Early warning indicators for hock burn in broiler flocks

Abstract

Hock burn is a common disease of broiler chickens affecting flock welfare and farmer income. Here we use hierarchical logistic regression (HLR) models to identify risk factors for hock burn using data from 5895 flocks, collected over 3.5 years by a large UK broiler company. The results suggest that at 2 weeks of age, weight and weight density may be useful predictors of flocks at risk of a high incidence of hock burn. In contrast, stocking density at placement is not. The use of these and other variables in disease prevention add value to routinely collected management data and can assist in improving broiler welfare and farm income.

Introduction

Hock burn is the name given to dermatitis of the plantar surface of the hocks of broiler chickens (Martland, 1984, 1985; Greene et al., 1985). Visible as brown or black marks on the hocks of birds at processing, it can range from mild to severe inflammation and may process skin ulceration (Greene et al., 1985) . Hock burn may be considered part of a syndrome of contact dermatitis, which includes footpad dermatitis and breast blisters (Greene et al., 1985; Martland, 1985; Haslam et al., 2007). The footpad is most commonly affected, followed by the hock and breast (Greene et al., 1985).

The aetiology of contact dermatitis has been investigated by experimental studies (Kjaer et al., 2006; Buijs et al., 2009) and field studies (Bruce et al., 1990; Haslam et al., 2007). Poor litter quality is consistently identified as a risk factor for hock burn. However, the role of other factors is less clear (Bruce et al., 1990; Stamp Dawkins et al., 2004; Kjaer et al., 2006; Haslam et al., 2007; Allain et al., 2009). Contact dermatitis is considered a measure of poor broiler welfare (SCAHAW, 2000; Allain et al., 2009) and flock prevalence has been identified as an important determinant of maximum allowable stocking density by the European Union (2007). Hock burn is also associated with failure to meet bird target weights and with reduced income (Bruce et al., 1990; Menzies et al., 1998). Broiler production in the European Union has traditionally involved large numbers of birds and very low profit margins (Sheppard, 2004), making the reduction of financial losses associated with hock burn important to the European broiler industry.

Broiler producers routinely collect large amounts of production and processing data. Currently these data are analysed to establish production targets and triggers for investigation or intervention of suspected poor health. Advances in electronic data-recording and information flows present the opportunity to use these data for disease prevention. In the present paper we examine the use of these data to identify the risk factors for hock burn.

The structures of broiler production systems pose analytical challenges to epidemiologists. Data concerning disease status are not independent. They are clustered; birds reared in one broiler house will be influenced by factors attributed to that particular house and are more similar to one another than birds reared in a different house. Similar effects apply to birds from the same farm owned and managed by the same company.

Most statistical techniques assume independence of data. The appropriate analysis of clustered data requires the use of hierarchical (“multilevel”) statistical analysis (Hox, 2002). Failure to recognize clustering can produce inaccurate model parameters and disease models (Hox, 2002). The use of hierarchical models to investigate broiler diseases such as hock burn and foot pad dermatitis is relatively uncommon and the contribution of factors operating at farm, house and flock levels remains uncharacterized.

In the present study we identify the risk factors for hock burn in broiler chickens using multivariable hierarchical statistical techniques to analyse production and processing data routinely collected by a large UK broiler company.

Materials and Methods

Data

Routine data relating to the management and processing of 5895 broiler flocks between 2005 and 2008 were obtained from a large UK broiler company. The individual epidemiological unit was a flock of birds placed in the same house on a farm at the same time, processed on the same day.

Case definition

Hock burn was defined as a brown or black discolouration greater than 5 mm in diameter, observed on either one or both hocks after slaughter.

Sampling

Five hundred birds were sampled from each consignment. The sample was increased to 1000 if hock burn levels exceeded 5% in the initial sample. This was not done for statistical reasons but because financial penalties were incurred by farmers with hock burn prevalences above 5%. The sampling strategy was to examine five or 10 groups of 100 consecutive birds, spaced at intervals along the processing line, from the start to finish of the consignment.

