Housing preweaned calves in groups and feeding them automatically is increasing in popularity worldwide. Advantages to this management system include the reallocation of calf labor, earlier socialization of the calf as well as the ability to feed more milk more easily. Unfortunately, housing calves in groups can lead to an increased incidence of morbidity and mortality and delays in disease detection. The use of precision dairy technologies, namely automatically captured feeding behavior and body temperature data, may aid producers in earlier disease detection and intervention. However, research to date suggests that changes in feeding behavior, as currently calculated and reported by autofeeders, are neither timely nor sensitive when used screening tool to detect morbidity in group-housed calves. Furthermore, studies are lacking to evaluate the utility of using automatically captured (sensor derived) body temperature data to detect illness in calves. The overall aim of this thesis was to improve our understanding of how sensor derived observations, such as feeding behavior or rumen temperature measures, vary in sick (vs healthy) calves, and to describe the diagnostic utility of individual animal data collected from precision dairy technologies as a tool to predict and/or detect disease in group housed automatically fed pre-weaned dairy calves. We proposed to apply a different statistical methodology to individual calf feeding behaviors when summarized at both the day and visit level, and to indwelling rumen temperature bolus measures, to determine if illness events could be detected in a sensitive and timely manner as compared to clinical diagnosis by trained farm personnel. Several objectives were set to accomplish this aim, the conclusions of which will be discussed in general terms in this chapter. Opportunities for future work in this area will be discussed at the conclusion of this chapter. The first objective of this study was to describe the use and utility of individual calf day level feeding behaviors to predict and detect disease. We conducted a prospective observational cohort study on 10 farms in Minnesota (n=4) and Virginia (n=6). Calves were enrolled upon entry to the group pen and exited the study at weaning. Study technicians visited the farms to collect enrollment, calf health data as recorded by farm personnel, and feeding behavior data from automatic feeder software. A matched pair analysis was performed to describe the difference in day level feeding behaviors and morbidity in the time before and during a treatment event. The results of this study show that calves drink less milk, drink more slowly, and visit the feeder without a milk meal (unrewarded visit) less frequently in the days surrounding a treatment event then age and pen matched healthy calves. There were no differences between sick and healthy calves when rewarded visits to the feeder were considered. These changes varied by clinical disease diagnosis by farm personnel, with the earliest and most consistent changes in calves diagnosed with diarrheal disease, followed by ill thrift calves, and finally calves diagnosed and treated for respiratory disease. We then investigated the diagnostic test characteristic and timing of statistical process control (SPC) techniques applied to individual animal daily average drinking speed, milk consumption, and unrewarded visit behavior to predict and detect clinical disease as compared to a farm personnel diagnosis. Self-starting CUSUM charts were parameterized for optimal sensitivity and timing in a test set of calves, then applied to all calves. The diagnostic test characteristic when evaluating single, two way and three way combinations of feeding behaviors were investigated. These results showed that the combination of drinking speed and milk consumption interpreted in parallel combination were the most sensitive (70.9%) and timely test to detect an illness event, signaling a sick calf an average of 3 days prior to a treatment event. However, none of the predictive values of any of the single, two way, or three way combinations of feeding behavior parameters had sufficient predictive ability to be used alone without daily observations by skilled calf caregivers. The results of objective one contribute to the knowledge of daily average feeding behavior in group housed dairy calves, and is the first attempt at investigating the utility of using signals generated by statistical process control to predict and detect disease. The use of drinking speed and milk consumption in combination provide the most sensitive test, but none of the predictive values were sufficient to use this method of detection alone. Calf caregivers with good observational skills are still necessary to detect sick calves in group housing systems. The second objective of this thesis was to describe the use and utility of visit (or meal) level feeding behaviors to predict and detect disease in automatically fed group housed preweaned dairy calves. Data collected from a subset of calves from objective one was used for this study, representing 8 farms in Minnesota (n=3) and Virginia (n=5). These eight farms had the institute function installed in automatic feeder hand held devices, which was used to record individual calf visit behavior. Visit level average behaviors were averaged into six hour increments (quarter day). A matched pair analysis was used to describe the difference in quarter day visit average feeding behaviors in sick and healthy calves around the time of an illness event. These results showed that sick calves had an increase in total drinking time at the feeder and a decrease in visit average drinking speed up to 24hrs prior to