Development of a novel clinical scoring system for on-farm diagnosis of bovine respiratory disease in pre-weaned dairy calves

Introduction

Bovine respiratory disease (BRD) is a major source of economic loss for the cattle industry (Panciera & Confer, 2010; Sischo et al., 1990; USDA, 2008). Respiratory disease is the leading cause of death in weaned dairy heifers and the second most common cause of mortality in pre-weaned calves in cattle herds in the United States. It is estimated that BRD is responsible for the loss of more than one million animals and approximately US $700 million annually (USDA, 2007; Wittum et al., 1996). Effective control of BRD has proven difficult in the North American dairy industry, at least in part due to the complexity of disease pathogenesis and the ubiquity of BRD-associated pathogens (Gorden & Plummer, 2010).

The healthy bovine respiratory tract has several mechanisms which prevent harmful microorganisms from colonizing exposed tissues, including mucous and cilia to trap and physically remove microbes and particulates, the mucosal immune response, and the maintenance of a symbiotic population of commensal bacteria (Ackermann, Derscheid & Roth, 2010). When infected by primary respiratory pathogens, such as bovine respiratory syncytial virus (BRSV) (Brodersen, 2010), bovine viral diarrhea virus (BVDV) (Ridpath, 2010), bovine herpesvirus type 1 (Jones & Chowdhury, 2010), or Parainfluenza 3 (PI3) virus (Ellis, 2010), the host’s respiratory defenses may become impaired (Ames, 2002; Caswell & Williams, 2007). Pasteurella multocida, Histophilus somni, Mannheimia haemolytica (Griffin et al., 2010), and Mycoplasma bovis (Caswell et al., 2010), may be naturally present in small numbers in the nasal passages of healthy cattle but can be opportunistically pathogenic when the host’s defenses become impaired. Other factors including nutritional status, stress, and air quality may also play roles in impairing host defense mechanisms (Ackermann, Derscheid & Roth, 2010; Gorden & Plummer, 2010).

Respiratory disease in calves may involve the upper or lower respiratory tract (Panciera & Confer, 2010). Infections of the upper respiratory tract such as rhinitis typically present with ocular and nasal discharge. Infections of the lower tract, in contrast, may be challenging to detect earlier in the disease course. In addition, both upper and lower tract infections may vary in severity. Despite the variability of presentation, the observation of clinical signs is the most common method used to identify cattle in need of treatment for BRD. However, the specific criteria used to detect BRD are subjective and vary widely among dairies and observers, leading to deleterious effects on animal welfare and unnecessary treatments with antimicrobial drugs (Kelly & Janzen, 1986; Radostits & Done, 2007). Identification of etiologic agents associated with specific cases of BRD based on clinical observation alone is not typically possible.

Respiratory disease can be confirmed using a variety of methods. Necropsy and diagnostic testing for BRD pathogens is the gold standard test to diagnose BRD. Imaging modalities, such as thoracic ultrasound and radiography, are also available to diagnose BRD ante mortem, but rely on expensive equipment that require specialized training to use and interpret (Abutarbush et al., 2012; Curtis et al., 1986; Masseau et al., 2008). Molecular and biochemical diagnostic tests, such as PCR and culture on selective media, respectively, are available for ante mortem diagnosis of BRD, but are prohibitively expensive and cannot provide results at the point of treatment (Cooper & Brodersen, 2010). Necropsy and molecular and biochemical methods may also be used to identify etiologic agents associated with cases of BRD; however, such identification is not a routine practice unless BRD has become epidemic in a herd.

A simple, objective clinical scoring system to improve and standardize BRD identification in dairy calves without the need for expensive equipment would be a useful tool for farm workers, clinicians, and researchers. Clinical scoring systems use information that can be rapidly collected from patients to assess patient health and prognosis and have been used in a variety of human and veterinary applications (Champion et al., 1981; Champion et al., 1989; Sullivan, Massaro & D’Agostino, 2004; Tollner, 1982). Scoring systems assign values to clinical signs, which are used to determine a total score. The patient’s total score, in turn, should correspond to their risk or likelihood of disease. Objective methods should be used to weight scores using clinical data to ensure that similar scores represent similar risks and to optimize score performance. Clinical signs that are difficult to measure with adequate precision or that require expensive or time-consuming methods to measure should not be included (Sullivan, Massaro & D’Agostino, 2004).