Outcome variable

A binary outcome variable representing “high” and “low” hock burn was defined by taking the 75th percentile of the flock prevalence of hock burn at slaughter as the cut-off point. The flock prevalence of hock burn was calculated for each slaughter date, using data aggregated over each consignment of birds from the flock processed on the same day. Company operatives under the guidance of the Official Veterinary Surgeon recorded the prevalence of hock burn in each consignment. Two operatives did most of the recording during the study period.

Independent variables

Independent variables consisted of growth and management factors characterized during the first 5 weeks. Original variables were replaced by denominator-based variables calculated from the data where appropriate. For example, daily water consumption was replaced by weekly water consumption per 1000 birds present at the start of that week. Numerical variables consisted of water consumption (l/1000 birds, and l/1000 birds/m² floor area), average weight (g), growth rate (g/day), stocking density (birds/m²), weight density (g/m²), mortality (%), leg culls (%), overall growth rate (g/day), final weight at slaughter (g) and the average age and number of the parent flocks. Categorical variables were reclassified using a single binary variable to represent each category (e.g. one variable each for the sex variables cockerels, pullets and “as hatched”). Observations containing outliers and missing values were removed. All data were normalized and highly correlated variables were excluded (pair-wise Pearson correlation coefficient > 0.8). After processing, the total number of variables and records was 45 and 6912, respectively.

Model-building

Univariable and multivariable hierarchical logistic regression (HLR) models were built using MLWiN 2.10 (Rasbash et al., 2009). Four-level HLR models were built using random effects at the levels of farm (Level 4), house (Level 3), flock placement date (Level 2), and individual (Level 1). Random effects and fixed effects were evaluated using the Wald statistic.

Initial variable screening was performed using univariable HLR analysis retaining variables with P < 0.25. Variance component models (VCM) were then constructed to provide information on the importance of data collected at each of the four levels. Fixed variables were then added and the final selection of variables made using backwards elimination and a probability cut off of 5% (P > 0.05).

Three multivariable models were built. The first model used data from all 5 weeks of the growing cycle. Models were also built using data from the first 2 weeks and the first 3 weeks of the growing cycle, to identify risk factors occurring at an early stage of the growing cycle. In the latter two models, three variables, which had been excluded by univariable hierarchical analysis, were included because of a priori evidence of their importance. These were stocking density at placement, birds culled due to leg weakness at 1 week and water consumption at 2 weeks.

The size of effect for each risk factor was expressed as the respective coefficient from the multivariable logistic regression equation (β coefficient), and as an odds ratio. As β coefficients were obtained from models built using normalized data, they were rescaled to calculate the odds ratios so that they could be related to the original data.

Model validation

The accuracy of the four-level HLR classifier built using data from 5 weeks was calculated using the area under the receiver operator characteristics curve, using a range of threshold parameters (0 < Π < 1). Both fixed effects and residuals were included in the prediction.

Results

Variance components model

This model was based on data from 112 broiler farms, 343 houses, 3084 flocks and 6912 hock burn observations. There was significant variance in hock burn at all levels in the model before the addition of fixed effects (Table 1).

Table 1.  Variance components model using data for all 5 weeks.

Level	Identifier	Variance	Standard error	Wald value	P value^a	VPC (%)
4	Farm	1.020	0.199	5.13	<0.05	15.2
3	House	0.208	0.086	2.42	<0.05	3.1
2	Flock date	2.17	0.164	13.23	<0.05	32.4
1	Hock burn	3.29^b				49.2
		Total 6.69
^aOne-tailed test. ^bUsing latent variable technique.

The largest proportion of variation in hock burn (proportion of the variance [VPC] = 49%) occurred at the lowest level. A relatively large proportion of variance (32%) was associated with the date of flock placement. Variation associated with the farm was 15% and variation attributed to house level variables was relatively minor (3%).

Univariable analysis

In univariable analysis, 32 of the 45 variables were associated with hock burn. These variables were used to build the final multivariable models.

