Impacts of heat stress on the accuracy of a noseband sensor for detection of eating and rumination behavior in confined cattle

Precision monitoring of feeding behaviors can aid in dairy herd management. Noseband sensors (RumiWatch System RW; Itin+Hoch GmbH) have been established as an automated gold standard for evaluating precision technologies in grazing cows, but more advanced algorithms have not been validated in confinement settings. Additionally, little is known regarding effects of environmental conditions on sensor performance. Therefore, accuracy of RW in quantifying eating and rumination time in confinement was evaluated using 2 versions of the analysis software algorithms (RW Converter V. 7. 3. 2 and V. 7. 3. 36) under both thermoneutral (TN; 21. 0°C, 64. 0% relative humidity RH, temperature-humidity index THI = 67) and heat stress conditions (HS; cyclical daily temperatures to mimic diurnal patterns; 0700 – 1900 h: 33. 6°C, 40. 0% RH, THI = 83. 5; 1900 – 0700 h: 23. 2°C, 70. 0% RH; THI = 70. 3). Nine individually-housed Holstein x Simmental cross steers were fitted with RW noseband sensors. Agreement for eating time reported by RW and visual observations (1-min scan sampling) was very high in TN regardless of software version (concordance correlation coefficient CCC: V. 7. 3. 2 = 0. 91; V. 7. 3. 36 = 0. 94), and remained high to very high (CCC: V. 7. 3. 2 = 0. 89; V. 7. 3. 36 = 0. 95) during HS. Agreement for rumination time was very high regardless of software version in both TN (CCC: V. 7. 3. 2 = 0. 93; V. 7. 3. 36 = 0. 99) and HS (CCC: V. 7. 3. 2 = 0. 91; V. 7. 3. 36 = 0. 99). Overall, RW accurately quantified eating and ruminating time in confined cattle, and noseband sensors retained accuracy during heat stress. These results indicate RW may serve as a benchmark for future precision technology validations in dairy cattle managed in confinement systems. Precision monitoring of feeding behaviors can aid in dairy herd management. Noseband sensors (RumiWatch System RW; Itin+Hoch GmbH) have been established as an automated gold standard for evaluating precision technologies in grazing cows, but more advanced algorithms have not been validated in confinement settings. Additionally, little is known regarding effects of environmental conditions on sensor performance. Therefore, accuracy of RW in quantifying eating and rumination time in confinement was evaluated using 2 versions of the analysis software algorithms (RW Converter V. 7. 3. 2 and V. 7. 3. 36) under both thermoneutral (TN; 21. 0°C, 64. 0% relative humidity RH, temperature-humidity index THI = 67) and heat stress conditions (HS; cyclical daily temperatures to mimic diurnal patterns; 0700 – 1900 h: 33. 6°C, 40. 0% RH, THI = 83. 5; 1900 – 0700 h: 23. 2°C, 70. 0% RH; THI = 70. 3). Nine individually-housed Holstein x Simmental cross steers were fitted with RW noseband sensors. Agreement for eating time reported by RW and visual observations (1-min scan sampling) was very high in TN regardless of software version (concordance correlation coefficient CCC: V. 7. 3. 2 = 0. 91; V. 7. 3. 36 = 0. 94), and remained high to very high (CCC: V. 7. 3. 2 = 0. 89; V. 7. 3. 36 = 0. 95) during HS. Agreement for rumination time was very high regardless of software version in both TN (CCC: V. 7. 3. 2 = 0. 93; V. 7. 3. 36 = 0. 99) and HS (CCC: V. 7. 3. 2 = 0. 91; V. 7. 3. 36 = 0. 