Optimal search strategy for clinicai trials in the Latin American and Caribbean Health Science

Objective: To define and disseminate the optimal search strategy for clinicai trials in the Latin American and Caribbean Health Science Literature (LILACS). This strategy was elaborated based on the optimal search strategy for MEDLINE recommended by Cochrane Collaboration for the identification of clinicai trials in electronic databases. Design: Technical information. Setting: Clinicai Trials and Meta-Analysis Unit, Federal University of São Paulo, in conjunction with the Brazilian Cochrane Center, São Paulo, Brazil. (http://www.epm.br/cochrane). Data.;. LILACS/CD-ROM (Latin American and Caribbean Health Science Information Database), 27th edition, January 1997, edited by BIREME (Latin American and Caribbean Health Science Information Center). LILACS Indexes 670 journals in the region, with abstracts in English, Portuguese or Spanish; only 41 overlap in the MEDLINE-EMBASE. Of the 168.902 citations since 1982, 104,016 are in human trials, and 38,261 citations are potentiality clinicai trials. Search strategy was elaborated combining headings with text word in three languages, adapting the interface of the LILACS. We will be working by locating clinicai trials in LILACS for Cochrane Controlled Trials Database. This effort is being coordinated by the Brazilian Cochrane Center.


INTRODUTION
A n unbiased search is the key to a relevant systematic review.To proteet against bias and ensure that alI relevant data are included in a review, it is important to use multiple sources to identify studies and a systematic approach to select among them I .
As for sources, Latin America and Caribbean have the LILACS 2 , an electronic database (CD-ROM and Online) which indexes regionalliterature from more than 670 Address for correspondence: Therefore, the use of LILACS for identifying clinicaI triaIs and systematic reviews should be a part of every search strategy.The Cochrane Schizophrenia Reviews Group (CSRG), injust one previous effoft, identified 1,391 trials using LILACS alone 3 • Our foeus was to adapt the optimal seareh strategy for randomized eontrolled trials proposed by the Cochrane Collaboration by incorporating ideas from the CSRG to formulate the optimal search strategy for using LILACS.

OPTIMAL SEARCH STRATEGY FOR CLINICAL TRIALS
In Table 1 we present the optimal search strategy to be üsed with LILACS/CD-ROM, 27th edition, January 1997.The strategy is divided into the following three stages as defined in Dickersen 4 et aI.and repJicated in the Cochrane Handbook: stage one (sets 1 -9) includes terms with high precision; stage two (sets 10 -28) includes terms with moderate precision; and stage three (sets 29 -39) includes terms with low precision but with optimal browser sensitivity.

DISCUSSION
U sing the optimal search strategy, are identifying 38,261 citations, in contrast with 3,505 citations retrieved by the CSRG have been identified search strategy.Only RCTs and CCTs will be selected from among these citations.
Under the auspices of the Cochrane Collaboration, with each new edition ofLILACS, we will be locating 5 new RCTs and CCTs for the Cochrane Controlled Trials Register in the Cochrane Library.
In addition, we will be working retrospectively by locating clinicaI trials in old records of LILACS, through the CLINT-LAC PROJECT (ClinicaI Trials in the Latin American and Caribbean Project).This effort is being coordinated by the Brazilian Cochrane Center (http:// www.epm.br/cochrane).
At present, of the 31 Cochrane Reviews Groups, only the CSRG is using LILACS.Just two others, the Hepato-Biliary and Peripheral Vascular Disease Reviews Groups, mention LILACS as an electronic Database that will be searched in the near future.We observed this after verifying the search strategy of each group6.
As non-English-Ianguage references are underrepresented in MEDLINE, and only published articles are included, there exists the potential for publication bias 4 ,7-10 and language bias ll • 12 • Thus, the inclusion of LILACS in any search strategy should be mandatory.
As for technology requirements, during installation of the DOS-based software for LILACS/CD-ROM, an .interface for three languages is available, so that opting to 1425 use English for the search is easy.The principal difficulty is that saving the strategy to floppy disk is impossible.Only expert knowledge will resolve this problem.
An update of the search strategy could be necessary.Analogously, it will be necessary to elaborate a search strategy for the disease and the intervention using headings and text free.The headings could verify the means, exploding if necessary.Text free (subject) will be synonyms, plural, different manner of writing, and correspondent words in English, Portuguese and Spanish languages will be indispensable.Obviously, further development is necessary in this area.
In conclusion, through constant efforts to elaborate and improve a search strategy for electronic databases, we are approaching the ideal.We believe that this collaborative w9rk is the best way to reduce bias in locating and selecting clinicaI trials.