DEV Community

dicaria
dicaria

Posted on Edited on Originally published at dicaria.com

Why are US drugs recalled? Mostly sterility and particles

One in five FDA drug recall records tells you almost nothing. In the 600 most recent entries, 114 say only "CGMP deviations" or similar — manufacturing practice failed, nothing more. No defect, no finding, no product detail. The field that looks like an explanation column mostly isn't one.

Read the reason field as a whole instead of row by row and a split appears: some reasons name what was found in the product — particles, microbes, undeclared drugs — while others only record what could not be proven about the process. The median reason text is 60 characters long. The top five texts cover 37.8% of the sample. And the biggest group of all, "lack of assurance of sterility", is a documentation claim, not a finding. Which kind you are reading decides what the recall means — and the dosage form decides which kind you get.

The short version

In a sample of the 600 most recent FDA drug recall entries (October 2025 to October 2026), the reasons group like this:

  • Sterility or microbial contamination: 159 (26.5%)
  • Manufacturing deviation with no specific defect named: 114 (19.0%)
  • Foreign particles or visible contaminants: 94 (15.7%)
  • Potency or dosage accuracy: 70 (11.7%)

The sample is recent, not the full FDA history, and the groups are our keyword classification of the free-text reason.

What are the reasons, in one table?

Reason group Entries Share Class I
Sterility / microbial contamination 159 26.5% 10
cGMP deviation, no defect named 114 19.0% 0
Foreign particles / visible contaminants 94 15.7% 15
Potency / dosage accuracy 70 11.7% 1
Impurities / chemical contamination 61 10.2% 1
Stability / dissolution / storage 57 9.5% 0
Container / closure / labeling 32 5.3% 2
Unapproved / undeclared ingredients 13 2.2% 4

Why do sterility and particles lead the list?

Reasons for 600 recent FDA drug recalls: sterility 159, cGMP 114, particles 94, potency 70, impurities 61, stability 57, container or labeling 32, unapproved 13

Sterility is the largest single group. Most of its rows use the same phrase, "lack of assurance of sterility", which means the firm could not show a product was free of microbes. It is a test-and-process claim. The product may be fine, but the firm cannot prove it.

Particles are a different kind of claim. The reason names something that was found in the product: glass, metal, fibre, plastic, or hair. Ninety-four of the 600 entries name a particle or visible foreign substance. Particles are also the largest Class I group in the sample: 15 of the 33 Class I entries name one.

Read together, these two groups make up 253 of 600 entries, or 42.2%. For a reader asking why a drug was pulled, the first answer in most cases is not "it did not work". It is "it may not be sterile" or "something was in it".

What about potency, impurities and stability?

Potency or dosage accuracy covers entries such as "subpotent drug" and "failed content uniformity". There were 70 of them. These recalls mean the amount of active ingredient in a tablet or vial is outside the labelled range. The patient may get less drug than the label says, or in a few cases more.

Impurities are a smaller group, 61 entries. Several name an N-nitroso compound above the acceptable intake limit. The reason text names the chemical, but the sample does not show how the impurity formed, and it cannot show whether this is a trend.

Stability and dissolution failures add 57 entries: products that test out of specification before expiry, or fail to release their drug at the expected rate. They are less visible than contamination, but with potency and impurities they make up about a third of the sample — failures a patient could not see or taste.

Why does one in five reasons name no defect?

About one in five entries, 114 of 600, says only "CGMP deviations" or a similar phrase about manufacturing practice. Some of those rows add a line about FDA inspection findings. None of them names the defect in the reason field.

For a buyer or an analyst, this is a real limit. A recall with a generic reason can mean a sterility failure, a labelling problem or a record-keeping gap. The field does not say which. You need the product description or the firm's letter to know.

Six hundred entries carry only 181 distinct reason texts once case and trailing punctuation are normalised. The top five texts alone account for 37.8% of the sample:

  • "Lack of assurance of sterility": 77 entries
  • "CGMP deviations": 68 entries
  • "Subpotent drug": 41 entries
  • "Presence of foreign substance: small metallic particles in chewable tablets": 23 entries
  • "Failed dissolution specifications": 18 entries

The median reason is 60 characters long. Two hundred and ten entries use 30 characters or fewer. Only 129 entries run past 120 characters, and those are the ones that describe what was found, where, or under what inspection finding.

This matters for anyone counting. Forty-one entries share one four-word text because they are the same kind of failure, not because one firm dominated. Counting reason text is a legitimate way to build a defect taxonomy, but only after the text is normalised. Counting raw strings gives you 181 "categories" and no picture.

Within the sterility group the same pattern is sharper: 142 of 159 entries lead with the bare "lack of assurance of sterility" phrase, and 87 entries in that group are 45 characters or shorter. The field tells you the category of failure and then stops.

Why is sterility so common? The dosage form decides

Sterility is only a requirement for products that enter the body without a further barrier: injections, infusions, eye drops, and other sterile liquids. That explains both the size of the group and who is in it.

Of the 159 sterility entries, 143 (89.9%) name a dosage form in the product description that is injectable or liquid, such as a vial, a prefilled syringe, an infusion bag or eye drops. The same keyword test catches 283 of the whole 600 entries, so liquid and injectable forms are less than half the sample but close to all of the sterility group.