Clinical scoring systems for BRD are not novel, and at least three scoring systems have previously been described to diagnose BRD in cattle. The first score published was developed by Thomas et al., as a research tool to quantitatively classify the severity of BRD in calves experimentally inoculated with BRSV or BVDV (Thomas et al., 1977). More recently, a scoring system was developed by veterinarians at the University of Wisconsin at Madison (McGuirk, 2008) and is based on five clinical signs to identify calves that should be treated for BRD. A third system known as DART (Depression, Appetite, Respiration and Temperature) was developed to identify beef cattle for BRD treatment in feedlots (Panciera & Confer, 2010).

The score described by Thomas et al. (1977) is ill-suited for field work because it uses 17 predictors, hematologic data, and requires observations specific to the pre-inoculation described in their study. The DART system does not appear in peer-reviewed literature and is difficult to standardize between locations because the clinical sign weights and decision points are not defined. The WI score is the most suitable of the three cited scores with published score weights and a decision rule. However, the WI score subdivides each of its clinical signs into 4 levels, which may have ambiguous overlap to inexperienced individuals making it difficult to appropriately classify clinical signs in calves. Until recently, information regarding these systems’ diagnostic performance in the field have not been published. A single published study has estimated the sensitivity and specificity of the WI scoring system as 55.4% and 58.0%, respectively (Buczinski et al., 2013). There is also no evidence that any of these BRD scores used quantitative methods to assign weights to clinical signs.

The objective of this study was to develop a simple scoring system with objectively assigned score weights for on-farm diagnosis of BRD in pre-weaned dairy calves. The scoring system developed in this study will be validated and used as part of a risk assessment tool under development. The risk assessment tool will be used to identify farm-specific management practices associated with BRD. A similar approach is being used to control and prevent the transmission of Johne’s disease in dairy herds (Berghaus et al., 2005).

Materials & Methods

Data used in this study were from a separate study performed to identify single nucleotide polymorphisms (SNPs) associated with BRD susceptibility. The original genetic marker research was a case-control study in which clinically ill cases were pair-matched to clinically healthy controls. The study was approved by the University of California, Davis Institutional Animal Care and Use Committee (protocol number 16431, approval date March 30, 2011).

Case definition

Clinical signs and diagnostic test results were both used to classify calves as BRD cases or controls. Histophilus somni, Pasteurella multocida, Bibersteinia trehalosi and Mannheimia haemolytica were considered as aerobic pathogens when categorizing cases and controls (Griffin et al., 2010). Calves that met any of the following three criteria were classified as cases: (1) positive for BRSV, BHV-1, or BVDV on PCR; (2) any aerobic pathogen detected on culture and WI Score ≥5; or (3) any Mycoplasma spp. detected on culture and WI Score ≥5. All other calves were classified as controls. Figure 1 depicts the algorithm for the classification of BRD cases and controls. Bovine coronavirus was not included as a criterion for case definition because current PCR assays could not differentiate between enteric and respiratory BCoV subtypes (Saif, 2010).

Results

The data set included clinical findings, viral PCR, and Mycoplasma and aerobic culture results for 2,030 calves. All calves enrolled in the study were Holsteins and were between 23 and 69 days of age when sampled. Bovine respiratory syncytial virus was detected in the upper respiratory tract of 169 (8.3%) calves. Bovine coronavirus was detected in 168 (8.3%) of calves. No calves tested positive for BVDV or BHV-1 virus. Pathogenic aerobes were cultured from the pharyngeal recess swabs of 911 calves, and Mycoplasma spp. were cultured from the pharyngeal recesses of 1,234 calves. At least one pathogen (BRSV, Mycoplasma spp., Histophilus somni, Pasteurella multocida, Bibersteinia trehalosi or Mannheimia haemolytica) was detected in 811 (87.2%) enrolled calves. A total of 932 calves (45.9%) had a WI BRD score of 5 or greater.

