A systematic review of studies that develop or evaluate search filters for the retrieval of diagnostic studies in MEDLINE

Tags: Oral
Beynon R1, Leeflang M2, Eisinga A3, McDonald S4, Mitchell R4, Whiting P1, Glanville J5
1School of Social and Community Medicine, University of Bristol, UK, 2Department of Clinical Epidemiology and Biostatistics, University of Amsterdam, Netherlands, 3UK Cochrane Centre, National Institute for Health Research, UK, 4Australasian Cochrane Centre, Monash Institute of Health Services Research, Australia, 5York Health Economics Consortium Ltd, University of York, UK

Background: Search filters are developed to improve the efficiency and effectiveness of searching and are created by identifying and combining search terms to retrieve records with a common feature (e.g. diagnostic accuracy).

Objectives: To determine whether any existing methodological search filters designed to retrieve diagnostic test accuracy (DTA) studies perform well enough to be recommended for use in Cochrane DTA reviews.

Methods: We conducted a systematic review of studies that developed and/or evaluated methodological filters designed to retrieve DTA studies in MEDLINE. Eligible studies had to report a measure of filter performance such as sensitivity or precision. Studies were identified through searching electronic databases and reference screening. Two review authors independently screened search results. Inclusion assessment, data extraction and methodological assessment were conducted by one author and checked by a second. Methodological assessment was undertaken using a modified checklist. Data to calculate sensitivity, specificity, precision and the number of abstracts needed to read (NNR) to identify one relevant study were extracted for each filter evaluation and were summarized in a forest plot. Results were analysed separately for studies that developed the original filters (development studies) and studies that evaluated existing filters (evaluation studies).

Results: Fifteen studies (16 papers) reporting on 56 filters were included: 9 development studies, 6 evaluation studies, and 2 studies that both developed and evaluated filters (Figure 1). None of the filters had consistently high sensitivity and low NNR in the evaluation studies. Estimates of sensitivity ranged from 29% to 100% and NNRs, where available, ranged from 9 to 100. Estimates of sensitivity in the development studies were higher than those reported in the evaluation studies in 28 of 33 comparisons, and estimates of NNR were lower in 5 out of 10 comparisons.

Conclusions: None of the filters evaluated showed sufficient sensitivity and precision to be recommended for identifying DTA studies in MEDLINE.