99). Overall, RW accurately quantified eating and ruminating time in confined cattle, and noseband sensors retained accuracy during heat stress. These results indicate RW may serve as a benchmark for future precision technology validations in dairy cattle managed in confinement systems. Precision monitoring of eating and/or rumination behavior can assist with detection of disease and illness, aid in reproductive management, and inform decision-making related to nutrition and feeding in dairy herds (Borchers and Bewley, 2015Borchers M. R. Bewley J. M. An assessment of producer precision dairy farming technology use, prepurchase considerations, and usefulness. J. Dairy Sci. 2015; 98 (25892693): 4198-4205https: //doi. org/10. 3168/jds. 2014-8963Abstract Full Text Full Text PDF PubMed Scopus (91) Google Scholar; Silva et al. , 2021Silva S. R. Araujo J. P. Guedes C. Silva F. Almeida M. Cerqueira J. L. Precision technologies to address dairy cattle welfare: Focus on lameness, mastitis and body condition. Animals (Basel). 2021; 11 (34438712): 2253https: //doi. org/10. 3390/ani11082253Crossref Scopus (34) Google Scholar). Traditional visual observation methods for validating precision technologies are time and labor intensive (Werner et al. , 2019Werner J. Umstatter C. Leso L. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'brien B. Evaluation and application potential of an accelerometer-based collar device for measuring grazing behavior of dairy cows. Animal. 2019; 13 (30739632): 2070-2079https: //doi. org/10. 1017/S1751731118003658Crossref PubMed Scopus (30) Google Scholar; Pereira et al. , 2020Pereira G. M. Heins B. J. O'Brien B. McDonagh A. Lidauer L. Kickinger F. Validation of an ear tag–based accelerometer system for detecting grazing behavior of dairy cows. J. Dairy Sci. 2020; 103 (32089298): 3529-3544https: //doi. org/10. 3168/jds. 2019-17269Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar). Therefore, identification and establishment of automated technologies as gold standards against which other emerging technologies may be compared is beneficial (Pereira et al. , 2021Pereira G. M. Sharpe K. T. Heins B. J. Evaluation of the RumiWatch system as a benchmark to monitor feeding and locomotion behaviors of grazing dairy cows. J. Dairy Sci. 2021; 104 (33455761): 3736-3750https: //doi. org/10. 3168/jds. 2020-18952Abstract Full Text Full Text PDF PubMed Scopus (16) Google Scholar). The noseband sensor-based technology, the RumiWatch System (RW; Itin + Hoch, GmbH, Liestal, Switzerland) has shown high precision and accuracy in reporting eating and rumination time in pasture-managed dairy cattle (Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar; Pereira et al. , 2021Pereira G. M. Sharpe K. T. Heins B. J. Evaluation of the RumiWatch system as a benchmark to monitor feeding and locomotion behaviors of grazing dairy cows. J. Dairy Sci. 2021; 104 (33455761): 3736-3750https: //doi. org/10. 3168/jds. 2020-18952Abstract Full Text Full Text PDF PubMed Scopus (16) Google Scholar). The RW has served as benchmark for validating ear tag and neck collar accelerometers in grazing cows (Werner et al. , 2019Werner J. Umstatter C. Leso L. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'brien B. Evaluation and application potential of an accelerometer-based collar device for measuring grazing behavior of dairy cows. Animal. 2019; 13 (30739632): 2070-2079https: //doi. org/10. 1017/S1751731118003658Crossref PubMed Scopus (30) Google Scholar; Pereira et al. , 2020Pereira G. M. Heins B. J. O'Brien B. McDonagh A. Lidauer L. Kickinger F. Validation of an ear tag–based accelerometer system for detecting grazing behavior of dairy cows. J. Dairy Sci. 2020; 103 (32089298): 3529-3544https: //doi. org/10. 