FDA Seeks Help Using Algorithms to Detect Adverse Event Anomalies

Regulatory NewsRegulatory News | 23 January 2020 |  By 

As it becomes more difficult for the US Food and Drug Administration (FDA) to decipher when a series of adverse events could actually be a sign of a more significant problem, the agency is calling on the public to develop computational algorithms for the automatic detection of adverse event anomalies using publicly available data.

The current system of tracking adverse events is passive, in that patients, patient guardians, health care providers and manufacturers submit voluntary reports of adverse events associated with products, which FDA then analyzes.

“FDA regulators use a variety of data mining methods and tools to analyze the volumes of adverse event reports and identify possible safety signals. Disproportionality methods, which identify unexpectedly high statistical associations between products and adverse events, serve as a primary method for identifying safety signals. Change-point analysis identifies changes in longitudinal adverse event patterns for products,” the agency explains.

Currently, the agency uses a few different databases to collect adverse event reports, including the FDA Adverse Event Reporting System (FAERS), which contains adverse event reports, medication error reports and product quality complaints, the Vaccine Adverse Event Reporting System (VAERS), and for device adverse events, the Manufacturer and User Facility Device Experience (MAUDE) database. Last summer, FDA also ended its Alternative Summary Reporting (ASR) program for devices, and made available on its website all adverse event reports received under ASR exemptions from 1999 to 2019.

And while FDA acknowledges that data mining and hands-on case reviews performed by medical officers have been crucial in detecting adverse event safety signals, the agency is also interested to see if machine learning and artificial intelligence may provide new insights into the data.

“Of particular interest is the automatic detection of anomalies in FDA adverse event data,” the agency says, noting that such anomalies can come in the form of drug-specific adverse event patterns (e.g. common adverse reactions that were not identified in a drug’s clinical trials or not identified on a label) or multi-drug adverse event patterns (e.g. certain manufacturers seeing adverse events or multiple drugs sharing adverse event patterns).

This automated anomaly detection is meant to help with the agency’s data mining and case reviews by enabling the unsupervised identification of novel potential safety signals. FDA said such algorithms should be able to detect anomalies automatically and without the use of known anomaly labeled training data.

Participants who want to be included in the challenge have until the end of next month to submit their ideas. And FDA says that select participants may be recognized, including through the involvement in future results manuscript(s), and could have the unique opportunity to sit side-by-side with FDA officials on a panel at the Modernizing FDA’s Data Strategy public meeting on 27 March.

PrecisionFDA Challenge


© 2022 Regulatory Affairs Professionals Society.

Discover more of what matters to you

No taxonomy