There is increasing global interest in improving animal disease surveillance to ensure timely detection of emerging disease outbreaks [1]. Syndromic surveillance is defined as surveillance using health-related data that precedes a diagnosis and signals a sufficient probability of an outbreak, to warrant further public or animal health response [2]. The use of pre-diagnostic data sources might detect naturally and non-naturally-occurring outbreaks sooner than would be possible with traditional diagnostic data [3]. Syndromic surveillance is recommended as an adjunct to traditional disease monitoring and surveillance programs. Although syndromic surveillance systems were originally developed to detect human disease outbreaks, these systems are increasingly being considered for their potential as early warning signals of disease outbreaks in livestock populations, for animal health and public health purposes [4–8].

Inspection of animals pre- and post-slaughter in abattoirs functions to ensure the safety of meat products for human consumption [9], and has also been used for active surveillance of animal and zoonotic diseases [10–12]. There is evidence supporting the sensitivity of meat inspection for detecting specific diseases in animals [13, 14], and condemnation rate information from abattoirs has been used to guide control strategies for common infectious diseases in pigs [15]. Provincially-inspected abattoirs in Ontario have the potential to provide information into a syndromic surveillance system for improved detection of disease outbreaks in Ontario finisher herds. However, the accuracy and validity of this type of data for detection of animal disease outbreaks has not been well-documented.

Swine herds in Ontario were affected by several important health events during recent years. Most notable was the emergence in the autumn, 2004 of a strain of porcine circovirus type II (PCV-2), restricted fragment length polymorphism (RFLP) pattern 321, which coincided with an outbreak of a severe form of Porcine Circovirus-Associated Disease (PCVAD). Up to 50% mortality in some grower-finisher pig herds was documented [16] and losses continued until the spring of 2006 when a new vaccine to protect against PCV-2 was implemented on Ontario pig farms [17]. Around the same time period, an outbreak of a more severe form of Porcine Reproductive and Respiratory Syndrome virus (PRRS) occurred, which continued until late 2006 [18], and in the spring of 2005, a triple reassortant subtype H3N2 of swine influenza type A virus (SIV) swept through Ontario pig herds by the summer of 2005 [19]. As a result of these major disease events, there has been attention directed at enhancing surveillance of swine herd health in this province.

The objective of our study was to explore the suitability of whole carcass condemnation data from provincially-inspected abattoirs in Ontario for incorporation into a syndromic disease surveillance system for early detection of new, emerging or re-emerging diseases in finisher hog herds. We examined whether the general proximity of swine operations to provincially-inspected abattoirs would support spatial analysis of abattoir data for disease surveillance. To better understand sources of bias or variables that would require consideration in condemnation data analysis for disease outbreak detection, non-disease factors that influence whole hog condemnations were investigated. In addition, we wanted to assess the performance of the scan statistic on unadjusted data to determine whether clusters of high whole carcass condemnations would be detected that were consistent with known large-scale historical health events.

The underlying causes for the decline seen in both the number of hogs processed and processing facilities have not been well-documented, however, over the course of the study time period there was also a decline in the number of operating farms within the province of Ontario [32]. These changes may have been a result of low hog prices and economic difficulties experienced by family-operated hog farms over this time period [32]. In contrast to the roughly 200 provincially-inspected abattoirs in Ontario for interprovincial meat trade, there are only 5 federally-inspected abattoirs, which are licensed to process meat for inter-provincial and international trade. Recent research confirms higher condemnation rates in provincial versus federal Ontario abattoirs [6, 8]. We also found supportive evidence that provincially-inspected abattoirs are generally located near farms of origin, and these abattoirs may therefore be a source of spatially-representative data for detecting localized outbreaks of disease in Ontario. Although there are guidelines and uniform evaluation criteria for meat inspectors to follow when deciding whether to condemn a carcass or its parts [33], varying whole carcass condemnation rates among abattoirs may reflect individual differences between meat inspectors, environmental conditions, abattoir-specific management practices, geographic differences and other unknown factors. If condemnation data are to be used for syndromic disease surveillance it is important to gain an understanding of the non-disease factors that influence these data.

The results from the negative binomial models revealed several variables that influence whole hog carcass condemnation rates in provincial abattoirs. The different condemnation patterns seen among agricultural regions during the years of the study may have been driven by disease outbreaks, however, regional differences in condemnation rates should also be considered when using these types of data for disease surveillance. Specifically, the higher rate of condemnations for Southern Ontario in comparison with other regions is consistent with trends observed in the crude condemnation rates (Table 3). Regional patterns in condemnations predicted by the model for Western and Central Ontario most closely resemble the trends expected as a result of the known health events that took place during the years of our study. Consideration of geographic differences in condemnation rate could aid interpretation of condemnation clusters and/or be used to set thresholds for detection of temporal or regional clusters within a syndromic surveillance system.

Previous studies have reported seasonal variations in partial condemnations in pigs [34], and condemnations secondary to specific pig diseases [35]. From our multivariable model, whole carcass condemnation rates were lower in the fall, in comparison to all other seasons. This finding could reflect the influence on disease related to the typical five month hog production cycle [36], whereby pigs born during the late spring and early summer seasons reach maturity for marketing in the fall. These pigs may have lower exposure to environmental stressors such as temperature fluctuations and relative humidity during weaning, compared to those weaned in the winter and summer [37, 38]. This may make their health more robust against multi-factorial respiratory diseases, which are associated with lower growth rates, reduced productivity, higher morbidity and mortality [38]. Previous findings of increased risk and severity of pneumonic lung lesions at slaughter seen in hogs infected with Mycoplasma hyopneumoniae at weaning age [39] demonstrates the influence of early disease on adult health status. Consideration of seasonal effects on condemnation rates is required when these data are interpreted for disease outbreak detection.

