Which GMP quality system was most often linked to FDA drug recalls since 2012? 

Redica’s Jerry Chapman asked the audience that question at the February Pharma Hot Topics Conference in San Diego during his presentation, “A Deep Dive Into Industry-Wide Recall Data: Leveraging an Expert Ontology and AI to Uncover Patterns and Insights.”

A senior GMP quality expert at Redica Systems, Chapman’s pointed out just how wrong conventional thinking — and analysis of raw FDA data — can really be.

Raw Data Figure 1
Figure 1 | Cleaned, organized, and classified, Redica Systems Enforcement Analytics data contradicts conventional thinking — production and quality systems were most often linked to FDA drug recalls from 2012–2021, not packaging and labeling.

“The overwhelming majority of you thought that the recalls are mostly packaging and labeling,” Chapman says. “Packaging and labeling was a distant third.” Instead, production and quality systems together accounted for an impressive 88% of drug recalls through Redica’s Enforcement Analytics.

Easy Mistakes to Make

Raw data is the problem. Without a dataset that’s cleaned, organized, and logically classified, your analysis and subsequent decisions could be misled. 

In the raw FDA data, the packaging and labeling quality system is indeed most often linked to recalls. There are more than 13,000 drug recalls in the full set, which is free and available for download from fda.gov.

You can’t count ‘label issue’ 538 times or you’re going to be wrong.

But there’s a separate event ID for each preparation or dosage of a particular drug in many, many recalls. And this is one of the problems Redica Systems solves.

“In 2014, there were two different companies that issued recalls for a label mix-up,” says Chapman. “So there’s 538 label-related recalls. Each was really a single recall. You can’t count ‘label issue’ 538 times or you’re going to be wrong.”

A Glimpse into Redica’s Data Cleanup and Classification

Correcting for the over-count, Redica Systems data reveals about 3,000 actual recall events, with much lower variability in the set overall.

Raw Data Figure 2
Figure 2 | Removing redundant data, Redica Systems reveals a much smaller, more consistent dataset.

With the help of its expert models and natural language processing, Redica Systems then classifies each event (not each event ID!) according to which drug GMP quality system it’s linked with:

  • Quality
  • Production
  • Facilities and Equipment
  • Packaging and Labeling
  • Materials
  • Laboratory

A Data Goldmine, Ready for Digging Into

Taking a closer look at this dataset reveals insights you can’t get from the raw data. 

For example, presenting the cleaned and classified data over time reveals that Production had been the major system linked to drug recalls until about 2018, when Quality took the lead spot.

Raw Data Figure 3
Figure 3 | In the cleaned-up FDA dataset of drug recalls, 2012–2021, a time series shows Quality-linked recalls becoming the most common type around 2018.

Remove More Dirt, Find More Gold

Taking only recalls linked to the Quality system (from among the six Drug CGMP Quality Systems), 79% involve CGMP deviations and 21% are recalls in which drugs had been marketed without an approved NDA or ANDA.

Raw Data Figure 4
Figure 4 | Of Quality System-related recalls, 79% feature CGMP Deviations, while 21% were marketed without an approved NDA or ANDA.

“Now if you look at GMP deviations, not all of them are classified,” says Chapman, “only about half.” “So within the FDA dataset, sometimes it’ll just say, ‘Recalled because of a deviation.’ But other times it will say ‘Recalled because of a deviation with an issue with discoloration of product,’ or whatever.”

Breaking these classified events into categories and presenting them in a time series reveals something strange: a 2018–2020 surge in recalls related to Laboratory deviations, followed by a huge surge in Materials deviations.

Figure 5
Figure 5 | With special causes removed, it’s much easier to see what caused CGMP deviation recalls in the 2012–2021 FDA dataset.

“That happens to be recalls due to nitrosamines,” Chapman says. “Came up in 2018, peaked in 2019, FDA comes out with a guidance in 2019, and then they go down in 2020.”

The Material system deviations resulted from a temperature deviation at a single CMO warehouse, causing the recall of 110 different products. 

Correcting for those special causes shows a truer picture, with Laboratory deviations rising from 2012 and falling almost to zero in 2020, among other interesting changes.

Digging Deeper for More Insight

Chapman focuses on a few more examples of how Redica Systems can explore this data. For example, among recalls related to the Production system, 21% resulted from the presence of particulates.

Raw Data Figure 6
Figure 6 | In the FDA 2012–2021 drug-recall dataset, 21% of those involving the Production System resulted from particulate contamination. Of those, the particulate itself was unknown in 75% of cases, with the remaining 25% broken down by type.

“75% of the time they’re unknown,” says Chapman. That’s not great, he says, but most will figure it out after the recall. Events with known particulates can help other sites determine where to focus their own efforts.

There’s Always More to Explore 

Redica Systems features an unmatched structured dataset of drug regulatory enforcement, as well as regulatory developments. Cleaned, organized, and classified, Redica Systems can alert your organization about potential problem areas in your supply chain. 

Contact Redica Systems today for a tour and learn how to use it in Inspection Preparation, Supplier Monitoring, and Regulatory Intelligence.

Get a Demo

We’ll can show you insights into any of your key suppliers, FDA investigators, inspection trends, and much more.

Request a Demo