Multiple and discordant systematic reviews in medical literature: an epidemiological cross sectional study across medical fields

Background: Systematic reviews (SRs) were originally developed to solve inconsistencies among primary studies. The growing number of SRs in scientific literature increases the likelihood of finding multiple overlapping SRs presenting discordant results or conclusions. This may repeat the ‘original sin’. Objectives: To capture a representative cross-sectional sample of published SRs across medical specialties, explore the epidemiology of multiple SRs and examine them in terms of results and conclusions discordances. Methods: To identify multiple SRs we used Clinical Evidence sections and related search strategies in topics related to oncology, neurology, cardiology and odonto-stomatology. Citations were screened to identify multiple SRs (that shared same objective, population, condition/pathology and intervention) published between 1997 and 2007. Clusters of multiple SRs sharing at least one outcome were created. Descriptive, reporting and bias-related aspects of the reviews were collected using specific tools (i.e. AMSTAR, QUOROM). Multiple SRs were classified as discordant by results or interpretation. The data were analyzed descriptively. Results: 4683 records have been screened thus far. Out of 431 SRs, 109 (25%) were multiple and 103 clusters were created. Preliminary analysis indicated that clusters were variable in terms of included multiple SRs (median 3, range from 2 to 7). 67 clusters (65%) were consistent and 36 clusters (35%) were discordant for one or more outcomes. Reporting and bias-related aspects were highly variable among and within clusters. Conclusions: Multiple SRs are now common in several medical fields with potential waste of research resources. Often multiple SRs reach discordant results. These discordances may decrease the perceived value of SRs as tool to solve inconsistencies and reduce uncertainty in the scientific enterprise.