Assessment of Diagnostic Tools for Identifying Cattle Shedding and Super-Shedding Escherichia coli O157:H7 in a Longitudinal Study of Naturally Infected Feedlot Steers in Ohio

Abstract

The objectives of this study were to compare the performance of different diagnostic protocols (rectoanal mucosal swabs and immunomagnetic separation [RAMS-IMS], fecal samples and IMS [fecal-IMS], and direct plating) to determine the prevalence of Escherichia coli O157:H7 and to evaluate the pattern of E. coli O157:H7 shedding and super-shedding (defined as having a direct plating count equal to or >10⁴ colony forming units of E. coli O157:H7 per gram of feces) in a longitudinal study of naturally infected feedlot steers. RAMS and fecal grab samples were obtained at 14-day intervals from 168 Angus-cross beef steers over a period of 22 weeks. Fecal samples were assessed by direct plating and IMS, whereas RAMS were tested only by enrichment followed by IMS to recover E. coli O157:H7. The period prevalence for shedding was high (62%) among feedlot steers and super-shedding was higher (23%) than anticipated. Although direct plating was the least sensitive method to detect E. coli O157:H7-positive samples, over 20% of high bacterial load samples were not detected by RAMS-IMS and/or fecal-IMS. The sensitivity of RAMS-IMS, fecal-IMS, and direct plating protocols was estimated using simple and multilevel mixed-effects logistic regression models, in which the dependent variable was the dichotomous results of each test and gold standard (i.e., parallel interpretation of the three protocols)-positive individuals were included as an independent variable along with other factors such as dietary supplements, time of sampling, and being exposed to a super-shedding pen-mate. The associations between these factors and the sensitivity of the diagnostic protocols were not statistically significant. In conclusion, differences in the reported impact of diet and probiotics on the shedding of E. coli O157:H7 in previous studies using RAMS-IMS or fecal-IMS were unlikely due to their impact on test performance.

Introduction

E scherichia coli O157:H7 is an important foodborne pathogen that causes hemorrhagic colitis and hemolytic uremic syndrome among humans (Karmali, 2004). Experimental and observational studies have shown that cattle shed this bacterium intermittently (Hancock et al., 1997; Shere et al., 1998; Sargeant et al., 2000). After ingestion of sufficient numbers, E. coli O157:H7 colonizes the intestinal tract of cattle and other ruminants without causing disease (Cray and Moon, 1995; Wray et al., 2000). The fact that cattle are asymptomatic shedders is important from a control standpoint since these animals carry the bacteria without showing any signs of clinical disease, and contribute to environmental contamination and dissemination of this pathogen to other herd mates.

It has been proposed that fecal shedding may be either transient (Faith et al., 1996; Paiba et al., 2003; Robinson et al., 2004), where cattle shed the pathogen briefly after exposure, or more long term, resulting from colonization of the gastrointestinal tract and, more precisely, the rectoanal mucosal junction (Naylor et al., 2003). Experimental studies have shown that colonization at the rectoanal mucosal junction is related to excretion of high levels of E. coli O157 in feces (Naylor et al., 2003; Low et al., 2005). Those animals that shed E. coli O157 at much higher concentrations than others have been called high-shedders or super-shedders (Low et al., 2005; Matthews et al., 2006a; Cobbold et al., 2007). Recently, Chase-Topping et al. (2008), on the basis of the available evidence, proposed a working definition of an E. coli O157 super-shedder as an animal that excretes >10⁴ colony forming units (CFU) per gram of feces. Super-shedders are purported to have a substantial impact on the on-farm epidemiology of E. coli O157:H7, because while they constitute a small proportion of cattle, it has been estimated that they may be responsible for >80% to 96% of the E. coli O157:H7 bacteria shed by cattle (Omikasin et al., 2003; Matthews et al., 2006b). However, many of the studies that have addressed the super-shedder concept have based their definitions and prevalence estimates on cross-sectional studies and/or simulation models (Omikasin et al., 2003; Fegan et al., 2004; Matthews et al., 2006a).

Culture methods to detect fecal shedding typically involve the use of an immunomagnetic separation (IMS) technique, an isolation procedure that includes an enrichment broth and the use of selective and differential media followed by molecular or biochemical methods for confirmation purposes. Rice and collaborators (2003), drawing on the observation that E. coli O157:H7 specifically colonizes the lymphoid follicle-dense mucosal epithelium at the terminal rectum, developed a rectoanal mucosal swab (RAMS) technique that involves sampling the mucosal surface of the rectoanal junction along with IMS to improve test performance. Different studies have arrived at different conclusions concerning the relative sensitivity of fecal-IMS and RAMS-IMS procedures for detecting the carriage of E. coli O157:H7 in cattle (Rice et al., 2003; Greenquist et al., 2005; Khaitsa et al., 2005; Fox et al., 2008; Niu et al., 2008). Rice et al.'s study (2003) suggested that the culture of swabs from the rectoanal junction mucosa was, in almost all cases, as sensitive as and usually more sensitive than culture of feces at detecting E. coli O157:H7 in cattle. However, fecal cultures were more sensitive than RAMS cultures at detecting E. coli O157:H7 initially after artificial exposure to E. coli O157:H7, which may indicate that RAMS appears to delineate colonized from transiently shedding cattle. Similarly, studies performed by Greenquist et al. (2005) and Fox et al. (2008) concluded that mucosal swab sampling was more sensitive than fecal sampling for determining the prevalence of E. coli O157 in feedlot cattle. Conversely, according to Khaitsa et al. (2005) and Niu et al. (2008) fecal sampling appears to be superior to RAMS for detection of E. coli O157:H7, although labor and animal restraint requirements for fecal collection were higher than those for the RAMS technique. Factors such as the type of study design (longitudinal vs. cross-sectional), the use of artificial versus naturally infected animals, the type of cattle production system targeted, and the existence of multiple different testing procedures pose problems when comparing test results from different studies. Further, previous results indicated that different dietary supplements affect the probability of testing positive to E. coli O157:H7 depending on the testing procedure (i.e., RAMS-IMS vs. fecal-IMS) (Cernicchiaro et al., 2010). However, it was unclear if the differences were due to the impact these factors have on the biology of E. coli O157:H7 in feces and RAMS, or on test performance.

The objectives of our study were the following: (i) to compare the performance of RAMS-IMS, fecal-IMS, and direct plating tests to determine the prevalence of E. coli O157:H7 in feedlot steers; (ii) to compare the level of agreement among RAMS-IMS, fecal-IMS, and direct plating at the animal and sample levels; (iii) to evaluate the impact of diet, time of sampling, and being exposed to a super-shedding pen-mate on the performance of these tests; and (iv) to determine the sensitivity of these procedures to detect super-shedding animals from a longitudinal study of naturally infected feedlot steers.

