UNIVERSITY OF CALGARY. Advances in Detection and Diagnosis of Bovine Respiratory Disease in Feedlot Cattle. Afra-Verena Freya Mang A THESIS

UNIVERSITY OF CALGARY Advances in Detection and Diagnosis of Bovine Respiratory Disease in Feedlot Cattle by Afra-Verena Freya Mang A THESIS SUBMITTED TO THE FACULTY OF GRADUATE STUDIES IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE GRADUATE PROGRAM IN VETERINARY MEDICAL SCIENCES CALGARY, ALBERTA SEPTEMBER, 2015 Afra-Verena Freya Mang 2015

Abstract Bovine Respiratory Disease (BRD) is one of the most significant health problems in the feedlot industry. The overall objective of this project was to improve BRD detection and diagnosis in commercial feedlots using technology. The first study aimed to: (i) describe changes in body temperature, physical activity, drinking and feeding behaviours associated with BRD; (ii) compare the diagnostic accuracy of these changes for the detection of cattle with BRD: and, (iii) define the diagnostic accuracy of several combinations of physiological and behavioural parameters for the detection of BRD. By equipping 561 feedlot steers with multiple healthmonitoring systems, we showed that cattle with BRD displayed a significant increase in rumen temperature, decrease in number of steps taken per day, and decrease in frequency and duration of feedbunk visits compared to healthy pen-mates. These changes were detected up to 7 days prior to clinical illness detected by pen checkers. The comparison of the diagnostic accuracy for BRD detection of these changes revealed that the most accurate were an increased rumen temperature, a decreased number of steps taken per day, and a decreased number of feedbunk visits per day. The combination in series (i.e., and rule) of an increased rumen temperature and a decreased number of feedbunk visits per day increased the specificity (Sp) of BRD detection up to 100% without decreasing the sensitivity (Se = 84%). The second study aimed to: (i) determine the level of agreement between a Computer Aided Lung Auscultation (CALA) system and lung auscultation by an experienced veterinarian; and, (ii) evaluate the diagnostic accuracy of CALA to diagnose BRD in feedlot cattle. Of the 561 steers, 35 were identified with visual BRD signs and 35 were selected as healthy controls. Comparison of veterinary auscultation and CALA revealed a substantial agreement (kappa = 0.77). Using latent class analysis, CALA had a relatively high Se (Se = 92.9%; 95% credible interval [CI] = 0.71-0.99) and Sp (89.6%; 95% CI ii

= 0.64-0.99) for diagnosing BRD compared to pen checking. In conclusion, the use of health monitoring and CALA systems can improve BRD detection and diagnosis in feedlot cattle, respectively. iii

Preface The work was done by Afra-Verena Mang under guidance and support of her supervisor, Edouard Timsit and co-supervisor, Carl Ribble. Chapter three is a modified manuscript accepted for publication in the Journal of Veterinary Internal Medicine. Text, associated tables and references have been formatted to follow thesis guidelines. All co-authors provided their expertise and reviewed the manuscript. In accordance with the University of Calgary s copyright guidelines, written permission for reproduction of the full article was obtained from all co-authors and publishers. The complete citation for this article is: Mang, A.V., Buczinski S., Booker C.W., and Timsit E. (2015) Evaluation of a computeraided lung auscultation system for diagnosis of bovine respiratory disease in feedlot cattle. J Vet Intern Med. 29: 1112-6. iv

Acknowledgements I would like to thank my supervisor, Dr. Edouard Timsit, for giving me the opportunity to work with him on this project for the last 2 years; it has been an amazing experience. Thank you for your support, guidance, and assistance. I would also like to thank my co-supervisor, Dr. Carl Ribble, for his guidance through my graduate program and providing his expertise for my thesis. It has been a pleasure learning from both of you. I also want to take the opportunity to thank my committee members: Dr. Claire Windeyer, Dr. Eugene Janzen and Dr. René Quiniou. Thank you for encouraging me during several meetings we had during the past two years. Thank you, Dr. Sebastién Buczinski, for providing your expertise and great support for the chapter on BRD diagnostics. Thank you, Dr. Calvin Booker, for supporting my fieldwork and opening the facilities of FHMS (Okotoks) for me. Many thanks goes to the staff of Chinook feeders who were actively involved in the fieldwork, and also to Geissler Cooperation for giving me the opportunity to experiment with CALA. Thank you, Dr. Barbara Wolfger, for all the private lectures in data analysis, brainstorming ideas, and your critiques. Thank you, Amanda Mirto, for your assistance with my writing and your time for reviewing my manuscripts. I would also like to thank the independent research group Te lo dice Calgary 1 for all the amazing adventures together while living in Canada. And thank you to all my colleagues and 1 Regina Krohn, Maria Sancho Gonzalez, Aline Bernardi, Matilde Tomaselli, Jaime Rodríguez Estival, Alysson Macedo, Diogo Câmara, Jagoš Radović v

friends I met during my time here, for all the moments shared, for the support in the program and during the project. Thank you for your time and help! Thank you, Elisabeth Maria Huba-Mang and Heinz-Peter Mang for encouraging me in my decisions and all the stimulating discussions and visits in Canada. Finally, I would like to thank, Walter Segura, for supporting me as I pursued my academic career even when this meant being separated by several kilometers and for long periods of time. I am looking forward to our upcoming adventures together as a family. vi

Epigraph It always seems impossible until it s done. Nelson Mandela vii

Table of Contents Abstract... ii Preface... iv Acknowledgements... v Epigraph... vii Table of Contents... viii List of Tables... xi List of Figures... xiii List of Symbols, Abbreviations and Nomenclature... xiv General Introduction... 1 1.1 Prevalence and performance impact of BRD in feedlot cattle... 2 1.1.1 Prevalence of BRD in feedlot cattle... 2 1.1.2 Performance impact of BRD... 5 1.2 Detection of cattle with BRD... 7 1.3 Diagnosis of feedlot cattle with BRD... 9 1.4 Objectives... 11 Evaluation of changes in body temperature, physical activity, drinking and feeding behaviours for the detection of Bovine Respiratory Disease in feedlot cattle... 12 2.1 Introduction... 12 2.2 Materials and methods... 13 2.2.1 Animals and housing... 14 viii

