Estimation of sensitivity and specificity of pregnancy diagnosis using transrectal ultrasonography and ELISA for pregnancy-associated glycoprotein in dairy cows using a Bayesian latent class model

Abstract

AIMS: To determine the sensitivity (Se) and specificity (Sp) of pregnancy diagnosis using transrectal ultrasonography and an ELISA for pregnancy-associated glycoprotein (PAG) in milk, in lactating dairy cows in seasonally calving herds approximately 85–100 days after the start of the herd’s breeding period.

METHODS: Paired results were used from pregnancy diagnosis using transrectal ultrasonography and ELISA for PAG in milk carried out approximately 85 and 100 days after the start of the breeding period, respectively, from 879 cows from four herds in Victoria, Australia. A Bayesian latent class model was used to estimate the proportion of cows pregnant, the Se and Sp of each test, and covariances between test results in pregnant and non-pregnant cows. Prior probability estimates were defined using beta distributions for the expected proportion of cows pregnant, Se and Sp for each test, and covariances between tests. Markov Chain Monte Carlo iterations identified posterior distributions for each of the unknown variables. Posterior distributions for each parameter were described using medians and 95% probability (i.e. credible) intervals (PrI). The posterior median estimates for Se and Sp for each test were used to estimate positive predictive and negative predictive values across a range of pregnancy proportions.

RESULTS: The estimate for proportion pregnant was 0.524 (95% PrI = 0.485–0.562). For pregnancy diagnosis using transrectal ultrasonography, Se and Sp were 0.939 (95% PrI = 0.890–0.974) and 0.943 (95% PrI = 0.885–0.984), respectively; for ELISA, Se and Sp were 0.963 (95% PrI = 0.919–0.990) and 0.870 (95% PrI = 0.806–0.931), respectively. The estimated covariance between test results was 0.033 (95% PrI = 0.008–0.046) and 0.035 (95% PrI = 0.018–0.078) for pregnant and non-pregnant cows, respectively. Pregnancy diagnosis results using transrectal ultrasonography had a higher positive predictive value but lower negative predictive value than results from the ELISA across the range of pregnancy proportions assessed.

CONCLUSIONS AND CLINICAL RELEVANCE: Pregnancy diagnosis using transrectal ultrasonography and ELISA for PAG in milk had similar Se but differed in predictive values. Pregnancy diagnosis in seasonally calving herds around 85–100 days after the start of the breeding period using the ELISA is expected to result in a higher negative predictive value but lower positive predictive value than pregnancy diagnosis using transrectal ultrasonography. Thus, with the ELISA, a higher proportion of the cows with negative results will be non-pregnant, relative to results from transrectal ultrasonography, but a lower proportion of cows with positive results will be pregnant.

KEY WORDS:

• Dairy cow
• pregnancy diagnosis
• ultrasonography
• pregnancy-associated glycoprotein
• ELISA
• latent class model
• sensitivity
• specificity
• Bayesian estimation

Acknowledgements

We thank the management and staff of ACE Farming Company for use of their herds and farms, Lisa Reynolds for veterinary field work, and the NZVJ reviewers who greatly assisted with their advice.

ORCID

RW Shephard http://orcid.org/0000-0003-1780-6847

Notes

1 https://doi.org/10.1080/00480169.2017.1391723

*Non-peer-reviewed

Additional information

Funding

This work was supported by Dairy Australia and Australian dairy farmers.