To assess the frequency, implications, and factors associated with reporting nonfinal analyses (NFAs) of randomized controlled trials (RCTs) as abstract publications.
We identified 138 consecutive reports of RCTs testing systemic therapy for lymphoma, breast, colorectal, or non–small-cell lung cancer published in six major journals between 2000 and 2004. We then searched proceedings of seven major cancer meetings, 1990 to 2004, for abstracts related to these publications which presented efficacy results. Articles and abstracts were compared for discordance in sample size, median follow-up, results, and conclusions. Abstracts were evaluated for statements explicitly noting or implying that results were not final. Factors associated with discordance were assessed by uni- and multivariate analyses.
We identified 303 related abstracts; 197 were eligible. In 86 abstracts (44%), results were stated or implied to be NFA; this was explicitly stated in 41 (21%). The NFAs included 12 where accrual was ongoing. Discordance with article was found in 124 abstracts (63%) and was more common with NFAs (67 of 86 [78%] v 57 of 111 [51%]; P = .0001). When compared with articles, authors' conclusions were substantively different in 17 abstracts (10%). Factors most associated with data discordance were lymphoma trial (odds ratio [OR], 3.8; 95% CI, 1.5 to 10.8), cooperative group trial (OR, 2.8; 95% CI, 1.4 to 5.6), and presentation of a NFA (OR, 2.9; 95% CI, 1.5 to 5.8).
Meeting abstracts often include NFAs and are frequently discordant with subsequent article publication.