2.2.2 Real-time monitoring of rumen temperature, feeding and drinking behaviours, and physical activity... 15 2.2.3 Health-monitoring... 15 2.2.4 Case definition and data analyses... 16 2.3 Results... 18 2.3.1 Health and performance data... 18 2.3.2 Changes in body temperature, physical activity, feeding and drinking behaviours associated with BRD... 19 2.3.3 Diagnostic performance of rumen temperature, physical activity, feeding and drinking behaviours for the detection of BRD... 22 2.3.4 Combination of several physiological and behavioural parameters for the detection of BRD... 27 2.4 Discussion... 28 Evaluation of a computer-aided lung auscultation system for diagnosis of Bovine Respiratory Disease in feedlot cattle... 32 3.1 Introduction... 32 3.2 Materials and Methods... 33 3.2.1 Animals... 33 3.2.2 Study design: case-control study... 34 3.2.3 Computer-aided Lung Auscultation... 35 3.2.4 Serum Haptoglobin Determination... 36 3.2.5 Data Analysis... 36 3.3 Results... 39 ix

3.4 Discussion... 42 General discussion... 46 4.1 Objectives and main results... 46 4.2 Strengths and weaknesses of the project... 47 4.3 Implications... 50 4.4 Conclusions and future directions... 51 References... 54 Appendix A: Copyright permission letter... 64 x

List of Tables Table 1-1: Bovine Respiratory Disease (BRD) morbidity rate, 1 st treatment relapse and mortality rate in large pen trials (>200 animals per pen) that occurred in Canadian commercial feedlots from 2000 to 2015.... 4 Table 1-2: Impact of Bovine Respiratory Disease (BRD) on individual body weight (BW), average daily gain (ADG), and carcass characteristics of 33,073 steers fed in large pens in two commercial western Canadian feedlots... 6 Table 1-3: Prevalence of lung lesions at slaughter in cattle treated and never treated for Bovine Respiratory Disease (BRD) during the feeding period... 8 Table 2-1: Clinical data in feedlot steers selected as true positive (TP) Bovine Respiratory Disease (BRD) cases (n = 34) or true negatives (TN) controls (n = 26).... 19 Table 2-2 Sensitivity (Se) and specificity (Sp) for each parameter 1 day prior to detection (PD) of Bovine Respiratory Disease in feedlot cattle.... 24 Table 2-3 Sensitivity (Se) and specificity (Sp) for each parameter 3 days prior to detection (PD) of Bovine Respiratory Disease in feedlot cattle.... 24 Table 2-4: Differences in area under the curves (AUC) of receiver operating characteristic (ROC) curves of several physiological and behavioural parameters 1 day prior to detection (PD) of Bovine Respiratory Disease in feedlot cattle.... 25 Table 2-5: Differences in area under the curves (AUC) of receiver operating characteristic (ROC) curves of several physiological and behavioural parameters 3 days prior to detection (PD) of Bovine Respiratory Disease (BRD) in feedlot cattle... 26 Table 2-6: Sensitivity (Se), specificity (Sp) and accuracy of several combination of increased rumen temperature, decreased number of steps per day and decreased feeding frequency for the detection of Bovine Respiratory Disease in feedlot cattle... 27 xi

Table 3-1: Prior distribution and posterior latent class estimates (median and 95% credibility interval [CI]) of Bovine Respiratory Disease (BRD) prevalence and test sensitivity (Se) and specificity (Sp) of pen-checkers and computer-aided lung auscultation (CALA) for BRD diagnosis.... 39 Table 3-2: Health data (rectal temperature, respiratory rate per minute, serum haptoglobin concentration) in feedlot steers selected as Bovine Respiratory Disease (BRD) cases (n = 35) or pen-matched healthy controls (n = 35) by pen-checkers.... 40 Table 3-3: Agreement between lung auscultation by an experienced veterinarian using a conventional stethoscope and computer-aided lung auscultation (CALA) for detection of abnormal lung sounds (e.g. increased bronchial sounds, crackles and wheezes) in feedlot cattle... 41 Table 3-4: Two-by-two table comparing diagnosis of Bovine Respiratory Disease (BRD) by pen checkers with BRD diagnosis by a computer-aided lung auscultation (CALA) system.... 42 xii

List of Figures Figure 2-1: Study population and sample sizes for true positive (TP) Bovine Respiratory Disease (BRD) cases and true negative (TN) controls.... 18 Figure 2-2: Changes in rumen temperature (A), number of steps (B), feeding frequency (C), feeding duration (D), drinking frequency (E), drinking duration (F), lying frequency (G) and lying duration (H) 10 days prior to pull for true positive (TP) Bovine Respiratory Disease (BRD) case ( n = 31) and true negative (TN; n = 24).... 21 Figure 2-3: Receiver operating characteristic (ROC) curves of several physiological and behavioural parameters 1 day (A) and 3 days prior detection (B) of Bovine Respiratory Disease (BRD) in feedlot cattle... 23 Figure 3-1: Distribution of lung scores obtained after Computer-aided Lung Auscultation (CALA) in a population of steers identified with visual signs of Bovine Respiratory Disease (BRD) by pen checkers (n = 35) and in a population of steers selected as healthy controls (n = 35).... 41 xiii

List of Symbols, Abbreviations and Nomenclature ADG AUC BRD CALA Average daily gain Area under the curve Bovine respiratory disease Computer-aided lung auscultation F Fahrenheit GEE HAP HCW PD Q1 Q3 SD Se Sp ROC Generalized estimating equation Serum haptoglobin Hot carcass weight Prior detection 25 th percentile 75 th percentile Standard deviation Sensitivity Specificity Receiver Operating Characteristic xiv

