Predicting sensitivity of repeated environmental sampling for Mycobacterium avium subsp. paratuberculosis in dairy herds using a Bayesian latent class model

Between-herd transmission of Mycobacterium avium subsp. paratuberculosis (MAP) by subclinically infected cattle is an important risk which can hamper effective control of paratuberculosis. Knowledge of herd status would substantially reduce this risk; MAP positive farms can be detected with environmental sampling. The objective of this study was to compare cumulative sensitivities of annual environmental sampling with two or four samples per sampling event without knowledge of true herd status and to calculate the number of sampling events to achieve a cumulative sensitivity of at least 0.9. Data from three repeated sampling events in two study populations, one with 55 herds (two samples/event) and another with 30 herds (four samples/event) including test results, herd and sample characteristics and prior prevalence estimates, were derived from the Alberta Johne's Disease Initiative (Alberta, Canada). A recursive Bayesian latent class model was used to predict the cumulative sensitivity of repeated environmental sampling events. A sampling scheme with four samples per sampling event had a higher cumulative sensitivity than an alternative scheme with two samples. To achieve a cumulative sensitivity of at least 0.9 with 95% probability, eight sampling events with two environmental samples per set, or four sampling events with four samples per set were required. Further model assessment demonstrated that these results can only be generalized to cattle populations with a similar withinherd prevalence to those studied here (approximately 0.08). Nonetheless, these results could help predict herdlevel prevalence in cattle populations after environmental testing and provide information regarding the uncertainty behind status estimates for herds repeatedly tested using environmental samples.