The U.S. Food and Drug Administration is working to harness fast-accumulating personal health data from the likes of Twitter, Facebook and wearable devices. But more than 90 percent of analysts’ efforts to capitalize on that data falls below the targeted efficiency level for the FDA’s Office of Surveillance and Biometrics in its Center for Device and Radiological Health, said Isaac Chang, who directs post-market surveillance for that office. “We have observations of signals and patterns,” he said. “But they’re one-off maps.”
The assessment helped fuel a University of Maryland workshop on Sept. 11, 2015, led by the Centers of Excellence in Regulatory Science and Innovation and Health Information and Decision Systems (CHIDS). Chang joined other academic, regulatory and medical-product industry experts to discuss the latest methods and challenges to extracting valid information from the explosion of social media and mobile health data to improve regulatory decision-making and ultimately, the overall health of patients.
“Big data took off around 2007,” said CHIDS co-director Gordon Gao, a professor at UMD’s Robert H. Smith School of Business. “Only in the past two years have we begun to close the gap in terms of grasping and acting on it.”
To reach targeted efficiency, Chang said his office must combine its analysis of separate signals and patterns into one solid signal. Execution, though, will require “transdisciplinary convergence” — the workshop’s keynote-speech theme from Henry “Skip” Francis, director of data mining for the FDA’s Center for Drug Evaluation and Research in the Office of Translational Sciences.
This means experts from different disciplines, such as computer and data science, psychology and business must work together with leaders from the pharmaceutical and medical products industries, said CHIDS founder Ritu Agarwal, senior associate dean for faculty research at the Smith School. “This convergence of the research, policy and practice communities would produce a very deep understanding, and leveraging, of a wealth of unique data sources.”
But pitfalls loom. “In addition to navigating privacy concerns, we need causal, more than correlational, evidence for actionable knowledge,” Gao said.
“It's easy to find spurious correlations in social media conversations,” Agarwal said. “We must be careful to be able to separate the ‘signal from the noise.’”
“Getting to the signal means drilling into specific communities for conversations sparked by trusted, influential sources and definable by such categories as location or baseline readings of trending topics,” said William Rand, assistant professor of marketing and director of Smith’s Center for Complexity in Business.
Trustworthy, high influencers maintain and bridge topical conversations between friends and followers in multiple networks over an extended period, Rand said. Given such a chain — with the right topic, "regulators could shape immunization educational messages with cues from conversations traversing from stay-at-home moms to school administrators."
Rand’s recent research into messaging surrounding Superstorm Sandy illustrates location’s significance in public health surveillance. Just ahead of the storm, conversations within 10 miles of New York City focused on concerns over food and water supplies. Residents outside that range talked about flooding and wind damage. Geographically separating those signals in real time, he said, could have heightened efficiency in relief efforts.
In addition to social media conversations, medical device data systems “are gauging patients’ perceptions and interpretations in a very deep and granular fashion like never before,” Agarwal said. “What's remarkable is that we can do all of this at scale and speed. We don’t have to wait 15 years for a randomized clinical trial to learn about the time to failure of a medical device. We can learn about how these things are working in real time.”
In addition to experts collaborating across disciplines, improving access to these data and to new tools will strengthen opportunities, said CHIDS Deputy Director Kenyon Crowley, one of the workshop organizers. "Achieving the grand vision still may take several years, but we will start to see more and more examples of findings and insight further demonstrating the value of these data,” he said.
Related CHIDS Activity
The center is collaborating with UMD’s School of Public Health to gauge and model information technology’s effect on public health and clinical care programs in Montgomery County, Md. Agarwal and Crowley discussed the project in a recent University of Kentucky-hosted webinar.
Collaborating with technology firm Inovalon, CHIDS researchers will investigate “patient churning” and Medicare Advantage performance in separate studies. The ‘churning’ study aims to reduce the involuntary movement of covered members from one health plan or coverage system to another. The other study seeks to identify which MA insurance plans are more likely to improve in the Centers for Medicare and Medicaid Services Five Star Rating system. The first-of-its-kind analysis will explore how patients benefit from plans that achieve higher ratings and examine how members respond to ratings, while making enrollment decisions. In addition, the analysis will identify factors that can help plans efficiently close gaps in care and improve star ratings. Read more, from UMD’s Office of Technology Commercialization…