General Introduction Bovine Respiratory Disease (BRD) is one of the most significant health problems in the beef industry (Miles, 2009). Despite substantial advances in antimicrobials and vaccines against respiratory pathogens, BRD remains an important cause of morbidity and mortality in beef cattle, causing considerable economic losses (>$500 millions per year in North America) and decreasing animal welfare (Miles, 2009). Cattle of all ages can suffer from BRD; however, cattle are most affected during the first 50 days following their arrival at the feedlot because they are exposed to a wide range of pathogens (as a result of commingling) at a time when various stresses negatively affect the immune system (Duff & Galyean, 2007). As an infectious disease, BRD is caused by several bacteria (Mannheimia haemolytica, Pasteurella multocida, Mycoplasma bovis, Histophilus somni) and viruses (Bovine respiratory syncytial virus (BRSV), Bovine herpesvirus type 1 (BoHV 1- IBR), Bovine influenza virus 3 (Pi3), Bovine viral diarrhea virus (BVDV)). However, it is a multifaceted problem with host (genetic, immunity, stress), management (weaning, transportation, commingling), and environmental (temperature variation) factors combining to predispose cattle to respiratory disease (Taylor et al., 2010a). 1

1.1 Prevalence and performance impact of BRD in feedlot cattle 1.1.1 Prevalence of BRD in feedlot cattle 1.1.1.1 BRD morbidity Bovine Respiratory Disease morbidity in feedlots is influenced by numerous factors including the type of cattle placed and the antimicrobial agent selected to control BRD on arrival (also called metaphylaxis). A recent US feedlot survey (USDA, 2013) revealed that 21.2 % of cattle weighing less than 700 lb. when placed at the feedlot subsequently developed BRD, whereas only 8.8% of cattle weighing 700 lb. or more when placed at the feedlot did. The higher prevalence of BRD among cattle weighing less than 700 lb. when placed is likely due to lighter cattle being younger and having less mature immune systems than cattle weighing 700 lb. or more (Taylor et al., 2010a). Lighter, younger cattle might also have been exposed to fewer pathogens than heavier, older cattle. In addition, lighter cattle are more likely to come directly from ranch sources (i.e., as opposed to being sourced from a stocker or backgrounder) and may not have been weaned prior to shipment to the feedlot, which likely induces more stress and impaired immune responses, possibly decreasing the cattle s resistance to disease (Aich et al., 2009a). As shown in Table 1-1, BRD morbidity in commercial feedlots in Alberta can range widely depending on the antimicrobial agent selected for metaphylaxis. It ranged from 3.4% to 28.8% within the same cattle population (high risk cattle; e.g., recently weaned, commingled, auction market derived) depending on the choice of metaphylaxis. Morbidity due to BRD in feedlot cattle should therefore be interpreted as a function of the risk category of cattle (high-risk versus moderate-to-low risk cattle) and the antibiotic treatment selected on arrival. \In western 2

Canadian feedlots, BRD morbidity typically ranges from 5 to 10 % in high-risk cattle when treated with tulathromycin (Draxxin, Zoetis, Kirkland, Qc, Canada) on arrival (Timsit; personal communication). However, it can be as high as 29% if a less efficacious drug is used (cf. Table 1-1). 3

Table 1-1: Bovine Respiratory Disease (BRD) morbidity rate, 1 st treatment relapse and mortality rate in large pen trials (>200 animals per pen) that occurred in Canadian commercial feedlots from 2000 to 2015. Study Type of animals Average initial BW 1 Sex (Kg) Risk Metaphylactic treatment Numbe r of animals BRD morbidity risk 4 BRD 1st relapse risk BRD mortality risk Schunicht et al. (2002) 288 Steers High LA 300 2 (30 mg/kg) 5495 22.55% 36.64% 1.55% tilmicosin (10 mg/kg) 5494 19.37% 38.53% 1.40% Schunicht et al. (2003) 259 Steers High LA 300 (30 mg/kg) 5163 19.25% 24.45% 1.88% Booker et al. (2006) 256 Steers High CCFA 3 (6.6 mg/kg) 3869 26.85% 60.44% 8.01% CCFA (6.6 mg/kg) 3866 27.96% 54.95% 8.17% tilmicosin (10 mg/kg) 3870 28.84% 63.89% 10.93% Booker et al. (2007) 278 Steers High tulathromycin (10 mg/kg) 3304 3.42% 23.01% 0.30% tilmicosin (10 mg/kg) 3304 14.04% 38.58% 1.88% LA 300 (30 mg/kg) 3302 17.02% 38.79% 2.54% Van Donkersgoed et al. (2008) 291 Heifer Moderate tilmicosin (10 mg/kg) 2250 13.02% 10.92% 0.22% tulathromycin (10 mg/kg) 2244 2.99% 8.96% 0.04% 1 BW = body weight; 2 LA 300 = long acting oxytetracyline 300 mg/ml; 3 CCFA= ceftiofur crystalline fee acid 300 mg/ml; 4 BRD case definition = visual signs of BRD + rectal temperature 40.5 C or 40 C. It is noteworthy that the prevalence of BRD reported in the literature is only a proxy of the true BRD morbidity as methods to detect and diagnose BRD in feedlot cattle are imperfect (White & Renter, 2009). Consequently, numerous cattle suffering from BRD are not detected (subclinical cases) and a high proportion of cattle diagnosed with BRD are actually affected by other diseases (Thompson et al., 2006; White & Renter, 2009). 4

1.1.1.2 BRD mortality Similar to BRD morbidity, BRD mortality is strongly influenced by the type of cattle placed and the antimicrobial selected for metaphylaxis (Table 1-1). Other factors also influence BRD mortality such as the antimicrobial selected for treatment and quality of BRD detection. Cattle pulled too late have a higher probability to die or not finish the production cycle (Apley, 1997). 1.1.2 Performance impact of BRD Compared to healthy pen mates, cattle treated for BRD have lower average daily gain (ADG) and lighter carcasses (i.e., lower hot carcass weight [HCW]) when the whole pen of cattle is marketed at one time (Cernicchiaro et al., 2013; Erickson et al., 2011). Cattle treated for BRD also tend to be leaner and have less marbling (Cernicchiaro et al., 2013; Erickson et al., 2011). As shown in Table 1-2, impacts of BRD increase with the number of treatments, with a greater impact in cattle treated twice or more (Table 1-2) (Cernicchiaro et al., 2013; Erickson et al., 2011). It is nevertheless important to note that cattle treated once or more for BRD can maintain their potential to produce carcasses with similar value to healthy animals given the additional days and intake energy required (i.e. compensatory response) (Holland et al., 2010). The impact of BRD on carcass quality can possibly be mitigated by isolating cattle with multiple BRD treatments and extending the days on feed for those animals. 5

