Validation of accelerometers to automatically record postures and number of steps in growing lambs

• Published in: Applied animal behaviour science 2020-08, Vol.229, p.105014, Article 105014
• Main Authors: Högberg, Niclas, Höglund, Johan, Carlsson, Annelie, Saint-Jeveint, Marie, Lidfors, Lena
• Format: Article
• Language: English
• Subjects: Activity; Animal and Dairy Science; Behaviour; Husdjursvetenskap; Lamb; Sensor; Validation
• ISSN: 0168-1591; 1872-9045; 1872-9045
• DOI: 10.1016/j.applanim.2020.105014

•Evaluation of two commercially available 3D-accelerometers for use in growing lambs.•Convincing agreement between direct observations and automated measurements of activity.•The 3D-accelerometers enables long-term monitoring of lamb activity.•Enables monitoring of lying and standing time and number of lying bouts. We validated the accuracy of two commercially available activity loggers for cattle in determining lying and standing durations, number of lying bouts and number of steps in growing lambs. Ten growing lambs divided into two weight classes were fitted with an IceTag on the right hind leg and an IceQube on the left hind leg. The IceTag reports activity per second, whereas the IceQube reports activity in 15-min periods. To enable comparison between loggers, IceTag data were also summarized in 15-min periods. Computed indications for the start of a lying bout of durations >10 s and >30 s was performed to enable filtering of lying bout data. Analyses of the lambs body posture and number of steps per second from 50 h of video recordings were used as a gold standard to determine the accuracy of the two loggers. Two observers scored the two different groups and inter-observer reliability was consistent for standing, lying and number of lying bouts (κ = 0.99). However, the observers defined step count differently and no agreement was found (κ = -0.05; -0.11). Based on Bland-Altman comparison both loggers can be used to record standing and lying time. The positive predictive value (PPV), sensitivity and specificity of the IceTag compared to video recordings per second for standing and lying were all > 91.5 %. The IceTag showed a poor PPV (< 44 %) and sensitivity (< 91 %) for lying bouts, whereas the IceQube showed a better PPV (< 92 %) but somewhat lower sensitivity (< 88 %). The performance improved with the computed indications for lying bouts, for IceTag (LB_10: PPV: 100 %; sensitivity: 89 %; LB_30: PPV: 100 %; sensitivity: 100 %) and IceQube (LB_10: PPV: 98 %; sensitivity: 89 %; LB_30: PPV: 100 %; sensitivity: 100 %)), respectively. However, based on Bland-Altman comparisons, no agreement between video recording and logger recordings could be found for step count. We conclude that both loggers are able to record standing and lying time accurately. However, the ability to record number of lying bouts is poorer for the IceTag than IceQube but increases if bouts < 30 s is disregarded. Furthermore, none of the loggers should be used for step count recordi