Association of clinical mastitis with milking behavior and performance of cows in an automatic milking system

The growth in the use of automatic milking systems (AMS) that provide detailed individual cow data creates the opportunity for continuous monitoring of deviations in milking behavior and performance that could be associated with diseases, such as mastitis. This study aimed to investigate the dynamics of milking behavior and performance in Holstein cows before and on the day of clinical mastitis diagnosis occurring during the first third of lactation and to compare these patterns with those observed in healthy cows under the same conditions. Data included a total of 2,913 lactations from cows that calved between January 2023 and April 2024, comprising 367,686 milking events from 2,710 cows (including 203 cows with 2 lactations) milked using 62 AMS. Milking behaviors included incomplete milking, milking unit kick-off, teat not found, and milking interval (h). Performance variables were milk yield (kg), box time (min), box time by milk yield ratio (min/kg), milk flow duration (min), mean milk flow rate (kg/min), and peak milk flow rate (kg/min). Mastitis cases were categorized by DIM at diagnosis into early (0 to 30 DIM; MAS1), mid (31 to 60 DIM; MAS31), and late (61 to 100 DIM; MAS61) periods. The early period analyzed milking data from 1 to 30 DIM, while for the MAS31 and MAS61, milking data included -14 to 0 d relative to the mastitis diagnosis. A healthy group was created as a reference for comparisons in each period. Multivariable mixed linear regression models and multivariable mixed logistic regression models were fitted using milking variables as the outcomes of interest. Compared with healthy controls, MAS1 cows had greater probabilities of incomplete milking and teat not found, and extended milking intervals. Affected cows had smaller milk yield, longer box time, greater box time by milk yield ratio, longer milk flow durations, and reduced mean and peak milk flow rates than the healthy group through the early period. Similarly, MAS31 and MAS61 had a significant increase in the probabilities of incomplete milking and teat not found, and extended milking intervals, box time, box time by milk yield ratio, and milk flow duration before and at the time of diagnosis. These 2 groups of cows also showed a sharp decline in milk yield, mean milk flow rate, and peak milk flow rate before and on the diagnosis day of clinical mastitis. Milking behavior and performance were altered in cows that experienced mastitis during the first third of lactation. These variables may serve as early indicators for detecting mastitis in cows managed under AMS.