Table 1-2: Impact of Bovine Respiratory Disease (BRD) on individual body weight (BW), average daily gain (ADG) and carcass characteristics of 33,073 steers fed in large pens in two commercial western Canadian feedlots a. BRD treatment 0 1 2+ P value Steers, n 30911 1823 339 DOF, Days 260.4 260.6 260.7 0.41 Initial BW, lb 630.8 632.6 640.3 0.04 Final BW,lb 1377.4 1374.6 1331.2 <0.001 ADG,lb 3.12 3.1 2.94 <0.001 HCW, lb 826.4 824.7 798.7 <0.001 Marbling 516.8 503 489.8 <0.001 LM area, in 12.7 12.7 12.4 <0.001 Fat depth, in 0.42 0.4 0.37 <0.001 USDA choice, % 51.3 42.1 36.9 <0.001 USDA YG 3.17 3.1 3.04 <0.001 a adapted from Erickson et al., 2011 The presence of severe lung lesions at slaughter irrespective of the presence of BRD treatment during the feeding period also decreases ADG (-0.07 kg compared to normal lung) and HCW (-7.1 kg compared to normal lung) (Rezac et al., 2014). The estimation of BRD impact on performance should therefore include not only the prevalence and severity of BRD clinical cases detected during the feeding period but also the prevalence of cattle with severe lung lesions never diagnosed with BRD (also called subclinical BRD cases) (Thompson et al., 2006). In summary, BRD is a very prevalent disease in feedlots particularly in high-risk cattle (e.g. recently weaned, commingled, auction-market derived). Its impact on performance is significant, especially if subclinical BRD cases are taken into account. 6

1.2 Detection of cattle with BRD Detection of cattle with BRD in feedlots is usually based on visual appraisal performed once or twice daily by pen checkers (Duff & Galyean, 2007). Clinical signs that are typically targeted include depression, anorexia/dysorexia, increased respiratory rate, nasal and ocular discharges (Duff & Galyean, 2007). Although strongly indicative of a disease state, these signs can be subtle or occur late in the disease process or both. As cattle are prey animals, they have a tendency to mask any sign of sickness as long as possible to protect themselves from predation (Noffsinger & Locatelli, 2004). To detect cattle with BRD based on visual appraisal is therefore difficult and numerous sick animals are either not detected at all or detected late in the disease process (Timsit et al., 2011b). Numerous studies compared BRD treatment records during the feeding period with lung lesions at slaughter (Table 1-3). These studies showed that the prevalence of lung lesions in feedlot cattle is typically high, ranging from 42% (Bryant et al., 1999) to 87% (Buhman et al., 2000) and that numerous cattle with lung lesions are never clinically diagnosed with BRD during the feeding period. By comparing lung lesions at slaughter to treatment records, White and Renter (2009) estimated the sensitivity (Se) of BRD detection by pen checkers at only 62%. 7

Table 1-3: Prevalence of lung lesions at slaughter in cattle treated and never treated for Bovine Respiratory Disease (BRD) during the feeding period. Number BRD Prevalence of lung lesions at slaughter (%) Study Country of animals treatment (%) All cattle Treated for BRD Never treated for BRD Wittum et al. (1996) USA 469 29 72 78 68 Bryant et al. (1999) USA 439 17 42 40 42 Gardner et al. (1999) USA 204 50 43 48 37 Buhman et al. (2000) USA 170 25 87 97 83 Thompson et al. (2006) South Africa 2036 23 43 55 39 Schneider et al. (2009) USA 1665 8 62 74 61 Tennant et al. (2014) USA 2336 8 64 84 62 Multiple studies also showed that cattle with BRD are detected by pen checkers two to six days after the beginning of fever (Schaefer et al., 2007; Timsit et al., 2011a). Early detection of diseased cattle is however a key component of effective BRD control in feedlots (Apley, 1997). It enables feedlot staff to provide treatment early in the disease process, maximizing the opportunities for clinical and bacterial cures, and thus, decreasing the impact of BRD on cattle performance and welfare (Janzen et al., 1984). Numerous systems have been recently commercialized to remotely and continuously monitor the physiology (body temperature) and behaviour (physical activity, drinking and feeding behaviours) of beef cattle. By detecting subtle changes in physiology or behaviour of sick cattle, these systems have the potential to significantly increase the timeliness and accuracy of BRD detection in feedlots. Using the Growsafe System (Growsafe Ltd., Airdrie, Alberta, Canada) that measures duration and frequency at the feed bunk, Quimby et al. (2001) were able to detect sick cattle over four days before pen checkers. Similarly, using rumen temperature 8

boluses or infrared thermography, Timsit et al. (2011a) and Schaefer et al. (2007; 2012) respectively detected sick animals up to 4-6 days prior to the onset of clinical symptoms detectable by visual appraisal. Finally, by using pedometers, White et al. (2012) showed that cattle with BRD have a significant decrease in physical activity. Unfortunately, it is not possible to define which health-monitoring system is the most promising to indicate cattle with BRD as diagnostic accuracies e.g. Se and specificity (Sp) for the detection of BRD of each change in physiology (body temperature) and behaviour (physical activity, drinking and feeding behaviours) was evaluated in different populations (e.g. feedlot cattle versus dairy calves), in different research settings (e.g. naturally occurring BRD versus experimental infection) or both. Furthermore, based on previous studies (Timsit et al., 2011b; Wolfger et al., 2012), each physiological or behavioural change alone suffers from a lack of Sp, therefore resulting in numerous false-positive BRD detections. Timsit et al. (2011) showed that up to 75% of the fever episodes detected by rumen temperature boluses lasted less than 47 hours even without any treatment. Similarly, Wolfger et al. (2012) showed that pulling cattle only based on abnormal feeding behaviour during the first weeks on feed could lead to pulling as much as 90% of healthy animals. Because adding multiple tests in series can improve the Sp of detection, combining multiple health parameters could improve the accuracy of BRD detection. However, these parameters have never been combined and the potential improvement in BRD detection is unknown. 1.3 Diagnosis of feedlot cattle with BRD The diagnosis of BRD in feedlots is currently based on visual signs of respiratory disease plus an increased rectal temperature with thresholds ranging from 103.5 to 105 F (Duff & 9

