Evaluation of Statistical Process Control Procedures to Monitor Feeding Behavior and Ruminal Temperature Changes Associated with Experimental Inoculation of Mannheimia Haemolytica

Objectives of this experiment were to determine if statistical process control (SPC) procedures coupled with remote collection of rumen temperature (RT) and feeding behavior (FB) patterns could accurately differentiate between animals experimentally inoculated with Mannheimia haemolytica (MH) or phosphate buffer solution (PBS), and determine if live yeast (LY) supplementation would mitigate responses to MH challenge. Thirty-six crossbred steers (BW=352 ± 23 kg) seronegative for MH were allocated within a 2X2 factorial treatment arrangement: Factor1 = roughage-based diet with or without LY (Saccharomyces cerevisiae boulardii.I-1079 at 1x10¹⁰ cfu/d, Lallemand), Factor2 = bronchoselective endoscopic inoculation with MH or PBS. Electronic feed bunks (GrowSafe) were used to measure DMI and FB traits, and ruminal thermo-boluses (Medria) used to measure RT at 5-min intervals. Data were collected 28-d prior to and following inoculation. Steers inoculated with MH exhibited elevated levels of haptoglobin, white blood cells and neutrophils (P<0.02), indicating that the MH challenge effectively stimulated immunologic responses. However, only 1 animal displayed overt clinical signs of disease. Shewhart charts (SPC procedure) were used in this analysis, and sensitivity, specificity and accuracy computed to evaluate univariate and multivariate models based on principal components analysis (PCA). Of the FB traits monitored, time to bunk had the highest model sensitivity (94%) and accuracy (94%), with model accuracies for head-down duration, and bunk visit duration and frequency being less (80, 79 and 56%, respectively). Model accuracy for DMI was intermediate at 85%. To address the diurnal nature of RT, data were averaged over 6-h intervals, and quarterly RT models evaluated separately. Model accuracy for the first quarter RT was more accurate (84%) then the other respective quarterly RT periods (82, 76, 79%). Two PCA models were constructed separately using all FB and RT traits. Model sensitivity and accuracy were relatively higher for the FB PCA (94 and 95%) then the RT PCA model (78 and 85%), with both PCA models performing slightly better than the best respective univariate trait model. The performance of a combined PCA model (all traits) was intermediate in accuracy at 91%. In this study, LY supplementation did not influence the sensitivity or accuracy of the univariate or PCA models. These results indicate that Shewhart procedures can effectively identify deviations in FB and RT patterns for the purpose of sub-clinical BRD detection. Furthermore, the PCA models were numerically more accurate than univariate traits and should be more robust in application due to their multivariate nature.