Eight hundred sixty-nine calves were classified as cases and 1,161 calves were classified as controls using the algorithm in Fig. 1. Only 809 pairs of the pairs enrolled contained both one case and one control and were valid to be included in the conditional logistic regression analysis. Twenty-eight pairs contained 2 cases and 166 pairs contained 2 controls. Twenty-four calves were not assigned to a pair in the data set.

Depression was noted in 530 (61.0%) cases, and 54 (4.4%) controls and the crude pair-matched OR for depression was 171.7 (95% CI: (55.2, 534.0), p < 0.0005). Diarrhea was observed in 26 cases and 3 controls, however, a pair-matched OR could not be estimated since no pairs had a control with and control without diarrhea. Six cases and zero controls were emaciated. A pair-matched OR for emaciation could not be estimated since no emaciated controls were sampled. Tachypnea was noted in 455 (52.4%) cases and 51 (4.4%) controls. The crude pair-matched OR for tachypnea was 218.5 (95% CI: (54.5, 876.4), p < 0.0005). Dyspnea was noted in 172 (19.8%) cases and 12 (1.0%) controls. The crude pair-matched OR for dyspnea was 164 (95% CI: (23, 1171), p < 0.0005). Abnormal respiration was noted in 467 (53.7%) cases and 52 (4.5%) controls. The crude pair-matched OR for abnormal respiration was 224.5 (95% CI: (56.0, 900.4), p < 0.0005). Diarrhea and emaciation were not further considered as candidates to be included in the model since pair-matched ORs could not be estimated.

Upon arrival to the facility, 332 (38.3%) cases and 329 (28.3%) controls had serum total protein concentrations equal to or less than 5.2 g/dl, which is consistent with failure of passive transfer of maternal antibodies. The crude pair-matched OR for FPT was 1.6 (95% CI: (1.3, 2.1), p < 0.0005).

Discussion

Three scoring systems were proposed for on-farm use to diagnose BRD in pre-weaned dairy calves. The score weights and cut-points of these scoring systems were selected using statistical estimation, in contrast to previously described systems in which score weights and cut-points were determined subjectively. All three scoring systems had excellent agreement with the WI score. The BRD1 scoring system correctly classified 91.5% of the calves in the study, but required handling calves without spontaneous cough in an attempt to elicit an induced cough. The BRD2 scoring system correctly classified 90.8% of the calves in the study and did not require additional calf handling to induce coughing, but had three levels of nasal discharge, instead of two. The BRD3 scoring system correctly classified 90.2% of the enrolled calves, did not require additional calf handling to attempt inducing a cough, and included only dichotomous predictors for all the clinical signs. Given similar performance to the WI scoring systems, our BRD scoring systems require less qualitative assessment of clinical signs and hence offer simpler algorithms for on-farm diagnosis of BRD in pre-weaned dairy calves. Specifically, BRD3 would be most feasible on dairies and calf raising facilities for daily observation of large calf numbers in intensive dairy systems, this is due to the simplicity of the dichotomized clinical signs and reduced calf handling required to obtain the final score. Specifically, the BRD3 scoring system will require handling a calf for rectal temperature measurement only when a calf’s total score based on all other clinical signs is equal to or greater than 4. At that time, a rectal temperature of 39.2°C (102.5°F) or greater will increase the calf’s score beyond the scoring system’s cut-point of 5. Nevertheless, the three scoring systems developed in this study are described and presented to demonstrate the selection process and allow end users to select the system that best suits their needs.

The systems presented here are the first clinical scoring systems for BRD published in peer-reviewed literature for which clinical data was used to weight scores and set cut-points. Prior to this study, the WI scoring system was the most widely accepted scoring system used in dairy medicine, but important pieces of information about this system, such as methods used to assign S_cp and S_p, sensitivity, specificity, predictive values and reliability, are absent from peer-reviewed literature. Further, the WI system uses 5 clinical predictors, each with 4 levels, and only uses four of these predictors to assign a score to the calf since the lower score of eyes and ears are dropped from the score. The systems presented here use 6 clinical predictors each with 2 levels, with the exception of nasal discharge in the BRD2 score, which has 3 levels, and use all 6 clinical predictors to assign total scores. It is anticipated that the inclusion of fewer levels of clinical predictors will improve the reliability and acceptance of the score.