Galyean, 2007). Unfortunately, this diagnostic method is not always accurate. By comparing lung lesions at slaughter with treatment records, White and Renter (2009) estimated the Sp of BRD diagnosis at 63%. Accurate BRD diagnosis is nevertheless crucial for effective BRD treatment and prevention strategies (Taylor et al., 2010b). It can maximize the success of treatment, reduce the costs associated with treatment, and ensure the prudent use of antibiotics (Taylor et al., 2010b). Several techniques could enhance the accuracy of BRD diagnosis in feedlots (Cooper & Brodersen, 2010; Duff & Galyean, 2007). For example, detection of respiratory pathogens in the upper or lower respiratory tract or both, such as Mannheimia haemolytica, Pasteurella multocida, Mycoplasma bovis could confirm the presence of a respiratory tract infection. However, numerous studies have shown that respiratory pathogens can also be found in healthy animals (Allen et al., 1992; Timsit et al., 2011b), which calls into question the significance of their detection for confirming BRD (Fulton & Confer, 2012). Determination of acute phase proteins (APPs) can also be used to confirm the presence of infection or inflammation in cattle (Murata et al., 2004). Among APPs, serum haptoglobin (Hap) has proven to be Se for indicating BRD and has been used in different studies to assess BRD morbidity (Aich et al., 2009b; Timsit et al., 2011a). However, because Hap can also increase in diseases other than BRD, it cannot be used alone to confirm BRD (Eckersall & Bell, 2010). Furthermore, it is noteworthy that both detection of pathogens and determination of Hap take time to complete, which is not compatible with a chute-side decision of treatment. Clinical examination of the respiratory tract such as thoracic ultrasonography and thoracic auscultation can also improve the accuracy of BRD diagnosis. Ultrasonography of the thoracic cavity has proven to be specific for the detection of lung lesions (Buczinski et al., 2014). 10

However, this examination lacks Se and requires highly trained staff and expensive equipment. Concerning lung auscultation, DeDonder et al. (2010) found a significant correlation between ante-mortem auscultation scores and post mortem lung lesions. Unfortunately, obtaining reliable results from auscultation requires a well-trained person with good acoustic abilities. To overcome these challenges, a computer-aided lung auscultation (CALA) system has recently been developed. By automatically classifying acoustic patterns in lung scores from 1 (normal) to 5 (chronic), this system requires minimal training and has the potential to significantly increase the Sp of BRD diagnosis in feedlots. However, its accuracy to diagnose BRD has yet to be evaluated. 1.4 Objectives In this project, we focused at enhancing, via the use of technologies, the ability of feedlot producers and veterinarians to accurately detect or diagnose feedlot cattle with BRD. More specifically, this project aimed: (1) To describe changes in physiology (rumen temperature) and behaviours (physical activity, feeding and drinking behaviour) associated with BRD, (2) To compare the diagnostic accuracy of these physiological and behavioural changes for the detection of BRD, (3) To define the diagnostic accuracy of several combinations of physiological and behavioural parameters for the detection of BRD and, (3) To evaluate the diagnostic accuracy of a computer-aided lung auscultation (CALA) system for the diagnosis of BRD. 11

Evaluation of changes in body temperature, physical activity, drinking and feeding behaviours for the detection of Bovine Respiratory Disease in feedlot cattle 2.1 Introduction Bovine Respiratory Disease (BRD) is one of the most significant health problems in the feedlot industry (Miles, 2009). A key to effectively controlling BRD is early and accurate identification of diseased cattle. This enables treatment earlier in the disease process, maximizing opportunities for clinical and bacterial cures (Apley, 1997). Unfortunately, current methods of disease detection based on visual appraisal do not favour early detection of BRD (Timsit et al., 2011b). Indeed, due to prey-predator behaviour, cattle often conceal signs of sickness, especially during early stages of disease (Weary et al., 2009). Consequently, a large proportion of feedlot cattle with BRD are detected and treated late in the course of the disease (Timsit et al., 2011b). Real-time monitoring of body temperature, feeding behaviour, drinking behaviour, or physical activity can facilitate early detection of feedlot cattle with BRD (Wolfger et al., 2015). Detecting early physiological and behavioural changes associated with BRD, namely fever, changes in appetite (dysorexia/anorexia) or drinking behaviour, or decreased physical activity, enabled several health-monitoring systems to detect cattle with BRD up to several days prior to the onset of BRD signs identified by visual appraisal (Buhman et al., 2000; Hanzlicek et al., 2010; Quimby et al., 2001; Schaefer et al., 2007; Timsit et al., 2011a). Unfortunately, it is not possible to define which changes are the most promising to indicate cattle with BRD as diagnostic accuracies e.g. sensitivity (Se) and specificity (Sp) for the detection of BRD of each 12

parameter were evaluated in different populations (e.g. feedlot cattle versus dairy calves), in different research settings (e.g. naturally occurring BRD versus experimental infection) or both. Furthermore, based on previous studies (Timsit et al., 2011a; Wolfger et al., 2012), each physiological or behavioural change alone can suffer from a lack of Sp, therefore resulting in numerous false-positive BRD detections. For example, Wolfger et al. (2012) showed that detecting sick cattle based only on abnormal feeding behaviour during the first weeks on feed could lead to detecting as many as 90% of healthy animals. To combine multiple physiological and behavioural parameters, for example, detecting cattle with fever and anorexia, could improve the accuracy of BRD detection as adding multiple tests in series (i.e., and rule) increases Sp. However, these parameters have never been combined and the potential improvement in BRD detection is unknown. The objectives were therefore (i) to describe the changes in body temperature, physical activity, feeding and drinking behaviours associated with BRD in feedlot cattle; (ii) to compare the diagnostic accuracy of these changes for the detection of BRD, and (iii) to define the diagnostic accuracy of several combinations of physiological and behavioural parameters for the detection of BRD. 2.2 Materials and methods This study was conducted in strict accordance with the recommendations of the Canadian Council of Animal Care (Olfert et al.,1993). The research protocol was reviewed and approved by the Animal Care Committee of the University of Calgary (AC13-0212). 13