Materials and Methods

Details on study population, sampling procedures, and laboratory protocols used in this study have been described elsewhere (Cernicchiaro et al., 2010). Briefly, 168 Angus-cross beef steers, initially weighing 250 to 340 kg, were introduced on October 5 and on November 2, 2005, into the beef feedlot facility located at the Ohio Agricultural Research and Development Center in Wooster, Ohio. Approximately 75% of cattle were supplied by stocker farms that usually provide calves to be finished in this research facility. The remaining 25% was provided by an Ohio cattle buyer. After acclimation of the animals to the facility, the trial began on November 21, 2005, and was conducted over a period of 22 weeks until April 25, 2006.

From each animal, one RAMS and at least 10 g of fecal grab samples were collected at 14-day intervals throughout the finishing period. RAMS samples were obtained by swabbing the rectoanal mucosa of each animal with a dry cotton-tipped 15-cm cleaning stick as described elsewhere (Rice et al., 2003; Khaitsa et al., 2005). Immediately after swabbing, rectal fecal grabs were obtained using a new disposable sleeve glove for each animal. Samples were transported within 1 hour after sampling to the Food Animal Health Research Program Laboratory in Wooster, Ohio.

For enumeration of colonies, a technique developed by LeJeune et al. (2006) based on Sorbitol MacConkey agar (SMAC) with nalidixic acid was employed in our study. Briefly, 10 g of fecal grab samples were homogenized in 90 mL of buffered peptone water (BPW) to make a 10⁻¹ dilution. Using 96-well plates containing 1 mL BPW, serial dilutions were made starting with the 10⁻¹ dilution aliquot up to 10⁻³. A 100-μL sample of each 10⁻³ dilution was spread-plated onto a 100 × 15 mm sorbitol MacConkey agar plate (CT-SMAC) containing cefixime (50 ng/mL) and potassium tellurite (2.5 μg/mL). Plates were incubated at 37°C for 18 h. The CT-SMAC plates were screened for colorless colony growth and plates were enumerated. A latex agglutination assay was used on up to 10 representative sorbitol-negative suspect colonies to confirm the O157 antigen (Oxoid Ltd., Nepean, Canada). The number of sorbitol-negative colonies was multiplied by the fraction of latex-positive colonies and the fecal dilution to estimate the number of E. coli O157 colonies present in each gram of feces.

The rest of the homogenate from the fecal grab samples and the RAMS suspended in BPW were enriched overnight at 42°C and processed by an IMS technique following the manufacturer's recommendations (Invitrogen Dynal A.S., Oslo, Norway). After automatic IMS, samples were plated onto CT-SMAC plates containing cefiximine (50 μL/mL) and potassium tellurite (2.5 mg/mL), and plates were incubated at 37°C for 24 h. Using sterile toothpicks, up to 5 colonies from IMS CT-SMAC plates were transferred to 96-well plates containing 180-μL of E. coli 4-methylumbelliferyl-β-D-glucuronide (EC Mug; EC media: Neogen-Acumedia Manufacturers Inc., Lansing, MI; MUG: Biosynth AG, Staad, Switzerland) agar. On each plate, one well was inoculated as a positive control and one well was inoculated for a negative control. Plates were incubated at 37°C for 24 h. The EC MUG fluorescence of 96-well plates was screened and colonies were transferred, using a 96-well replicator, to 150 × 15 mm MAC plates. A latex agglutination assay was used to confirm the presence of the O157 antigen of lactose-positive, MUG-negative colonies.

Descriptive statistics

The period prevalence of E. coli O157:H7 at the sample level was estimated as the proportion of samples testing positive by a particular diagnostic protocol (RAMS-IMS, fecal-IMS, or direct plating) divided by the total number of samples tested for that specific protocol. Period prevalence of E. coli O157:H7 at the animal level was determined as the proportion of steers testing positive at least once by a particular diagnostic protocol divided by the total number of cattle tested by that protocol. Further, pen period prevalence of E. coli O157:H7 was estimated as the proportion of samples testing positive in each pen divided by the total number of samples tested per pen throughout the entire study. In addition, these prevalences were also estimated based on a parallel interpretation of the protocols where a positive sample, animal, or pen was based on a positive result from any protocol.

Duration of shedding was estimated by determining the number of sampling dates (bi-weekly intervals) between the first and last consecutive sample visits that yielded E. coli O157-positive samples detected by the different diagnostic protocols.

Comparison of diagnostic protocols

The Cohen's kappa statistic (Cohen, 1960), McNemar's Chi-square test (McNemar, 1947), and prevalence-adjusted bias-adjusted kappa (PABAK) (Byrt et al., 1993; Mak et al., 2004) were estimated to assess the overall agreement beyond chance on the presence or absence of E. coli O157:H7 in fecal samples at the sample and animal levels by pair-wise comparison of the proportion of positives detected by the three diagnostic protocols.

The kappa statistic is biased if the prevalence of the outcome is very low or very high (outside 20% to 80%), or McNemar's indicates that the proportions between two tests are different (Zwick, 1988; Feinstein and Cicchetti, 1990; Dohoo et al., 2009). However, to date, the use of alternatives to kappa is uncommon in the veterinary literature (Petersen et al., 2004; Thomsen and Baadsgaard, 2006). PABAK was developed to adjust the measures of agreement by the potential effect of prevalence and bias; thus, if in fact the prevalence is impacting kappa or there is bias on the measurements, PABAK will reflect a less-biased estimate of agreement than kappa. In addition, relatively large differences between kappa and PABAK estimates indicate that there was in fact evidence of bias in the agreement.

The degree of agreement was interpreted based on the following scale, originally intended for subjective levels of agreement, described by Landis and Koch (1977): ≤0, poor agreement; 0.01–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement.

To estimate the relative sensitivities of the different diagnostic protocols, we considered being positive to at least one protocol our gold standard, as the interpretation of the three protocols in parallel would increase the sensitivity of detection. Relative sensitivity for each of the three diagnostic protocols was estimated at the sample and animal levels as the proportion of samples, or animals, testing positive for each protocol divided by the number of samples, or animals, testing positive by at least one of these protocols (parallel interpretation of tests). Further, the specificity of all protocols at the sample and animal level was treated as 100%, assuming that no false-positives can arise from fecal culture-based tests.

Direct plating results and E. coli O157:H7 super-shedders

Positive direct plating results were quantified and the amount of bacteria was expressed as the number of CFU/g of feces. Super-shedders were defined as those animals producing counts of 10⁴ CFU of E. coli O157:H7 per gram of feces or higher by direct plating onto CT-SMAC plates, in accordance to the working definition of Chase-Topping et al. (2008). In addition, to determine the ability of the RAMS-IMS and fecal-IMS protocols to detect super-shedding events, as defined by the direct plating method, the results of these two protocols were compared to the direct plating test at the sample and animal levels and the Cohen's kappa statistic, McNemar's Chi-square test, and PABAK were estimated.

