Dufour Simon, Durocher Jean, Dubuc Jocelyn, Dendukuri Nandini, Hassan Shereen, Buczinski Sébastien
Département de pathologie et microbiologie, Faculté de médecine vétérinaire, Université de Montréal, C. P. 5000, Saint-Hyacinthe, QC, J2S 7C6, Canada.
Valacta, 555 boul. des Anciens-Combattants, Sainte-Anne-de Bellevue, QC, H9X 3R4, Canada.
Prev Vet Med. 2017 May 1;140:122-133. doi: 10.1016/j.prevetmed.2017.03.008. Epub 2017 Mar 28.
Using a milk sample for pregnancy diagnosis in dairy cattle is extremely convenient due to the low technical inputs required for collection of biological materials. Determining accuracy of a novel pregnancy diagnostic test that relies on a milk sample is, however, difficult since no gold standard test is available for comparison. The objective of the current study was to estimate diagnostic accuracy of the milk PAG-based ELISA and of transrectal ultrasonographic (TUS) exam for determining pregnancy status of individual dairy cows using a methodology suited for test validation in the absence of gold standard. Secondary objectives were to evaluate whether test accuracy varies with cow's characteristics and to identify the optimal ELISA optical density threshold for PAG test interpretation. Cows (n=519) from 18 commercial dairies tested with both TUS and PAG between 28 and 45days following breeding were included in the study. Other covariates (number of days since breeding, parity, and daily milk production) hypothesized to affect TUS or PAG test accuracy were measured. A Bayesian hierarchical latent class model (LCM) methodology assuming conditional independence between tests was used to obtain estimates of tests' sensitivities (Se) and specificities (Sp), to evaluate impact of covariates on these, and to compute misclassification costs across a range of ELISA thresholds. Very little disagreement was observed between tests with only 23 cows yielding discordant results. Using the LCM model with non-informative priors for tests accuracy parameters, median (95% credibility intervals [CI]) TUS Se and Sp estimates of 0.96 (0.91, 1.00) and 0.99 (0.97, 1.0) were obtained. For the PAG test, median (95% CI) Se of 0.99 (0.98, 1.00) and Sp of 0.95 (0.89, 1.0) were observed. The impact of adjusting for conditional dependence between tests was negligible. Test accuracy of the PAG test varied slightly by parity number. When assuming false negative to false positive costs ratio≥3:1, the optimal ELISA optical density threshold allowing minimization of misclassification costs was 0.25. In conclusion, both TUS and PAG showed excellent accuracy for pregnancy diagnosis in dairy cows. When using the PAG test, a threshold of 0.25 could be used for test interpretation.
由于采集生物材料所需的技术投入较低，使用牛奶样本对奶牛进行妊娠诊断极为方便。然而，由于没有可用于比较的金标准检测方法，确定一种依赖牛奶样本的新型妊娠诊断检测的准确性很困难。本研究的目的是使用一种适用于在没有金标准的情况下进行检测验证的方法，估计基于牛奶PAG的ELISA和经直肠超声检查（TUS）对个体奶牛妊娠状态诊断的准确性。次要目标是评估检测准确性是否随奶牛特征而变化，并确定用于PAG检测结果判读的最佳ELISA光密度阈值。本研究纳入了来自18个商业奶牛场的519头奶牛，在配种后28至45天同时进行了TUS和PAG检测。测量了其他假设会影响TUS或PAG检测准确性的协变量（配种后天数、胎次和每日产奶量）。采用一种假设检测之间条件独立的贝叶斯分层潜在类别模型（LCM）方法，以获得检测敏感性（Se）和特异性（Sp）的估计值，评估协变量对这些指标的影响，并计算一系列ELISA阈值下的错误分类成本。检测之间观察到的不一致很少，只有23头奶牛的检测结果不一致。使用对检测准确性参数具有非信息先验的LCM模型，获得TUS的Se和Sp估计值中位数（95%可信区间[CI]）分别为0.96（0.91，1.00）和0.99（0.97，1.0）。对于PAG检测，观察到Se中位数（95%CI）为0.99（0.98，1.00），Sp为0.95（0.89，1.0）。调整检测之间的条件依赖性的影响可忽略不计。PAG检测的准确性随胎次略有变化。当假设假阴性与假阳性成本比≥3:1时，使错误分类成本最小化的最佳ELISA光密度阈值为0.25。总之，TUS和PAG在奶牛妊娠诊断中均显示出优异的准确性。使用PAG检测时，可将阈值0.25用于检测结果判读。