Risk factors identified using data from the first 5 weeks of growth period

Seventeen risk factors for high hock burn after 5 weeks were identified. Eight risk factors—the months October to January, rearing pullets, stocking density at 5 weeks, water consumption at 5 weeks, and the average weight of the birds at slaughter—were associated with an increased risk. Nine risk factors were associated with a decreased hock burn risk: the months March to August, cockerels, automatic water meters, and average weight at 2 weeks. This model had accuracy of 0.95 as measured by the area under the receiver operator characteristics curve.

Risk factors identified using data from the first 5 weeks of growth period

Thirteen variables were associated with hock burn in broiler flocks after 2 weeks (Table 2). Four were associated with a higher risk of hock burn: placement in the months of November and December, single-sex rearing of cockerels, and weight density at 2 weeks. Nine variables were associated with a lower risk of hock burn: the months March to August, automatic water meters, stocking density at placement, and average weight at 2 weeks.

Table 2.  Risk factors for hock burn identified by hierarchical multivariable analysis.

	2 weeks				3 weeks				5 weeks
Risk factor	β coeff.	SE	Rescaled coeff.	OR	β coeff.	SE	Rescaled coeff.	OR	β coeff.	SE	Rescaled coeff.	OR
Increased hock burn
November	0.162	0.043	0.617	1.853	0.166	0.043	0.632	1.882	0.254	0.055	0.977	2.656
December	0.181	0.043	0.616	1.851	0.173	0.042	0.589	1.802	0.313	0.056	1.079	2.943
January	NS	NS	NS	NS	NS	NS	NS	NS	0.148	0.059	0.493	1.638
October	NS	NS	NS	NS	NS	NS	NS	NS	0.151	0.057	0.521	1.683
Pullets	NS	NS	NS	NS	0.099	0.045	0.276	1.318	0.171	0.054	0.475	1.608
Weight density at two weeks	1.609	0.542	0.001	1.001	1.465	0.54	0.001	1.001	NS	NS	NS	NS
Water consumption increase weeks 2 to 3	n/a	n/a	n/a	n/a	0.335	0.054	0.005	1.005	NS	NS	NS	NS
Stocking density at 5 weeks	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.244	0.074	0.079	1.083
Water consumption at 5 weeks	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.66	0.068	0.004	1.004
Average weight at slaughter	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.952	0.048	0.003	1.003
Risk changes over time
Cockerels	0.131	0.047	0.294	1.342	NS	NS	NS	NS	−0.246	0.061	−0.547	0.579
Reduced hock burn
August	−0.375	0.053	−1.35	0.259	−0.388	0.053	−1.397	0.247	−0.397	0.067	−1.418	0.242
May	−0.278	0.052	−1.071	0.343	−0.291	0.051	−1.121	0.326	−0.293	0.064	−1.127	0.324
March	−0.238	0.051	−0.901	0.406	−0.227	0.051	−0.859	0.424	−0.262	0.064	−1.008	0.365
April	−0.202	0.052	−0.837	0.433	−0.222	0.051	−0.92	0.399	−0.257	0.063	−1.071	0.343
June	−0.183	0.049	−0.673	0.51	−0.207	0.049	−0.761	0.467	−0.228	0.062	−0.844	0.43
Automatic water meter	−0.283	0.084	−0.644	0.525	−0.272	0.082	−0.619	0.539	−0.279	0.092	−0.634	0.53
Stocking density at placement	−1.318	0.439	−0.551	0.576	−1.182	0.437	−0.495	0.61	NS	NS	NS	NS
Average weight at 2 weeks	−1.432	0.433	−0.029	0.972	−1.327	0.431	−0.026	0.974	−0.244	0.072	−0.005	0.995
coeff., coefficient; n/a, not available; NS, not significant; OR, odds ratio; SE, standard error.

