Background: Clinicians and systematic reviewers rely on the publication of all clinical trials, including trials that have been used by regulatory agencies in assessing the quality and safety of a drug. However, not all trials submitted to regulatory agencies are published.
Objectives: To identify and characterize discrepancies, if any, between clinical trial data submitted to the Food and Drug Administration (FDA) in new drug applications (NDAs) and the corresponding published trials. Specifically: 1) Are the results of clinical trials submitted to the FDA published in the scientific literature? 2) What is the concordance between the data submitted to the FDA and the data published in scientific literature?
Methods: All NDAs for New Molecular Entities (NMEs) approved by the FDA from January 2001-December 2002 were identified. FDA officer reviews of each NDA were searched to identify all efficacy trials submitted in the NDA. Publications corresponding to each of the NDA trials were identified by searching Pubmed and the Cochrane Library. In addition, the reference lists of identified articles were scanned and authors and drug companies were contacted. Data were collected to assess publication status, trial and publication characteristics, concordance between the NDA trials and published data, and differences in framing of the results of data submitted to the FDA and data published. A subset of information for each trial was double coded.
Results: Thirty-three NDAs containing 168 efficacy trials met inclusion criteria for the study. Of the 168 trials, 76 were placebo-controlled trials, 8 were crossover studies, and the remaining 84 were active-controlled trials. Seventy-seven percent of trials (129/168) were published, with 2 trials having duplicate publications, for a total of 131 published reports found in 120 papers. Fifteen percent (25/168) were either published incompletely as pooled publications or abstracts or were confirmed not published. No information was found for the remaining 8% (14/168). Out of the 120 papers, 21% (25/120) did not disclose any funding source, 2% (2/120) disclosed government funding, and the remaining 78% (93/120) reported industry funding. Outcomes and conclusions were compared between NDA trials that were published and not published and all published reports. For outcome assessment, trials were grouped by number of primary outcomes reported and statistical significance of the outcomes. Only 83% (109/131) of the published reports specified a primary outcome.

Statistical significance of primary outcome (trials with only one primary outcome)

o NDA Trials, published (n=91): Positive 82% (75/91), Negative/Null 11% (10/91), Unknown 7% (6/91)
o NDA Trials, not published (n=29): Positive 60% (18/29), Negative/Null 23% (7/29), Unknown 13% (4/29)
o Published reports (n=84): Positive 96% (81/84), Negative/Null 4% (3/84), Unknown 0% (0/84)
Statistical Significance of primary outcome (trials with multiple primary outcomes)
o NDA Trials, published (n=38): Positive 47% (18/38), Negative/Null 47% (18/38), Unknown 5% (2/38)
o NDA Trials, not published (n=10): Positive 30% (3/10), Negative/Null 70% (7/10), Unknown 0% (0/10)
o Published reports (n=25): Positive 48% (12/25), Negative/Null 15% (13/25), Unknown 0% (0/25) Conclusions
o NDA Trials, published (n=129): Favorable 64% (83/129), Neutral 7% (9/129), Not Favorable 2% (3/129), Unknown/Mixed 26% (34/129)
o NDA Trials, not published (n=39): Favorable 54% (21/39), Neutral 18% (7/39), Not Favorable 2% (3/39), Unknown/Mixed 20% (8/39)
o Published Reports (n=131): Favorable 95% (125/131), Neutral 2% (2/131), Not Favorable <1% (1/131), Unknown/Mixed 2% (3/131)
Conclusions: Some trial data submitted to the FDA are not published, and trials with only positive primary outcomes and/or those with favorable conclusions are more likely to be published. These discrepancies between published data and data accessible only through the FDA argue strongly for mandatory clinical trials database that includes results reported in a systematic fashion.