Estimation of sensitivity of diagnostic protocols using logistic regression models

To determine if the characteristics of the different diagnostic protocols varied with diet, time of sampling, and exposure to a super-shedding pen-mate, the sensitivities of RAMS-IMS, fecal-IMS, and direct plating protocols were estimated using logistic regression models, as proposed by Coughlin et al. (1992) and applied by different authors (Greiner and Gardner, 2000; Lindberg et al., 2001; Dohoo et al., 2009). Time of sampling, measured as the time since the start of the trial in days (range = 0–154 days), was included as an independent variable and modeled on a continuous scale. In addition, the following treatments included in a concurrent study intended to examine the effects of different dietary supplements on the prevalence of E. coli O157:H7 on the same study units (Cernicchiaro et al., 2010) were included in the analysis: two types of corn (high moisture corn vs. dry-whole shelled corn); two natural feed additives (Aspergillus oryzae [Amaferm] vs. Saccharomyces cerevisiae boulardii [Levucell] vs. no supplementation [control group]); and two levels of vitamin A (2200 IU/kg of feed vs. no supplementation). Experimental details on these interventions were described elsewhere (Cernicchiaro et al., 2010). In addition, exposure to a super-shedding animal in the pen was modeled as a dichotomous variable and it was defined as having a super-shedding pen-mate at the specific time of sampling. The actual super-shedder was coded as unexposed if there was no other super-shedder in the same pen on the same sampling date or exposed if another super-shedding animal was present.

On the basis of Coughlin et al.'s approach (1992), estimates of sensitivity and specificity can be derived by including the dichotomous results of the diagnostic test as the dependent variable and the results of the gold standard (true disease status) as an independent variable after adjusting for different covariates. In our study the variable being positive to E. coli O157:H7 by at least one diagnostic protocol was considered our gold standard. In addition, the hierarchical structure of the data can be taken into account in these models by including random effects for each level of clustering (Lindberg et al., 2001; Dohoo et al., 2009).

Baseline observations (time of sampling = 0) were removed from the dataset to perform this analysis because no baseline measurements were available for the fecal-IMS protocols and only one sample was detected positive by direct plating. The interpretation of the three protocols in parallel, our gold standard test, would have primarily reflected the results of the RAMS-IMS protocol, making this reference test not appropriate to estimate the sensitivity of the fecal-IMS and direct plating protocols and likely overestimating the sensitivity of the RAMS-IMS in that particular time period.

Initially, all predictors were examined in a univariable screen. The linearity assumption between the log odds of the different responses and time of sampling was assessed using graphical methods (i.e., lowess smoothing of the logit of the outcome on the continuous predictor). If the assumption was not met, depending on the shape of these relationships, the predictor variable was categorized unless it was more appropriately transformed (e.g., natural logarithm) or modeled with the addition of a quadratic term (Dohoo et al., 2009).

A multivariable main effects model including all predictors was built manually using a backward elimination procedure. One by one, each nonsignificant variable (p > 0.05) was removed from the model to determine their potential confounding effect. A confounding variable was defined as any nonintervening variable that resulted in a 20% or greater change in the coefficient of a statistically significant variable when it was removed from the model (Dohoo et al., 2009). All possible two- and three-way interaction terms were created between all predictors and tested at a 5% significance level by adding them to the model one at a time. Main effects that were part of significant interaction terms (p < 0.05) were retained in the model regardless of their individual significance. After the models were fitted using simple logistic regression models, random effects were added. Multi-level mixed-effects logistic regression models using adaptive quadrature were fitted in Stata 10 (StataCorp LP, College Station, TX) using the “xtmelogit” procedure (Rabe-Hesketh et al., 2002). The random-effects model had a three-level hierarchical structure of samples nested in cattle and of cattle nested within pens. A manual model building approach was employed to fit the following two types of models: (1) multivariable multi-level logistic regression models with random intercepts for pen and animal, and (2) multivariable multi-level logistic regression models with random intercepts for pen and animal and random slope for time of sampling with an unstructured covariance structure. The different models regressed each response variable independently (RAMS-IMS, fecal-IMS, and direct platting) when the results of the gold standard were set to one, to obtain the modeled estimates of sensitivity (Dohoo et al., 2009). Thus, sensitivity was estimated by restricting our analysis to gold standard-positive individuals. For the random-intercept models, intra-class correlation coefficients were computed at the pen and animal levels using a latent-variable technique (Dohoo et al., 2009). Information criteria (Akaike's Information Criterion [AIC] and the Schwarz's Bayesian Information Criterion [BIC]) were employed to compare the models, and the model with the lowest value of these criteria was deemed to be the best fitting model.

The sensitivities of the different diagnostic methods and their respective 95% confidence intervals (CI) were estimated based on the model predictors according to the formula proposed by Coughlin et al. (1992): \documentclass{aastex} \usepackage{amsbsy} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{bm} \usepackage{mathrsfs} \usepackage{pifont} \usepackage{stmaryrd} \usepackage{textcomp} \usepackage{portland, xspace} \usepackage{amsmath, amsxtra} \pagestyle{empty} \DeclareMathSizes {10} {9} {7} {6} \begin{document} \begin{align*}(1) \ {\rm Sensitivity} = \frac{1} {1 + {\rm exp} \left[ - \left(\alpha \,+ \sum \nolimits_{k = 1}^{k} \beta_{k}X_{k} \right) \right.},\\ \quad\hbox{when gold standard} = 1 \end{align*} \end{document} where α corresponds to the intercept of the model and Σ^k β_kX_k denotes the linear function of (1, …, k) explanatory variables, one of which corresponds to the binary results of the gold standard along with their coefficients (β).

Mixed models provide cluster-specific parameters that correspond to values estimated for a particular cluster (in our case, pen and animal random effects), as opposed to population-averaged or marginal estimates (Dohoo et al., 2009). The latter compares the results of a sample being detected positive or negative by a particular diagnostic protocol from any pen and any animal in the study. Therefore, to compare the results of conditional and marginal models employed in our study, which differ in their ability to model the variance of the data, population-averaged parameters were estimated from cluster-specific parameters using the following formula: \documentclass{aastex} \usepackage{amsbsy} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{bm} \usepackage{mathrsfs} \usepackage{pifont} \usepackage{stmaryrd} \usepackage{textcomp} \usepackage{portland, xspace} \usepackage{amsmath, amsxtra} \pagestyle{empty} \DeclareMathSizes {10} {9} {7} {6} \begin{document} \begin{align*} \beta^ { \rm PA } = \frac { \beta^ { \rm SS } } { \sqrt { 1 \,+\, 0.346 \sigma^ { 2 } } } \end{align*} \end{document} where β ^SS is the cluster- or subject-specific coefficient (on the logit scale) and σ ² is the sum of all variance components (Dohoo et al., 2009).

Results

Baseline prevalence

On the basis of the RAMS-IMS method, E. coli O157:H7 was isolated from 4.6% (7/153) of the steers during the baseline sampling during the initial weighing of the cattle. For direct plating, only one animal tested positive for E. coli O157:H7 (0.8%, 1/129); the same sample also tested positive by the RAMS-IMS method and was deemed to belong to a super-shedding animal (>10⁴ CFU/g of feces). Unfortunately, because fecal grab samples were not tested by fecal-IMS on the first date of sampling (day 0), baseline prevalence was not available for this protocol.