3168/jds. 2019-17269Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar;). Earlier RW converter algorithms were capable of accurately reporting feeding behavior of confined cattle (Ruuska et al. , 2016Ruuska S. Kajava S. Mughal M. Zehner N. Mononen J. Validation of a pressure sensor-based system for measuring eating, rumination and drinking behaviour of dairy cattle. Appl. Anim. Behav. Sci. 2016; 174: 19-23https: //doi. org/10. 1016/j. applanim. 2015. 11. 005Crossref Google Scholar; Zehner et al. , 2017Zehner N. Umstätter C. Niederhauser J. J. Schick M. System specification and validation of a noseband pressure sensor for measurement of ruminating and eating behavior in stable-fed cows. Comput. Electron. Agric. 2017; 136: 31-41https: //doi. org/10. 1016/j. compag. 2017. 02. 021Crossref Scopus (112) Google Scholar), but development of the more recent V. 7. 3. 36 utilized in the above-mentioned studies focused on improved performance in grazing applications (Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar). This newer version has the advantage of differentiating time spent eating with the head up or down. Validating accuracy of these advanced algorithms in confinement could provide greater insights when using RW as a benchmark while also generating more comprehensive information regarding feeding behaviors. Additionally, little is known about impacts of environmental conditions on accuracy of precision management technologies. Cattle subjected to elevated temperature-humidity indices THI may exhibit changes in head position and movement as well as patterns of feeding behavior (Islam et al. , 2020Islam M. A. Lomax S. Doughty A. K. Islam M. R. Clark C. E. Automated monitoring of panting for feedlot cattle: sensor system accuracy and individual variability. Animals (Basel). 2020; 10 (32867225): 1518https: //doi. org/10. 3390/ani10091518Crossref Scopus (12) Google Scholar; Islam et al. , 2021Islam M. A. Lomax S. Doughty A. K. Islam M. R. Jay O. Thomson P. Clark C. E. Automated monitoring of cattle heat stress and its mitigation. Front. Anim. Sci. 2021; 2: 60https: //doi. org/10. 3389/fanim. 2021. 737213Crossref Scopus (13) Google Scholar). Robustness of precision technologies to heat stress is of particular relevance when viewed in the context of rising global temperature and humidity due to climate change (Intergovernmental Panel on Climate Change (IPCC), 2021Intergovernmental Panel on Climate Change (IPCC) Summary for Policymakers. in: Masson-Delmotte V. Zhai P. Pirani A. Connors S. L. Péan C. Berger S. Caud N. Chen Y. Goldfarb L. Gomis M. I. Huang M. Leitzell K. Lonnoy E. Matthews J. B. R. Maycock T. K. Waterfield T. Yelekçi O. Yu R. Zhou B. Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom2021Available online at: https: //www. ipcc. ch/report/ar6/wg1/downloads/report/IPCCAR6WGIFullReport. pdfDate accessed: November 9, 2023Google Scholar). Therefore, the objective of this study was to evaluate accuracy of the RW noseband sensor technology in quantifying eating and rumination time in confined cattle under both thermoneutral and heat stress conditions. It was hypothesized that sensor performance would be robust, even when cattle were subjected to heat stress. Procedures and use of animals were approved by the University of Kentucky Institutional Animal Care and Use Committee (Protocol # 2023–4282). Housing and animal care were consistent with the Guide to Care and Use of Agricultural Animals in Research and Teaching (FASS, 2020FASS Guide for the care and use of agricultural animals in research and teaching. FASS Inc. , Champaign, IL, USA2020Google Scholar). Nine Holstein x Simmental cross steers (488 ± 9 kg) were housed in a temperature and humidity-controlled room with randomly assigned individual (3 × 3 m) pens within the Intensive Research Building at the C. Oran Little Research Center in Versailles, KY (October 2023). Before the study, steers were adapted (14 d) to both housing and diet (corn silage + dried distillers' grains with solubles, ad libitum). Steers were subjected to 2 consecutive 7-d periods: 1. ) thermoneutral (TN; 21. 