Agreement between different sleep states and behaviour indicators in dairy cows

• We examined if behaviour estimates can be used to identify sleep for dairy cows.
• Data from behaviour observations were compared with electrophysiological data.
• Behavioural indicators greatly overestimated total sleep time in dairy cows.
• Behavioural indicators for REM sleep could identify REM sleep with good precision.
• There is a risk that REM sleep indicator underestimated total REM sleep time.

Conclusive data regarding behavioural indicators of different sleep states in adult dairy cows are lacking, i.e. agreement between behavioural indicators of sleep and corresponding electrophysiological measures. Behavioural estimates for quantifying total sleep time in calves have been developed, so this study examined whether these behavioural estimates also apply for adult cows.Behaviour observations and electrophysiological readings were recorded for a total of 13 cows during one recording session per cow lasting on average 4 h 22 min. Recording started when the cow was fully awake and finished when at least one probable sleep bout had been recorded. The behavioural estimates used in the study were: ‘lying with head lifted and still’ for non-rapid eye movement (NREM) sleep, ‘lying with head resting’ for rapid eye movement (REM) sleep and ‘lying with head lifted and moving’ for awakeness. As statistical measures of agreement between behavioural estimates and electrophysiological status (both recorded at 30 s intervals), Cohen's kappa as well as sensitivity and specificity measures were calculated. Additionally, misclassifications were evaluated to better understand agreement between the behaviour and electrophysiological sleep classification. Since interval length might have affected the agreement, the output data were also aggregated into 60, 90 and 120 s intervals and analysed using Wilcoxon sign-rank test to determine the most appropriate interval length.It was found that the behavioural estimates for assessing total sleep time in calves could not be applied to adult cows as they markedly overestimated NREM and REM sleep time. Behavioural estimates for NREM and REM sleep time were on average 124 ± 17 and 14 ± 4 min per cow, respectively, while the electrophysiological estimate for NREM and REM was on average 20 ± 5 and 10 ± 3 min per cow, respectively. Using the behavioural estimate ‘lying with head resting’, REM sleep could be identified with moderate precision, but this indicator alone likely underestimates total duration of REM sleep. Behavioural estimates for NREM sleep showed high sensitivity (81%) but low specificity (6%) while the behavioural estimates for REM sleep showed high sensitivity (70%) and moderate specificity (41%). For both categories, both sensitivity and specificity increased with increasing measurement interval length. Drowsing as identified from electrophysiological data was present mainly when the cows were lying with head lifted and still, whereas awakeness was present when the cows were lying with head lifted and moving.