Acceptance by the aga is the single best indicator of subsequent full publication of research

Abstract

We studied publication rates and determinants of publication in GI-research. In particular, our concern was the possibility of publication bias, i.e. publication based on the statistical significance of study results. Methods. A random sample (n=l,000, stratified by study type) of all abstracts submitted to the 1992 to 1995 AGA-meetings, was followed up by a computerized literature search (Medline), as well as by postal questionnaires to abstract authors. In addition, abstracts were evaluated for study design features, statistical significance of study results, sample size, and research topic. Abstract evaluation was blinded with respect to acceptance, authors and origin. A quality score was assigned to all abstracts, based on completeness of reporting and measures undertaken to improve internal validity. High impact journals were defined as scientific impact factors (IF) > 5. Differences in publication were tested using log rank tests and Cox regression analysis. Results. Information is available on research presented in 232 abstracts submitted to the 1992-1994 meetings. Within the follow-up period of 3-5 years, 121 research projects (52%) were published in full, 63% of the subgroup of controlled clinical trials (CCT), 57% of basic science studies (BSS). Thirty percent of all abstracts were published within one year of the meeting. The majority of publications (60%) were in GI&Hepatology journals (71% of CCT publications, most frequently Dig Dis Sci, Am J Gastroenterol and Gastroenterology), followed by basic science journals (51% of BSS, most frequently Am J Physiol). Only 2% of all projects presented as abstracts were subsequently published in general medical journals, 19% of all publications were in high IF-journals, 98% were in English. Log ranks tests showed significantly better outcomes, when abstracts had been accepted for AGApresentation (p=.01). Also, outcome was dependent on study type, with CCT and BSS being more likely to be published than epidemiological or other research on humans (p=.03). With respect to topics, research on IBD had the highest publication rates (86%), research on upper GI-problems the lowest (43%), but overall, the research area was not a significant predictor (p=.l). There was no difference in publication rates, or in the percentage of high IF publications between North-American, Australian, and European research. However, although not statistically significant, countries other than these had markedly less overall publications (30% as compared to 54%, p=0.07), and none in high IF-journals. Statistical significance of study results and sample size could not be shown to influence publication rates, nor was the formal quality of the respective AGA abstract predictive. Cox regression analysis retained abstract acceptance as the only significant predictor of subsequent full publication (HR 1,9, CI 1.2-3.0). Preliminary results from the postal survey found lack of time or loss of interest the most frequent reasons for non publication. In 43% of unpublished projects, a manuscript had not or not yet been prepared. Conclusions: Acceptance for abstract presentation was found to be the only significant predictor of full publication of research projects initially submitted as abstracts to DDW meetings. There is, in addition, some evidence, that the country of origin, research type and research topic may be of importance. Bias based on the statistical significance of the study findings (publication bias) does not seem to be an issue. *Fellow of the German Academic Exchange Service (DAAD). Research funded by the Calgary Regional Health Authority (CRHA).