0°C, 64. 0% relative humidity RH, THI = 67) and 2. ) heat stress (HS; cyclical daily temperatures to mimic diurnal patterns; 0700 – 1900 h: 33. 6°C, 40. 0% RH, THI = 83. 5; 1900 – 0700 h: 23. 2°C, 70. 0% RH; THI = 70. 3). The THI was calculated using the following equation: THI = 0. 81 tdb + RH (tdb – 14. 4) + 46. 4, where tbd = dry bulb temperature (Hahn, 1999Hahn G. L. Dynamic responses of cattle to thermal heat loads. J. Anim. Sci. 1999; 77 (15526777): 10-20https: //doi. org/10. 2527/1997. 77suppl₂10xCrossref PubMed Google Scholar; Islam et al. , 2021Islam M. A. Lomax S. Doughty A. K. Islam M. R. Jay O. Thomson P. Clark C. E. Automated monitoring of cattle heat stress and its mitigation. Front. Anim. Sci. 2021; 2: 60https: //doi. org/10. 3389/fanim. 2021. 737213Crossref Scopus (13) Google Scholar), with THI threshholds for heat stress as previously defined (Eigenberg et al. , 2005Eigenberg R. A. Brown-Brandl T. M. Nienaber J. A. Hahn G. L. Dynamic response indicators of heat stress in shaded and non-shaded feedlot cattle, Part 2: Predictive relationships. Biosyst. Eng. 2005; 91: 111-118https: //doi. org/10. 1016/j. biosystemseng. 2005. 02. 001Crossref Scopus (127) Google Scholar; Hahn et al. , 2009Hahn G. L. Gaughan J. B. Mader T. L. Eigenberg R. A. Thermal indices and their applications for livestock environments. in: DeShazer J. A. Livestock Energetics and Thermal Environment Management. American Society of Agricultural and Biological Engineers, St. Joseph, MI, USA2009: 113-130Crossref Google Scholar). Feed dry matter intake (DMI) and respiration rates (RR) were recorded daily during TN (DMI: 2. 10 ± 0. 02% BW; RR: 54 ± 2 breaths/min) and HS (DMI: 1. 62 ± 0. 04% BW; RR: 105 ± 2 breaths/min). Each steer was fitted (as described by Rombach et al. , 2018Rombach M. Münger A. Niederhauser J. Südekum K. H. Schori F. Evaluation and validation of an automatic jaw movement recorder (RumiWatch) for ingestive and rumination behaviors of dairy cows during grazing and supplementation. J. Dairy Sci. 2018; 101 (29290426): 2463-2475https: //doi. org/10. 3168/jds. 2016-12305Abstract Full Text Full Text PDF PubMed Google Scholar) with a RW halter (firmware V. 2. 29) 3 d before the study began. Five observers were trained to conduct visual observations using 1-min scan sampling (Büchel and Sundrum, 2014Büchel S. Sundrum A. Evaluation of a new system for measuring feeding behavior of dairy cows. Comput. Electron. Agric. 2014; 108: 12-16https: //doi. org/10. 1016/j. compag. 2014. 06. 010Crossref Scopus (0) Google Scholar). Inter-observer agreement was assessed over 30-min periods (Cohen's Kappa κ = 0. 96 ± 0. 