ISSG structured abstract for:
Montori VM, Wilczynski NL, Morgan D, Haynes RB. Optimal search strategies for retrieving systematic reviews from MEDLINE: analytical survey. BMJ 2005;330(7482):68.
Structured abstract prepared by: Sue Bayliss
Objective:
This filter is designed as an optimal search strategy for retrieving systematic reviews in MEDLINE (Ovid).
Methods:
The authors identified a gold standard (GS) of 753 records by handsearching 161 clinical journals in 2000. Two validation sets were used to test filters: one database of 161 high yield journals including the Cochrane Database of Systematic Reviews (CDSR) and one excluding CDSR. The search terms for the filters were collected from consulting widely with experts, examining other groups’ strategies and from relevant records.
Results:
In the full validation set (including CDSR) the most sensitive filter scored 99.9% for sensitivity (Confidence Interval 99.6 to 100) and had a specificity of 52.0% (CI 51.6 to 53.5), the most precise scored 57.1% (CI 53.9 to 60.3) with a sensitivity of 71.2% (CI 68.0 to 74.4). The filter providing the best compromise between sensitivity and specificity scored 98.0% (CI 97.0 to 99.0) sensitivity and 90.8% specificity (CI 90.5 to 91.1) with 14.2% precision (CI 13.4 to 15.2).
Discussion:
The authors note that by using a small subset of the database to derive strategies, the performance of the identified strategies may have been overestimated.
ISSG commentary:
The restriction to one year may affect the future currency of the filter, given likely changes in terminology and indexing terms. The description of the methodology, particularly with regard to building the gold standard dataset, could be clearer.