Meta-analytical equivalence studies on diagnostic tests for bovine brucellosis allowing assessment of a test against a group of comparative tests.

In the assessment of diagnostic tests the task may arise to show that a candidate test is non-inferior compared to a comparative (standard) test with regard to the diagnostic sensitivity or specificity. This setting is known as "one-sided equivalence" and has been applied to a single comparison between two diagnostic tests (Chen et al., 2003). Recently, the approach has been extended into a meta-analytical framework (EFSA, 2006), allowing for the difference between the sensitivity (or specificity) of two diagnostic tests to be estimated using information gathered through systematic literature review. Using this approach, confounding factors are adjusted by matching of parameter estimates on study population and preferred levels of the confounding factors. However, the power of this approach was found to be limited and therefore Markov chain Monte Carlo logistic regression (MCMCLR) models that allow adjustment for confounding variables have been developed (EFSA, 2006). We report here a refinement of the statistical inference based on the latter approach. The objective was to generate a posterior distribution of the meta-analytical difference statistic for the candidate test and a set of comparative tests. The algorithm for this purpose uses Monte Carlo sampling from the posterior distributions of sensitivity (or specificity) and, for each iteration, (i) identifies the least performant comparative test, (ii) establishes the difference statistics for this test and the candidate test and (iii) compares the difference statistic with a critical threshold value. The proportion of iterations in which the critical threshold was exceeded is then interpreted as the P-value for the one-sided equivalence test for the candidate versus the set of comparative tests. We illustrate and discuss the method using a case study on tests for bovine brucellosis.