This study estimated the test characteristics of a commercially available PCR assay, PathoProof ™ Mastitis PCR Assay, and BC to identify IMI with S. aureus in dairy cows at dry off. The study was done using LCA to avoid the assumption of an available perfect reference test. The findings were ambiguous. PCR had lower Se and higher Sp than BC for low PCR Ct-value cut-offs, but it was reversed for higher PCR Ct-value cut-offs. Surprisingly, Se and Sp of BC were affected by the choice of the PCR Ct-value cut-off. Increasing prevalence at increasing PCR Ct-value cut offs support the assumption of a quantitative relationship between Ct-values and infection load in the sample .
A PCR assay detects DNA from both viable and non-viable bacteria while BC is only capable of detecting viable bacteria . In cases where the infection is cured, either due to treatment or self-cure, bacterial DNA can still be present in the udder and the cow can be falsely diagnosed as positive by the PCR assay . When increasing the PCR Ct-value cut-off, the Se of BC decreases, indicating that the bacteria detected by PCR could be non-viable, in amounts lower than the detection threshold of the BC or derived from other sources than the milk. When using a high PCR Ct-value cut-off, more animals may be false positive by PCR due to different reasons for example non-viable bacteria in low concentrations. Since S. aureus commonly inhabits cow skin , the bacteria could furthermore be a result of contamination, teat canal infections or teat skin infections. Carry-over in the milking system could also be a possibility . The results also indicate that the disease definition is changing depending on the PCR Ct-value cut-off. The effect is shown in the prevalence estimates where a PCR Ct-value cut-off of 32 results in a disease prevalence of 23% in subpopulation Jutland/Fyn. When the PCR Ct-value cut-off is increased to 37, the disease prevalence has increased to 37% in the same population. One could argue that a PCR Ct-value cut-off at 32 indicates the presence of a high amount of bacteria in the milk, which most likely are associated with an active IMI. When the amount of bacteria in the milk is high, we increase the probability that viable bacteria are present, which can be detected by BC. When increasing the PCR Ct-value cut-off, the disease definition is apparently changing, reflected by an increasing prevalence. Higher PCR Ct-values may indicate that the amount of bacteria is below the detection threshold of BC, that the bacteria present are non-viable and/or that the bacteria derive from other sources than the infected mammary gland. Under the assumption that the disease definition changes dependent on the chosen PCR Ct-value cut-off, changes in test estimates for both tests at different PCR Ct-value cut-off for the real-time PCR may be a logical consequence.
In this study however, both BC and the real-time PCR assay had relatively high Se and Sp, depending on the PCR Ct-value cut-off, indicating that both tests can serve as useful tools in diagnosing IMI caused by S. aureus. Friendship et al., 2010  evaluated the PathoProof ™ Mastitis PCR Assay in detecting S. aureus. All cows in the study originated from herds with a known history of intramammary infections due to S. aureus. The study found a Se for the PCR Assay of 0.94 and Sp of 0.95 when using BC as gold standard. Koskinen et al. 2010 , found the Se and Sp for the PathoProof ™ Mastitis PCR Assay to be 0.862 and 0.916 respectively when evaluated for all pathogens . These Se and Sp are similar to those found in this study, where Se and Sp for the PCR assay were estimated to 0.93 and 0.95 respectively at PCR Ct-value cut-off 37. The later article argues that the use of BC as reference standard most likely results in an underestimation of test estimates for the PCR due to inaccuracy of the reference test. Our study can support this assumption and support the use of LCA in situations where no perfect reference test is available.
All milk samples used for BC in this study were frozen prior to analysis. Freezing has been shown to both increase the likelihood for obtaining S. aureus in the milk sample [36–38] and decrease or to show no effect in detection rates [39–41]. Evaluating the effect of freezing on BC results was beyond the scope of our study. Thus, we cannot exclude that freezing might have affected the results.
PCR is shown to be able to detect mastitis bacteria in samples that yielded no growth in bacteriological culture , which is an advantage in cases where the aetiology of the mastitis is of special interest. Another advantage is the speed of the real-time PCR analysis compared to BC; there is a possibility for the same-day results.
When interpreting the Ct-value given by the PCR assay, one should consider the type of sample and the sampling conditions. The samples for PCR were in this study collected under non-sterile conditions and metered by the milking equipment. In these samples there is a possibility for contamination from the previous cows and other sources than the milk. Pre-milking practices in the herd and condition of the milking equipment may have an influence on the obtained Ct-values. However, differences between the included herds made the estimation of these factors not possible. The results given in form of a Ct-value should furthermore be interpreted with all other available information for the cow, such as history, clinical signs and SCC in the milk. This should however also be the practice with results from BC.
An implied assumption of the LCA is that the two tests are conditionally independent given disease status. Given that there is no culturing involved in the PCR procedure, this assumption is justified, however, only when accepting the underlying disease definitions discussed above.
In this study, the population was divided based on geographical location, Jutland/Fyn and Zealand. The difference in prevalence found here was not large, but as the herds were selected to present problems, we could not expect large differences. In Toft et al. 2005 , the importance of the difference in true prevalence was examined. Although smaller differences might infer some bias, it is more important that the assumption about constant Se and Sp across populations is justified.
Both farms using AMS and conventional milking systems were included in the study. A total of 25 cows originated from AMS herds. A study by Lovendahl and Bjerring , indicates that carryover problems are not negligible in the AMS system compared to conventional systems. It could be assumed that a higher proportion of cows could be classified as false positive due to carryover. Excluding the 25 cows originating from AMS herds did not result in any notable differences on the estimates of test accuracy.