Overall prevalence estimates of E. coli O157:H7 at the sample, animal, and pen levels

RAMS-IMS and fecal-IMS showed similar capabilities in detecting samples and animals positive to E. coli O157:H7, whereas direct plating was the least sensitive method. Overall, sample period prevalence was 10.3%, 10.7%, 4.1%, and 13.8% for RAMS-IMS, fecal-IMS, direct plating, and by any protocol, respectively. Animal period prevalence was 54.2%, 48.2%, 35.1%, and 61.9% for RAMS-IMS, fecal-IMS, direct plating, and by any protocol, respectively, whereas period prevalence at the pen level was 87.5% for RAMS-IMS and fecal-IMS, 75.2% for direct plating, and 95.8% for any protocol (Table 1).

Table 1.

Detection of Escherichia coli O157:H7 in Fecal Samples Using Rectoanal Mucosal Swabs-Immunomagnetic Separation, Fecal-Immunomagnetic Separation, and Direct Plating Protocols at the Sample, Animal, and Pen Levels from a Longitudinal Study of Naturally Infected Feedlot Steers in Ohio

	Proportion of positive samples ¹			Proportion of positive animals ²			Proportion of positive pens ³
Protocol	Prop.	% ⁴	95% CI (%)	Prop.	% ⁴	95% CI (%)	Prop.	% ⁴	95% CI (%)
RAMS-IMS	187/1821	10.3^a	8.9–11.7	91/168	54.2^a	46.7–61.7	21/24	87.5^a	67.6–97.3
Fecal-IMS	170/1591	10.7^a	9.2–12.2	81/168	48.2^a	40.6–55.8	21/24	87.5^a	67.6–97.3
Direct plating	74/1784	4.1^b	3.2–5.0	59/168	35.1^b	27.9–42.3	19/24	79.2^a	57.8–92.9
Positive to any test	254/1836	13.8	12.2–15.4	104/168	61.9	54.6–69.2	23/24	95.8	78.9–99.9

Estimated as the number of positive samples divided by the total number of samples tested throughout the study period.

Estimated as the number of animals with at least one positive test result throughout the study divided by the total number of animals tested.

Estimated as the number of pens with at least one positive test result throughout the study divided by the total number of pens tested.

Values in each column followed by different letters were significantly different among the three tests being compared (p < 0.05, as determined by a Chi-square test).

RAMS-IMS, rectoanal mucosal swabs-immunomagnetic separation; Fecal-IMS, fecal grabs-immunomagnetic separation; CI, confidence intervals.

At the animal level, prevalence ranged between 0% and 35.1% on any sampling date with a mean of 8.5% of animals shedding E. coli O157:H7 per sample date based on any test. Moreover, 18 animals (11%) showed positive samples on three or more consecutive sampling dates during the study period. The maximum duration of shedding lasted eight consecutive sampling dates based on fecal-IMS testing (n = 1). On the basis of RAMS-IMS, the maximum number of sampling dates with uninterrupted positive results was seven (n = 1), whereas positive samples for four consecutive dates were the maximum number detected by the direct plating method (n = 1). This last animal also tested positive on three of the four sampling dates to both RAMS-IMS and fecal-IMS tests. In contrast, a total of 42 animals had at least one episode of intermittent shedding (two or more positive samples separated by one or more negative samples) within a specific diagnostic protocol throughout the entire feeding period.

At the pen level, 96% (23/24) of pens were considered positive at least once to any protocol throughout the study period. Sample level prevalence by pen, estimated considering all samples collected in each pen during the entire study period, ranged from 0% to 30.9% by RAMS-IMS, 0% to 34.4% by fecal-IMS, and 0% to 17.3% by direct plating. Pen prevalence ranged from 0% to 41.2% by at least one diagnostic protocol. Throughout the entire study period, pens 5, 6, and 22 remained negative based on RAMS-IMS. Pens 5, 22, and 23 did not produce positive results by fecal-IMS, and pens 6, 13, 19, 20, and 22 tested negative by direct plating. All samples from pen 22 remained negative during the study period based on the three diagnostic protocols.

Comparison of diagnostic protocols

The McNemar's Chi-square tests for the comparison between RAMS-IMS versus direct plating and fecal-IMS versus direct plating at the sample and animal levels were statistically significant (p < 0.05), suggesting a serious disagreement between the protocols and evidence of bias in our kappa statistics (Table 2). PABAK indices showed that there was in fact evidence of bias in the kappa statistic at the sample level, as indicated by the relatively large difference between the kappa and PABAK estimates. PABAK estimates at the sample level were as follows: 0.84, 0.83, and 0.86 for the comparisons between RAMS-IMS versus fecal-IMS, RAMS-IMS versus direct plating and between fecal-IMS and direct plating, respectively. However, at the animal level, the values for kappa and PABAK estimates were almost identical, suggesting that the bias in the kappa estimates was minimal at this level and kappa was also a suitable measure of agreement (Table 2). Estimates of the PABAK statistic at the animal level were 0.68, 0.50, and 0.60 for the comparisons between RAMS-IMS versus fecal-IMS, RAMS-IMS versus direct plating, and between fecal-IMS and direct plating, respectively, indicating moderate to substantial agreement (Landis and Koch, 1977) (Table 2).

Table 2.

Cohen's Kappa Statistic and Prevalence-Adjusted Bias-Adjusted Kappa Among Diagnostic Protocols for Detection of Escherichia coli O157:H7 Shedding at the Sample and Animal Levels from a Longitudinal Study of Naturally Infected Steers in Ohio

	RAMS-IMS versus Fecal-IMS		RAMS-IMS versus Direct		Fecal-IMS versus Direct
Statistic	Sample	Animal	Sample	Animal	Sample	Animal
Kappa^a	0.59	0.68	0.41	0.51	0.51	0.60
Kappa 95% CI	0.54–0.64	0.53–0.84	0.36–0.45	0.37–0.58	0.47–0.56	0.45–0.74
p-Value	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
McNemar's χ ²	0.20	1.38	81.20	20.51	84.77	16.03
Prob > χ ²	0.6547	0.2393	<0.0001	<0.0001	<0.0001	0.0001
PABAK^b	0.84	0.68	0.83	0.50	0.86	0.60
PABAK 95% CI	0.82–0.87	0.55–0.82	0.80–0.86	0.37–0.63	0.84–0.89	0.48–0.72
p-Value	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001

Degree of agreement based on the Cohen's kappa statistic as proposed by Landis and Koch (1977): ≤0, poor agreement; 0.01–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; 0.81–1.00 almost perfect agreement.

Prevalence-adjusted bias-adjusted kappa (Byrt et al., 1993).

