Background & Aims: The published risk of adenocarcinoma in the setting of B
arrett's esophagus (BE) varies. Publication bias, the selective reporting o
f studies featuring positive or extreme results, may result in overestimati
on of this cancer risk in the literature. The aim of this study was to asse
ss those publications reporting a cancer risk in BE for evidence of publica
tion bias. Methods: A MEDLINE search for all published estimates between 19
66 and 1998 of cancer risk in BE was performed. All studies reporting a can
cer risk expressible in cancers per patient-year of follow-up were retrieve
d. Bibliographies of these studies were surveyed for additional estimates.
All publications that required an initial endoscopy with histologic confirm
ation of BE and any cancer were included. The relationship of reported canc
er risk to size of the study was assessed, Multivariable regression control
ling for differences in definition of BE, as well as other study characteri
stics, was performed. The data were also analyzed by means of a funnel diag
ram, an epidemiologic method to assess publication bias. Results: Five hund
red fifty-four abstracts were reviewed. Twenty-seven publications met the s
tated criteria for inclusion. There was a strong correlation between cancer
risk and the size of the study, with small studies reporting much higher r
isks of cancer than larger studies. This association persisted when differe
nces in the definition of BE, retrospective vs. prospective nature of the s
tudy, surveillance interval, and the effect of cancer detected in the first
year were considered. The funnel diagram analysis suggested publication bi
as. Conclusions: The cancer risk in BE may be overestimated in the literatu
re due to publication bias.