Estimation of sensitivity and specificity of bulk tank milk PCR and 2 antibody ELISA tests for herd-level diagnosis of Mycoplasma bovis infection using Bayesian latent class analysis.

ABSTRACT

Mycoplasmosis (due to infection with Mycoplasma bovis) is a serious disease of beef and dairy cattle that can adversely impact health, welfare and productivity (Maunsell et al. (2011)). Mycoplasmosis can lead to a range of often severe, clinical presentations. Mycoplasma bovis (M. bovis) infection can present either clinically or subclinically, with the potential for recrudescence of shedding in association with stressful periods. Infection can be maintained within herds because of intermittent shedding (Calcutt et al., 2018, Hazelton et al., 2018). M. bovis is recognized as poorly responsive to treatment which represents a major challenge for control in infected herds. Given this, particular focus is needed on biosecurity measures to prevent introduction into uninfected herds in the first place. A robust and reliable laboratory test for surveillance is important both for herd-level prevention and control. The objective of this study was to estimate the sensitivity and specificity of 3 diagnostic tests (one PCR and 2 ELISA tests) on bulk tank milk, for the herd-level detection of M. bovis using Bayesian latent class analysis. In autumn 2018, bulk tank milk samples from 11,807 herds, covering the majority of the main dairy regions in Ireland had been submitted to the Department of Agriculture testing laboratory for routine surveillance were made available. A stratified random sample approach was used to select a cohort of herds for testing from this larger sample set. A final study population of 728 herds had bulk tank milk samples analyzed using a Bio-X ELISA (ELISA 1), an IDvet ELISA (ELISA 2) and a PCR test. A Bayesian latent class analysis (BLCA) was conducted to estimate the sensitivity (Se) and specificity (Sp) of the 3 diagnostic tests applied to bulk tank milk (BTM) for the detection of the herd-level infection. An overall LCA was conducted on all herds within a single population (a 3-test, 1-population model). The herds were also split into 2 populations based on herd size (small herds had < 82 cattle) (a 3-test, 2-population model) and separately into 3 regions in Ireland (Leinster, Munster and Connacht/Ulster) (a 3-test, 3-population model). The latent variable of interest was the herd-level M. bovis infection status. In total, 363/728 (50%) were large herds, 7 (1.0%) were positive on PCR, 88 (12%) positive on ELISA 1, and 406 (56%) positive on ELISA 2. Based on the 2-population model, the sensitivity (95% Bayesian credible interval (BCI) was 0.03 (0.02, 0.05), 0.22 (0.18, 0.27), 0.94 (0.88, 0.98) for PCR, ELISA 1 and ELISA 2 respectively. The specificity (95% BCI) was 0.99 (0.99, 1.0), 0.97 (0.95, 0.99), and 0.92 (0.86, 0.97) for PCR, ELISA 1 and ELISA 2 respectively. The herd-level true prevalence was estimated at 0.43 (BCI 0.35, 0.5) for smaller herds. The true prevalence was estimated at 0.62 (BCI 0.55, 0.69) for larger herds. The true prevalence was estimated at 0.56 (BCI 0.49, 0.463) in the 1-population model. For the 3-population model, the sensitivity (95% BCI) was 0.03 (0.02, 0.05), 0.24 (0.18, 0.29), 0.95 (0.9, 0.98) for PCR, ELISA 1 and ELISA 2 respectively. The specificity (95% BCI) was 0.99 (0.99, 1.0), 0.98 (0.96, 0.99), and 0.88 (0.79, 0.95) for PCR, ELISA 1 and ELISA 2 respectively. The herd-level true prevalence (95% BCI) was estimated at 0.65 (0.56, 0.73), 0.38 (0.28, 0.46) and 0.53 (0.4, 0.65) for population 1, 2, 3 respectively. Across all 3 models, the range in true prevalence was 38% to 65% of Irish dairy herds infected with M. bovis. The operating characteristics vary substantially between tests. The IDvet ELISA had a relatively high Se (the highest Se of the 3 tests studied) but it was estimated at 0.95 at its highest in 3-test, 3-population model. This test may be an appropriate test for herd-level screening or prevalence estimation within the context of the endemically infected Irish dairy cattle population. Further work is required to optimize this test and its interpretation when applied at herd-level to offset concerns related to the lower than optimal test Sp.