The observed (crude) factory submission risk ranged from 0 to 52 per 10,000 animals slaughtered (Table 1). There was a five-fold range in submission risk, after excluding factories submitting fewer than 10 animals. The average crude submission risk (25 per 10,000) during 2005-07 was higher than found in previous reports (21.8 per 10,000 ; 22 per 10,000, ). The variation in crude submission risk among factories (5-fold) was lower than reported previously. There was a 7-fold difference  after controlling for year, month and animal type, and a 9-fold difference  after controlling for the same confounding factors included in our study). The factory confirmation risk ranged from 30.2% to 90.1%, with an average of 68.5% (this average was slightly higher than the 64.4% previously reported by Frankena et al. ).
When interpreting these results, care should be taken when drawing direct comparisons between the two study periods. Fewer factories were operational when the later study was conducted (42 for the period 2003-2004 vs. 37 for the period 2005-2007). Furthermore, it is worth noting that the age profile for animals with complete information on confounding factors during the period 2005-2007 differs from those animals in the period 2003-2004. For example, the study population during 2003-2004  included only 0.49% animals older than 8 years old, while in our study, 2.6% of animals were older than 8 years. This may affect the observed (crude) submission and confirmation risk, because age is associated with both lesion submission and confirmation risks . In our study this was confirmed by the increased odds ratios as age of the slaughtered animals increased (data not shown). This difference is because a new EU cattle identification system was introduced in Ireland in 1996 whereby cattle born from that year were registered centrally and also had their birth dates recorded, with older animals retaining their existing old ID (which did not include birth details) until they exited the national herd due to being slaughtered or via natural wastage. Thus, we conducted an additional analysis excluding animals 9-10 years old in which the average submission risk was 24 per 10,000 and the confirmation risk was 67.51%. Additionally we calculated the Pearson correlation coefficients for both the crude submission risk and crude ranking for the 35 factories present in both studies (2003-2004 and 2005-2007). Correlations were 0.76 (p < 0.001) and 0.71 (p < 0.001) respectively. These results indicate that factories with higher crude submission risk (ranking) during 2003-2004 tended to remain with higher crude submission risk (ranking) during the period 2005-2007 and vice versa.
The adjusted risk is the most appropriate descriptive summary of a factory's relative effectiveness (when compared to other factories rankings). The variations in the risk profile of the animals among the factories were substantially less than expected, as shown by the close agreement between the crude and adjusted estimates of the factory risk. Therefore, it can be concluded that the animal- and farm-related factors did not substantially contribute to the variation in factory level submission and confirmation risk that was observed. A similar conclusion was drawn in the earlier study . Collins  suggested that the variation in factory surveillance effectiveness may be due to factory-related factors, for example, line speed and light intensity, and/or other factors related to veterinary inspectors, for example their experience, interest, motivation and workload.
There was a negative correlation between the submission risk and the confirmation risk; as the number of submissions increased, the percentage of the lesions that were confirmed as TB decreased. However, we do not think that the proportion of submitted lesions that are confirmed as bovine TB should be a metric for assessing surveillance effectiveness.
The detection of gross bovine TB lesions in cattle at slaughter, coupled with successful trace-back of these animals to the herd of origin, is critical to the detection of infected herds, and for the progress of the national bovine TB control and eradication program in Ireland. Between the years 2005-2007, approximately 30% of new herd breakdowns were identified by means of bovine TB slaughter surveillance . Inadequate inspection of carcasses to find gross (visible) lesions could delay the successful eradication of bovine TB in Ireland and increase the cost of the eradication program. Improved factory surveillance would contribute to national efforts to control bovine TB. The identification of infected herds before the scheduled tuberculin test may help in minimizing the size of major breakdowns in an index herd and the spread of infection from an index herd to contiguous herds.
In conclusion, during the period 2005-2007 an increase in bovine TB lesion submission and confirmation risk was observed when compared to the period 2003-2004. However, substantial variation in both the submission and the confirmation risks of TB lesions among factories remains, indicating that the practices applied in detecting and submitting lesions are not uniform. We suggest that studies be conducted to identify the critical factors (variables) present among factories with "high" and "low" rankings. We also recommend continuing monitoring of the effectiveness of slaughter surveillance in Ireland as part of quality control in the national programme. Emphasis should be placed on efforts to improve bovine TB surveillance in factories identified with lower rankings in this study. Training programmes should be considered that would reduce the variability in submission and confirmation risk measured between individual factories.