Background: The accuracy of a diagnostic test should be compared to the accuracy of its alternatives. Direct comparisons of tests, in the same patients and against the same reference standard, offer the most valid study design, but are not always available. Comparative systematic reviews are therefore bound to rely on indirect comparisons. As the results from these comparisons can be biased, we investigated ways to correct for indirectness.

Methods: From a large systematic review about the accuracy of D-Dimer testing for venous thromboembolism, we selected those comparisons between two assays that contained three or more direct comparisons and four or more indirect comparisons or single assay studies. Each comparison was analyzed using the bivariate random effects meta-regression model with assay-type and directness as covariates in the model. In comparisons with a significant effect of directness on sensitivity or specificity, we included the following study features to correct for these differences: referral filter, consecutive enrolment, time-interval, type of reference standard and verification.

Results: Of the seventeen eligible comparisons, five showed a significant effect of direct comparisons on logit-sensitivity or logit-specificity. A first analysis showed that in three of these analyses, the addition of at least one of the study features decreased the impact of directness as a covariate. For example, in one analysis directness had a significant effect on specificity, but addition of time-interval or reference standard removed this effect (from P = 0.048 to P = 0.22 resp. P = 0.12). Further analyses will be presented at the Colloquium.

Conclusions: Adjusting the effect of directness for study features seems to be possible in some instances, but no systematic effects were found. Study characteristics that may be influential in one comparison, may have no influence at all in another comparisons.

Knowledge translation and communicating evidence Day 22 Oct (P3A272-P3B345)