01). Steers were divided into 2 groups (n = 4 or 5 steers/group) based on proximity of pens. Each group was observed for 14 h total beginning on d 3 of each period (selected due to documented cumulative effects of heat stress Ominsky et al. , 2002). Observational data were collected from 0700 – 0800 and 0800 – 0900 h (daily removal of orts and feeding between 0700 and 0800 h). Additional time blocks included 0930 – 1030, 1100 – 1200, 1330 – 1430, 1500 – 1600, and 1630 – 1730 h. Steers were observed in each time block on 2 d within each period, resulting in a maximum of 126 h of observations per period. Observers were randomly assigned to animal groups and observational periods such that each group was observed by one observer during each observational period. Observers classified behavior exhibited by steers in each minute as eating, ruminating, or other behavior. Definitions were modified from those described by Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar: eating – head down with muzzle located near or above feed-level in the bunk (bunk height = 18 cm) or head up with chewing motion, and rumination – regurgitation, chewing, salivation and swallowing of ingested feed. All minutes in which animals were exhibiting neither eating nor rumination were recorded as other behavior. The RW physical and software components were as previously described (Zehner et al. , 2017Zehner N. Umstätter C. Niederhauser J. J. Schick M. System specification and validation of a noseband pressure sensor for measurement of ruminating and eating behavior in stable-fed cows. Comput. Electron. Agric. 2017; 136: 31-41https: //doi. org/10. 1016/j. compag. 2017. 02. 021Crossref Scopus (112) Google Scholar; Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar). Data were processed using 2 versions of RumiWatch Converter software (V. 7. 3. 2 and V. 7. 3. 36), utilizing the following system parameters: EAT1TIME (head down), EAT2TIME (head up), RUMINATIONTIME. Total eating time was calculated by summing EAT1TIME and EAT2TIME. It should be noted that V. 7. 3. 2 cannot differentiate between head up vs. down (Zehner et al. , 2017Zehner N. Umstätter C. Niederhauser J. J. Schick M. System specification and validation of a noseband pressure sensor for measurement of ruminating and eating behavior in stable-fed cows. Comput. Electron. Agric. 2017; 136: 31-41https: //doi. org/10. 1016/j. compag. 2017. 02. 021Crossref Scopus (112) Google Scholar). Thus, 100% of eating time is reported as EAT1TIME, whereas algorithms for V. 7. 3. 36 enable reporting of both EAT1TIME and EAT2TIME (Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar). Based on means and standard deviations reported in previous studies (Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar; Werner et al. , 2019Werner J. Umstatter C. Leso L. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'brien B. Evaluation and application potential of an accelerometer-based collar device for measuring grazing behavior of dairy cows. Animal. 2019; 13 (30739632): 2070-2079https: //doi. org/10. 1017/S1751731118003658Crossref PubMed Scopus (30) Google Scholar), power analysis determined 8 animals were needed to detect differences in sensor performance between periods (power = 0. 80, 95% confidence level; Friedman, 1982Friedman H. Simplified determinations of statistical power, magnitude of effect and research sample sizes. Educ. Psychol. Meas. 1982; 42: 521-526https: //doi. org/10. 1177/001316448204200214Crossref Google Scholar). The number of observations required to evaluate agreement between methods was calculated using the SimplyAgree package (Caldwell, 2022Caldwell A. R. SimplyAgree: an R package and jamovi module for simplifying agreement and reliability analyses. J. Open Source Softw. 2022; 7: 4148https: //doi. org/10. 21105/joss. 04148Crossref Google Scholar, Carrasco et al. , 2013Carrasco J. L. Phillips B. R. Puig-Martinez J. King T. S. Chinchilli V. M. Estimation of the concordance correlation coefficient for repeated measures using SAS and R. Comput. Methods Programs Biomed. 