On June 29, 2026, the New England Journal of Medicine retracted a paper it had published five years earlier. The notice runs three sentences. Here is the middle one: “without the knowledge of these two authors, the primary end-point assessments in nine patients were readjudicated after database lock and trial unblinding.”

Read it slowly, because every clause carries weight. The trial had ended. The database was locked. The blind was broken, so whoever changed the scores knew which patients had received the drug and which had received the comparator. Then, in nine patients, the primary outcome was scored again. None of that appeared in the published paper. And that paper was not one study among many. It was the study the FDA leaned on to approve the drug.

In more than two hundred years of publishing, the Journal had never retracted a registrational trial, the kind of study a drug approval is built on. This was the first. The FDA approved avacopan, branded Tavneos, in October 2021 on the strength of ADVOCATE’s two primary endpoints: that avacopan was non-inferior to a prednisone taper at inducing remission at week 26, and superior to it at sustaining remission at week 52. The week-52 superiority claim is what separated a steroid-sparing breakthrough from an expensive add-on riding on top of the same steroids everybody was already taking. It is also the claim the nine re-scored patients touched.

The margin was always thin

The people in the room in 2021 could see how narrow this was. When the FDA’s Arthritis Advisory Committee met that May, it split 9 to 9 on whether the efficacy data supported approval. A tie. The committee’s stated worry was the prednisone comparison itself: 86 percent of participants had received glucocorticoids outside the study’s design, which made it hard to argue that avacopan was replacing steroids rather than running alongside them. The panel cleared the drug on benefit-risk by a single vote, 10 to 8, and the agency followed.

How thin was the week-52 result once you looked under it? Rheumatologist Mike Putman, reading the FDA’s filing, reports that the endpoint came in at p = 0.1025 before the reassessment, short of statistical significance, and that a post-hoc look found flipping the outcomes of just 5 patients would have carried it across the line. Nine files were reopened. The published superiority claim is what came out the other side.

The FDA’s own conclusion is not hedged. The agency now states that it “can no longer conclude that there is, or has ever been, valid demonstration that Tavneos is effective.” Not that the evidence weakened. That there may never have been evidence at all.

Who did the scoring, and who owned the drug

This is where the documents allow names, and where they demand care. The paper carried four authors. Two were academics who led the trial, Professor David Jayne and Professor Peter Merkel; they are the two who requested the retraction and who, the notice says, did not know the reassessment had happened. The other two were employees of ChemoCentryx, the company that developed avacopan and ran ADVOCATE. Putman’s read of the FDA notice identifies the company’s chief medical officer, Dr. Pirow Bekker, and its director of biostatistics, Dr. Huibin Yue, as the individuals the agency ties to the readjudication and to what it characterizes as data manipulation, material omissions, and untrue statements.

Those are the FDA’s allegations, set out in a notice of opportunity for a hearing, not adjudicated findings, and the distinction matters. What is not in dispute, because the journal has now conceded it, is that the scoring changed after unblinding and that the change went unmentioned.

Follow the ownership. Avacopan was ChemoCentryx’s lead asset, and a year after approval, in August 2022, Amgen bought the company for $3.7 billion. It bought the drug, the ADVOCATE data, and the liability. By the time the questions surfaced, Tavneos was a revenue line worth defending: U.S. sales roughly doubled to $283 million in 2024, up 111 percent, and kept climbing in 2025. That is the sum now sitting on the other side of the FDA’s effectiveness finding.

TAVNEOS U.S. SALES, 2024
$283 millionup 111 percent
Amgen's revenue from the drug kept climbing even as the trial behind its approval came under fire. Source: Amgen Q4 2024 financial results

Europe moved. Washington is still arguing.

The regulators split on speed. On June 26, 2026, the European Medicines Agency’s CHMP recommended revoking Tavneos’ marketing authorisation outright, concluding that ADVOCATE “was conducted in breach of good clinical practice (GCP) principles” and that its data were “incorrect and misleading,” that no new patient should start the drug, and that existing patients should be switched to alternatives. The EMA also spelled out the safety side of the ledger the efficacy fight had overshadowed: Tavneos is associated with drug-induced liver injury and vanishing bile duct syndrome, “including cases with a fatal outcome.”

The FDA asked Amgen to pull the drug on January 16, 2026. Amgen refused. Chief executive Robert Bradway called Tavneos an “important and effective medicine” with a “favourable benefit-risk profile,” and the company tells prescribers it stays “confident in the overall benefit-risk profile” on the strength of nearly five years of use and more than 8,000 treated patients in the United States. Amgen has commissioned the Duke Clinical Research Institute to run a fresh, fully blinded readjudication of the endpoints, and has asked for a hearing rather than accept withdrawal. So a company is now paying an outside institute to re-score the outcomes of a trial whose last re-scoring is the reason the paper was retracted.

Here is who has acted and who has not. The Journal has retracted. Europe’s regulator has moved to revoke. The FDA has stated its conclusion and offered a hearing. The hearing has not happened. Until it does, Tavneos stays on U.S. shelves, prescribed for a serious autoimmune disease, on the strength of a trial its own publisher will no longer stand behind. The records that would show whether nine files were reopened to rescue a failing endpoint, or for some defensible reason, sit with the people who reopened them. The agency has scheduled the argument. It has not yet made anyone answer for it.

Sources