FDA Mines Billing Data for Drug Interaction Insight
The Food and Drug Administration is five years into a pilot program aimed at identifying hazardous drug interactions by mining the medical billing records of millions of Americans. The program, dubbed Mini-Sentinel, is a creative application of big data technologies that has the potential to improve people’s lives.
The FDA started funding the Mini-Sentinel project in 2009 with the goal of coming up a better way to monitor for unintended side effects of prescription drugs. It’s impossible to eliminate all adverse reactions to medications. But by closely monitoring the impact of newly approved drugs, the FDA may be able to avoid situations like Merck’s 2004 withdrawal of the anti-inflammatory drug Vioxx, which was blamed in the deaths of 60,000 people.
The FDA faced difficult problems with its existing reporting system. According to a recent report on the FDA’s new reporting system at NPR.org, the existing system relied heavily on voluntary reports about adverse reactions sent in by doctors, pharmacists, and patients. The FDA receives about 1 million such reports a year. But with millions of lives at stake, a voluntary system was seen as too inaccurate, slow, and random.
The FDA craved a new reporting system that could yield faster insights about the safety of drugs after they’ve hit the market. That’s the genesis behind Mini-Sentinel, a five-year, $116-million project that’s part of the FDA’s larger Sentinel program to provide more active surveillance of regulated medical products.
Mini-Sentinel pulls most of its data from billing records, specifically the diagnostic and test codes that doctors, hospitals, and insurance companies use to determine how the costs will be reimbursed. The insurance companies Aetna, Humana, and Kaiser Permanente are data partners in the project, representing about 150 million covered Americans.
The distributed Mini-Sentinel database can survey more than 350 million person-years worth of observation, including 4 billion drug prescriptions and 4.1 billion patient encounters, Dr. Janet Woodcock, director of the FDA’s Center for Drug Evaluation and Research, told Health IT Outcomes. The program, which is headed by Dr. Richard Platt of Harvard Medical School’s Department of Population Medicine, is anonymized to prevent the chance of privacy infractions.
The FDA is doing hundreds of queries a year with Mini-Sentinel, and it’s showing some good outcomes. According to the NPR.org story, the FDA was able to spot a correlation between a blood pressure drug called olmesartan and intestinal problems, as reported in the diagnostic codes of billing records. The insight led the FDA to add a warning label to the blood pressure drug.
However, not everybody is convinced this approach is a good one. Aside from the obvious privacy issues, there are questions about the underlying data science. Billing records were not designed to be used in this way, the critics say, and scientists need to be very careful in how they craft their queries. They should also be careful how much faith people should put into the results.
One critic of Mini-Sentinel is says Thomas Moore, a senior scientist at the Institute for Safe Medication Practices. Moore points to an instance where a Mini-Sentinel analysis failed to find a correlation between the anticoagulant drug dabigatran and excessive and fatal bleeding, as they’re reported in the billing records. In an April 2013 article in the New England Journal of Medicine, researchers pointed to flawed data science in the Mini-Sentinel program.
“There are limitations to the Mini-Sentinel analysis, including lack of adjustment for confounding variables and lack of a detailed medical record review (to verify whether the claim code reflected an actual bleeding occurrence),” the authors write in the NEJM. “To address some of these limitations, we are now conducting two protocol-based assessments, using claims data from Mini-Sentinel and other claims databases, in which adjustments will be made for confounding factors.”
The FDA’s Woodcock admits that Mini-Sentinel does have limitations. “Things may change depending on what claims data set you use, or how you run your definitions, how you set up your parameters and so forth,” Woodcock says in the NPR.org article. “And there’s no doubt these studies are vulnerable to all these changes.”
The future of Mini-Sentinel is unclear. The program’s five-year lifespan is up in September, and the FDA has yet to say whether it will continue funding it. Whether or not the FDA’s search for drug side effects continues in its current form, you can expect the federal agency to continue its search for a better reporting system.