Estimation of sensitivity of RAMS-IMS, fecal-IMS, and direct plating protocols

At the sample level, the relative sensitivities were slightly lower than at the animal level, at 72.9% (95% CI: 66.9%–78.3%), 70.2% (95% CI: 64.1%–75.9%), and 29.6% (95% CI: 23.9%–35.7%) for RAMS-IMS, fecal-IMS, and direct plating protocols, respectively. At the animal level, without considering observations from baseline, the sensitivity was 86.6% (95% CI: 78.2%–92.7%) for the RAMS-IMS protocol, 83.5% (95% CI: 74.6%–90.3%) for the fecal-IMS protocol, and 59.8% (95% CI: 49.3%–69.6%) for the direct plating protocol.

Estimation of sensitivity of diagnostic protocols using logistic regression models

Among the random effects models, the AIC and BIC showed lower values for the multi-level logistic regression models with random intercepts for pen and animal compared to the random slope models. Although information measures were often lower for the simple logistic regression models than for the random effects models, we provided the estimates from both models since the latter accounts for clustering derived from animals sharing a common environment (i.e., pen) and repeated measurements within individual animals (Table 3). Clustering was negligible for the estimation of sensitivity for the RAMS-IMS protocol given the small values obtained for the variance components for pen and animal (Table 3). Variance at the animal level was minimal for direct plating and the intra-class correlation coefficient for both levels of clustering was low (2.4%). However, for fecal-IMS, the correlation between samples of different animals in the same pen was 12.4%, whereas the correlation between samples in the same animal was 20.1% (Table 3). Neither the main predictors of interest (time components and exposure to a super-shedding pen-mate) nor the dietary treatments (corn type, feed additives, and vitamin A) were significantly associated with the sensitivity of the diagnostic protocols as estimated from the coefficients of the simple and multi-level logistic regression models (Table 3).

Table 3.

Sensitivity Estimates for Rectoanal Mucosal Swabs-Immunomagnetic Separation, Fecal-Immunomagnetic Separation, and Direct Plating Protocols for Observations from Day 14 to 154 on the Feeding Trial Estimated Based on Intercept-Only Simple and Mixed-Effect Logistic Regression Models from a Longitudinal Study of Naturally Infected Feedlot Steers in Ohio

	Simple logistic regression		Mixed-level logistic regression ^a
Outcome	Estimate	95% CI	Estimate	95% CI
RAMS-IMS
Sensitivity^b (%)	72.9	70.0–78.0	72.9	70.0–78.0
AIC	290.75		294.75
BIC	294.25		305.27
Variance components			Variance	SE
Pen	na		8.24 × 10⁻²⁰	0.40
Animal	na		2.37 × 10⁻¹⁶	0.43
Fecal-IMS
Sensitivity^b (%)	70.2	64.2–75.7	71.9	60.9–80.7
AIC	296.63		293.85
BIC	300.12		304.31
Variance components			Variance	SE
Pen	na		0.51	0.28
Animal	na		0.32	0.42
Direct plating
Sensitivity^b (%)	29.6	19.5–35.5	29.1	23.1–36.0
AIC	301.88		305.37
BIC	305.39		315.89
Variance components			Variance	SE
Pen	na		0.08	0.23
Animal	na		9.00 × 10⁻¹³	0.48

Mixed-effects logistic regression model with random intercepts for pen and animal.

On the basis of Coughlin et al.'s formula (Coughlin et al., 1992).

AIC, Akaike's information criterion; BIC, Schwarz's Bayesian information criterion; na, not applicable; SE, standard error.

On the basis of simple logistic regression models considering no explanatory variables (intercept-only models) and no random effects, the sensitivities for the RAMS-IMS, fecal-IMS, and direct plating protocols were 72.9%, 70.2%, and 29.6%, respectively, which are the same values we obtained by estimating the sensitivity of these methods at the sample level using contingency tables. However, when random effects were added to account for clustering at the pen and animal levels, sample level sensitivities were 72.9%, 71.9%, and 29.1% for RAMS-IMS, fecal-IMS, and direct plating, respectively. In addition, these models yielded slightly different 95% CI for sensitivity compared to the ones produced by simple logistic regression models (Table 3). In addition, the population-averaged sensitivities derived from the random effects models for the RAMS-IMS, fecal-IMS, and direct plating protocols were 72.9%, 68.6%, and 29.9%, respectively. Given the small between-pen and between-animal variation found in these models, differences between cluster-specific and population-averaged estimates were nonexistent or minimal. Observed differences in sensitivity estimates derived from multilevel models, even after conversion to population averaged estimates, can also be affected by varying group sizes, but this was not an issue due to the balanced nature of our study (Dohoo et al., 2009).

Sensitivity of super-shedding detection

Thirty-five percent of steers (59/168; 95% CI: 27.9%–42.3%) had at least one direct positive test result throughout the entire study period (range of counts/sample = 10^2.0–10^4.7 CFU/g). Eighty-one percent of these animals had only one positive test, whereas the remainder had more than one positive test based on direct plating with the following positive counts: 2 (n = 8); 3 (n = 2); 4 (n = 1). Counts decreased over time except for one animal where the second sample showed an increase and one where the count remained the same. On the basis of direct plating, ∼1% of animals (1/168; 95% CI: 0.01%–3.3%) showed counts between 10² and <10³ CFU of E. coli O157:H7 per gram of feces, 12% (20/168; 95% CI: 7.4%–17.8%) showed at least one count equal or higher than 10³ CFU/g but <10⁴ CFU/g, and 22.6% (38/168; 95% CI: 16.5%–29.7%) of the steers experienced at least one super-shedding event throughout the entire feeding period (Table 4). In addition, overall, 2.2% (39/1784; 95% CI: 1.6%–3.0%) of samples showed counts equal or higher than 10⁴ CFU/g of feces. One to five super-shedding steers were identified per pen over the entire feeding period with 71% (95% CI: 48.9%–87.4%) of pens having at least one animal experiencing a super-shedding event. Only 78.9% (95% CI: 62.7%–90.4%) of super-shedding animals were detected positive by RAMS-IMS, whereas 89.5% (95% CI: 78.6%–98.3%) of super-shedders tested positive by fecal-IMS, at any time during the study period.

Table 4.

Ability of Direct Plating, Rectoanal Mucosal Swabs-Immunomagnetic Separation, and Fecal-Immunomagnetic Separation to Detect Super-Shedding Events at the Sample and Animal Levels

	Concentration of bacteria (CFU/g)
	10²–10³	10³–10⁴	≥ 10⁴ ^a
Sample level
No. of samples^b	7	28	39
% of samples (95% CI)^c	0.4 (1.6–8.1)	1.6 (1.0–2.3)	2.2 (1.6–3.0)
Animal level
No. of animals^d	1	20	38
% of animals (95% CI)^e	0.6 (0.02–3.3)	11.9 (7.4–17.8)	22.6 (16.5–29.7)
Sample level Se (%)^f
RAMS-IMS	57.1 (18.4–90.1)	78.6 (59.0–91.7)	79.5 (63.5–90.7)
Fecal-IMS	85.7 (42.1–99.6)	89.3 (71.8–97.7)	89.7 (75.8–97.1)
Animal level Se (%)^g
RAMS-IMS	0 (0–97.5)	80 (56.3–94.3)	78.9 (62.7–90.4)
Fecal-IMS	100 (2.5–100)	95 (75.1–99.9)	89.5 (75.2–97.1)

Super-shedding events according to the Chase-Topping et al.'s definition (Chase-Topping et al., 2008).

