Searching in the dark: why diagnostic test performance is essential to interpreting surveillance results, using <i>Phytophthora agathidicida</i> as a case study

<p dir="ltr">Accurate and precise estimations of the performance of diagnostic tests are needed to design and interpret surveillance. Kauri (<i>Agathis australis</i>), a keystone forest species, is threatened by kauri dieback, caused by <i>Phytophthora agathidicida</i>. Test performance is measured by the diagnostic sensitivity (probability a tree that does have <i>P. agathidicida</i> returns a positive test) and diagnostic specificity (probability a tree that does not have <i>P. agathidicida</i> returns a negative test). In the absence of a perfect test, sensitivity and specificity can be calculated using Bayesian latent class analysis (BLCA). A cross-sectional study of the baseline prevalence of <i>P. agathidicida</i> in 761 randomly selected kauri trees in the Waitākere Ranges was undertaken in 2021. Visual assessment of symptoms and morphological identification from soil bioassay baiting and culture were used as ‘tests’ to infer presence of <i>P. agathidicida</i>. A BLCA model was built using prior expert opinion from collaborators on the tests’ performance, and on high versus low pathogen prevalence areas within the study area. For visual assessment, the estimated sensitivity was 41.0% (95% PI 29.8-53.3) and specificity was 87.0% (95% PI 84.0-89.8). For the morphological test the sensitivity was 63.2% (95% PI 42.6-88.1) and specificity was 98.7% (95% PI 96.8-99.8). Morphological test performance values can be used to estimate the true prevalence of <i>P. agathidicida</i> based on apparent prevalence, understand areas of pathogen freedom, interpret historical results and account for low test sensitivity when designing surveillance programmes and calculating sample sizes.</p>