Test Characteristics and Biases

A diagnostic test can be characterized by its diagnostic sensitivity (Se) and diagnostic speci­ficity (Sp). The standard definitions are that SE describes the probability that a test can detect an animal with a given target condition, and Sp describes the probability that a test is nega­tive given that the target condition is absent.

First, target conditions vary with the purpose of testing, and therefore the purpose should be specified in any test validation to en­sure that the test evaluation is fit for purpose (OIE, 2003). Second, the test should be evalu­ated in a population similar to the one in which it is used in order to avoid a substantial risk of spectrum bias (Ransohoff and Feinstein, 19 78). Spectrum bias is a particular concern in diagnostic test evaluations for chronic dis­eases, where case-control designs have been used. A thorough description of the population in which a given test has been evaluated, along with the selection criteria for the cases and controls, are vital to the interpretation of any test results used subsequently. Lack of such a description can easily lead to scenarios where estimates presented as diagnostic sensitivities and specificities have an interpretation that is very different to the standard interpretation (Table 20.2). If the reader is not cautious, then the estimates can be misinterpreted. For exam­ple, Clark et al. (2008) stated that, ‘direct fecal PCR was 70.2% sensitive and 85.3% specific when using culture as the gold standard', i.e. the target condition considered by Clark et al. (2008) may have been ‘infectious with MAP’ or ‘shedding of MAP’. However, this was inter­preted by Magombedze et al. (2017) as sensi­tivity to detect MAP infection, which is a very different target condition. Therefore, the con­clusions of Magombedze et al. (2017) must be reconsidered. Even the same target condition can be presented with different case definitions and consequently different meanings within the same studies (Table 20.2). Further cau­tion is therefore warranted when estimates are interpreted.

Se and Sp estimates are generally incon­sistently reported, with highly variable target conditions. A summary of Se and Sp estimates for ante-mortem tests, stratified by target condi­tion, was reported by Nielsen and Toft (2008). Because these estimates related to the purpose of testing and relied on the occurrence of im­mune responses, these aspects will be discussed further.

20.4