FDA Warning Letter Breakdown: Inside a GLP Collapse at a Chinese Testing Facility
Data integrity problems reveal what happens when compliance becomes theater and why the firm voluntarily suspended all FDA-related non-clinical testing through December 2027.
This breakdown is available for paid subscribers. Only paid subscribers get regular full access to our breakdowns and other analyses. If you’re not already a paid subscriber, you can upgrade here.
Want to stay out of our warning letter breakdowns? Contact us to access our global network of 3,250+ consultants and 250+ former FDA employees. We run audits, mock inspections, and remediation for 17 of the top 25 life science firms.
Today, we’re looking at what one might call a catastrophic GLP failure at a Chinese contract testing facility that performed safety testing for medical devices. The violations are so severe and systemic that the company voluntarily suspended all FDA-related testing for three years rather than attempt remediation.
On July 11, the FDA’s CDRH issued a warning letter following a Jan 15–24, 2025 inspection by the FDA’s Office of Bioresearch Monitoring (OBMI) Foreign Inspection Cadre. The facility conducted GLP nonclinical safety studies, including guinea pig maximization sensitization (GPMT), acute systemic toxicity (AST), rabbit pyrogen, muscle implantation, hemolysis, and cytotoxicity tests.
The inspection reviewed 67 studies, with 63 conducted between 2023 and 2024.
The violations revealed a testing operation where data integrity was essentially fictional—from retroactive data creation to physically impossible concurrent activities to animals that mysteriously never died in quarantine. The company's February 21, 2025 response acknowledged they couldn't sustain "the basic costs associated with full system compliance" and suspended operations through at least December 31, 2027.
Below, we've dissected the eight major violations to extract critical lessons for any facility conducting GLP studies. While this case is extreme, it still offers some valuable lessons for everyone in this space.
Study directors who weren't directing
The FDA's first violation targeted what reads to be a more or less complete abdication of study director responsibilities across multiple studies and directors. The agency describes fundamental failures to ensure protocols were followed, data were accurate, and GLP regulations were met.
Maybe the most damning finding here: FDA investigators directly observed study personnel retroactively completing records for an ongoing subacute systemic toxicity study, documenting "clinical observations and animal weight data from Day 1 of the study up to Day 26" all at once.
When asked for proof that animals were actually weighed, the records showed only 5 out of 20 weights for week 2 and zero weights for week 3.
Even more troubling were the physical impossibilities in the data. The same person reportedly "shaved all three rabbits for study [redacted] and weighed all three rabbits for study [redacted] at exactly 17:19" on the same date. Two different people were documented using the same scale at overlapping times to weigh different animals.
The firm’s SOP required daily animal observations, but no daily observation raw data existed in the study records. Twelve intracutaneous reactivity studies used previously used rabbits despite protocols requiring "only healthy, previously unused animals." Guinea pigs were started on studies after only 3 days of acclimatization, when the SOP required at least 5 days.
Ask yourself:
Can our study directors physically access and review raw data in real-time, or only at study completion?
Do we flag biologically impossible data entries (same person, multiple locations)?
When we say "previously unused animals," do we have a system that actually prevents reuse?
Test article handling: when identity becomes fiction
Another area of deficiency exposed a breakdown in test and control article handling.
The facility had no documentation of lot/batch number verification during sample receipt or preparation.
Sample quantities were recorded in meaningless units like "3 packs" with no indication of actual needle counts.
After study completion, study directors routinely changed test article information in final reports at the sponsors' request. Twelve studies showed post-hoc changes to lot numbers, materials, and sizes, with some even swapping out test article photos. This suggests test article identity wasn't maintained—it was retroactively invented.
Ask yourself:
Could someone change test article information in a final report without triggering a deviation or investigation?
Do our sample prep forms require mandatory lot number verification fields that can't be bypassed?
If asked to prove what exact article was tested 6 months ago, could we do it?
An ineffective Quality Assurance Unit
The Quality Assurance Unit's failures were breathtaking in scope. It failed to identify any of the data irregularities FDA found—the impossible timestamps, the retroactive data creation, the protocol violations.
This wasn't an inadequate QAU function; it was functionally a non-existent QAU.
Of 63 studies from 2023-2024, only 3 had records of raw data audits despite final reports claiming QAU reviewed everything.
Thirty studies had no record of the reported final report audits.
One study still in draft form already included a signed QAU statement claiming audits were complete on specific dates—but when the FDA requested those audit reports, they could only produce one protocol audit with a different date than claimed.
Ask yourself:
If the FDA asked for proof of a specific QAU audit mentioned in a final report, could we produce it immediately?
Does our QAU actually look at raw data, or just check that forms are filled out?
Can a final report be issued with QAU signatures before QAU audits are actually performed?
Two sets of (conflicting) SOPs
The facility operated under two conflicting sets of SOPs—"domestic" procedures and "GLP" procedures—with no guidance on which to use. The domestic SOPs required four separate forms for clinical observations. The GLP SOPs required only one. Neither set was adequate.
The GPMT SOP lacked photos demonstrating Magnusson Kligman scoring. The abnormal conditions SOP provided no species-specific guidance despite testing on mice, rats, and rabbits. SOPs didn't include critical safety instructions like ensuring animals' breathing wasn't restricted during occlusive dressing application.
Multiple studies showed deviations from SOPs that weren't authorized by study directors or documented in raw data. The facility simply picked whichever SOP was convenient or ignored both.
Ask yourself:
Do we have any conflicting or duplicate SOPs that could create confusion about requirements?
Would a new technician be able to correctly perform a procedure using only our SOP?
Animal isolation and a mathematical impossibility
The FDA found the facility's animal quarantine claims defied both biology and mathematics.
Over five months, they received some number of guinea pigs with 100% recorded as "normal" with zero quarantine deaths. During the inspection, 6 guinea pigs died in 2 days.
They also received 180 rabbits on one date for quarantine, but their total rabbit quarantine capacity across all rooms was only 141.
The math simply didn't work.
Ask yourself:
Do our animal receipt records reflect biological reality (some animals arrive sick)?
Before accepting a shipment, do we verify we have physical space for proper quarantine?
Would our quarantine records withstand basic mathematical scrutiny?
Every animal is #1
The facility's approach to animal identification made mix-ups inevitable. Control and test animals used identical ID ranges reused across multiple concurrent studies, creating inevitable mix-ups and traceability gaps.
Records showed "guinea pigs with identical IDs from multiple studies concurrently, in the same room, and on the same scale."
There was no way to ensure the right animal received the right treatment or that results were attributed correctly.
Ask yourself:
Is it physically possible in our system for two animals to have the same ID at the same time?