The objective of this study was to analyze whether changes in behavior can be a good early predictor of sickness in calves. Friesian males calves (n = 325; 30 ± 9 d of age; 65 ± 15 kg) were monitored with an activity-monitoring device from 30 to 90 d of life in 4 periods corresponding to 4 seasons. The activity-monitoring device measured number of steps, number of lying bouts, lying time, and frequency and time of visits to the feed bunk. Calf health status was monitored daily and all incidences were recorded. To compare sick and healthy calves, a matched pair design was used to assign calves into the healthy group. Day 0 was defined as the day of sickness diagnosis. For each sick calf, 3 calves with no signs of sickness during the entire period (healthy calves) on the same date, in the same season, and of similar age (±4 d) and weight at entry were identified. A multivariate linear mixed model was used from d -10 to +10 relative to the sickness diagnosis to describe differences between sick and healthy calves. A multivariate logistic regression model was used for predicting sick calves on the days before the diagnosis. Significance was declared at P < 0.05. Daily, healthy calves had 1,476 ± 195 steps, spent 185 ± 32.5 min at the feed bunk, consumed 10 ± 1.1 meals, had 19.5 ± 1.8 lying bouts, and spent an average of 978 ± 30.5 min lying. The difference in behavior between sick (n = 33) and healthy calves (n = 99) began to be evident on d -10. Sick calves had fewer steps and numbers of visits to the feed bunk on d -1 and 0 and spent less time at the feed bunk on d -10 and -1 compared with healthy calves. From d -2 to d 9, sick calves had 15% fewer lying bouts, with no difference in lying time except on d -10, when sick calves spent more time lying. The best prediction model was for d -1 and included season and age at entry as qualifying variables, and frequency of visits to the feed bunk, steps, and lying time as behavior predictors (69% sensitivity, 72% specificity, 72% accuracy, 55% false discovery rate, and 12% false omission rate). However, an earlier prediction would be more useful to reduce the negative effect of sickness on production and welfare. The prediction model for d -10 had 67% sensitivity, 67% specificity, 67% accuracy, 60% false discovery rate, and 14% false omission rate. Results indicate that the occurrence of sickness can be predicted in advance, and an automated alarm system could be used to identify calves at risk of becoming sick and apply a preventive treatment.
Read full abstract