Abbreviations: ROC, characteristic of the receiver`s operation; AUC, an area below the ROC curve; IC, confidence interval; LRTI, lower respiratory tract infection; NPA, negative approval rate; NPV, negative predictive value; AAE, positive percentage agreement; The APA, positive forecast value; ROC, the operating characteristic of the receiver; RPD, retrospective medical diagnosis; Sir.B. 99% sensitivity or negative forecast value, experimental interpretation must be extremely cautious, even if low levels of uncertainty may be present in the means of comparison. In such cases, a seemingly reasonable condition for robust test performance leads to the refusal, in almost all cases, of a perfect test, since the effects of this document are not taken into account. Parties interested in very high performance (i.e. 99% sensitivity or NPV) should bear in mind that these high performances can only be demonstrated in practice for a near-perfect comparison method. In the absence of an almost flawless comparison method, it will not be possible to validate such high test performance characteristics and attempts to do so may lead to an underestimation of candidate performance. The effects of an imperfect comparison on high-performance tests are analyzed quantitatively in S7 Supporting Information. For validation, the FDA recommends a “clinical agreement study,” as well as loD detection and cross-reactivity studies. We focus here on clinical matching, which generally compares the results of two different methods.
The FDA states that “designed clinical samples” can be used, which means that it is acceptable to peak with a high-quality (preferably inactive) control material. The FDA recommendation applies to 30 reactive samples (20 low reagents for 1 to 2 times loD and 10 times higher, covering the test area) and 30 non-reactive samples. The FDA also requires that the first 5 positive and first 5 negative results of the patient be confirmed by a previously approved EEA method. Figure 5 simulates a screening test in a low-prevalence environment, where negative ground truth is significantly more common than truth-positive ground. An example of this scenario is cervical cancer screening through Pap striped cytology, where significant positive rates (cell abnormalities of unknown importance) can be expected and positive test results do not necessarily give high confidence in the presence of a high-level disease [28,29].