The lower condemnation rates associated with larger abattoirs may reflect differences in herd health on the farms of origin that ship to these facilities. This finding agrees with prior research investigating abattoir size and specific organ condemnations in pigs [8]. Another explanation for the association between abattoir size and condemnation rate, may reflect a difference in the quality of inspection related to a reduced length of time available for examination of each hog. A study of broiler flocks in France found higher condemnation rates associated with low slaughter-line processing speeds [40]. Currently, information on slaughter processing speed for the various abattoirs in our study is unavailable to test this hypothesis.

The decision by a producer to ship animals for slaughter may be associated with market value or other trade considerations irrespective of animal health. We did not find an association between market hog stock price and condemnation rates in the multivariable model. However, market hog stock prices may not entirely reflect the profitability to the producer and the decision to ship particular hogs for marketing. The complex relationship between hog stock prices and decisions to ship may be affected by cyclical hog cycles, feed prices and other production costs, currency exchange rates, pork wholesale and retail prices and the producer’s resulting share of profits [41].

We detected clusters of whole carcass condemnations in space, time, and space-time for the entire study period and spatially when the data were analyzed by individual years. Spatial clusters were detected only in Southern, Western and Central Ontario regions when the entire study period was analyzed. When the data was scanned spatiotemporally, clusters were only detected in Southern and Western Ontario. This supports the findings from the descriptive statistics and the multilevel model demonstrating higher condemnation rates in these regions, particularly Southern Ontario. Upon examination of the spatial scan analysis run for individual years, it is apparent that there are some similar spatial clusters that were detected during multiple years, indicating that stable underlying factors may have contributed to these clusters. These factors may reflect abattoir characteristics, including differences between meat inspectors at individual abattoirs, or farm-level factors, including biosecurity, which may influence the health status of pigs in these areas. There were no documented outbreaks in Northern Ontario during 2003 to adequately explain the model-based increase in condemnation rates, however, greater variability in condemnation rates within this region due to fewer hogs processed may explain the variation seen overall within this region across the years.

The temporal cluster of high condemnation rates detected by scan statistics were consistent with the historical emergence of PCVAD, PRRS and swine influenza documented by the AHL. Results of the space-time scan analysis did not detect clusters that could have indicated these outbreaks with the exception of the third most likely cluster. This space-time cluster containing 3 abattoirs, was detected between April 2004 and April 2005, when the aforementioned disease outbreaks were beginning to affect farms in the province [16, 18, 19]. However, due to the retrospective nature of the study and the lack of source farm information due to privacy constraints, validation of these clusters was not possible. Future prospective studies could evaluate the sensitivity and specificity of cluster detection for disease surveillance of pigs using whole carcass condemnation data by validating clusters with on-farm and laboratory data.

While space-time permutation models have the advantage of controlling for pre-existing purely spatial and temporal clusters, it assumes the background population to be stable in space-time. Using data from all abattoirs across the province, where baseline numbers of animals processed may increase or decrease faster in some areas over others, could have resulted in biased P-values, favoring cluster detection. This issue, referred to as population shift bias [25], may be more of a problem when the changes in disease patterns are being examined over longer periods of time, in this study over a number of 5-month production cycles. Employing adjusted space-time scan analysis based on models that account for background population may offer advantages over space-time permutation models on unadjusted data [42].

The most likely temporal cluster of high condemnations was detected during the fall of 2004 until the fall of 2006, which, interestingly, corresponds with the timeframe during which known outbreaks of respiratory illnesses took place in Ontario herds. The introduction of a vaccine to protect against PCV-2 in late 2006 also coincides with the end of the high condemnation rate temporal cluster. These findings are consistent with previous results of temporal scans of PCV-2 positive cases diagnosed at the Animal Health Laboratory, which reported a temporal cluster of high rates among suspected cases tested during 2005 [30], and clusters of low rates of PCV-2 positive cases among those tested during the fall and winter of 2006 for polymerase chain reaction-positive cases [30]. Because it is not clear whether the underlying reasons for the whole carcass condemnations were secondary to lesions or systemic signs of PCVAD, PRRSv and/or swine influenza infection, it is not possible to confirm the results from the spatial, temporal and space-time. The findings of the temporal scan analysis in particular, do appear to correspond with the wide-spread disease event, and suggest that data collected from abattoir inspections may be useful for detecting disease outbreaks earlier than by traditional laboratory diagnostic data.

This preliminary study shows that whole hog carcass condemnation data from provincial abattoirs condemnations may provide useful information for inclusion in a syndromic surveillance system for pigs in the province of Ontario. Interestingly, a study investigating partial condemnations due to lung and kidney pathology did not clearly demonstrate improvement in potential disease detection compared with whole hog carcass condemnation data [16]. In comparison with data pertaining to individual organ condemnations, whole carcass condemnation data may provide a better indicator of overall health status in a population of pigs, making it more suitable for detection of multifactorial disease syndromes. Furthermore, as discussed in our previous paper [16], an important difference between the two data types relates to the requirement for a veterinary inspector to oversee all whole carcass condemnations in provincial abattoirs, whereas a trade inspector is licensed to condemn partial carcasses without direct veterinary oversight [43]. The performance and practicality of incorporating whole carcass condemnation data in a surveillance program warrants further assessment. A prospective study with real-time data that could be validated using laboratory disease incidence data would be beneficial to further investigate the utility of whole carcass condemnation data for disease outbreak detection.