CHARACTERISTICS OF META-ANALYSES
SUBMITTED TO A MEDICAL JOURNAL
Donna F Stroup,1 Stephen B Thacker,1 Carin M Olson,2,3 and Richard M Glass3
1Epidemiology Program Office, Centers for Disease Control and Prevention, Atlanta, GA, 30333, USA; 2University of Washington, Seattle, WA, USA; 3JAMA, 515 N State St, Chicago, IL 60610, USA
Objective: To assess whether the methodologic quality of meta-analyses affects publication; and whether specific quality characteristics predict publication.
Design: Case series of all meta-analyses submitted to JAMA, a weekly general medical journal with guidelines for submitting meta-analyses. One investigator
identified meta-analyses according to predetermined criteria. Two other investigators, blinded to the authors' identity, abstracted standard methodologic
characteristics from the meta-analyses and rated their quality.
Results: Seventy-six meta-analyses were evaluated from January 1, 1996, through June 30, 1997. The following percentages demonstrated the characteristic with
high or medium quality: 100% addressed appropriateness of pooling data; 99% generalized conclusions appropriately and defined the problem or stated a hypothesis;
93% considered alternative explanations for results; 92% specified the basis for selecting and coding data; 90% stated explicit inclusion and exclusion criteria;
89% used an appropriate statistical analysis; 89% showed an effort to include all available studies; 73% tested for heterogeneity; 63% provided guidelines for
future research; 60% documented how data were classified and coded; 51% coded objectively; 42% assessed publication bias; 34% assessed study quality; and 32%
gave enough detail for a reader to replicate results. Only 13% assessed the comparability of cases and controls. Among the meta-analyses, 18% displayed all
but 1 or 2 of the characteristics; only 9% had less than half the characteristics. To date, publication decisions have been made on 55 meta-analyses.
Because of the small number accepted (6/55 or 11%), preliminary results associating quality and publication would be unreliable and are not presented.
Conclusions: Meta-analyses submitted to a weekly general medical journal meet most methodologic standards.
Return to Session Information