2.2.1 Animals and housing Two successive replicates of 276 and 285 Angus-cross calves at high-risk of developing BRD (i.e. recently weaned, co-mingled, and auction-market derived) were studied during the first 50 days on feed (DOF) following their arrival at a southern Alberta commercial feedlot from October 2013 to February 2014. Their initial body weight was 246 ± 45 kg. Upon arrival, steers were allowed to rest for at least 12 h (with ad libitum access to hay and water) before processing. At processing, steers received a subcutaneous injection of a long-acting macrolide (tulathromycin; Draxxin, Zoetis, Kirkland, QC, Canada), were weighed and vaccinated against infectious bovine herpes virus-1, bovine viral diarrhea virus (types I and II), bovine parainfluenza-3, bovine respiratory syncytial virus, Mannheimia haemolytica (Pyramid FP 5 + Presponse SQ, Boehringer Ingelheim, Burlington, ON, Canada), and Histophilus somni and clostridial pathogens (Ultrabac 7/Somnubac, Zoetis). Steers were also dewormed with pour-on ivermectin solution (Bimectine, Bimeda-MTC, Cambridge, ON, Canada). The two replicates were fed in large outdoor dirt-floor pens (67 61 m with a 64-m fence-line concrete feed bunk). Steers were fed twice daily, at 0630 and 1430 h, a 55 to 63% concentrate receiving/growing diet formulated to meet or exceed (NRC, 2000) nutrient requirements. Each morning before feeding, bunks were visually evaluated and feed deliveries were adjusted to ensure that feed was offered for ad libitum consumption. On d 50, steers were individually weighed, revaccinated (Bovi-shield Gold 5, Zoetis) and implanted (Synovex Choice, Zoetis). 14

2.2.2 Real-time monitoring of rumen temperature, feeding and drinking behaviours, and physical activity At on-arrival processing, steers were equipped with rumen temperature boluses (Medria, Chateaugiron, France) as described (Timsit et al., 2011a) to monitor body temperature and drinking behaviour (frequency and volume of drinking bouts). They were also equipped with three-axis accelerometers (ENGS, Rosh Pina, Israel) as described (Wolfger et al., 2015) to monitor physical activity (number of steps, frequency and duration of lying bouts). Finally, the pen housing the cattle was equipped with an active antenna in front of the bunk that emitted a magnetic field every 10 seconds within 30 cm of the bunk (ENGS system, Rosh Pina, Israel). This antenna combined with the accelerometers enabled us to monitor feeding behaviour (frequency and duration of bunk attendance) (Wolfger et al., 2015). During the study period neither staff nor investigator had access to the data measured by the health-monitoring devices (blinded study). 2.2.3 Health-monitoring Cattle were observed daily by pen checkers. Steers with visual signs of BRD (e.g. one or more of the following: depression, nasal or ocular discharge, cough, increased respiratory rate, labored breathing) were removed from the pen and clinically examined by an experienced veterinarian. For each steer suspected of being affected by BRD, one apparently healthy steer (no visual signs of BRD or other disease) was conveniently selected (e.g. close to the gate, close to the apparently sick animal, etc.) as a control and similarly examined. Clinical examination included measurement of respiratory rate, rectal temperature, and lung auscultation using a 15

conventional stethoscope to detect abnormal lung sounds (e.g. increased bronchial sounds, crackles and wheezes). In addition to clinical examination, a blood sample was collected from each animal to detect inflammation via determination of serum haptoglobin (HAP) concentration. HAP concentrations were determined in duplicate using a commercial kit as described (Timsit et al., 2011a). Steers with visual BRD signs and a rectal temperature 40 C received, via subcutaneous injection, flunixin meglumin and florfenicol (2 ml/15 kg BW, Resflor, Merck Animal Health, Kirkland, QC, Canada). 2.2.4 Case definition and data analyses Steers with one or more visual BRD signs, abnormal lung sounds at veterinary auscultation, and a HAP concentration 0.25 g/l were defined as true positive (TP) BRD cases (Timsit et al., 2009). True negatives (TN) were defined as steers with no BRD signs, no abnormal lung sounds (based on veterinary auscultation) and a HAP concentration < 0.25 g/l. To avoid incorporation bias, rectal temperature was not included in the case definition as it is highly correlated with rumen temperature. Clinical data (respiratory rate per min, rectal temperature, HAP concentrations) and early ADG (calculated between d 0 and d 50) were compared between TP and TN using the non-parametric (Mann-Whitney U test) and parametric tests (Student s t-test) in Stata Version 13.1 (StataCorp LP, College station, TX). To describe changes in rumen temperature, frequency and duration of feed bunk visits, frequency and duration of drinking bouts, number of steps, and frequency and duration of lying 16

bouts associated with BRD, Generalized Estimating Equation (GEE) models clustered for repeated measurements using an autoregressive correlation matrix were built in Stata Version 13.1. These models compared each physiological or behavioural parameter recorded in TP and TN steers during the 10 days prior to clinical examination. For these models, all parameters were aggregated by day and the unit of observation was the individual steer. The autoregressive structure assumed a steady decay in correlation with increasing time between the observations. The replicate effect was taken into account in the models. The diagnostic accuracies for the detection of BRD of all parameters 1 day and 3 days prior to BRD detection by visual appraisal were compared using the Receiver Operating Characteristic (ROC) curves (Greiner et al., 2000). Two ROC curves were built for each parameter and Area Under the ROC Curve (AUC) of each parameter were compared using a nonparametric approach (DeLong et al., 1988) in MedCalc (MedCalc Software, Ostend, Belgium). The AUC s were interpreted using a guideline suggested by Swets (1988): parameters with an AUC = 0.5 considered as uninformative, parameters with 0.5 <AUC 0.7 considered as less accurate, parameters with 0.7 <AUC 0.9 considered as moderately accurate, and parameters with 0.9 <AUC <1 considered as highly accurate. The physiological and behavioural parameters with the highest AUC on day -1 were then combined in series (i.e., and rule; only animals positive for both tests are considered positive) and the Se, Sp and overall accuracy (accuracy = prevalence*se + [1-prevalence]*Sp) of each combination were compared. The cut-off used for each parameters were derived from the Youden index (J = Se + Sp 1), which maximized both Se and Sp. Statistical significance was declared at P < 0.05. 17

