Article type
Year
Abstract
Background: Often, a search for a review of diagnostic test accuracy studies (DTA review) results in a high number of references to screen. However, if the search could be limited to only the freely accessible database MEDLINE, the number of references needed to screen may be reduced and costs may be spared.
Objectives: To assess the number of studies that will be missed when a search for diagnostic accuracy studies is restricted to MEDLINE and whether this will introduce bias in a meta-analysis.
Methods: We collected diagnostic reviews published in high impact journals between 2006 and February 2011, that included a search in MEDLINE and in at least one other biomedical database, stated the search date, and did not apply language restrictions. For each meta-analysis, we identified the number of studies that could not be found in MEDLINE. When the meta-analysis included ≥10 studies, we repeated the meta-analysis, but including only those studies that were indentified in MEDLINE. We calculated the ratio of the diagnostic odds ratios (RDOR) based on the MEDLINE only studies versus all studies.
Results: We identified 615 reviews of which 44 reviews with 76 meta-analyses were included. In 65 meta-analyses (85.5%) the studies were all identified in MEDLINE. In the remaining 11 meta-analyses, 87.5 to 99.0% of the studies were identified in MEDLINE. Of those, 10 meta-analyses included ≥ 10 studies. However, due to the very low proportion of non-MEDLINE studies we did not find repeating meta-analysis helpful. Therefore no RDOR was calculated.
Conclusions: Surprisingly, the extra yield of a search for primary diagnostic accuracy studies in other databases than MEDLINE was very low. Based on our relatively small sample, searching in databases other than MEDLINE hardly seems to affect the outcome of a DTA review.
Objectives: To assess the number of studies that will be missed when a search for diagnostic accuracy studies is restricted to MEDLINE and whether this will introduce bias in a meta-analysis.
Methods: We collected diagnostic reviews published in high impact journals between 2006 and February 2011, that included a search in MEDLINE and in at least one other biomedical database, stated the search date, and did not apply language restrictions. For each meta-analysis, we identified the number of studies that could not be found in MEDLINE. When the meta-analysis included ≥10 studies, we repeated the meta-analysis, but including only those studies that were indentified in MEDLINE. We calculated the ratio of the diagnostic odds ratios (RDOR) based on the MEDLINE only studies versus all studies.
Results: We identified 615 reviews of which 44 reviews with 76 meta-analyses were included. In 65 meta-analyses (85.5%) the studies were all identified in MEDLINE. In the remaining 11 meta-analyses, 87.5 to 99.0% of the studies were identified in MEDLINE. Of those, 10 meta-analyses included ≥ 10 studies. However, due to the very low proportion of non-MEDLINE studies we did not find repeating meta-analysis helpful. Therefore no RDOR was calculated.
Conclusions: Surprisingly, the extra yield of a search for primary diagnostic accuracy studies in other databases than MEDLINE was very low. Based on our relatively small sample, searching in databases other than MEDLINE hardly seems to affect the outcome of a DTA review.