Risk factors identified using data from the first 3 weeks of growth period

Fourteen variables were associated with hock burn after 3 weeks (Table 2). Five were associated with an increased risk: the months of November and December, pullets, weight density at 2 weeks, and water consumption increment at 3 weeks. Nine variables were associated with a lower risk of hock burn: the months March to August, automatic water meters, stocking density at placement, and average weight at 2 weeks.

Discussion

The results of the present study offer options for using routine production data to reduce the risk of hock burn. Stocking density at 5 weeks of age and weight at slaughter were associated with being in the top quartile of hock burn prevalence. Although these variables occur too late in the growing period to be valuable targets for intervention, the association between hock burn and weight at 2 weeks offers potential for its control. This relationship appears complex. Average weight at 2 weeks was associated with a decreased risk of hock burn but weight density (g/m²) was associated with an increased risk. The most plausible biological interpretation of this is that weight per se is a measure of the health of individual birds. Weight at a particular age is a common measure of health in all animals. However, increasing weight density, even at 2 weeks, indicates that the birds are likely to reach slaughter weight earlier and will therefore have increased weight density at the end of the growing period. This has been associated with reduced welfare parameters such as litter quality, bird behaviour patterns, physiological parameters, body weight, and carcass quality, including contact dermatitis (Dozier et al., 2005, 2006; Estevez, 2007; Buijs et al., 2009).

Evidence from the present study suggests that weight density at 2 weeks is a proxy measure for effects at the end of the growing period. This variable disappears from models built using data from the whole growing period and appears to be replaced by stocking density at 5 weeks and average weight at slaughter. Thinning (i.e. the removal of a proportion of the birds from the house) is commonly used at the end of the growing period to provide the birds with more space and to reduce pressure on the environment.

The results presented here offer producers a potentially valuable practical tool to reduce hock burn by managing the end of the growing period. Rather than using fixed times for slaughter, algorithms could be developed that allow the optimal time and degree of thinning to be estimated using information collected during earlier growth. These risk factors are therefore potential lead welfare indicators (Manning et al., 2007a). In addition, weighing the birds more frequently towards the end of the growing period, and the accurate selection of heavier birds for removal during thinning, could also help reduce hock burn by removing those birds at greatest risk.

In contrast to data measured at 2 weeks, the results of the present study suggest that the use of limits on stocking density at placement, to reduce the risk of hock burn, may be less valuable. There was a negative association between stocking density at placement and hock burn (i.e. the higher stocking densities reduced the risk of hock burn). This was surprising as other research has suggested that increased contact dermatitis may be associated with stocking density at placement and final weight density as predicted by placement density (McIlroy et al., 1987; Buijs et al., 2009). Why the apparent contradiction? It is possible that higher stocking densities at placement have a direct effect on bird health in the early weeks. One study found chicks achieved greater body weight and food conversion at 15 days at higher stocking densities, but this reduced thereafter (Dozier et al., 2006). However, if stocking density had a direct effect on contact dermatitis then it would be most likely to happen at the end of the growing period, when the increased size or weight density of the birds may become more important than their number. This is confirmed by the results of this study, in which stocking density at 5 weeks and weight at slaughter show positive associations with hock burn, replacing stocking density at placement and 2 weeks.

High placement stocking densities may also be a proxy measure for good management (i.e. better farm managers are confident that they can achieve good health and productivity results even at higher stocking densities). The occurrence of higher mortality or culling in flocks stocked more densely at placement is also a hypothesis. This might reduce weight density of these birds throughout the crop when compared with birds placed at lower stocking density. However, this is refuted by the data as both mortality and culling were measured and there is no evidence that they were associated with hock burn.

Although there have been a number of other studies of hock burn, few of these have incorporated the structure of the broiler industry using hierarchical logistic regression models (Bruce et al., 1990; Pagazaurtundua & Warriss, 2006; Haslam et al., 2007). Variance component models allowed us to measure the effect of farm-level, house-level and flock-level variables on the variation within the data. Factors associated with the date of flock placement were associated with the largest proportion of the variance (VPC = 32%) at these levels. This is consistent with the effect of placement month identified in this and previous studies (McIlroy et al., 1987; Stamp Dawkins et al., 2004). The increased risk in winter months and the reciprocal decrease in risk in the summer months identified here may reflect the management conflict in winter months between heat conservation and ventilation. Reduced broiler house ventilation in colder weather may result in reduced litter quality and an increased risk of contact dermatitis. In previous studies we have demonstrated an association between wet litter and winter months (Hermans et al., 2006).

