Version 1.1 (2026-05-29): Corrected a typographical error in the author ORCID iD shown in the manuscript PDF (now 0009-0002-6226-440X). No changes to data, methods, results, or conclusions. This study quantifies the prevalence and predictors of primary outcome discrepancies between clinical trial registrations and publications at unprecedented scale, analysing all 46,365 completed interventional trials with posted results on ClinicalTrials.gov (2015–2025). Key Findings 59.4% of trials (27,555/46,365) were matched to at least one publication Automated screening flagged 24.7% (3,353/13,595) of confirmed trial-publication pairs as possible discrepancies Manual validation of 123 pairs found 6 true discrepancies (4.9%; 95% CI 1.6–8.9%) Discrepancies were predominantly minor modifications, not substantive outcome switching No association between primary outcome statistical significance and discrepancy flagging (21.0% vs 20.6%; p=0.745) Phase 1 trials had the highest automated flag rate (47.4%), Phase 4 the lowest (15.6%) Implications: Primary outcome discrepancies affect approximately 5% of clinical trials and are predominantly minor rather than substantive. Critically, discrepancy rates are not associated with result significance, suggesting that outcome switching is not systematically driven by unfavourable results at the population level. Automated screening can efficiently triage trial populations but lacks sensitivity for subtle discrepancies. Contents: Main manuscript (PDF and DOCX), supplementary materials (PDF), and high-resolution main figures (3 PNG).
Hayden Farquhar (2026) studied this question.