Accuracy of leukocyte esterase test, endometrial cytology and vaginal discharge score for diagnosing postpartum reproductive tract health status in dairy cows at the moment of sampling, using a latent class model fit within a Bayesian framework

The objectives of this retrospective study were: 1) to determine the sensitivity (Se) and specificity (Sp) of leukocyte esterase test (LE), endometrial cytology (CYTO) and vaginal discharge score (VDS) for diagnosis of reproductive tract diseases in dairy cows at the moment of sampling; 2) to assess the impact of different thresholds on test accuracy and misclassification costs; and 3) to quantify herd prevalence of reproductive tract diseases in dairy farms from Quebec, Canada. Data from 2092 cows (39 herds) enrolled in two randomized control trials were included. Cows were examined at 35 (+/- 7) days in milk using LE, CYTO, and VDS to determine their reproductive tract health status. A latent class model assuming conditional dependence of CYTO and LE was fit within a Bayesian framework. Non informative priors were used for the Se and Sp of LE, CYTO, and VDS, while prior information for disease prevalence was obtained from expert opinions (mode = 20%, 5th percentile = 10%). Posterior inferences (median and 95% Bayesian credible intervals; BCI) were obtained using the WinBUGS statistical freeware. An initial model was built using thresholds of >= 1 (small amount of leukocytes), of >= 6%, and of >= 3 (mucopurulent discharge) for the LE, CYTO, and VDS, respectively. Then, the impact on tests accuracy and misclassification costs of using different thresholds was explored. Optimal thresholds balancing the need for good antimicrobial stewardship and animal health considerations were proposed. The optimal thresholds obtained in the final model were: LE >= 2 (moderate amount of leukocytes), CYTO >= 6%, and VDS >= 2 (mucus with flecks of pus). In the final model, median (95% BCI) Se for LE, CYTO and VDS were 42.6% (38.8-47.0), 45.9% (41.9-50.7), and 64.2% (59.1-70.3), respectively. Median Sp was 90.9% (88.0-93.6) for LE, 92.2% (89.2-94.9) for CYTO and 96.6% (91.3-99.8) for VDS. Median within-herd prevalence of reproductive tract disease was estimated at 47.9% (39.0-56.7). At a threshold >= 2 (mucus with flecks of pus), VDS had the highest Se and Sp among the tests evaluated. The LE is an interesting diagnostic alternative for detecting reproductive tract disease in dairy cows. At a threshold >= 2 (moderate amount of leukocytes), LE Se and Sp were comparable to those of CYTO. This is the first study reporting the accuracy of LE, CYTO, and VDS for diagnosing reproductive tract diseases in dairy cows at the moment of sampling.