Detection of subclinical mastitis from on-line milking parlor data

Abstract

A model, based on automatically collected data, was developed for detection of subclinical mastitis. The logistic regression model was based on the following variables: milk electrical conductivity, milk production, parity, and DIM. Subclinical mastitis was defined as a minimal period of 1 wk in which the SCC was > 500 x 10(3) cells/ml. In contrast, periods were defined as healthy if the SCC was < 200 x 10(3) cells/ml. The resulting model had a sensitivity of 55% and specificity of 90% for individual milkings. For periods of 14 milkings, sensitivity was 54% and specificity 92% when the threshold for that period was > 6 electrical conductivity signals for high SCC. Based on these test characteristics, the model could be used as an initial screening tool in a herd with a high incidence of subclinical mastitis. Cows with a signal would have a higher probability of being diseased than the total population. In such herds, separation of milk from the signaled cows might be a possible management strategy to reduce the SCC in the bulk milk tank.