[Skip to Content]
[Skip to Content Landing]
September 11, 1991

A Meta-analysis of Estrogen Replacement and Breast Cancer

Author Affiliations

San Antonio, Tex

San Antonio, Tex

JAMA. 1991;266(10):1359. doi:10.1001/jama.1991.03470100050020

To the Editor.  —Steinberg et al admit to including only published studies in their meta-analysis because "most studies represent a large enough investment to warrant some effort to publish, whether results are positive or negative, and most credible work will be published."1 Available data on publication bias suggest that, despite the effort invested in a study, investigators often do not submit them for publication due to negative results2 and that the study quality of unpublished work may be as good or better than published work.3Although Sacks et al4 do not actually recommend the inclusion of unpublished studies in all meta-analyses, they do emphasize the need for calculation of a failsafe N to provide the reader with the number of negative studies not included that would be required to overturn a significant result. If we use the 15-year RR of 1.3 reported by Steinberg et al obtained