2013; 109 (23031487): 293-304https: //doi. org/10. 1016/j. cmpb. 2012. 09. 002Crossref PubMed Scopus (93) Google Scholar, Cohen, 1960Cohen J. A coefficient of agreement for nominal scales. Educ. Psychol. Meas. 1960; 20: 37-46https: //doi. org/10. 1177/001316446002000104Crossref Google Scholar) in R (version 4. 3. 1; R Core Team, 2020R Core Team R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria2020Retrieved from https: //www. R-project. org/Google Scholar), determining 80 observations were needed per period (power = 0. 80, 95% confidence level; Bland and Altman, 1999Bland J. M. Altman D. G. Measuring agreement in method comparison studies. Stat. Methods Med. Res. 1999; 8 (10501650): 15-60https: //doi. org/10. 1177/096228029900800204Crossref Google Scholar; Lu et al. , 2016Lu M. -J. Zhong W. -H. Liu Y. -X. Miao H. -Z. Li Y. -C. Ji M. -H. Sample size for assessing agreement between two methods of measurement by Bland-Altman method. Int. J. Biostat. 2016; 12 (27838682) 20150039https: //doi. org/10. 1515/ijb-2015-0039Crossref PubMed Scopus (158) Google Scholar, Ominski et al. , 2002Ominski K. H. Kennedy A. D. Wittenberg K. M. Nia S. M. Physiological and production responses to feeding schedule in lactating dairy cows exposed to short-term, moderate heat stress. J. Dairy Sci. 2002; 85 (12018417): 730-737https: //doi. org/10. 3168/jds. S0022-0302 (02) 74130-1Abstract Full Text PDF PubMed Google Scholar). An additional steer and hours of observation were added in the event of potential data loss and to provide adequate coverage given predominant behaviors were likely to vary across the day. A small amount of data loss (<7%) did occur due primarily to issues with battery connections in RW units, but retained data (n = 123 and 112 observations TN and HS, respectively). Agreement between visual observations and the automated system were evaluated by repeated measures concordance correlation coefficients (CCC; Carasco et al. , 2013) and Bland-Altman biases (mean differences) and 95% limits of agreement for repeated measures (Zou, 2011Zou G. Confidence interval estimation for the Bland-Altman limits of agreement with multiple observations per individual. Stat. Methods Med. Res. 2011; 22 (21705434): 630-642https: //doi. org/10. 1177/0962280211402548Crossref Scopus (148) Google Scholar), with animal set as the subject (SimplyAgree; Caldwell, 2022Caldwell A. R. SimplyAgree: an R package and jamovi module for simplifying agreement and reliability analyses. J. Open Source Softw. 2022; 7: 4148https: //doi. org/10. 21105/joss. 04148Crossref Google Scholar). The CCC and Bland-Altman statistics are presented along with respective 95%-confidence intervals (CI). The CCC values were interpreted using previously defined criteria wherein: negligible = 0. 0–0. 3, low = 0. 3–0. 5, moderate = 0. 5–0. 7, high = 0. 7–0. 9, and very high = 0. 9–1. 00 (Hinkle et al. , 2003Hinkle D. E. Wiersma W. Jurs S. G. Applied Statistics for the Behavioral Sciences. Houghton Mifflin College Division, Boston, MA, USA2003Google Scholar). Under- or over-estimation of the bias was considered significant if the line of equality (difference between measurements = 0) was not within the 95%-CI (Hinkle et al. , 2003Hinkle D. E. Wiersma W. Jurs S. G. Applied Statistics for the Behavioral Sciences. Houghton Mifflin College Division, Boston, MA, USA2003Google Scholar; Giavarina, 2015Giavarina D. Understanding bland altman analysis. Biochem. Med. (Zagreb). 2015; 25 (26110027): 141-151https: //doi. org/10. 11613/BM. 2015. 