2.3 Results 2.3.1 Health and performance data Of the 561 steers studied during their first 50 days on feed, 35 (6.2%) were detected with visual BRD signs by pen checkers and 34 (6.1%) were defined as TP after clinical examinations performed by a trained veterinarian (Fig 2-1). Forty-nine (8.7%) steers were selected as pen-mate controls but only 26 (4.6%) were defined as TN after clinical examination; 7 (1.2%) steers had abnormal lung sounds at veterinary auscultation and 16 (2.9%) had a serum Hap concentration 0.25g/L. Clinical data obtained during clinical examinations of TP and TN steers are indicated in Table 2-1. Average daily gain of TP steers between d 0 and d 50 (1.74 ± 1.14 lbs/d) was significantly lower (P <0.05) than TN steers (2.72 ± 0.61 lbs/d). Figure 2-1: Study population and sample sizes for true positive (TP) Bovine Respiratory Disease (BRD) cases and true negative (TN) controls. 18

Table 2-1: Clinical data - in feedlot steers selected as true positive (TP) Bovine Respiratory Disease (BRD) cases (n = 34) or true negatives (TN) controls (n = 26). TP-BRD ( n = 34 ) TN (n = 26) Mean (±SD) Median (Q1-Q3) Mean (±SD) Median (Q1-Q3) Respiration rate/min 46 (±5) a 46 (42-50) 33 (±3) b 31 (30-36) Rectal temperature ( F) 105.5 (±0.8) a 105.5 (105.0-106.2) 102.7 (±0.9) b 102. 7 (102.4-103.4) Serum haptoglobin (g/l) 1.53 (±0.64) 1.51 (1.07-1.91) a 0.14 (±0.06) 0.16 (.09-.18) b Q1, 25 th percentile; Q3, 75 th percentile; SD, Standard deviation Within a row, means or medians without a common superscript differed (P <0.05) F, Fahrenheit 2.3.2 Changes in body temperature, physical activity, feeding and drinking behaviours associated with BRD Due to extreme weather or inadequate data backup, only 31 of 34 TP and 24 of 26 TN steers had complete recordings of rumen temperature, physical activity, and drinking and feeding behaviours (Fig. 2-2). Changes in body temperature, physical activity, feeding and drinking behaviours observed between TP and TN during the 10 days prior to clinical examination (i.e., pull by the pen checkers) are presented in Figure 2-2. Significant differences (GEE model; P< 0.05) were observed between TP and TN steers for rumen temperature (increased in TP versus TN), number of steps taken per day (decreased in TP versus TN) and frequency and duration of feedbunk visits per day (decreased in TP versus TN). Although TP steers had a tendency to drink less (both frequency and volume of drinking bouts) and lie less (both duration and frequency of lying bouts) compared with TN steers, no significant differences were observed for these behavioural parameters in the GEE model. 19

In TP steers, increase in rumen temperature and decrease in number of steps taken per day relative to TN steers, were detected up to 7 days prior to detection by visual appraisal whereas differences in feeding behaviour (decreased frequency and duration of feedbunk visits) were detected mainly during the 2 days prior to detection. 20

Figure 2-2: Changes in rumen temperature (A), number of steps (B), feeding frequency (C), feeding duration (D), drinking frequency (E), drinking duration (F), lying frequency (G) and lying duration (H) 10 day prior to pull for true positive (TP) Bovine Respiratory Disease (BRD) case (solid-lines; n = 31) and true negative (TN; dashed-lines; n = 24); grey-vertical-lines /=/ mean ± standard deviation (SD). 21

2.3.3 Diagnostic performance of rumen temperature, physical activity, feeding and drinking behaviours for the detection of BRD Receiver Operating Characteristic (ROC) curves of all physiological and behavioural parameters for the detection of BRD 1 and 3 days prior to detection by pen checkers are represented in Figure 2-3. On day -1 and -3, parameters with highest AUCs (i.e., best overall diagnostic performance) were increased rumen temperature with AUCs of 0.89 (95% CI: 0.78-0.96) and 0.75 (95% CI: 0.61-0.85), respectively, and decreased number of steps taken per day with AUCs of 0.83 (95% CI: 0.71-0.92) and 0.67 (95% CI: 0.54-0.79) (Tables 2-2 and 2-3). It is noteworthy that differences in AUCs between these parameters were not significant (Table 2-4 and 2-5). Based on their AUCs, both increased rumen temperature and decreased number of steps taken per day were considered as moderately accurate to detect BRD (i.e. 0.7< AUC 0.9). Decreased number of feedbunk visits per day was the only other moderately accurate behavioural parameter to detect BRD with an AUC of 0.78 (95% CI: 0.65-0.88) on day -1. All remaining parameters were considered as less accurate with AUCs 0.65. 22

Figure 2-3: Receiver Operating Characteristic (ROC) curves of several physiological and behavioural parameters 1 day (A) and 3 days prior detection (B) of Bovine Respiratory Disease (BRD) in feedlot cattle). 23

Table 2-2 Sensitivity (Se) and specificity (Sp) for parameters 1 day prior to detection (PD) of Bovine Respiratory Disease (BRD) in feedlot cattle. Parameter Day PD Youden s J Cut-point Sensitivity Specificity AUC (95% CI) Rumen temperature a 1 40 84 83 0.89 (0.78-0.96) Number of steps b 1 2303 81 79 0.83 (0.71-0.92) Feeding per day b Duration 1 1.26 92 39 0.65 (0.51-0.78) Frequency 1 4 81 58 0.78 (0.65 0.88) Drinking per day a Duration 1 9 58 77 0.65 (0.51-0.78) Frequency 1 4 25 86 0.52 (0.37-0.66) Lying per day a Duration 1 11.2 51 68 0.56(0.42-0.68) Frequency 1 5 44 75 0.54(0.40-0.67) a Parameter was increased in true positive cases of BRD compared to true negative healthy controls. b Parameter was decreased in true positive cases of BRD compared to true negative healthy controls. Table 2-3 Sensitivity (Se) and specificity (Sp) for parameters 3 days prior to detection (PD) of Bovine Respiratory Disease (BRD) in feedlot cattle. Parameter Day PD Youden s J Cut-point Sensitivity Specificity AUC (95% CI) Rumen temperature a 3 40 61 92 0.75 (0.61-0.85) Number of steps b 3 2153 77 61 0.67 (0.54 0.79) Feeding per day b Duration 3 2 81 33 0.51 (0.38 0.65) Frequency 3 7 67 50 0.61 (0.48-0.74) Drinking per day a Duration 3 9 67 57 0.53 (0.39 0.67) Frequency 3 3 37 70 0.51 (0.36-0.65) Lying per day a Duration 3 14 21 92 0.53 (0.39-0.66) Frequency 3 7 64 60 0.65 (0.51-0.77) a Parameter was increased in true positive cases of BRD compared to true negative healthy controls. b Parameter was decreased in true positive cases of BRD compared to true negative healthy controls. 24

