Graduate Student Literature Review: Detecting health disorders using data from automatic milking systems and associated technologies

This review synthesizes a range of research findings regarding behavioral and production responses to health disorders and subsequent illness detection for herds using automatic (robotic) milking systems (AMS). We discuss the effects of health disorders on cow behavior and production, specifically those variables that are routinely recorded by AMS and associated technologies. This information is used to inform the resultant use of behavior and production variables and to summarize and critique current illness detection studies. For conventional and AMS herds separately, we examined research from the past 20 yr and those variables recorded automatically on-farm that may respond to development of illness and lameness. The main variables identified were milk yield, rumination time, activity, and body weight, in addition to frequency of successful, refused, and fetched (involuntary) milkings in AMS herds. Whether making comparisons within cow or between sick and healthy cows, consistent reductions in activity, rumination time, and milk yield are observed. Lameness, however, had obvious negative effects on milk yield but not necessarily on rumination time or activity. Finally, we discuss detection models for identifying lameness and other health disorders using routinely collected data in AMS, specifically focusing on their scientific validation and any study limitations that create a need for further research. Of the current studies that have worked toward disease detection, many data have been excluded or separated for isolated models (i.e., fresh cows, certain lactation groups, and cows with multiple illnesses or moderate cases). Thus, future studies should (1) incorporate the entire lactating herd while accounting for stage of lactation and parity of each animal; (2) evaluate the deviations that cows exhibit from their own baseline trajectories and relative to healthy contemporaries; (3) combine the use of several variables into health alerts; and (4) differentiate the probable type of health disorder. Most importantly, no model or software currently exists to integrate data and directly support decision-making, which requires further research to bridge the gap between technology and herd health management.