The Filtered Record

A drug sponsor submits the results of every trial in a new drug application to the FDA, positive or negative. The journal literature receives only the trials that authors submit and editors accept. A comparison of the two records measures what publication removed.

Turner's group could not determine whether sponsors, authors or journals were responsible for the missing publications. The comparison measured the completeness of the published record and did not test whether the drugs worked.

In 2022 the same group repeated the comparison for 30 trials of four newer antidepressants, involving 13,747 patients. All 15 trials the FDA judged positive were published transparently. Of the 15 negative trials, seven were published transparently, six went unpublished and two were presented as positive. Transparent publication of negative trials rose from 11 percent in the older cohort to 47 percent, and the effect-size gap narrowed to 0.24 in FDA data against 0.29 in journals. The 2022 study examined efficacy and did not test safety reporting.

The three counts used different observation periods and coding. The reanalysis covered the taper phase as well as the acute phase and included one possible case in each comparison group. The table counts patients with recorded behaviors; it does not count deaths. The reanalysis authors acknowledged that classifying adverse events requires judgment.

Keller and colleagues replied that they added the secondary measures before the blind was broken because the original depression scale measured adolescents' symptoms poorly, and they disputed the reanalysis's methods and its authors' impartiality.

On 30 September 2025 the Journal of the American Academy of Child and Adolescent Psychiatry published an expression of concern about the 2001 article online. The notice appeared in the June 2026 issue. An expression of concern alerts readers and leaves the article in the literature. The article had been in print for twenty-four years when the notice was attached.

Under the Final Rule of the FDA Amendments Act, the sponsor of a covered clinical trial must submit summary results to ClinicalTrials.gov by a set deadline.

FDA's register of notices of noncompliance with the results-reporting requirement lists Acceleron Pharma as its earliest entry, with a notice dated 27 April 2021 and a response dated 13 December 2021. The register lists further notices through October 2026 and records no penalty outcomes.

In the same cohort, industry sponsors had higher adjusted odds of reporting on time than non-industry, non-government sponsors (odds ratio 3.08, 95% CI 2.52–3.77).

Accelerated approval allows the FDA to approve a drug for a serious condition on a surrogate endpoint, such as tumor shrinkage, judged reasonably likely to predict clinical benefit. The sponsor must then run confirmatory trials. A failed confirmation can lead to withdrawal or a change to the label. Until confirmation, the drug is sold and prescribed on the surrogate evidence.

Liu, Kesselheim and Cliff followed 46 cancer drug–indication pairs granted accelerated approval from 2013 to 2017, with records through 2023. Twenty showed a survival or quality-of-life benefit. Nine converted to regular approval without showing either. Ten were withdrawn, and seven remained pending.

In 2025 the Inspector General reviewed 24 accelerated approvals, ten chosen because they raised concern and fourteen chosen at random. It found unplanned analyses, internal disagreements or incomplete meeting records in three: aducanumab, hydroxyprogesterone caproate and eteplirsen. The Inspector General stated that the review was not a compliance review. By 18 May 2026 the FDA had implemented the recommendation on meeting documentation and had not implemented the recommendation on criteria for referral to its review council.

The Consolidated Appropriations Act, 2023, gave the FDA authority to require a confirmatory trial to be underway before accelerated approval or within a set period after it. FDA's January 2025 guidance on that authority describes how the agency interprets "underway." The guidance is marked "Draft. Not for implementation. Contains non-binding recommendations."

In March 2015 the International Agency for Research on Cancer (IARC) classified glyphosate, the active ingredient in Monsanto's Roundup, as probably carcinogenic to humans. Chapter 5 records the regulatory and legal history that followed (§5.7).

Kroes died in 2006 and Munro in 2011. In 2017 New York Medical College told Science that it had found "no evidence" that Williams violated its ghostwriting policy.

EPA's 2020 glyphosate assessment relied on the review for its data and for its summaries of 16 other studies. The European Food Safety Authority cited it in 2023. According to Mother Jones, EPA knew of the Inspector General's finding and did not disclose it. Kirkland denies ghostwriting, and Bayer, Monsanto's owner, says the memo drew on limited documents. Taylor & Francis opened an investigation after Oreskes and Kaurov requested a retraction.

In 2016 the Brussels office of FleishmanHillard compiled stakeholder lists for Monsanto. Bayer's outside counsel, Sidley Austin, reported on 5 September 2019 that 14 list documents held 1,208 names, covering France, Germany, Italy, the Netherlands, Poland, Spain, the United Kingdom and EU-institution lists, and that 1,475 individuals had been notified. Sidley found no support for the allegation that the lists were illegal. According to press reports, a French list of about 200 people, including journalists and politicians such as Ségolène Royal, led Le Monde to file a complaint and a French prosecutor to open an inquiry in May 2019, and Bayer suspended the agency.

Chapter 3 records the Guinea-Bissau research on DTP vaccination and infant mortality (Finding 3.7). A 2026 article in Vaccine listed ten registered trials by the Bandim group whose primary outcomes were partly or wholly unpublished. One was a DTP-related trial reported to have enrolled 6,534 children.

Chapter 3 also records the Henry Ford Health study comparing vaccinated and unvaccinated children (Finding 3.5). The study was completed around 2020, was not submitted for publication, and entered the public record in September 2025 through a Senate subcommittee.

The filters described above operate at different points between a completed study and its reader: the decision to publish, the choice of outcomes to report, the deadline for posting results, the approval granted before confirmation, and the names on a review. The corrections came from an academic comparison with FDA files, an independent reanalysis, the HHS and EPA Inspectors General, documents produced in litigation, an outside audit of registered trials, and journals acting years or decades after publication.

Some filters have narrowed. Transparent publication of negative antidepressant trials rose from 11 to 47 percent in the newer cohort, the FDA has issued results-reporting notices since 2021, and Congress strengthened the confirmatory-trial requirement through the Consolidated Appropriations Act, 2023. In November 2026 a reader of the literature found the Study 329 article in print with an expression of concern, the Kier and Kirkland review under a publisher investigation, and the primary outcomes of ten Bandim trials partly or wholly unpublished.