The task of accurately classifying calves as BRD-positive cases or BRD-negative controls is difficult without a reference test. Identification of cases and controls without a gold-standard is a common challenge in epidemiologic studies. Case definitions that are based on multiple criteria are employed frequently and considered suitable so long as the criteria are appropriate for the goals of the study (Coggon et al., 2005). The algorithm, shown in Fig. 1, was developed to best identify cases and controls based on the available data. The detection of a primary pathogen for BRD, such as BRSV, using a sensitive and specific method, such as qPCR, indicates a high index of suspicion that the pathogen is present and the calf has BRD. The isolation of opportunistically pathogenic organisms, such as P. multocida and Mycoplasma bovis, is a much less specific criterion for BRD because these bacteria are commonly isolated from cattle without BRD. The specificity of bacterial isolation was improved by incorporating clinical information quantified by the WI scores. The WI system can be a useful method to quantify clinical BRD signs and was effectively employed in this study because two of the researchers were veterinarians with substantial calf experience which allowed for consistent evaluation and scoring of calves for BRD at enrollment. Classification as a case using the combined results of the WI score and bacterial isolation is a form of serial test interpretation, a strategy used to improve test specificity and reduce false positive results (Thurmond & Johnson, 2004).

As diagnostic tools, BRD scoring systems are most informative when estimates of the test sensitivity and specificity are known (Dohoo, Martin & Stryhn, 2010). The conditional probability that S_i,total ≥ S_cp given i was a case, and that S_i,total < S_cp given that i was a control, could be interpreted as cut-point specific estimates of sensitivity and specificity, respectively. However, these values were not interpreted as sensitivity and specificity in this study because of concerns that these estimates may be biased compared to the true sensitivity and specificity, due to two potential sources of selection bias. The first source arises from the use of the WI score in the case definition because the clinical signs used by the WI score were also included as predictors in the models for the BRD1, BRD2, and BRD3 systems. Hence, sampling of the cases and controls may not have been independent of the predictors in the model, resulting in biased parameter estimates. Although it is expected that the incorporation of microbiological results would reduce such a bias, caution is required when interpreting the estimated coefficients as causal measures of association. However, since the estimated coefficients were used to assign relative score weights, not to quantify causal associations, the method is acceptable for this application. The second source of bias arises from the pair-matched design of original study, which may have caused the sampled controls to be more similar to the cases and not representative of the referent population of all healthy calves on the facility. The CLR models account for the artificial similarities between cases and controls, but the ROC analysis used to select the optimal cut-points does not. Further research is needed to estimate the sensitivity and specificity of all three presented BRD scoring systems. Diagnostic methods such as thoracic ultrasound, auscultation, and hematology should be used to diagnose BRD in calves and estimate the sensitivity and specificity. While the methods listed are not gold-standards for BRD diagnosis, methods of estimating test sensitivity and specificity using imperfect tests have been described (Enøe, Georgiadis & Johnson, 2000). Furthermore, any proposed scoring systems should be validated using an independent data set.

Bovine coronavirus was not included in the case assignment algorithm due to the PCR lack of specificity for the respiratory BCoV subtype and the unclear role of BCoV in BRD in calves. While multiple subtypes of BCoV in cattle have been described, including respiratory and enteric calf diarrhea subtypes, based on the clinical presentation of infected cattle, no antigenic or genetic markers have been found to consistently differentiate the subtypes (Saif, 2010). Previous research has also established that animals infected with either subtype will shed viral particles in nasal secretions (Cho et al., 2001a; Cho et al., 2001b; Hasoksuz et al., 2002). Hence, the results of the BCoV PCR cannot be relied on to distinguish respiratory BCoV from enteric strains being shed from nasal mucosa. Inclusion of the BCoV term as a predictor did not significantly improve the fit of the model ( $\Delta {\mathrm{G}}_{\mathrm{BRD}1}^2=0.51$ , p = 0.439; $\Delta {\mathrm{G}}_{\mathrm{BRD}2}^2=0.10$ , p = 0.75; $\Delta {\mathrm{G}}_{\mathrm{BRD}3}^2=0.10$ , p = 0.75) and did not meaningfully change the values of the other coefficients. Additionally, the role of BCoV in calf BRD is unclear. Experimentally, BCoV has been demonstrated to induce respiratory signs following oronasal inoculation in calves in some studies (Kapil et al., 1991; McNulty et al., 1984) and not in others (Reynolds et al., 1985; Saif et al., 1986). Similarly, some observational studies have found significant associations between BCoV exposure and BRD (Storz et al., 2000), while others were unable to detect an association between serologic evidence of exposure to BCoV and incidence of BRD (Martin et al., 1998; Plummer et al., 2004).

