So Far, FDA’s Sentinel Drug Safety Monitoring Initiative Has Not Delivered as Hoped | Pharma Industry Regulation |

With access to claims data on about 200 million patients and 5.5 billion patient encounters across the United States, the US Food and Drug Administration’s (FDA’s) Sentinel Initiative has grown into a juggernaut of patient experience since its pilot program launched in 2008 (Kuehn BM. JAMA. 2008;300[2]:156-157).


The pilot phase, Mini Sentinel, met a Food and Drug Administration Amendments Act (FDAAA) requirement to create an active drug safety monitoring system. Created as a public-private partnership, the system would give the FDA the ability to query patient data while still protecting patient privacy. The FDA contracted the Harvard Pilgrim Health Care Institute to be the initiative’s operating center, which interacts directly with Sentinel’s 18 data partners, predominantly major health insurers recruited to provide access to data on large swaths of US patients. The center has also worked to develop and refine the methods for querying partners’ data.


In the past year, the system has transitioned out of the “mini” phase, Janet Woodcock, MD, director of the FDA’s Center for Drug Evaluation and Research (CDER), said during the annual Sentinel Initiative Public Workshop hosted in February by the Duke-Margolis Center for Health Policy.


“Sentinel has become an important part of our routine safety surveillance,” Woodcock said at the workshop.


Many herald the developments so far as major achievements. But only a handful of FDA safety actions have resulted from Sentinel, and it has not yet become the active drug safety surveillance system many hoped it would be. For example, the hope by some that Sentinel would be able to flag previously unknown safety problems with drugs hasn’t yet panned out.


“It’s very ambitious, and I congratulate everyone working on it,” Doris Peter, PhD, director of Consumer Reports Best Buy Drugs project, said at the workshop. “The original use case of finding [drug safety problems]… hasn’t been delivered on yet, and we don’t want the FDA to lose sight of that.”


At the time of the interim analysis, Sentinel queries had resulted in no medical product withdrawals or recalls, 2 label changes, and 3 safety communications, according to the FDAAA-mandated report. In many cases, Sentinel has been used not as the primary source but as an adjunct to other agency tools.


Some Sentinel findings have been questioned. Reports of serious and fatal bleeding-related adverse events associated with a newer anticoagulant, dabigatran, around 2011 led to a Sentinel query. Based on an analysis of Sentinel data, the FDA concluded the bleeding rates associated with dabigatran were not higher than those with the long-time anticoagulant warfarin and that dabigatran’s new drug status may have artificially driven up adverse event reports (Southworth MR et al. N Engl J Med. 2013;368[14]:1272-1274). The agency issued a 2012 safety communication about the Sentinel-derived results (


However, Sentinel captured only 19 cases of gastrointestinal bleeding, and data on risk factors were missing (Moore TJ et al. BMJ. 2014;349:g4517). What’s more, contrary to Sentinel findings, a more recent meta-analysis of clinical trials comparing the 2 drugs found an increased risk of bleeding-related adverse events with dabigatran compared with warfarin (Sipahi I et al. JAMA Intern Med. 2014;174[1]:150-151).


“[The experience with dabigatran] teaches us something about the dawn of the era of electronic surveillance,” said Thomas Moore, senior scientist at the Institute for Safe Medication Practices. The system is not yet reliable enough to become a primary source or to identify unknown safety problems, he noted. Moore and his colleagues have argued that more work is needed to validate the reliability and reproducibility of studies using electronic health data and to determine what types of questions the studies can answer (Moore TJ and Furberg CD. Drug Saf. 2015;38[7]:601-610).


“It’s a good thing that FDA has invested heavily in learning what we can do with electronic health records at the present time,” Moore said. “Unfortunately, the lesson is not nearly as much as we’d hoped.”