Negative percentage agreement (NPA): the percentage of negative calls from the comparator that the test to be evaluated is considered negative. This value is calculated in the same way as specificity. However, the NPA is used in place of specificity to recognize that, because of the uncertain comparison, this measure should not be interpreted in such a way as to accurately reflect the measure of specificity. CLSI EP12: User Protocol for Evaluation of Qualitative Test Performance Protocol describes the terms of the Positive Percentage Agreement (AEA) and the Negative Performance Agreement (NPA). If you have two binary diagnostic tests to compare, you can use an agreement study to calculate these statistics. Because specificity/APA reflects the ability to accurately identify negative controls, which are more widely available than patient samples, IC tends to be narrower for these metrics than in sensitivity/AAE, allowing for consideration of the proportion of positive cases a test can find. Uncertainty in patient classification can be measured in different ways, most often using statistics from inter-observer agreements such as Cohens Kappa or correlation terms in a multitrait matrix. These statistics, as well as the statistics associated with them, assess the extent of matching in the classification of the same patients or samples by different tests or examiners, in relation to the extent of compliance that would be accidentally expected. Cohen`s Kappa goes from 0 to 1. Value 1 indicates perfect match and values below 0.65 are generally interpreted as having a high degree of variability when classifying the same patients or samples. Kappa values are frequently used to describe reliability between patients (i.e.
the same patients between physicians) and the reliability of intra-rater service (i.e. the same patient with the same physician on different days). Kappa values can also be used to estimate the variability of .B measurements at home. Variability in patient classification can also be recorded directly as probability, as in the standard Bayesic analysis. Regardless of the measurement used to measure variability in classification, there is a direct correspondence between the variability measured in a test or a means of comparison, the thought-out uncertainty to that extent, and the erroneous classifications resulting from that uncertainty. A comparator may have an intrinsic property or boundary that allows it to render a broad distribution of values relative to the truth ground that is measured. An example is shown in Figure 1. Suppose a particular condition is characterized by a variable with a continuous normal distribution at the ground truth level, and that cutoffs have been defined to identify rare events (positive or negative calls) at the tail of the distribution. Suppose a comparator used to repeatedly measure this condition returns a Cauchy distribution of values.
Then, the distribution of measured values will have extensive tails that are not present at the level of the ground truth, which could lead to false negative calls or false negatives by means of comparison.