Adverse Event Reviews: FDA Needs To Boost ‘Accuracy And Efficiency’
This article was originally published in The Pink Sheet Daily
Executive Summary
Agency seeks academic institutions and commercial organizations with tools that can conduct automated analysis of case reports from FAERS data.
FDA would like to cut back on the amount of time staffers have to spend poring over millions of adverse event case reports. To that end, the agency is looking for academic institutions or commercial organizations with technology that can run an automated analysis of the data.
FDA posted
a Request for Information “seeking tools and methods which will increase the accuracy and efficiency of AE case report reviews performed by FDA safety evaluators.” The agency said these tools would be used to triage and review postmarketing adverse event drug product case reports submitted to the FDA Adverse Event Reporting System (FAERS).
Specifically, FDA is looking for a TM/NLP/ML (Text Mining/Natural Language Processing/Machine Learning) system that can be applied to FAERS data.
The agency explained that safety evaluators in the Office of Surveillance and Epidemiology review AE case reports and conduct causality assessments to determine whether a drug may have caused the reported adverse event. FDA said this assessment relies heavily upon the quality of information and key data elements in the case reports.
“Regardless of the method used, conducting manual AE case report reviews for causality assessment is labor- and time-intensive,” the Request for Information, issued Dec. 22, states.
Is Automated Review Possible?
The agency noted that it received more than one million drug product AE case reports in 2013 and that the number is increasing annually. The reports are put into the FAERS database, which currently contains approximately nine million AE reports.
Last year, FDA noted that the 2013 tally was 30% more than the previous year, an unprecedented increase, and contributed to a data entry backlog in late June 2013 (Also see "FDA Faces More Tech Hurdles After Clearing AERs Backlog, And Congress Is Watching" - Pink Sheet, 2 Jun, 2014.).
The Request for Information says FDA will assemble “a dataset of case reports to serve as a gold standard and learning dataset for pilot testing” and that the applicant will provide the agency in-house access to the TM/NLP/ML system.
Brian Overstreet, co-founder and president of AdverseEvents Inc., said that in analyzing an AE report one often cannot determine if there is a solid link between the drug and adverse event without taking the accompanying narrative into account and someone has to read this information to make that assessment. He said an automated system that scans large blocks of text in the supporting information and pulls out key information that can prove or disprove causality could reduce the time FDA staff have to spend looking at the reports.
In putting out the request, FDA is “trying to fish to see if this is even possible,” Overstreet said. “If they get enough positive responses they will go the next step to a request for proposal for competitive bidding.”
Adverse Events uses its own system to filter and analyze the FAERS data. Launched three years ago, the company has used the system to compare the adverse events of drugs in a particular class (Also see "Drug Safety Business: Start-Up With AERS Filter Eyes Bisphosphonate Events" - Pink Sheet, 10 Oct, 2011.).
Overstreet said his company focuses on the back end, filtering the data after FDA publishes an event and is encouraged by the agency’s move towards automated review.
“The more FDA can do on the front end, the more utility and usefulness will come out of FAERS,” he stated.