Full publication of results initially presented in abstracts

Abstract

Background Results initially presented at scientific meetings are usually summarized as abstracts. These abstracts may be published in conference proceedings or in journal supplements, and thus are not generally available. Because results presented in abstracts may be preliminary or imprecise, and because these results may not be subsequently published in full, publication bias may result; i.e. the decision for publication is based on the magnitude or direction of study results. Publication bias, in turn, creates problems for those completing systematic reviews and those relying on the published literature for evidence. Objectives To determine the rate at which results in abstracts are subsequently published in full, and the time interval between presentation at a meeting and full publication. To assess the association between study characteristics and full publication. Search methods We searched MEDLINE, Science Citation Index, reference lists, and author files. Date of most recent search: June 2000. Selection criteria We included all reports that examined the subsequent rate of full publication of results related to the life sciences that were initially presented as abstracts or in summary form, with follow‐up of at least two years. Data collection and analysis One reviewer extracted data from each report onto a standard form. We calculated the weighted mean rate of publication and the range, median, and mean time to publication. Dichotomous variables were analysed using relative risks. Main results Combining data from 46 reports (15,985 studies reported as abstracts) resulted in a weighted mean rate of subsequent full publication of 44.8% (95% confidence interval (CI) 44.0 to 45.6). Data from five reports examining randomized controlled trials (573 studies reported as abstracts) resulted in a weighted mean rate of subsequent full publication of 55.8% (CI 51.8 to 59.9). The median of the mean time to publication determined in nine reports was 19.7 months, and the median of the median time to publication determined in eleven reports was 17 months. Most abstracts that were published in full appeared within three years of presentation at the meeting. Three reports, in which authors defined 'positive' results as a 'significant' result favoring any treatment arm over another, showed an association between 'positive' results and full publication (RR = 1.51; CI 1.27 to 1.79). No association was detected between 'positive' results and full publication in four reports in which authors defined a 'positive' result as a result favoring the experimental treatment compared to the control treatment (RR = 0.97; CI 0.81 to 1.18). Sample size equal to or above the median was shown to be associated with full publication (RR = 1.24; CI 1.07 to 1.45) as was oral (versus poster) presentation (RR = 1.13; CI 1.00 to 1.27). Abstracts accepted for presentation at a meeting were more likely to be followed by full publication than those rejected (RR = 1.84; CI 1.69 to 2.00). Abstracts with results describing clinical research were less likely to be followed by full publication than abstracts with results describing basic research (RR = 0.79; CI 0.71 to 0.88) whereas abstracts describing randomized controlled trials appear to be published more often than other types of study designs (RR = 1.29; CI 0.95 to 1.75). Authors' conclusions Full publication of studies initially appearing as abstracts appears to be biased in that 'significant' results were more frequently published than non‐significant results. More studies, especially those examining full publication of abstracts describing results from randomized controlled trials, need to be completed to confirm these observations.