Table 2-4: Differences in area under the curves (AUC) of receiver operating characteristic (ROC) curves detection of several physiological and behavioural parameters 1 day prior to detection (PD) of Bovine Respiratory Disease (BRD) in feedlot cattle. Day 1 PD Rumen temperature Number of steps Feeding duration Feeding frequency Drinking duration Drinking frequency Lying duration Lying frequency Rumen temperature X 0.09 0.26 * 0.10 0.24 * 0.37 ** 0.33** 0.37 ** Number of steps 0.09 X 0.17 0.01 0.15 0.28 * 0.22 0.26 * Feeding duration 0.26* 0.17 X 0.16 * 0.02 0.11 0.06 0.10 Feeding frequency 0.10 0.01 0.16 * X 0.14 0.27 * 0.21 0.25 * Drinking duration 0.24 * 0.15 0.02 0.14 X 0.13 0.11 0.13 Drinking frequency 0.37 ** 0.28 * 0.11 0.27* 0.13 X 0.06 0.04 Lying duration 0.33 ** 0.22 0.06 0.21 0.11 0.06 X 0.03 Lying frequency 0.37 ** 0.26 * 0.10 0.25* 0.13 0.04 0.03 X * = significant differences at P<0.05 ** = significant differences at P<0.001 25

Table 2-5: Differences in area under the curves (AUC) of receiver operating characteristic (ROC) curves for several physiological and behavioural parameters 3 days prior to detection (PD) of Bovine Respiratory Disease (BRD) in feedlot cattle. Day 3 PD Rumen temperature Number of steps Feeding duration Feeding frequency Drinking duration Drinking frequency Lying duration Lying frequency Rumen temperature X 0.04 0.22 * 0.11 0.21* 0.18 0.25 * 0.14 Number of steps 0.04 X 0.19 0.08 0.17 0.14 0.21 * 0.10 Feeding duration 0.22* 0.19 X 0.11 0.02 0.043 0.04 0.07 Feeding frequency 0.11 0.08 0.11 X 0.09 0.07 0.14 0.03 Drinking duration 0.21 * 0.17 0.02 0.09 X 0.02 0.26 0.26 Drinking frequency 0.18 0.14 0.04 0.07 0.02 X 0.05 0.05 Lying duration 0.25 * 0.21 * 0.04 0.14 0.26 0.05 X 0.11 Lying frequency 0.14 0.10 0.07 0.03 0.03 0.05 0.11 X * = significant differences at P<0.05 ** = significant differences at P<0.001 26

2.3.4 Combination of several physiological and behavioural parameters for the detection of BRD Sensitivity, Sp, and overall accuracy of several combinations of increased rumen temperature, decreased number of steps taken per day, and decreased frequency of feedbunk visits per day are presented in the Table 2-6. All combinations resulted in a Sp of 100% (i.e., no false positive detection). The combination with the best accuracy (91%) was increased rumen temperature and decreased feedbunk visit frequency with a combined Se of 84% and Sp of 100%. Decreased number of steps per day combined with decreased frequency of feedbunk visit also had a good diagnostic accuracy (87%). Table 2-6: Sensitivity (Se), specificity (Sp) and accuracy of several combinations of increased rumen temperature, decreased number of steps per day, and decreased feeding frequency for the detection of Bovine Respiratory Disease (BRD) in feedlot cattle. Disease /test Parameters +/+ +/- -/+ -/- Se (95 CI) % Sp (95 CI) % Accuracy % (95 CI) Rumen temperature (1) 26 5 4 20 84 (66-95) 83 (63-95) 84 (61-100) Number of steps per day (2) 25 6 5 19 81 (63-93) 79 (58-93) 80 (58-100) Feeding frequency per day (3) 25 6 10 14 81 (63-93) 58 (34-78) 71 (50-97) 1 + 2 22 9 0 24 71 (52-86) 100 (86-100) 84 (61-100) 1 + 3 26 5 0 24 84 (66-95) 100 (86-100) 91 (67-100) 2 + 3 24 7 0 24 77 (59-86) 100 (86-100) 87 (64-100) 27

2.4 Discussion The simultaneous monitoring of multiple physiological and behavioural parameters enabled, for the first time, a comparison of the diagnostic accuracies of changes in body temperature, physical activity, drinking and feeding behaviours for the detection of BRD in feedlot cattle. In the present study, cattle with BRD had significant increases in rumen temperature, decreases in number of steps taken per day, and decreases in frequency and duration of feedbunk visits compared to healthy pen-mates up to 7 days prior to clinical illness detected by pen checkers. However, no significant differences were observed between sick and healthy steers in drinking and lying behaviours. The comparison of AUCs under ROC curves showed that the most accurate parameters to detect BRD, 1 and 3 days prior to clinical illness, were an increased rumen temperature and either a decreased number of feedbunk visits per day or decreased number of steps taken per day, both with an AUC under the ROC curve above 0.75. All other parameters were less accurate with AUC under the ROC curve below 0.75, meaning that less than 75% of the healthy and sick cattle were correctly classified with these parameters. The combination of increased rumen temperature and decreased frequency of feedbunk visits increased Sp of BRD detection up to 100% without decreasing its sensitivity. Because BRD is an infectious disease caused by viruses (bovine herpes virus 1, bovine respiratory syncytial virus, etc.) and bacteria (Mannheimia haemolytica, Pasteurella multocida, Mycoplasma bovis, etc.), cattle affected with this condition typically develop fever early in the disease process (Fajt et al., 2004). Based on experimental trials (Fajt et al., 2004; Rose-Dye et al., 2011) and field studies (Schaefer et al., 2007; Timsit et al., 2011a), 28