Use of SCC in genetic selection for reduced incidence of mastitis: A mixture model approach

Mixture distributions consist of two or more sub-distributions (groups), where group memberships of the observations are unknown. The udder health status of each cow at the test-days on which SCC are recorded are usually unknown. Further, as SCC increases as a result of udder infection, the observed SCC/SCS can be assumed being a two (or more)- component mixture depending on mastitis status (e.g, healthy/mastitic). Using a mixture model for statistical analysis of observed test-day SCS implies that the observations can be categorized into putative disease categories according to the probability of disease, given the observed data and all other parameters in the model. A Bayesian hierarchical twocomponent normal mixture model for SCS was developed, where baseline SCS is associated with some fixed and random effects, and test-day health status is assumed fully determined by an underlying liability to mastitis. The prior probability of mastitis may vary between different sub-groups, such as herd, stage of lactation, lactation number, etc. Hence, the liability may be associated by some fixed and random effects distinct from those affecting baseline SCS. Based on analysis of simulated data, the model seemingly gives unbiased estimates of all parameters, and the predicted breeding values for liability to mastitis seem better suited for genetic selection than crudely selecting for lower SCS. The model also provides estimates of test-day probability of mastitis, based on the recorded SCS, which may be used for disease detection. The proposed model could easily be extended to handle a wider range of problems related to statistical analyses of mixture traits.