Formal Correction of Pivotal Trial Publications Following FDA Safety Drug Withdrawals: A Systematic Case Series (1990 to 2024)

Technical Summary: Formal Correction of Pivotal Trial Publications Following FDA Safety Drug Withdrawals

Problem Statement
When the FDA withdraws a drug for safety reasons, the pivotal randomized clinical trials that supported its original approval often remain in the peer-reviewed literature without formal correction. While regulatory agencies revoke approvals and clinical guidelines update recommendations, it is unclear whether the specific publications serving as the evidentiary basis for those approvals receive machine-discoverable annotations (retractions, expressions of concern, or corrections) alerting readers to the drug's altered status. The peer-reviewed literature's correction mechanisms are designed to address fraud or error within a paper's own claims, not to annotate papers whose clinical utility has been overtaken by later regulatory actions based on data the original trials did not contain. This study investigates the frequency with which such formal corrections occur.

Methodology
The study employed a systematic case series design analyzing US drugs withdrawn for safety reasons between 1990 and 2024.

• Cohort Derivation: The primary cohort consisted of 29 drugs meeting strict inclusion criteria (US market withdrawal for safety, 1990–2024), derived from the Onakpoya et al. registry and cross-referenced with FDA Federal Register notices. Two border-case drugs (LAAM/ORLAAM and nefazodone) were held out for sensitivity analysis, and four reintroduced drugs were tagged separately.
• Pivotal Paper Identification: For each drug, pivotal trial publications were classified into three tiers:
  • Tier 1: Explicitly cited in FDA medical reviews or approval documents.
  • Tier 2: Identified via authoritative retrospective reviews (necessary for pre-1995 approvals with incomplete FDA archives).
  • Tier 3: Drugs where pivotal trials were never substantially published in peer-reviewed literature.
• Correction Assessment: Correction status was determined by manual inspection of live PubMed records, Crossref metadata, the Retraction Watch database, and Google Scholar. The primary endpoint was a "safety-related paper-level action" (retraction, expression of concern, or correction/erratum specifically addressing safety, efficacy, or risk-benefit evidence). Secondary endpoints included any formal action regardless of content and any machine-discoverable withdrawal-status annotation.
• Pattern Typology: Withdrawals were classified by the relationship between the withdrawal reason and the pivotal paper's claims:
  • Pattern A: Direct contamination of the pivotal paper (misrepresentation/omission of safety data).
  • Pattern B: Failure of a surrogate (B1) or non-replication of a clinical outcome (B2).
  • Pattern C: Harm in a domain the pivotal trial was not designed to measure.
• Audit: A pre-specified audit of 36 paper pairs (38% of the cohort) was conducted to verify correction status and classification accuracy, with a re-audit trigger set at >5% error.

Key Results

• Correction Rate: Across the primary cohort of 29 drugs, 63 indexed pivotal efficacy papers were identified. Only one paper received a safety-related paper-level action: an Expression of Concern issued for the VIGOR trial (rofecoxib) in 2005. This represents a rate of approximately 1.6% (1/63).
• Retractions and Annotations: There were zero retractions of pivotal papers. Furthermore, zero pivotal papers carried a machine-discoverable annotation indicating the drug had been withdrawn, suspended, or reintroduced.
• Pattern Distribution:
  • Pattern C (harm in an unmeasured domain) accounted for 22 of the 29 withdrawals.
  • Pattern A (direct paper contamination) accounted for only one case (rofecoxib), which was the sole instance of formal correction.
  • Pattern B (surrogate failure or non-replication) accounted for the remaining cases, including the PROWESS trial (drotrecogin alfa) and CAST trials (encainide), none of which received formal corrections despite the failure of their clinical rationale.
• Audit Findings: The audit confirmed zero discordance on the primary endpoint. The single discrepancy found (a routine erratum for troglitazone) affected the secondary endpoint only, not the primary safety-related endpoint.

Key Contributions

1. Quantification of the Gap: The study provides the first systematic measurement of the rate of formal correction for pivotal trials following FDA safety withdrawals, establishing a baseline of ~1.6%.
2. Structural Typology: It introduces a pattern typology (A, B1, B2, C) that explains why corrections are rare. The data suggests that most withdrawals (Patterns B and C) arise from evidence that does not constitute a "misrepresentation" of the original paper's specific claims, thereby falling outside the scope of standard correction mechanisms.
3. Infrastructure Analysis: It demonstrates that the peer-reviewed literature lacks a durable, machine-discoverable signal of post-approval drug withdrawal. Downstream consumers (clinicians, systematic reviewers, AI tools) relying on standard bibliographic metadata encounter no indication that a drug's pivotal evidence has been superseded by a safety withdrawal.
4. Case Illustrations: The paper details specific high-profile cases (e.g., PROWESS, CAST, Natalizumab) where the clinical rationale was overturned or the drug was suspended, yet the original pivotal papers remained uncorrected and unannotated in the literature.

Significance and Claims
The paper concludes that the lack of formal correction is not primarily a failure of the correction infrastructure or peer review, but a structural mismatch between the mechanisms of drug withdrawal and scientific correction.

• Structural Mismatch: Withdrawals are typically triggered by harms in domains the pivotal trials were not designed to measure (Pattern C) or by the failure of surrogate endpoints (Pattern B). Formal correction, however, requires demonstrated misrepresentation of the paper's own claims. Consequently, the correction mechanism "does not attach" to the vast majority of withdrawal events.
• Institutional Gap: The study identifies a gap in the institutional division of labor: journals are not surveillance systems, regulators issue withdrawal notices rather than paper annotations, and pharmacovigilance databases record adverse events rather than bibliographic signals. No single actor is currently responsible for annotating the peer-reviewed record when a drug's regulatory status changes.
• Implication: The peer-reviewed medical literature currently carries no machine-discoverable signal of post-approval drug withdrawal. While the gap is technically closable (e.g., via structured regulatory-status fields in metadata), the study posits that whether such a mechanism should be built is a normative and infrastructural question beyond the scope of this descriptive analysis. The findings suggest that reliance on the "correction" mechanism alone is insufficient for maintaining an accurate record of drug safety post-approval.