials—which collect data on carefully selected populations under ideal conditions—pragmatic clinical trials take place in everyday settings and take advantage of existing data sources, such as electronic health records (EHRs). It’s an approach that’s both practical and efficient—and one that wouldn’t be possible without the innovative statistical methods behind it.

These methods will take center stage this September at the First Seattle Symposium on Health Care Data Analytics, sponsored by the Biostatistics Unit at Group Health Research Institute (GHRI) and co-hosted by the University of Washington (UW) Department of Biostatistics. This first-of-its-kind event will bring together scientists from across the country to discuss the statistical challenges of using EHR data for health research and to explore new methods developed to overcome those challenges.

“Without new methods being developed, interpreting findings from pragmatic clinical trials wouldn’t be feasible,” explains GHRI Biostatistician and Associate Investigator Andrea Cook, PhD. “Fixing an entire health care system means figuring out how to test interventions that will work in a real-world setting. You need different methods and a whole new approach to trial design to do that.”

Dr. Cook is organizing the symposium alongside GHRI Biostatistician and Senior Investigator Jennifer Nelson, PhD, and their longtime colleague Patrick Heagerty, PhD, associate chair of Biostatistics at the UW and a GHRI affiliate investigator. The trio will use the symposium to share new methods they and others have developed as part of national collaborations that rely on large health care databases to accelerate research that improves public health.

One such project is the Food and Drug Administration’s Mini-Sentinel, in which all three symposium organizers are helping develop methods to support post-marketing safety surveillance of new drugs and devices (see 2012 feature story). Another is the National Institutes of Health (NIH)’s Health Care Systems Research Collaboratory, which is co-led by GHRI Executive Director Eric B. Larson, MD, MPH, and researchers from Duke Clinical Research Institute and Harvard Pilgrim Health Care Institute. The Collaboratory currently supports seven pragmatic clinical trials designed to determine the best methods for mining the real-world data resources of large health systems like Group Health.

“A key factor in the successful use of existing data such as EHRs and billing records for research is to have a clear understanding of how the data arise,” explains Dr. Nelson. “Different medical systems may not record patients’ use of health care in exactly the same way. And in everyday health care settings, people use and receive care differently. If this underlying context is not accounted for in analyses, our answers to research questions will be biased.”

“For example, less healthy people might go to the doctor more often, so they will have more outcomes recorded,” adds Dr. Cook. “On the other hand, if an intervention is effective, it might make people healthier and less likely to visit the doctor, meaning they will have fewer outcomes. Without methods to control such variances in volume of outcomes, you can get inaccurate results.”

It’s a challenge that GHRI biostatisticians have been tackling for years. For instance, in a 2009 study published in The Lancet, Dr. Nelson and GHRI colleagues showed that prior research suggesting the flu vaccine decreased mortality among seniors had missed a key puzzle piece called “healthy user bias” (see 2009 feature story.)

“People who were able to get out of the house to go get a flu vaccine were healthier to begin with,” she says. “Methods used in prior database studies did not fully account for these differences, and this led to misleading findings. Using existing data helps make research more practical and efficient. But you need to ensure that the methods are also scientifically sound.”

With the recent rapid growth in pragmatic clinical trials and the use of health care databases for research, such methodological rigor is more important than ever. That’s why Drs. Cook, Nelson, and Heagerty decided to organize the upcoming symposium.

“We’re excited to bring together scientists from across the country to share ideas and learn about the best ways to use health care databases for research.” says Dr. Heagerty. “The symposium will present a variety of expert perspectives on what future studies in this arena will look like and how EHRs can be best leveraged to answer important research questions.”

In addition to pragmatic clinical trial design, the symposium will feature innovative methods for drug and vaccine safety surveillance, and inference and prediction using EHR data. Visit the symposium website to learn more and register. Space is limited—reserve your spot today!