While BHV-1, BVD, and BRSV are all BRD pathogens, no calves from the study tested positive for BHV-1 or BVDV. Several potential explanations exist for why these viruses were not detected. A greater proportion of cattle in the western US are vaccinated for BHV-1 and BVDV than other parts of the United States (USDA, 2007). It has been shown that increasing the proportion of vaccinated individuals causes an increase in age at infection thereby lowering disease prevalence in younger populations (Keeling & Rohani, 2008). A 2003 study of dairy calves in this region of California found only two calves out of 434 (0.5%) from two herds to be persistently infected (PI) with BVDV at birth (Munoz-Zanzi et al., 2003). It is expected that the prevalence of persistently infected calves on this facility in 2011 was even lower due to the facility’s selectivity of clients and the increased mortality of PI calves during the first few weeks of life. Since PCR assays for each calf were performed using material collected and stored in a single vial, the presence of positive results for BRSV and BCoV (both enveloped RNA viruses) would seem to indicate sample handling was also sufficient for BHV-1 and BVDV (enveloped DNA and RNA viruses, respectively) had they been present. The absence of BHV-1 and BVDV in this population, and subsequent omission of these viruses from the case definition algorithm, may limit the validity of the presented systems to detect cases of BRD in other populations where these pathogens may be more prevalent. Further studies in calves infected with BHV-1 and BVD viruses are required to validate the accuracy of the current scoring systems.

The current study was performed in the United States, where rectal temperature is typically measured in degrees Fahrenheit. While rectal temperature was dichotomized using the reported upper physiological limit in dairy cattle (39.2°C or 102.5°F) (Andersson & Jonasson, 1993), other temperature cut-points were also evaluated, including less than, or greater than to or equal to 38.3°C (101.0°F), 38.6°C (101.5°F), 38.9°C (102.0°F), and 39.4°C (103.0°F). However, none of the models with these alternate temperature cut-points fit models as well as 39.2°C or 102.5°F (data not shown).

The values of S_p in this study were assigned based on the values of the CLR coefficients. This approach to S_p values assignment was adapted from a method described previously (Segev et al., 2008), which assigned S_p values based on odds ratios (ORs) estimated by exponentiating logistic regression coefficients. This change was made because logistic regression models assume that the joint effects of variables relate to the odds of an outcome in a multiplicative, not additive, manner (Breslow & Day, 1980). Therefore, the sum of the ORs as determined by multivariable logistic regression does not necessarily represent the OR of disease in an individual with multiple exposures compared to an individual with no exposures. Another method that determines S_p values based on logistic regression coefficients instead of ORs has been described (Sullivan, Massaro & D’Agostino, 2004); however the pair-matched case-control design of the current study was incompatible with Sullivan’s method.

Conclusion

The BRD scoring systems developed in this study provide three options to assess the BRD status of pre-weaned dairy calves. The scoring systems utilize objective and easy to obtain criteria thereby reducing subjectivity. For ease of use, an investigator may assess the presence or absence of ocular discharge, nasal discharge, ear droop or head tilt, respiratory quality and spontaneous coughing. Calves with abnormal ear or head carriage, or calves with nasal discharge and one other clinical sign, or calves that have any three clinical signs are BRD cases based on the BRD3 scoring system. Only calves with nasal discharge or calves with two other clinical signs (spontaneous coughing, ocular discharge or abnormal respiratory) would require handling the calf to measure its rectal temperature and confirm BRD status if the temperature is ≥39.2°C or 102.5°F.