Automated milking systems: using productivity and behavioral data to detect illness in dairy cows

Abstract Text:

To develop better ways of using milking activity, productivity, and behavioral data to detect illness, we collected longitudinal data throughout the lactation of 57 Holstein dairy cows (19 PP, 38 MP; 3.1±1.1 lactations) housed in a free-stall barn equipped with an automated milking system (AMS). Cases of illness were recorded, including sub-clinical ketosis (SCK; n=23), calving-related disease (CRD; n=14), hoof disorders and severe lameness (n=16), pneumonia (n=8), and gastrointestinal issues and displaced abomasum (DA; n=7). We collected continuous milking activity data from the AMS. Lying, rumination, and activity data were recorded by electronic data loggers. Data were analyzed in repeated measures mixed linear regression models. Days relative to the day of diagnosis/treatment (day 0) were analyzed as a fixed effect for each illness separately, with data extending back to day -14. Analyses were performed between: (a) the day from which each outcome variable deviated significantly from baseline production/behavior (Tukey’s tests were used to make day-by-day comparisons), and (b) day -1, since recovery had begun following treatment on day 0. Outcome variables tested were milk yield (3-d rolling average), daily rumination time, activity (unit-less measure of head and neck motion), and lying behavior (lying time, bout frequency, bout length). Mean milk production declined by 4.3 and 4.1 kg/d from day -4 to diagnosis of DA (P<0.001) and pneumonia (P=0.01), respectively. From day -14 to diagnosis of hoof disorders, production steadily declined by 0.6 kg/d (P<0.001). Mean rumination time declined by 54 and 55 min/d from day -5 to diagnosis of DA (P<0.001) and pneumonia (P=0.03), respectively. Before SCK diagnosis (2 tests/fresh cow ~1 week apart), rumination decreased by 13 min/d from day -6 to diagnosis (P=0.05); this was most drastic day -3 to -1 (-34 min/d; P=0.001). Activity levels declined by 40 units/d from day -4 to diagnosis of DA (P<0.001), but decreased gradually from day -14 to diagnosis of SCK (-15 units/d; P<0.001) and CRD (-23 units/d; P<0.001). Lying behavior was less predictive of illness, as it did not vary until the day of diagnosis of any illness. These results suggest that the effects of illness on rumination, activity, and productivity are apparent several days before diagnosis and could be used to earlier identify illness in AMS herds. Since behavior and productivity appear to respond differently to various types of illness, it is possible that certain parameters may be illness-specific.

Keywords:

automated milking, dairy cow behavior, illness