FDA Looks to Big Data to Find Adverse Event Signals Among Reporting Noise
Posted 03 April 2013 | By
Two new safety tools proposed by the US Food and Drug Administration (FDA) would change the way regulators collect pharmacoepidemiological and adverse event data, making it easier for FDA to identify and address safety problems at their earliest points of emergence.
FDA's Science Board has a considerable amount of pull within the agency. When, for example, it released a 2007 report on the agency's scientific and regulatory deficiencies, the agency used the report to enact sweeping changes at the agency meant to make improvements.
So in 2010, when one of its reports called on the agency's Center for Drug Evaluation and Research (CDER) to widen its approaches to risk detection beyond pharmacoepidemiology-the assessment of the effects of pharmaceutical use in a population-regulators listened.
In particular the report explained that "pharmacoepidemiology can be expected to detect only the tip of the iceberg when seeking an increase in the incidence of common conditions."
"The need to integrate expertise in human and basic pharmacology is also evident in studies of comparative effectiveness, dominated so far by epidemiology and clinical trials uninformed by this discipline," it added.
An Age-Old Problem
The problem FDA is faced with is something of an age-old one: At the time of a drug's approval, very little is known about it. Even if the drug is subject to thousands of patients-10,000 for example-it still might not catch a one-in-15,000 fatal event. While this might not be a concern if the entire patient population is only 15,000 people, use in 15 million patients would statistically lead to 1,000 deaths. Additionally, patients enrolled in clinical trials are often the "ideal" patients, while real-world use will subject a product to a nearly infinite list of unique circumstances and uses.
Traditionally, regulators have been faced with two approaches: require more patients to be enrolled in trials in an attempt to identify those risks, or move those studies to a postmarketing setting where specific issues can be probed in real-world conditions. But FDA is now looking at something of a third approach, one predicated on making better use of already-existing data to help it find signals in the shadows of the noise.
A New Approach and a New Program
To accomplish this, FDA says it's adopting the Science Board's recommendations in the form of a new safety program within CDER's Office of Clinical Pharmacology (OCP), a sub-office of the Office of Translational Sciences (OTS). That program, called the "Pharmacological Mechanism-Based Drug Safety Prediction" (PMDSP) program, is intended to mine data and recognize patterns through the use of technology in order to develop predictions for as-yet unrecognized safety signals.
"An important goal is to allow the development of prospective hypotheses to provide to Office of Surveillance and Epidemiology (OSE) and Office of New Drugs (OND) to inform their evaluation of potential safety signals with the goal of reducing time and effort spent on false positive signals," FDA wrote in a notice soliciting outside support for the program.
The program is slated to conduct 10-12 pilot safety prediction projects in its first year, and FDA said it's ready to continue the program for at least the next five years.
Big Data to Find Small Events
A second program floated concurrently but separately from the PMDSP program is one FDA says will boost its ability to mine data contained in the MEDLINE database for drug adverse events. MEDLINE is maintained by the National Library of Medicine (NLM), and is a massive, 20 million-article database of biomedical abstracts, articles and citations.
FDA said it is already in the process of conducting a pilot program it expects to complete by Q3 2013 that will allow it to "identify disproportionate reporting of drug-adverse event pairs." That program is one that FDA now wishes to expand upon, and has sent out a solicitation looking for companies with the capability to build a technology platform that can mine data and "efficiently distinguish real signals from the background noise in huge pharmacovigilance databases."
The project would be slated for completion some time in 2014, with additional operations continuing thereafter.