The pharmaceutical world is currently reeling from the news that the FDA’s Center for Drug Evaluation and Research (CDER) has proposed a full market withdrawal of Tavneos (avacopan). The drug manufacturer, Amgen, is embroiled with the FDA in the dispute about the withdrawal of Tavneos.
- FDA alleges ‘manipulated’ data supported approval of Amgen’s autoimmune drug
- The FDA Cracks Down on Tavneos
- Amgen Updates Tavneos Label While US FDA Weighs Withdrawal Proposal
While drug recalls for safety are not uncommon, this case is unique: the FDA alleges that the drug’s original developer, ChemoCentryx, engaged in deliberate data manipulation to secure its 2021 approval. Tavneos was approved in 2021 to treat a rare autoimmune disease called ANCA-associated vasculitis. Amgen scooped it up the following year in its $4 billion acquisition of ChemoCentryx.
The "how" behind this manipulation is a masterclass in statistical engineering and the violation of clinical trial integrity. Based on the internal "Walton Report" and subsequent investigations, here is exactly how the results of the pivotal ADVOCATE trial were doctored.
The 15-Day Manipulation Window
The integrity of a clinical trial rests on the "blind"—the rule that researchers don't know who is getting the drug and who is getting the placebo until the results are finalized. In a standard clinical trial, once the database is locked, the assignments are unblinded, and the statistical analysis begins. Any changes after this point are typically forbidden unless they involve the correction of clear, documented clerical errors identified through blinded quality control.
In the ADVOCATE trial, the data was officially locked on November 5, 2019 and sponsor (ChemoCentryx at that time) performed two step processes for the database lock (i.e., the initial lock and the database relock – final lock). The misconduct to manipulate the data occurred during the initial database lock and the database relock. The timeline of the alleged misconduct is listed below.
An initial analysis conducted immediately after this lock showed a devastating result: Tavneos failed to meet its primary endpoint of superiority for sustained remission at 52 weeks. According to the FDA, this was a "not statistically significant" result that should have ended the drug's path to approval. The sponsor then ‘manipulated’ the data following the steps below:
Step 1: Breaking the Blind
On November 6, 2019, two high-ranking officials—Director of Biostatistics Mr. Yue and Chief Medical Officer Dr. Bekker—were granted access to the unblinded data. Instead of accepting the failed result, the FDA alleges these individuals specifically hunted for patient cases that could be altered to flip the p-value into the significant range.
Step 2: The "Hypothetical Dataset"
Before making any official changes, Yue and Bekker reportedly created a "hypothetical dataset". This was essentially a simulation to see exactly how many patients needed their status changed from "non-responder" to "responder" to achieve statistical superiority. They identified nine specific subjects for "readjudication"—six in the Tavneos arm and three in the control arm.
Step 3: Targeted Reclassification
The manipulation targeted patients where protocol rules regarding "missing data" or "glucocorticoid use" could be creatively reinterpreted.
In total, five of the six targeted Tavneos patients were successfully changed to "responders". Crucially, not a single patient in the control arm was reclassified. This asymmetric adjustment pushed the p-value to p < 0.001, creating the illusion of a highly successful drug.
Step 4: The Omission of Truth
When the New Drug Application (NDA) was submitted to the FDA in 2020, ChemoCentryx included only the second, manipulated analysis. They failed to disclose that an original, failing analysis had ever existed or that unblinded personnel had "corrected" the data after the fact. This is why the FDA now cites "untrue statements of material fact" as a primary reason for withdrawal.
The Human Cost
This wasn't just a victimless statistical crime. By manufacturing proof of effectiveness, the developers downplayed the drug's severe risks. Since its approval, the FDA has flagged 76 cases of drug-induced liver injury (DILI) causally linked to Tavneos, including 8 deaths. Some patients developed Vanishing Bile Duct Syndrome (VBDS), a rare condition that progressively destroys the liver’s bile ducts.
Current Status
Amgen, which acquired ChemoCentryx for $3.7 billion in 2022, maintains that it is not aware of any issues with the underlying patient data and stands by the drug's safety profile. However, the FDA has issued a Notice of Opportunity for a Hearing (NOOH), a formal step toward stripping the drug of its marketing rights.
Until a final ruling is made, Tavneos remains on the market, leaving thousands of patients and doctors in a difficult position: weighing a "breakthrough" that may have been a statistical mirage against the risk of fatal liver failure.
Conclusion
The basic principles in the randomized, double blinded clinical trial prohibit the data changes after the study is unblinded. The study unblinding must occur after the database is locked.
Two step database lock process is flawed and should be abandoned.
Any data review and endpoint adjudication should be conducted in the blinded fashion and must be completed before the database lock/study unblinding.
Any violation of these principles will cause the suspicion of the data manipulation and the data integrity is compromised.