Identifying studies on diagnostic tests in oncology: the prostate-specific antigen (PSA) experience

Tags: Poster
Franceschini R, Trevisiol C, Valentini M, Pistotti V

Background: Search strategies developed from one set of disciplines might not be optimal for retrieving strong evidence for other healthcare professions because of the variation in indexing quality in MEDLINE.

Objectives: To assess the sensitivity and the specificity of the existing search strategies in MEDLINE for identifying potentially relevant studies related to PSA. To evaluate which single or combination of terms offer a higher sensitivity and/or a good balance between sensitivity and specificity adequate enough to use regularly.

Methods: We compared search strategies for PubMed with the final list of studies (gold standard) on PSA and diagnosis indexed in MEDLINE (1966-2004). This database was developed by using an initial list of citations found by a "topic only" search strategy in MEDLINE, EMBASE, The Cochrane Library and was selected by two trained reviewers by applying inclusion/exclusion criteria. A single or combination of terms of the existing strategies in the selected database was tested.

Results: Haynes' and Vincent's sensitive methodological filters remain to be the most sensitive MEDLINE search strategies also in this setting (sensitivity: 99.4%), appropriate for systematic reviews. However, a new combination of terms chosen by the different existing search strategies, offers a good balance between sensitivity and specificity (sensitivity: 94,2%; specificity: 60,9%) which is useful for busy clinicians who need to be updated regularly.

Conclusions: Validation of existing methodological filters for studies on diagnostic accuracy for application in other medical areas is of utmost importance.