Background:

It is desirable to have easy access to systematic reviews (SRs) instead of narrative reviews (NRs). Databases’ methodological filters are good in identifying reviews of the literature but SRs are gathered together with NRs. Considering current limitations of most databases to classify reviews methodologically, we developed a collaborative web-based software to select SRs in LILACS.

Objectives:

• To identify SRs in LILACS using DARE (Database of Abstracts of Reviews of Effects) and other relevant criteria • To assess the accurateness of some criteria on title/abstract for SRs identification.

Methods:

A search strategy (based on Montori 2005, DARE and the Ecuador Cochrane Center strategies) was conducted from January 2008 to February 2011. Study selection was performed through EROS (Early Review Organizing Software), a web-based software to serve in the initial phases of a SR process. Pairs of researchers independently assessed criteria of SR by titles/abstracts and then by full text, considering DARE criteria as the reference test for confirmed SR (Box 1). Discrepancies were solved by consensus or by a third reviewer if consensus were not reached.

Results:

The search strategy retrieved 2241 references with 1786 full-texts available. In total, 15% of the assessed full-texts were classified as SRs based on DARE criteria, while only 1.3% using Oxman & Guyatt and 1% using Cochrane criteria. The accuracy of title/abstract criteria to select SRs compared to DARE criteria are presented in Table 1.

Conclusion:

• This collaborative model will allow LILAC’s users to easily access to confirmed SRs by incorporating labels into LILACS. • The proportion of confirmed SRs out of references retrieved by Montori filter was still low. • Studies not calling themselves as a SRs or meta-analysis in the title/abstract had the highest Negative Predictive Value (97%) and those performing a meta-analysis had the highest Positive Predictive Value (88%).