Article type
Year
Abstract
Background: Methodological filters for diagnostic test accuracy (DTA) studies may reduce the number of citations that must be manually screened by authors of DTA systematic reviews, whilst maintaining high search sensitivity.
Objectives: To characterise the performance of methodological filters for DTA studies and assess whether filter performance has improved since the publication of the STAndards of Reporting for Diagnostic studies (STARD) statement in 2003.
Methods: Two investigators independently handsearched three high-ranking nephrology journals for the years 2002–2003 and 2009–2010 to create a reference set of DTA studies. Data was extracted on type of test evaluated, target condition and study population. Seven methodological filters for DTA studies were run in MEDLINE with terms to restrict the results to the journals and years of interest. Filter performance was calculated as sensitivity, specificity and number needed to screen (NNS) with 95% confidence intervals. Differences in filter performance pre/post STARD were tested using the two proportion Z-test.
Results: 103 DTA studies were identified by handsearch, accounting for 2.1% of all studies published. DTA studies were predominantly conducted in dialysis patients (37/103, 35.9%), assessed biochemical tests (34/103, 33.0%) and tests designed to assess kidney function (30/103, 29.1%). Overall, the most sensitive filter was van der Weijden 1997; sensitivity 0.95 (95%CI 0.91–0.99) and specificity 0.53 (95%CI 0.53–0.56) (Fig. 1). At a DTA study prevalence of 2.0%, the van der Weijden 1997 filter reduced the NNS by 26/50 (52.0%) (Table 1). Changes in filter performance pre to post STARD were variable, see Figure 1.
Conclusions: Methodological filters for DTA studies may reduce the NNS by authors of DTA systematic reviews in nephrology, whilst maintaining acceptable search sensitivity. Further research is required to assess filter performance across a broader range of journals.
Objectives: To characterise the performance of methodological filters for DTA studies and assess whether filter performance has improved since the publication of the STAndards of Reporting for Diagnostic studies (STARD) statement in 2003.
Methods: Two investigators independently handsearched three high-ranking nephrology journals for the years 2002–2003 and 2009–2010 to create a reference set of DTA studies. Data was extracted on type of test evaluated, target condition and study population. Seven methodological filters for DTA studies were run in MEDLINE with terms to restrict the results to the journals and years of interest. Filter performance was calculated as sensitivity, specificity and number needed to screen (NNS) with 95% confidence intervals. Differences in filter performance pre/post STARD were tested using the two proportion Z-test.
Results: 103 DTA studies were identified by handsearch, accounting for 2.1% of all studies published. DTA studies were predominantly conducted in dialysis patients (37/103, 35.9%), assessed biochemical tests (34/103, 33.0%) and tests designed to assess kidney function (30/103, 29.1%). Overall, the most sensitive filter was van der Weijden 1997; sensitivity 0.95 (95%CI 0.91–0.99) and specificity 0.53 (95%CI 0.53–0.56) (Fig. 1). At a DTA study prevalence of 2.0%, the van der Weijden 1997 filter reduced the NNS by 26/50 (52.0%) (Table 1). Changes in filter performance pre to post STARD were variable, see Figure 1.
Conclusions: Methodological filters for DTA studies may reduce the NNS by authors of DTA systematic reviews in nephrology, whilst maintaining acceptable search sensitivity. Further research is required to assess filter performance across a broader range of journals.
Images
