Use of a hidden Markov model for interpretation of serial cow milk paratuberculosis antibody enzyme-linked immunosorbent assay results adjusted for milk yield and quality.

Abstract

Paratuberculosis (Johne's disease), caused by Mycobacterium avium subsp. paratuberculosis (MAP), is a common, economically-important and potentially zoonotic contagious disease of cattle, with worldwide distribution. Disease management relies on identification of animals which are at high-risk of being infected or infectious. The disease is chronic in nature, and infected animals may be infectious in the absence of overt clinical signs. Coupled with limited sensitivity of available diagnostic tests, this creates difficulties in identifying high-risk animals. In some disease-control programmes, dairy cows are classified with regards to risk according to the results of serial tests which quantify MAP antibodies in milk samples. Such classification systems are limited by the influence of non-disease factors on test results, dichotomisation of continuous results into "positive" or "negative" according to an imperfect threshold, and subjectivity in defining which patterns of serial test results indicate different risk-categories. An unsupervised learning (clustering) approach was applied to paratuberculosis test results and milk-recording data collated from 47 farms over an approximately ten-year period between 2010 and 2021. Paratuberculosis test results were first adjusted according to influential non-disease factors using linear models. Continuous-time hidden Markov models were fit to the adjusted test results. The final model revealed four distinct latent states (clusters). Examination of the distribution of adjusted test results associated with each latent state suggested that states were ordinal and aligned with disease progression. Model transition probabilities demonstrated that the probability of an animal progressing to the highest state was dependent on its current state. Of particular note was the existence of a latent state, characterised by paratuberculosis test results below the conventional test-positive threshold, which was associated with a relatively high probability of progression to the highest cluster. This research has led to objective classification of animals according to serial test results, and furthermore suggests the presence of groups of different disease risk amongst animals whose test results fall below the routinely used test-positive threshold. Identification of such groups could be used to better manage disease on farms, through implementation of management practices which limit disease transmission from high-risk animals.