Number of samples detected by direct plating at each specified concentration (based on the maximum count).

The percentage of samples at each concentration was estimated by dividing the number of samples detected by direct plating at each concentration by the total number of samples tested by direct plating (n = 1784).

Number of animals with at least one direct plating sample detected by direct plating at each specified concentration (based on the maximum count).

The percentage of animals at each concentration was estimated by dividing the number of animals detected by direct plating at each concentration by the total number of animals tested by direct plating (n = 168).

Sample-level sensitivity was estimated as the number of samples detected positive by either RAMS-IMS or fecal-IMS protocol at each concentration (as detected by direct plating) divided by the total number of samples detected by direct plating at that concentration.

Animal-level sensitivity was estimated as the number of animals detected positive by either RAMS-IMS or fecal-IMS protocol at each concentration (as detected by direct plating) divided by the total number of animals detected by direct plating at that concentration.

The McNemar's Chi-square tests indicated that there were significant differences in the proportion of positives in all comparisons among protocols, which may indicate that the results of the kappa statistics were biased. When PABAK was estimated, it showed that the kappa statistics at the sample level were actually biased and they resulted in much higher estimates of agreement beyond chance after adjusting for the potential effect of extreme prevalence and inter-test bias. PABAK values for comparisons at the sample level were 0.81, 0.83, and 0.75 for RAMS-IMS versus direct plating, fecal-IMS versus direct plating, and any positive test versus direct plating, respectively. At the animal level, however, the estimates of PABAK were almost identical to those of the unadjusted kappa, indicating that there was only a small degree of bias in the kappa statistics. PABAK values were 0.26, 0.39, and 0.21 for RAMS-IMS versus direct plating, fecal-IMS versus direct plating, and any positive versus direct plating, respectively.

Discussion

The results of our study indicate that shedding of E. coli O157:H7 was common and that RAMS-IMS and fecal-IMS protocols showed similar capabilities in detecting positive samples among this population of feedlot steers. In addition, test sensitivities were estimated using simple and multilevel logistic regression models to account for different covariates and for the hierarchical structure of our data. The sensitivity of these diagnostic protocols was not influenced by dietary treatments, time of sampling, or being exposed to a super-shedding pen-mate. Super-shedding in this group of cattle was higher than previously reported (Matthews et al., 2006a), and that although direct plating was the least sensitive method to detect E. coli O157:H7-positive samples, over 20% of high bacterial load samples were not detected by RAMS-IMS and/or fecal-IMS. These results suggest that more analytical studies should be performed to investigate whether the outcomes derived from RAMS versus fecal grabs provide a consistent picture of the epidemiology of E. coli O157:H7 shedding and super-shedding in cattle.

In contrast with previous studies that have favored one method over the other (Rice et al., 2003; Greenquist et al., 2005; Khaitsa et al., 2005; Niu et al., 2008), we found that RAMS-IMS and fecal-IMS tests had similar sensitivities for E. coli O157:H7 detection at both sample and animal levels. Variation among studies most likely results from the effect of potential confounding factors such as size and type of samples, consistency of feces and swabbing technique and laboratory protocols, which might influence the sensitivity of swabs or fecal samples in recovering E. coli O157:H7. Thus, future efforts to standardize sampling and diagnostic procedures are desirable to permit better comparison among studies.

When the capabilities of RAMS-IMS, fecal-IMS, and direct plating tests to detect E. coli O157:H7 shedding were compared based on the Cohen's kappa statistic, only the comparisons between RAMS-IMS and fecal-IMS were valid, based on the McNemar's Chi-square test, and suggested a moderate to substantial agreement at the sample and animal levels, based on the Landis and Koch scale (Landis and Koch, 1977). However, this scale should be interpreted with caution since it was designed for the interpretation of agreement between subjective tests and it may exaggerate the magnitude of agreement for objective tests. Even though there were suggestions that the kappa statistics may not have been valid, significant PABAK estimates supported the existence of statistically significant moderate to substantial agreement beyond chance, especially at the sample level. At the animal level, the kappa estimates were subject to less bias and the estimates of agreement beyond chance were similar for the kappa and PABAK statistics; this was more likely the result of the prevalence estimate being close to 50%. However, it is important to note that both kappa and PABAK lack the ability to account for other covariates (e.g., diet and management factors) or the influence of autocorrelation in the data. In our study, the effect of bias from autocorrelated data in our estimates of the kappa and PABAK statistics were likely small considering the little difference between the random and fixed effects models used for estimating the sensitivity of these tests.

Previous studies indicated that diet does affect E. coli O157:H7 populations in cattle before slaughter; however, the direction and magnitude of the effect of different dietary interventions on E. coli O157:H7 shedding in cattle is still controversial (Buchko et al., 2000; Bach et al., 2005; Fox et al., 2007; Jacob et al., 2008). Further, additional sources of variability, such as the existence of different testing strategies, should be accounted for when explaining the differences among study results. Therefore, the application of methods such as logistic regression models, which evaluate test performance while accounting for management factors or other potential sources of variation (e.g., clustering), could be advantageous. Although mixed models accounting for clustering at the pen and animal levels were performed, in our study there were no epidemiologically significant differences in sensitivity and in the interval estimates between simple and multi-level models. The variance components for the fecal-IMS models were much greater than the negligible values seen when mixed models were applied to the RAMS-IMS and direct plating data. It is uncertain why clustering was much more pronounced for the fecal-IMS protocol than the RAMS-IMS or direct plating protocols. However, the difference in values for the information criteria (AIC and BIC) between models was marginal, suggesting that the results of these three diagnostic protocols can be treated as independent observations. Given that the results from the mixed models were not substantially different from those resulting from single level models, we favored results from the latter for the sake of simplicity and clarity. However, results from both models were presented to demonstrate the similarity between them.

Time of sampling was not an issue in our logistic models employed to evaluate test performance, indicating that test sensitivity seems to be independent of changing prevalence over time. Our previous research indicated that the impact of different dietary supplements on the probability of testing positive to E. coli O157:H7 varied with the diagnostic protocol used (Cernicchiaro et al., 2010). However, it was not clear if the contradictory effect of dietary treatments on the probability of testing positive for E. coli O157:H7 by RAMS-IMS and fecal-IMS was related to the potential effect of diet on the sensitivity of these diagnostic protocols or to the effect of these diets on the bacterial population present in feces and RAMS. In the current study, test sensitivities were not influenced by dietary treatments, time of sampling, or exposure to super-shedding pen-mates. Thus, we can conclude that in this study diet did not affect the sensitivity of the RAMS-IMS, fecal-IMS, and direct plating protocols, and we hypothesize that diet, instead, might impact other bacterial species that affect preferential binding of E. coli O157:H7 at particular gastrointestinal locations. Therefore, efforts should be made to identify the existence of potential biological differences among bacteria present in fecal and RAMS samples that could explain discrepancies in results obtained using the different diagnostic protocols.