015Crossref PubMed Google Scholar). Agreement for total eating and rumination time was evaluated for both converter versions in each period as well as for combined data across both TN and HS. EAT1TIME and EAT2TIME (V. 7. 3. 36) were compared with visually-observed total eat time (Werner et al. , 2018Werner J. Leso L. Umstätter C. Niederhauser J. J. Kennedy E. Geoghegan A. Shalloo L. Schick M. O'Brien B. Evaluation of the RumiWatchSystem for measuring grazing behaviour of cows. J. Neurosci. Methods. 2018; 300 (28842192): 138-146https: //doi. org/10. 1016/j. jneumeth. 2017. 08. 022Crossref PubMed Scopus (70) Google Scholar). To compare the bias across periods, differences between observations and RW were then analyzed using PROC MIXED in SAS (V. 9. 4, SAS Inst. Inc. , Cary, NC) with period as the fixed effect, animal as the random variable, and day as the repeated measure. Normality of residuals was assessed using a Shapiro–Wilk test. Results were considered significant at P ≤ 0. 05. Summary statistics for eating and rumination times recorded via visual observation and RW are presented in Table 1. Eating times ranged from 0 to 49 min for visual observations and 0 to 51. 47 min for RW. Visually-observed rumination times ranged from 0 to 58 min, with 0 to 58. 73 min for RW. Bland-Altman plots for total eating and rumination time across both TN and HS (combined data) are shown in Figure 1a-d. When combined data for TN and HS were analyzed, CCC for total eating time (EAT1TIME + EAT2TIME; V. 7. 3. 2: 0. 91 0. 89, 0. 93; V. 7. 3. 36: 0. 95 0. 94, 0. 96) and rumination time (V. 7. 3. 2: 0. 92 0. 90; 0. 94; V. 7. 3. 36: 0. 99 0. 99, 0. 99) were within ranges indicating very high agreement. There was no significant under- or over-estimation of total eating time (bias: −0. 73 -2. 65, 1. 19 min/h) when combined data were analyzed with V. 7. 3. 2, but there was an under-estimation of rumination time (bias: −2. 39 -4. 06, −0. 71 min/h). Conversely, when data were analyzed with V. 7. 3. 36, there was an underestimation of eating (bias: −3. 07 -3. 84, −2. 30 min/h) but not rumination time (bias: 0. 28 -3. 47, 4. 04 min/h). Table 1Visual observations and automated measurements (RumiWatch System; RW; Converter software V. 7. 3. 2 and V. 7. 3. 36) collected under thermoneutral (n = 123 observations) and heat stress (n = 112 observations) conditions (n = 9 steers) Behavior (min/h) Visual ObservationsV. 7. 3. 2V. 7. 3. 36Min1Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. LQ1Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. Med1Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. UQ1Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. Max1Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. MeanMinLQMedUQMaxMeanMinLQMedUQMaxMeanThermoneutralTotal Eating Time0. 002. 0012. 0022. 0049. 0014. 740. 004. 3012. 5823. 8748. 9215. 710. 005. 7316. 3329. 2351. 4718. 47EAT1TIME2EAT1TIME (head down) and EAT2TIME (head up) cannot be differentiated with V. 7. 3. 2. ————————————0. 001. 958. 7315. 9029. 959. 92EAT2TIME2EAT1TIME (head down) and EAT2TIME (head up) cannot be differentiated with V. 7. 3. 2. ————————————0. 002. 707. 4511. 6039. 188. 55Rumination Time0. 000. 0015. 0028. 0055. 0016. 520. 006. 6717. 9829. 7053. 6019. 180. 000. 0014. 1826. 8753. 6016. 10Heat StressTotal Eating Time0. 000. 004. 5011. 0044. 008. 130. 001. 184. 8312. 6839. 358. 420. 002. 175. 7014. 4845. 5310. 18EAT1TIME————————————0. 000. 141. 965. 9724. 704. 04EAT2TIME————————————0. 000. 893. 449. 7526. 436. 14Rumination Time0. 005. 7517. 0027. 0058. 0016. 670. 0010. 1818. 6127. 