The RxView Blog

Real world data, analytics, and insights for healthcare decision makers

Advera Health Analytics, Inc.

A Crystal Ball for FDA Action: How we Validated RxSignal

Posted by Dr. Keith Hoffman on March 11, 2016

rxsignal_drug_safety_cover.pngWe are pleased to report that our latest major publication, “A Pharmacovigilance Signaling System Based on FDA Regulatory Action and Post-Marketing Adverse Event Reports” was released this week in the esteemed journal “Drug Safety.”  It was published as an online-first feature and will also appear in the April copy of the print version of Drug Safety.

What was the goal of the RxSignal system? That was straightforward - develop an analytic that could accurately predict post-marketing adverse event label changes for FDA approved drugs.

However, in “Big Data,” implementation of straightforward goals abruptly faces the reality of multiple considerations. Which FDA actions to assess? Which drugs to look at first? There are millions upon millions of potential drug / adverse event pairs to analyze within FDA’s Adverse Event Reporting System (FAERS) - which should we focus on? How should we define a signal? Can we sort out which adverse events FDA preferentially acts upon, and, alternately which they seem to rarely act on? What inclusion and exclusion criteria will produce reliable - but also broad and actionable - data for our clients? How do we objectively test the method once we decide on the parameters above?

None of the questions above were easy, but,we systematically, objectively, and thoroughly (painstakingly so) attacked each point above.

After many months of highs and instructive lows, the dust finally settled. I can now say confidently, well, we “nailed it.”

The issue of focus was paramount. We narrowed the vast sea of millions of potential drug / adverse event pairs down considerably. In our test group, the model selected just 97 drug/AE pairs as potential signals out of 41,834 possible pairs.

Beyond considerable focus, it would be meaningless unless the signals provided accurate predictions. Our back-testing found that 73% of future label changes in an independent group of drugs (i.e. a group of drugs that had nothing to do with the establishment of our inclusion and exclusion criteria, etc.) were accurately predicted.

73%. I’m really happy with that, especially given some of vagaries of how/when/why FDA adds warnings to drug labels. At the very start of the project I wasn’t confident that we could create a quantitative method that could deliver that high of a success rate.

But we did it.

Our RxSignal system, at least to my knowledge, is the world’s only method to identify serious adverse events that are likely to be added to a drug’s label.

I’m proud to say that we are at the forefront of inventing and developing actionable analytics that provide our clients with powerful insights into drug safety.

Contact us directly to request a full copy of the Drug Safety publication.

 

keith_headshot4.png

Keith Hoffman, PhD | VP of Scientific Affairs

Topics: FDA, FAERS, RxSignal

Subscribe to Email Updates