In this study, it was uncommon to observe animals shedding E. coli O157 at any level over multiple sampling periods. We recognize, however, that evaluating shedding by bi-weekly intervals may ignore the impact of different phenomena involved in the dynamics of E. coli O157:H7 shedding by cattle, such as intermittent shedding, episodes of transient shedding followed by re-infection or simply shedding at low concentrations that are below the detection limit of these diagnostic protocols.

Although direct plating was the least sensitive method to detect E. coli O157:H7–positive samples, over 20% of high bacterial load samples were not detected by RAMS-IMS and/or fecal-IMS. Similarly, when these protocols were compared to evaluate their ability to detect super-shedding events at the sample level, higher levels of agreement were noted with PABAK. Therefore, adjustment of kappa was appropriate to minimize bias particularly at the sample level for detection of E. coli O157:H7 shedding and super-shedding events. The estimates of sensitivity of the RAMS-IMS and fecal-IMS protocols to detect super-shedding events are probably biased upward given that these estimates were based on direct plating as a gold standard and some super-shedders could still have been missed by this testing procedure. The reasons why super-shedding events do not always test positive by fecal-IMS and RAMS-IMS protocols are unclear. The use of different enrichment times or type of beads employed in the IMS procedure (Verstraete et al., 2010) as well as the antibiotics present in the enrichment broth (Vidovic et al., 2007) may affect the sensitivity of detection of E. coli O157. The role of these and other potential factors such as the frequency and magnitude of shedding on the performance of the protocols should be further investigated.

Our results suggest that the classification of super-shedders might depend on the frequency and duration of sampling. Yet, previous reports were based on simulation models and cross-sectional prevalence data in various cattle production systems (e.g., cow-calf and pastured cattle). Consequently, it is likely that their estimates are underrepresented, as animals should not be classified as nonshedders on the results of a single visit given shedding in cattle is transient and new episodes of carriage or re-infection can take place (Matthews et al., 2006b). More longitudinal studies in different populations of cattle are needed to disentangle the dilemma of being a super-shedder with a long-term, high fecal excretion of E. coli O157:H7 versus a super-shedding animal that sheds high amounts of bacteria transiently. It is crucial to determine the causes of high-level shedding so that control measures targeting these events can be developed. If individual animals are super-shedders, then testing, culling, selective breeding, or isolation may be effective control measures. In contrast, if all or most animals exhibit super-shedding events during the normal course of infection, most of these strategies for control would not be practical or economically viable strategies to control E. coli O157:H7 on the farm. Instead, interventions aimed at reductions of bacterial counts in the environment or from all animals would be more feasible.

Footnotes

Acknowledgments

The original study was funded by the Beef Checkoff. The lead author was supported by funds provided by the U.S. Department of Agriculture through their National Research Initiative (USDA-NRI), Epidemiological Approaches to Food Safety Grant # 2006-01227. The computational infrastructure for this research was obtained with the support of the Canada Foundation for Innovation and the Ontario Ministry of Research and Innovation through a grant to D.L. Pearl.

Disclosure Statement

No competing financial interests exist.

References

Bach

, Stanford

, McAllister

. Survival of Escherichia coli O157:H7 in feces from corn- and barley-fed steers. FEMS Microbiol Lett, 2005; 252:25–33.

Buchko

, Holley

, Olson

, Gannon

VPJ

, Veira

. The effect of different grain diets on fecal shedding of Escherichia coli O157:H7 by steers. J Food Prot, 2000; 63:1467–1474.

Byrt

, Bishop

, Carlin

. Bias, prevalence and kappa. J Clin Epidemiol, 1993; 46:423–429.

Cernicchiaro

, Pearl

, McEwen

, Zerby

, Fluharty

, Loerch

, Kauffman

, Bard

, LeJeune

. A randomized controlled trial to assess the impact of dietary energy sources, feed supplements, and the presence of super-shedders on the detection of Escherichia coli O157:H7 in feedlot cattle using different diagnostic procedures. Foodborne Pathog Dis, 2010; 7:1071–1081.

Chase-Topping

, Gally

, Low

, Matthews

, Woolhouse

. Super-shedding and the link between human infection and livestock carriage of Escherichia coli O157. Nat Rev Microbiol, 2008; 6:904–912.

Cobbold

, Hancock

, Rice

, Berg

, Stillborn

, Hovde

, Besser

. Rectoanal junction colonization of feedlot cattle by Escherichia coli O157:H7 and its association with supershedders and excretion dynamics. Appl Environ Microbiol, 2007; 73:1563–1568.

Cohen

. A coefficient of agreement for nominal scales. Educ Psychol Meas, 1960; 20:37–46.

Coughlin

, Trock

, Criqui

, Pickle

, Browner

, Tefft

. The logistic modeling of sensitivity, specificity, and predictive value of a diagnostic test. J Clin Epidemiol, 1992; 45:1–7.

Cray

, Moon

. Experimental infection of calves and adult cattle with Escherichia coli O157:H7. Appl Environ Microbiol, 1995; 61:1586–1590.

10.

Dohoo

, Martin

, Stryhn

. Veterinary Epidemiologic Research, 2nd. Charlottetown, Prince Edward Island: AVC, Inc., 2009.

11.

Faith

, Shere

, Brosch

, Arnold

, Ansay

, Lee

, Luchansky

, Kaspar

. Prevalence and clonal nature of Escherichia coli O157:H7 on dairy farms in Wisconsin. Appl Environ Microbiol, 1996; 62:1519–1525.

12.

Fegan

, Vanderlinde

, Higgs

, Desmarchelier

. The prevalence and concentration of Escherichia coli O157 in faeces of cattle from different production systems at slaughter. J Appl Microbiol, 2004; 97:362–370.

13.

Feinstein

, Cicchetti

. High agreement but low kappa: I. The problems of two paradoxes. J Clin Epidemiol, 1990; 43:543–548.

14.

Fox

, Depenbusch

, Drouillard

, Nagaraja

. Dry-rolled or steam-flaked grain-based diets and fecal shedding of Escherichia coli O157 infeedlot cattle. J Anim Sci, 2007; 85:1207–1212.

15.

Fox

, Shi

, Nagaraja

. Escherichia coli O157 in the rectoanal mucosal region of cattle. Foodborne Pathog Dis, 2008; 5:69–77.

16.