6758. 7318. 750. 005. 8916. 7525. 5858. 7216. 631 Min = minimum; LQ = lower quartile; Med = median; UQ = upper quartile; Max = maximum. 2 EAT1TIME (head down) and EAT2TIME (head up) cannot be differentiated with V. 7. 3. 2. Open table in a new tab Detailed agreement statistics for eating time during TN and HS are shown in Table 2. When data for TN and HS were analyzed separately, the CCC for total eating time were both very high during TN (0. 94) regardless of software version. However, while the CCC for V. 7. 3. 36 remained very high during HS, the CCC for V. 7. 3. 2 fell to within the range indicating only high agreement (0. 89). When eating time data were analyzed using V. 7. 3. 2, there were no significant under- or over-estimations in either period. However, there was a significant over-estimation of eating time in both TN (3. 75 min/h) and HS (2. 06 min/h) with V. 7. 3. 36. Accordingly, mixed model analysis revealed no difference in the bias by period for total eating time with V. 7. 3. 2 (P = 0. 48), but the bias was greater for TN in comparison with HS when data were analyzed with V. 7. 3. 36 (P < 0. 01). Table 2Agreement statistics comparing automated measurements (RumiWatch System; Converter software V. 7. 3. 2 and V. 7. 3. 36) to visual observations under thermoneutral (TN) and heat stress (HS) conditions (n = 9 steers) TN (n = 121 observations) HS (n = 112 observations) V. 7. 3. 2V. 7. 3. 36V. 7. 3. 2V. 7. 3. 36Total Eating TimeCCC1CCC = repeated measures concordance correlation coefficient; BA = Bland-Altman bias = mean difference; LOA = 95% limit of agreement. Data are presented as agreement (95%-confidence interval). 0. 91 (0. 88, 0. 93) 0. 94 (0. 92, 0. 95) 0. 89 (0. 84, 0. 93) 0. 95 (0. 93, 0. 97) BA1CCC = repeated measures concordance correlation coefficient; BA = Bland-Altman bias = mean difference; LOA = 95% limit of agreement. Data are presented as agreement (95%-confidence interval). , 2Expressed in min/h. – bias0. 89 (−1. 29, 3. 07) 3. 75 (2. 86, 4. 64) 0. 30 (−1. 58, 2. 18) 2. 06 (1. 37, 2. 74) BA1CCC = repeated measures concordance correlation coefficient; BA = Bland-Altman bias = mean difference; LOA = 95% limit of agreement. Data are presented as agreement (95%-confidence interval). , 2Expressed in min/h. – lower LOA−10. 53 (−14. 28, −8. 33) −2. 80 (−4. 19, −1. 84) −8. 89 (−12. 31, −7. 01) −2. 96 (−4. 06, 2. 22) BA1CCC = repeated measures concordance correlation coefficient; BA = Bland-Altman bias = mean difference; LOA = 95% limit of agreement. Data are presented as agreement (95%-confidence interval). , 2Expressed in min/h. – upper LOA12. 31 (10. 11, 16. 06) 10. 30 (9. 34, 11. 69) 9. 48 (7. 60, 12. 90) 7. 07 (6. 33, 8. 17) EAT1TIME3EAT1TIME (head down) and EAT2TIME (head up) cannot be differentiated with V. 7. 3. 2. CCC—0. 75 (0. 71, 0. 79) —0. 64 (0. 55, 0. 71) BA – Bias—−4. 84 (−6. 25, −3. 43) —−4. 09 (−5. 15, −3. 03) BA – lower LOA—−17. 93 (−20. 16, 16. 25) —−15. 95 (−17. 82, −14. 51) BA – upper LOA—8. 25 (6. 56, 10. 48) —7. 78 (6. 33, 9. 65) EAT2TIME3EAT1TIME (head down) and EAT2TIME (head up) cannot be differentiated with V. 7. 3. 2. CCC—0. 64 (0. 58, 0. 70) —0. 82 (0. 78, 0. 85) BA – Bias—−6. 22 (−8. 40, −4. 03) —−1. 98 (−2. 74, −1. 23) BA – Lower LOA—−21. 34 (−24. 77, −19. 04) —−11. 52 (−12. 95, −10. 40) BA – upper LOA—8. 91 (6. 61, 12. 34) —7. 56 (6. 43, 8. 98) Rumination TimeCCC0. 93 (0. 90, 0. 95) 0. 99 (0. 99, 0. 99) 0. 91 (0. 87, 0. 94) 0. 99 (0. 98, 0. 99) BA – bias2. 738 (0. 92, 4. 56) −0. 43 (−0. 94, 0. 08) 2. 08 (0. 31, 3. 86) −0. 04 (−0. 41, 0. 32) BA – lower LOA−6. 74 (−9. 89,