The FDA has the antidepressant trial data. It won't publish it.

On 8 July 2026, an X post by the user @cremieuxrecueil summarised an update to a previously published study: older antidepressants' apparent effect sizes had been inflated by the systematic exclusion of weak, unpublished results from the published literature. The agency that holds the underlying trial data — the United States Food and Drug Administration — does not publish it. The gap between what the regulator can see and what the public, the prescribing physician, and the patient can see, is the gap the post is pointing at.

That gap is not new. What is newly legible, courtesy of the post, is the specific mechanism: studies that fail to show a clear effect tend not to make it into journals, while studies that do make it into journals tend to be the ones the regulator has already accepted as evidence of efficacy. The published literature, in other words, is filtered twice — once by the sponsor's decision to submit, and once by the journal's appetite for positive findings. The FDA sits above both filters with the raw data, but is under no statutory obligation to put it on the public record.

What the underlying review actually showed

The original review, by Kirsch and colleagues and circulated widely under the shorthand of the Kirsch analysis, compiled the data submitted to the FDA for the six most widely prescribed antidepressants approved between 1987 and 1999. Because the FDA required sponsors to submit all trials conducted for an indication — not just the published ones — the compiled dataset gave researchers an unusually complete picture. The published literature, by contrast, showed effect sizes roughly a third larger than the full FDA dataset suggested. The headline finding, often repeated and often overstated, was that the mean effect size of these antidepressants in patients with moderate-to-severe depression was below the threshold conventionally used to declare a drug "minimally effective."

The 2024 update referenced in the 8 July post tested whether publication bias had improved since the original review. The answer, in summary form, was partial: newer antidepressants showed smaller gaps between published and FDA-dataset effect sizes, suggesting sponsors had become less aggressive about burying negative results. But the underlying problem — the FDA holding data that the public cannot see, with no requirement to release it — had not been resolved. If anything, the update moved the question upstream: why is it left to academic reviewers, working years after the fact, to reconstruct a dataset the regulator has always had?

What the FDA does and does not publish

The FDA's Center for Drug Evaluation and Research publishes approval packages for drugs it greenlights. Those packages contain the medical and statistical reviews, the labelling, and a summary of the clinical trial evidence the agency relied on. What they do not routinely contain is the patient-level data underlying the trials — the kind of dataset a researcher would need to run an independent re-analysis, check a sponsor's coding decisions, or test a subgroup hypothesis.

Two pathways exist for outside researchers to get closer to that data. The first is the Freedom of Information Act request, which can produce documents but is slow, partial, and frequently redacted on commercial-confidentiality grounds. The second is the sponsor itself, which can choose — and increasingly, under pressure from journal editors and trial-registry rules, does choose — to make patient-level data available through platforms such as Vivli or the Yale University Open Data Access project. Neither pathway substitutes for a presumption of publication.

The European Medicines Agency, for contrast, took a different route in 2014 and again in 2016, adopting a policy that makes clinical-trial data submitted in support of marketing-authorisation applications publicly available after a regulator review process. The EMA's policy has its own limits — it covers applications received after a given date, and the review process can take months — but it places the default on disclosure rather than on request. The FDA has not moved in that direction.

The structural frame: a regulator that audits but does not broadcast

The pattern here is a familiar one in American regulatory architecture. The FDA's job, as Congress has defined it, is to make a binary decision: approve or do not approve. Once that decision is made, the agency treats the underlying data as belonging to the sponsor, with disclosure governed by the sponsor's discretion and by FOIA law rather than by a default of openness. The researcher who wants to replicate, the journalist who wants to verify, and the patient who wants to weigh benefits against harms all face the same downstream constraint: they can read the agency's summary of the evidence, but not the evidence itself.

This is not corruption. It is design. The FDA does not conceal the data so much as it declines to publish it, and the legal architecture that defines its discretion — the Federal Food, Drug, and Cosmetic Act, the FDA Amendments Act of 2007's trial-registration provisions, and the agency's own regulations — leaves that discretion broad. What the 8 July post surfaces, then, is less a scandal than an accountability gap: a public body that adjudicates on the basis of evidence it will not itself place in the public record.

What changes, what doesn't, and who has standing to push

Two reform paths are plausible. The first is legislative — a Congress sufficiently concerned about publication bias to require the FDA to publish patient-level data for approved drugs, with appropriate safeguards for personal health information. The history of similar proposals is not encouraging; pharmaceutical-industry opposition tends to be well organised and well funded, and trial-transparency legislation has stalled in successive Congresses. The second is regulatory — an FDA rule, issued under existing statutory authority, that places a presumption of disclosure on the data the agency has already reviewed.

The standing question is who pushes. Academic researchers have an interest but limited leverage. Patient-advocacy organisations have moral standing but uneven access to the technical arguments. The pharmaceutical industry has both the leverage and a strong interest in the status quo. The EMA's policy suggests that a regulator with the political cover to act can act unilaterally; whether a future FDA commissioner has both the will and the cover is a question the agency's recent leadership has not been eager to answer in public.

In the meantime, the specific complaint in the 8 July post — that the FDA has the data and does not publish it — remains accurate. It will remain accurate after this article publishes. And it will remain accurate as long as the regulator's default posture is to audit quietly rather than to publish, even when the audited record would, in many cases, vindicate the regulator's own decisions.

This article distinguishes itself from the wire's typical coverage of antidepressant efficacy by foregrounding the disclosure question rather than the efficacy question. The original Kirsch finding has been extensively reported; what is less commonly traced is the institutional reason the underlying data remains, by default, in the regulator's drawer.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://en.wikipedia.org/wiki/Kirsch_antidepressant_meta-analysis
• https://en.wikipedia.org/wiki/European_Medicines_Agency
• https://en.wikipedia.org/wiki/FDA_Amendments_Act_of_2007
• https://en.wikipedia.org/wiki/Freedom_of_Information_Act_(United_States)
• https://en.wikipedia.org/wiki/Center_for_Drug_Evaluation_and_Research