Journal of Clinical Epidemiology 58 (2005) 867–873 REVIEW ARTICLES Searching one or two databases was insufficient for meta-analysis of observational studies Adina R. Lemeshow a , Robin E. Blum a, * , Jesse A. Berlin b , Michael A. Stoto c , Graham A. Colditz a,d a Department of Epidemiology, Harvard School of Public Health, 677 Huntington Avenue, Boston, MA 02115, USA b Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, PA, USA c RAND Corporation, Arlington, VA, USA d Channing Laboratory, Department of Medicine, Brigham and Women’s Hospital, and Harvard Medical School, Boston, MA, USA Accepted 8 March 2005 Abstract Objective: To address methodologic issues in searching for observational studies by presenting database search methods and results. Study Design and Setting: Results of two literature searches for publications reporting on observational studies of alcohol consumption and the risk of breast cancer and large bowel cancer were compared, to evaluate the sensitivity of various bibliographic databases and search strategies, including hand-searching reviews and meta-analyses. Results: The target sensitivity of 90% of publications in the breast cancer search was achieved by starting with Medline, then adding Biosis, Embase, and SCI EXPANDED–SSCI, which provided a total of 72 (91%) of the 79 relevant publications. To reach a similar 89% sensitivity for large bowel cancer, at least Biosis, Dissertation Abstracts Online, Embase, ETOH, and Medline had to be searched, with the addition of hand search of reviews and meta-analyses. Conclusion: Limiting a search to one or two databases when conducting meta-analyses of observational studies will not provide a thorough summary of the existing literature. The findings support recommendations to implement a comprehensive search of electronic databases and the reference lists of recent review articles and meta-analyses.  2005 Elsevier Inc. All rights reserved. Keywords: Meta-analysis; Bibliographic databases; Epidemiologic studies; Sensitivity; Cancer; Alcohol drinking 1. Introduction In 1904, Karl Pearson performed the first reported meta- analysis. He derived a quantitative summary estimate for the efficacy of a vaccine against enteric fever [1]. One hundred years later, researchers continue to wrestle with methodologic issues in the conduct and interpretation of meta-analyses and their application in public health. Al- though the general approach to the conduct of meta-analysis has been extensively described [2], most work addressing key issues has focused on statistical methods rather than data collection. Of those investigators who have examined methodologic issues pertaining to data collection, most have relied solely on Medline and have limited consideration to randomized control trials [2]. Recently, however, more meta-analyses have drawn on observational data [3]. In 2000, Stroup et al. [4] published guidelines for the reporting of meta-analyses of observational studies (although they did * Corresponding author. Tel.: 617-432-2723; fax: 617-566-7805. E-mail address: rblum@hsph.harvard.edu (R.E. Blum). 0895-4356/05/$ – see front matter  2005 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2005.03.004 not specifically address searching and data collection meth- ods). To date, there still are no standard guidelines for creat- ing a search strategy, defining exclusion–inclusion criteria, determining which electronic databases to search, or re- porting the results of a literature search. With the present study, we address the issue of searching for observational studies by considering database search methods and results. We compared the results of two literature searches for publi- cations reporting on observational studies of alcohol con- sumption and the risk of breast cancer and large bowel cancer, with the particular aim of exploring the sensitivity of various bibliographic databases and search strategies, in- cluding hand searching of reviews and meta-analyses. 2. Methods Following standard practices for conducting meta-analysis [5], we identified publications on the relationship between alcohol consumption and breast and large bowel cancer, respectively, by searching several electronic databases and checking reference lists of meta-analyses and reviews.