Farm level accounted for 15% of the total variance (VPC = 15%), suggesting significant variation between farms, even within the same company. The variance associated with the broiler house was very small (VPC = 3%). This may reflect the uniformity of house construction management of heating, ventilation, and feeding within individual houses on the same farm. Knowledge of variance components could inform future priorities in study design and data collection in this research area.

One of the hypotheses for the aetiology of hock burn is that it is a contact dermatitis, associated with the interaction between hock skin and litter. Poor quality litter or an increased time spent sitting down due to weakness or ill-health may be contributory factors. Because of the potential importance of wet litter, we analysed weekly water consumption in these flocks. The only risk factors identified were an increased risk of hock burn associated with increased water consumption in the fifth week, and a decreased risk associated with the use of automatic water meters. Increased water consumption may reduce the quality of the litter both directly through increased excretion or indirectly, for example by spillage. The association between the use of automatic water consumption meters and reduced hock burn is an interesting finding, perhaps reflecting an ability to accurately monitor and control the water intake, allowing better stockmanship. In a previous study, increased water consumption on certain days in the broiler cycle was associated with foot pad dermatitis (Manning et al., 2007b). Further characterization and clarification of the effects of water consumption on contact dermatitis would be helpful.

In addition to novel modifiable risk factors identified in the first half of the flock cycle, several additional risk factors were identified at 5 weeks. The effect of the sex of the bird on hock burn is unclear (McIlroy et al., 1987; Kjaer et al., 2006; Haslam et al., 2007). Our results suggest that pullets are more susceptible to hock burn, consistent with reports that the skin of cockerels contains more collagen than the skin of pullets (Granot et al., 1991), and has increased skin puncture strength (Bilgili et al., 1993). The effect of weight at slaughter on contact dermatitis is also unclear (Martland, 1984) but our results are in agreement with the study by Kjaer who found a positive association between body weight and hock burn (Kjaer et al., 2006).

We have added value to routinely collected data by identifying risk factors for hock burn, and highlighted those that may signal an “early warning” for flocks at risk of hock burn. We have also quantified the variance in hock burn at important levels in the broiler lifecycle using techniques robust to data clustering. The use of data management systems to reduce disease levels in broiler flocks has been reported previously (McIlroy et al., 1988). However, the analysis of on-farm data stored in these systems appears undeveloped.

Analysis of secondary data l imited the variables available for analysis to those collected by the company. Several variables relating to important aspects of the broiler house environment are not routinely collected, including relative humidity, gaseous composition, and litter quality. We also had no control over the individuals who may have measured them. These difficulties may be overcome by well-designed studies involving data collection by a small number of well trained operators (Haslam et al., 2007). The range of definitions for hock burn is wide (Allain et al., 2009), which also makes comparison between individual studies more difficult. However, all flocks in this analysis were processed at the same abattoir, and hock burn was measured by a limited number of people. Inaccuracies and inconsistencies in hock burn measurements may therefore be randomly distributed amongst the different consignments, reducing any bias effect of individual operatives.

This analysis was based on data from one large UK broiler company only, and care should be taken in extrapolating the conclusions concerning the aetiology of hock burn to the whole UK broiler population. However, the early warning signals for hock burn identified here could be used in a regression equation at 2 weeks to predict flocks at risk of high hock burn. The validity of this model could be tested on flocks from both this and other companies using retrospective data.

Acknowledgements

The authors wish to thank the poultry company that participated in the present study. This work was funded by the Biotechnology and Biological Sciences Research Council (Grant No. BB/DO12627/1).