Greenquist

, Drouillard

, Sargeant

, Depenbusch

, Shi

, Lechtenberg

, Nagaraja

. Comparison of rectoanal mucosal swab cultures and fecal cultures for determining prevalence of Escherichia coli O157:H7 in feedlot cattle. Appl Environ Microbiol, 2005; 71:6431–6433.

17.

Greiner

, Gardner

. Epidemiologic issues in the validation of veterinary diagnostic tests. Prev Vet Med, 2000; 45:3–22.

18.

Hancock

, Besser

, Rice

, Herriott

, Tarr

. A longitudinal study of Escherichia coli O157 in fourteen dairy herds. Epidemiol Infect, 1997; 118:193–195.

19.

Jacob

, Fox

, Narayanan

, Drouillard

, Renter

, Nagaraja

. Effects of feeding wet corn distiller's grains with solubles with or without monensin and tylosin on the prevalence and antimicrobial susceptibilities of fecal food-borne pathogenic and commensal bacteria in feedlot cattle. J Anim Sci, 2008; 86:1182–1190.

20.

Karmali

. Infection by Shiga-toxin producing Escherichia coli. An overview. Mol Biotechnol, 2004; 26:117–122.

21.

Khaitsa

, Bauer

, Gibbs

, Lardy

, Doetkott

, Kegode

. Comparison of two sampling methods for Escherichia coli O157:H7 detection in feedlot cattle. J Food Prot, 2005; 68:1724–1728.

22.

Landis

, Koch

. The measurement of observer agreement for categorical data. Biometrics, 1977; 33:159–174.

23.

LeJeune

, Hancock

, Besser

. Sensitivity of Escherichia coli O157 detection in bovine feces assessed by broth enrichment followed by immunomagnetic separation and direct plating methodologies. J Clin Microbiol, 2006; 44:872–875.

24.

Lindberg

, Groenedaal

, Alenius

, Emanuelson

. Validation of a test for dams carrying foetuses persistently infected with bovine viral-diarrhoea virus based on determination of antibody levels in late pregnancy. Prev Vet Med, 2001; 51:199–214.

25.

Low

, McKendrick

, McKechnie

, Fenlon

, Naylor

, Currie

, Smith

DGE

, Allison

, Gally

. Rectal carriage of enterohemorrhagic Escherichia coli O157 in slaughtered cattle. Appl Environ Microbiol, 2005; 71:93–97.

26.

Mak

HKF

, Yau

KKW

, Chan

BPL

. Prevalence-adjusted bias-adjusted κ values as additional indicators to measure observer agreement. Radiology, 2004; 232:302–303.

27.

Matthews

, Low

, Gally

, Pearce

, Mellor

, Heesterbeek

JAP

, Chase-Topping

, Naylor

, Shaw

, Reid

SWJ

, Gunn

, Woolhouse

MEJ

. Heterogeneous shedding of Escherichia coli O157 in cattle and its implications for control. PNAS, 2006a. 103:547–552.

28.

Matthews

, McKendrick

, Ternent

, Gunn

, Synge

, Woolhouse

. Super-shedding cattle and the transmission dynamics of Escherichia coli O157. Epidemiol Infect, 2006b. 134:131–142.

29.

McNemar

. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 1947; 12:153–157.

30.

Naylor

, Low

, Besser

, Mahajan

, Gunnm

, Pearce

, McKendrick

, Smith

, Gally

. Lymphoid follicle-dense mucosa at the terminal rectum is the principal site of colonization of enterohemorrhagic Escherichia coli O157:H7 in the bovine host. Infect Immun, 2003; 71:1505–1512.

31.

Niu

, McAllister

, Rozema

, Stephens

, Bach

, Johnson

, Standford

. Comparison of fecal versus rectoanal mucosal swab sampling for detecting Escherichia coli O157:H7 in experimentally inoculated cattle used in assessing bacteriophage as a mitigation strategy. J Food Prot, 2008; 71:691–698.

32.

Omikasin

, MacRae

, Ogden

, Strachan

NJC

. Concentration and prevalence of Escherichia coli O157 in cattle feces at slaughter. Appl Environ Microbiol, 2003; 69:2444–2447.

33.

Paiba

, Wilesmith

, Evans

, Pascoe

, Smith

, Kidd

, Ryan

, McLaren

, Chappell

, Willshaw

, Cheasty

, French

, Jones

, Buchanan

, Challoner

, Colloff

, Cranwell

, Daniel

, Davies

, Duff

, Hogg

, Kirby

, Millar

, Monies

, Nicholls

, Payne

. Prevalence of faecal excretion of verocytotoxigenic Escherichia coli O157 in England and Wales. Vet Rec, 2003; 153:347–353.

34.

Petersen

, Enoe

, Nielsen

. Observer agreement on pen level prevalence of clinical signs in finishing pigs. Prev Vet Med, 2004; 64:147–156.

35.

Rabe-Hesketh

, Skrondal

, Pickles

. Reliable estimation of generalized linear mixed models using adaptive quadrature. Stata J, 2002; 2:1–21.

36.

Rice

, Sheng

, Wynia

, Hovde

. Rectoanal mucosal swab culture is more sensitive than fecal culture and distinguishes Escherichia coli O157:H7-colonized cattle and those transiently shedding the same organism. J Clin Microbiol, 2003; 41:4921–4929.

37.

Robinson

, Wright

, Hart

, Bennet

, French

. Intermittent and persistent shedding of Escherichia coli O157 in cohorts of naturally infected calves. J Appl Microbiol, 2004; 97:1045–1053.

38.

Sargeant

, Gillespie

, Oberst

, Phebus

, Hyatt

, Bohra

, Galland

. Results of a longitudinal study of the prevalence of Escherichia coli O157:H7 on cow-calf farms. Am J Vet Res, 2000; 61:1375–1379.

39.

Shere

, Bartlett

, Kaspar

. Longitudinal study of Escherichia coli O157:H7 dissemination on four dairy farms in Wisconsin. Appl Environ Microbiol, 1998; 64:1390–1399.

40.

Thomsen

, Baadsgaard

. Intra- and inter-observer agreement of a protocol for clinical examination of dairy cows. Prev Vet Med, 2006; 75:133–139.

41.

Verstraete

, De Zutter

, Messens

, Herman

, Heyndrickx

, De Reu

. Effect of the enrichment time and immunomagnetic separation on the detection of Shiga toxin-producing Escherichia coli O26, O103, O111, O145 and sorbitol positive O157 from artificially inoculated cattle faeces. Vet Microbiol, 2010; 145:106–112.

42.

Vidovic

, Germida

, Korber

. Sensitivity of two techniques to detect Escherichia coli O157 in naturally-infected bovine fecal samples. Food Microbiol, 2007; 24:633–639.

43.

Wray

, McLaren

, Randall

, Pearson

. Natural and experimental infection of normal cattle with Escherichia coli O157. Vet Rec, 2000; 147:65–68.

44.

Zwick

. Another look at interrater agreement. Psychol Bull, 1988; 103:374–378.