The pattern repeats in Class I. Most of the 15 particle recalls are also injections or infusions: saline bags, imaging contrast, chemotherapy preparations. A particle in a tablet is a quality problem. A particle in an infusion that goes straight into a vein is a different class of problem, and FDA classifies it that way.

This is the closest the sample comes to an explanation. It is a structural one, about which products can fail a sterility test at all, not about which firms run worse plants. The sample has no field for plant, line, or inspection history, so it cannot support the second kind of claim.

How serious are these recalls? Mostly Class II

The FDA classifies each recall by risk. In the sample, 520 of 600 entries (86.7%) are Class II, meaning the use of the product "may cause temporary or medically reversible adverse health consequences" or the probability of serious harm is remote. Only 33 entries (5.5%) are Class I, which the FDA defines as a reasonable probability of serious adverse health consequences or death. Class III is 46 entries (7.7%).

Class and reason are different questions. Sterility entries are mostly Class II (148 of 159), with 10 Class I. Particle entries are 15 Class I out of 94 (16.0%), the largest Class I count of any group. Unapproved or undeclared ingredients have a higher Class I share (4 of 13, 30.8%), but 13 entries is too few to draw a rule from. So the reason does not tell you the risk level. Read the class.

How the sample was pulled

The recall data is openFDA's drug enforcement endpoint. No API key is needed:

# the 600 most recent entries by report date
curl "https://api.fda.gov/drug/enforcement.json?limit=600&sort=report_date:desc"

# or only Class I, newest first
curl "https://api.fda.gov/drug/enforcement.json?search=classification:%22Class%20I%22&limit=100&sort=report_date:desc"
Enter fullscreen mode Exit fullscreen mode

Two caveats if you rerun this. The sample in this article is a snapshot from 4 November 2026, and the window moves as new entries are posted — rerunning the same query today returns a different, more recent 600. And the reason groups here are a keyword classification of the free-text reason_for_recall field: 600 entries collapse to 181 distinct texts only after normalising case and trailing punctuation, and the grouping is ours, not an FDA category. We read the Class I slice of the same sample in a companion post.

What this data cannot answer

  • Whether this is the full history. The sample is the 600 most recent entries by report date, out of roughly 18,000. It shows the current mix, not the long-term mix.
  • Why contamination happened. The reason text records what was found. It does not record the root cause, such as equipment, supplier or process.
  • How many patients were affected. The entries have no patient-count field.
  • Which products are most at risk. Recall counts show what was caught and reported, not the rate of defects in the market.
  • Whether recall volume is rising or falling. Nearly all of the sample falls in 2026, so no year-on-year trend is claimed.

If you are holding a lot number, the fields matter in this order: classification gives the risk level and is the only field FDA assigns; reason_for_recall gives the category of failure, and only past about 120 characters does it name the substance or the finding; product_description tells you whether your dosage form and strength are covered; distribution_pattern says whether the lot went nationwide; status separates an open recall from a terminated one. Readers usually start with the reason, which is the least specific field — a generic reason with a Class II code is a different situation from a named particle with a Class I code.

Frequently asked questions

Q. Why was my medicine recalled?

A. The FDA record gives a reason for each recall. In the sample, the most common are sterility problems (26.5%) and foreign particles (15.7%). Check the product description and the reason field for your lot.

Q. Does a recall mean the drug is dangerous?

A. Not always. In this sample 86.7% of recalls are Class II, which the FDA says may cause temporary or medically reversible harm. Only 5.5% are Class I, which means a reasonable probability of serious harm or death.

Q. Are most recalls for contamination?

A. About 42% of the sample is sterility or foreign-particle reasons combined. The rest are potency, impurity, stability, packaging, and generic manufacturing reasons.

Q. Which products are most often recalled for sterility?

A. Injections and other sterile liquids. Of the 159 sterility entries, 143 (89.9%) name an injectable or liquid dosage form, such as a vial, a prefilled syringe, an infusion bag or eye drops.

Data notes

  • Source: FDA Drug Recalls endpoint, accessed through the agentdata SDK on 4 November 2026. Fields used: recalled_on, classification, status, reason_for_recall, distribution_pattern.
  • Sample: the 600 most recent entries by report date, not a random sample. Dates run from 8 October 2025 to 13 October 2026. 592 of 600 are dated 2026.
  • Classification of reasons: derived by keyword rules applied to the free-text reason, first match wins. Spot checks on five random entries per group read correctly after two fixes. The groups are our classification, not an FDA category.
  • Class: taken directly from the classification field. Class definitions are from FDA's recall background page.
  • Derived figures: all shares and the 42.2% combined share are computed from the 600-row sample.
  • Excluded: entries are not named by firm in this article. The firm field is not used for any statement here.

What we're building on this endpoint. I'm with dicaria — a data API for AI agents, currently in beta. The FDA recall endpoint above is one of the sources we serve: same records, with the reason groups, class and status normalized into filterable fields, and every row carrying its official recall number, so an agent can cite what it returns. If that's your stack, the quickstart is two steps — register for a key, then run one MCP command.

No API key is needed to reproduce anything in this post: the raw endpoint is free and public, and the queries above are complete.

Originally published on dicaria.com.

Top comments (0)