Who Marks the Trial Record When a Drug Is Withdrawn? Regulatory-Status Metadata on Pivotal Publications for US Safety Withdrawals, 1990 to 2024

<title>Abstract</title> <p> <bold>Background:</bold> When the FDA withdraws a drug for safety reasons, the clinical trials that supported its approval remain in the peer-reviewed literature. Clinicians, systematic reviewers, guideline developers, and automated evidence-synthesis tools retrieve those trials through standard bibliographic databases. We asked a question prior to any question of correction or blame: when the regulatory record and the bibliographic record are compared, does the pivotal trial record of a withdrawn drug carry any machine-discoverable signal of the withdrawal, or are the two records discordant? A machine-discoverable signal is a formally linked or structured notice returned with the paper’s record, such as a retraction, an expression of concern, or a correction. <bold>Methods:</bold> We assembled a cohort of US drugs withdrawn for safety reasons between 1990 and 2024 using the Onakpoya withdrawal registry, FDA Federal Register notices, and confirmatory regulatory sources. For each drug, pivotal trial publications were classified into three identification tiers: FDA-confirmed (Tier 1), retrospective-review-inferred (Tier 2), and no identifiable published pivotal paper (Tier 3). Correction status for each pivotal paper was determined by manual inspection of the live PubMed record, with Crossref, the Retraction Watch database, and Google Scholar as parallel checks. The primary endpoint was a safety-related paper-level action; two secondary endpoints captured any formal action regardless of content, and any machine-discoverable withdrawal-status annotation. Each drug's withdrawal was classified with a pattern typology distinguishing contamination of the pivotal paper itself (Pattern A), failure of a surrogate or clinical-outcome claim on re-examination (Pattern B), and harm in a domain the pivotal trial was not designed to measure (Pattern C). <bold>Results:</bold> The primary cohort comprised 29 drugs spanning three decades, seventeen therapeutic classes, and five regulatory pathways, yielding 63 indexed pivotal efficacy papers. One paper, the VIGOR trial of rofecoxib, received an Expression of Concern, the only action meeting the primary endpoint, a rate of one in 63 (approximately 1.6 percent), with zero retractions and no withdrawal-status annotation on any pivotal paper. Most withdrawals, 22 of 29, were Pattern C. In two cases the approval-relevant interpretation of the pivotal evidence was later reversed or no longer sustained, encainide and drotrecogin alfa, yet no signal attached to the trial record. <bold>Conclusions:</bold> The peer-reviewed record carried no structured withdrawal-status annotation on any pivotal paper, and only one safety-related formal action among 63. The regulatory record and the bibliographic record are therefore discordant: the regulatory record documents withdrawal, while the pivotal-paper record carries no corresponding status signal, and the two are not linked. A reader who retrieves the pivotal trial of a withdrawn drug finds the paper unchanged, with no marker of the withdrawal. This is not an editorial failure. We found no routinely implemented mechanism that placed withdrawal-status metadata on the records examined, and responsibility for placing one appears fragmented across journals, regulators, and bibliographic services. The low correction rate is the measurement behind this finding, not a claim that correction was owed. The gap now matters more, because automated systems that rely principally on indexed article-level metadata may not detect a withdrawal that is absent from that metadata. </p>