“A sample of other CGMP records” — how FDA sizes a retrospective review
483 response filed 15 December 2025 · Citations 21 CFR 211.160(a), 211.192, 211.113(b), 211.42(c)(10)
A sterile drug manufacturer in Greece was inspected in November 2025. Investigators found microbial plates incubating without the raw data, sample preparation records, or batch identification that give a plate its meaning. The firm went looking on its own, found fourteen more batches in the same condition, and invalidated the sterility tests for all sixteen.
That is a firm doing more than it was asked to do. FDA found the response inadequate anyway. A retrospective review is not a sampling exercise: its size is set by how far the questionable practice reached, not by what a statistician would call a representative draw.
Earlier issues dealt with what a response says about a finding. This one is about what it claims not to need to look at.
What FDA found
The observation is small and entirely ordinary. Microbial plates — sterility, environmental monitoring, water samples — sat in the incubators without the records that connect a result to a batch. FDA is specific: the firm “did not have raw data and information related to sample preparation and the batch/sample identification number available for review,” and “did not document the associated data and information contemporaneously.”
A second thread runs alongside it. Critical CGMP forms used to record test results — the sterility test report among them — were printed from the firm’s IQVIA software and were not adequately controlled.
The rest of the letter establishes why that matters here rather than somewhere less consequential. Smoke studies showed air bouncing off an operator’s chest and re-entering the filling line; gram-negative organisms appeared repeatedly in ISO 5 air samples; the filling rooms held lower pressure than the rooms next to them.
- 10 – 21 Nov 2025
- FDA inspects the Rodopi facility. Plates found incubating without contemporaneous records.
- Nov 2025
- Firm suspends manufacturing at the site.
- 15 Dec 2025
- 483 response. Worksheet review finds 14 further batches; all 16 sterility tests invalidated.
- 23 Apr 2026
- Import Alert 66-40.
- 27 May 2026
- Warning letter. Every response addressed is found inadequate.
Set against a five-year history of sterility and media-fill failures, a sterility record is not a paperwork question. It is the only evidence the batch has.
What the firm said
The response reads as a model of self-disclosure, and deserves to be described that way. The firm went back through its in-process microbiological worksheets, found the problem was larger than investigators had seen — fourteen further batches — invalidated the tests for all sixteen, then proposed independent review of everything else.
“You also commit to engaging an independent third-party consultant to review the integrity of data by conducting a complete review of all batch release records and a sample of other CGMP records using statistical methods.” FDA’s summary of the firm’s 483 response, Warning Letter 320-26-80
Read that as a quality professional and it is a defensible plan: complete coverage where risk concentrates, valid sampling everywhere else. FDA rejected it.
Why FDA rejected it
The scope was bounded by document type, not by practice
“Your response is limited to discarding sterility tests you conducted during our inspection and that were found to have inadequate CGMP documentation.” The further fourteen came from searching the same worksheets for the same gap. Nothing asked when uncontrolled printing of test forms began, or which other microbiological activities were documented that way.
Invalidating the test removed the evidence, not the question
Sixteen results were withdrawn. The batches they released still exist, their sterility now unestablished rather than unfavourable, and the response does not say what follows for product in distribution. FDA’s phrasing points past the documents: the full scope of the impact on microbiological test reliability.
“You did not holistically assess microbiological laboratory practices to determine the full scope of the impact on microbiological test reliability. Additionally, your proposal for a third-party data integrity review using a statistical sampling method is inadequate, as the degree of retrospective review is not sufficient given the significance of the questionable practices documented during the inspection.” FDA, Warning Letter 320-26-80
A sample estimates a rate; a review has to find a boundary
Sampling works when you are characterising a population you believe behaves consistently. A retrospective review exists because that belief has been withdrawn. The operative clause is given the significance of the questionable practices: depth is set by the seriousness of the practice, not by what a sampling plan would consider sufficient.
FDA’s counter-scope is one word long
The requested action asks for assessment of documentation systems used throughout manufacturing and laboratory operations — not a representative subset. Elsewhere in the letter, proposed CAPAs are rejected because they “lack sufficient detail and a description of how effectiveness of these actions will be evaluated.”
The pattern worth carrying forward
Sampling is the instinct the profession trains hardest, and it is correct almost everywhere. The exception is narrow: once a control failure has called a record-keeping practice into question, the population you would sample from is the thing in dispute. A percentage drawn from it inherits the doubt instead of resolving it.
The second half of this response is the more expensive habit. Discarding an unreliable result is correct and insufficient — the record goes away, the disposition it supported does not.
- What is the practice, and when did it start? The boundary of the review is the boundary of the practice — not of what the investigator happened to open. If you cannot date the start, it runs back to the earliest record the practice could have touched.
- Are we estimating a rate, or finding an edge? Sampling answers the first well and the second not at all. If what FDA needs to know is how far this reached, a percentage reads as avoidance however sound the statistics are.
- If the record is unreliable, what happens to the batch it released? Invalidating a test removes the evidence, not the product. Name the affected lots, their distribution status, and what you are doing about them — before FDA has to ask.
A defensible response would have started with the practice rather than the plates: when uncontrolled forms entered use, and which tests they touched. Full coverage of that population, whatever its size, then a lot-by-lot statement of what sixteen withdrawn results mean for product in commerce.
A sample tells you how often. After a documentation failure, FDA is asking how far.
Veritas Quality Consultants works with pharmaceutical, medical device, and biologics manufacturers on Form 483 responses, root cause analysis, and CAPA development — built to FDA expectations inside the 15-day window.
Request a consultation