clinical disease diagnosis by farm personnel. Visit average milk consumption and total time at the feeder was only different between sick and healthy calves in the 6 hour time period prior to clinical diagnosis. Statistical process control techniques were then applied to these same visit average feeding behaviors to understand the diagnostic test characteristics and timing of using this method to detect a sick calf. Self-starting CUSUM chart parameters were first optimized for sensitivity and timing in a testing subset of calves, then optimal parameters were applied to all calves in the data set. Diagnostic test characteristics and timing for visit average feeding behaviors were analyzed alone and in combination. A positive alert on a combination of drinking speed, total drinking visit time, and/or milk consumption provided a sensitively to 89% and was able to detect as sick calf an average of 6.5d prior to detection by farm personnel. However, the specificity was very poor (7.7%) and predictive values for all single and combination visit average feeding behaviors were also poor, with negative predictive values ranging from 41 – 48% and positive predictive ability ranging from 50 – 54%. The results of objective two contribute to the knowledge of visit (meal) average feeding behavior in group housed dairy calves. Overall, the use of visit average feeding behaviors had improved sensitivity and timing when compared to the aforementioned evaluation of daily average feeding behaviors. However, predictive ability of the test was not improved, suggesting that neither day-level of visit (meal) level feeding behavior data are sufficient to predict or detect disease when used as the sole method of detection. As such, daily visual observation by trained personnel will still be necessary to detect illness in calves. The third objective of this thesis was to investigate the diagnostic utility of an indwelling calf rumen temperature bolus system. As a first step, a validation study was performed to describe the performance of the bolus as compared to two reference standards. First, the bolus temperature was compared to a known water bath temperature. The bolus was well correlated to the water bath temperature over a range of biologically plausible temperatures. Second, a prospective cross sectional study was performed that compared the bolus temperature measurement to the rectal temperature in growing heifer calves and described the diagnostic test characteristics to detect a rectal temperature ≥ 39.5ºC. The bolus underestimated the rectal temperature of growing heifer calves by an average of 0.33ºC and had a poor sensitivity (29%) and positive predictive value (17%) to detect a rectal temperature ≥ 39.5ºC. As a second step in this investigation, a field study was conducted to describe the use and utility of an indwelling rumen temperature bolus system to predict and detect disease in automatically fed group housed preweaned dairy calves. A prospective cohort study was performed on two farms in MN utilizing group housing and automatic feeding. Enrolled calves were administered boluses at birth and their temperatures were automatically captured during the time they were in the group pen. Temperatures were averaged by both hour and six hour time periods. We reported a monophasic diurnal pattern of individual calf bolus temperature measurements over a 24 hour period, which varied by farm and season. Results of a matched pair analysis showed that bolus temperature was elevated 24hrs prior to clinical diagnosis by farm personnel as compared to healthy control calves, though this varied by type of disease present. When specific diseases were investigated, calves diagnosed with pneumonia and ill thrift had a bolus temperature that was elevated 24hr prior to clinical diagnosis, but calves diagnosed with diarrhea did not have different bolus temperature measures than their healthy matched control calves. Statistical process control techniques as well as threshold and deviation limits were then applied to individual calf temperature data to learn if these methods of data analysis could be applied to these bolus temperature measures to predict and detect disease in an accurate and timely manner. Results showed that no detection technique had a sufficient combination of acceptable diagnostic test characteristics and timing to be applied directly in the field. Positive and negative predictive ability of all detection techniques were poor, indicating that caution should be used in considering these methods as the false positive rate may be unacceptable for producers using these systems. In addition to the poor diagnostic test characteristics, a high rate of bolus loss (23% of calves) would also limit the utility of this system if adopted on commercial dairy farms. The results of objectives three represent the first study investigating the use and utility of an indwelling rumen temperature bolus for prediction and detection of morbidity in group housed pre weaned dairy calves. No detection method provided test characteristics that were sufficient to predict or detect disease. An unexpected result from this field study was the difference in diurnal variation in RTB measures by farm. More studies on more farms are needed to understand how ambient temperature and barn temperature are associated with calf body temperature, performance, and health.
University of Minnesota Ph.D. dissertation. September 2017. Major: Veterinary Medicine. Advisor: Sandra Godden. 1 computer file (PDF); ix, 221 pages.
The Application of Precision Dairy Technologies to Detect Disease in Group Housed Automatically Fed Preweaned Dairy Calves.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.