206. Comparison of Rumination Time Estimates from Two Solar-powered Smart GPS Devices in Beef Cattle on Pasture.

Abstract This study evaluated whether two solar-powered wearable sensors, the 701x, model xTpro ear tag (701x, Fargo, ND, USA) and the Innogando, model RUMI Beef neck collar (Innogando S. L., Arneiro, Lugo, Spain), produced comparable estimates of hourly rumination time (min/h) activity for beef cattle on a pasture environment. Eight mature Beefmaster cows (mean BW = 1,202 kg; mean age = 1,115 d) were initially enrolled in the study, with only four animals meeting the data-quality criteria for the final hour-level paired analysis after device-related exclusions. Both devices were fitted simultaneously to the same animal, and the final dataset comprised 4,176 matched animal-date-hour observations for direct within-animal comparison. Data were summarized and analyzed in R version 4.5.2 using the packages lme4, lmerTest, emmeans, and irr. Analyses included paired t-tests, Wilcoxon signed-rank tests, Pearson correlation, the intraclass correlation coefficient ICC(2,1), Bland-Altman agreement analysis, and linear mixed-effects models. The principal mixed models evaluated fixed effects of device, hour, and a device-by-hour interaction, incorporating random intercepts for animal and date, while a secondary model assessed temporal drift via a device-by-day interaction. Overall means were similar for the 701x and Innogando devices (18.89 ± 0.315 vs. 18.59 ± 0.299 min/h, respectively), as were mixed-model least-squares means (18.69 ± 0.74 vs. 18.38 ± 0.74 min/h; P = 0.400). The paired mean difference (701x minus Innogando) was 0.308 min/h (SE = 0.357; t = 0.864; 95% CI = -0.392 to 1.008; P = 0.388), with no systematic median shift detected by the Wilcoxon signed-rank test (P = 0.973). However, hourly agreement between the two devices was low, as evidenced by a Pearson correlation of 0.326, an ICC(2,1) of 0.325 (95% CI = 0.293 to 0.357), and wide Bland-Altman limits of agreement (-38.01 to 38.63 min/h). Rumination output varied significantly by hour (P < 0.001), and a highly significant device-by-hour interaction was observed (P < 0.001). Model-based hourly contrasts revealed device disagreement across the hours of the day, ranging from 701x reading lower than Innogando at hour 7 (-17.07 ± 1.74 min/h; P < 0.001; Cohen’s dz = -1.06) to 701x reading higher at hour 20 (17.59 ± 1.74 min/h; P < 0.001; Cohen’s dz = 0.88). No significant device-by-date drift was detected (P = 0.905). Although overall rumination estimates were similar, poor hour-specific agreement showed that 701x and Innogando characterized within-day rumination patterns differently. Thus, hourly rumination data from these platforms should not be assumed to be directly comparable across studies, and predictive or decision-support models developed from one device should be transferred to the other only cautiously, because device-specific calibration and classification algorithms may alter both the magnitude and temporal pattern of inferred rumination behavior.