- Email: [email protected]
Evaluation of a Prototype On-Line Electrical Conductivity System for Detection of Subclinical Mastitis
OUR INDUSTRY TODAY
Evaluation of a Prototype On-Line Electrical Conductivity System for Detection of Subclinical Mastitis L.M.T.E. LANSBERGEN, M. NIELEN,1 T.J.G.M. LAM, A. PENGOY,2 and Y. H. SCHUKKEN Department of Herd Health and Reproduction Utrecht University Yalelaan 7 3584 CL Utrecht, The Netherlands K. MAATJE IYO-DLO Institute for Animal Production PO Box 501 3700 AM Zeist, The Netherlands ABSTRACT
A prototype on-line system for measurement of electrical conductivity of quarter milk was evaluated for accuracy in detection of subclinical mastitis compared with that of bacteriological culture and SCC of sampled quarters. Because of the low quarter prevalence of mastitis, quarters were sampled conditionally from the signals of the on-line system. All signaled quarters and a random selection of the nonsignaled quarters were sampled. To calculate sensitivity and specificity, the total number of nonsignaled quarters was extrapolated. The system identified correctly 18 out of 23 subclinical quarters and 521 out of 555 healthy quarters. Quarter prevalence was about 1%. Predictive value of a positive test (35%) and the predictive value of a negative test (99%) were calculated, as well as sensitivity (25%) and specificity (99%), after extrapolation of the total number of nonsignaled quarters. Because of repeated measurements, sensitivity may be underestimated. When signaled quarters were defined by repeated signals within 14 d, predictive value positive increased to 48%. The prototype on-line system did not detect subclinical mastitis very ac-
Received August 30. 1993. Accepted November 3, 1993. lCorresponding author. 2Yeterinary Microbiological Diagnostic Centre. 1994 J Dairy Sci 77:1132-1140
curately because of suboptimal test characteristics. (Key words: on-line electrical conductivity, subclinical mastitis detection) Abbreviation key: EC = electrical conductivity, QMEC = quarter milk EC, RA = running average. INTRODUCTION
Modem, large dairy herds often use automated management systems. The costs of mastitis (26), combined with stricter legislation for bulk SCC, indicate a potential use for an efficient and direct (on-line) method for the detection of subclinical and clinical mastitis. At present, on-line mastitis detection systems based on electrical conductivity (EC) are becoming commercially available. According to Kitchen (11), mastitis increased the EC of milk because of changes in ionic concentrations. As a result of the damage of udder tissue, concentrations of lactose and K+ decreased, and concentrations of Na+ and Cl- increased. Several studies have evaluated measurement of milk EC as a method for detecting subclinical or clinical mastitis (20). Most investigators (4, 5, 6, 12, 14, 21, 24) considered milk EC to be accurate for the detection of mastitis; others (I, 2, 17) considered it to be less accurate. In the future, more on-line mastitis detection systems probably will be developed that need evaluation before commercial use. One of the problems in evaluation is the definition of subclinical mastitis (20). Also, analysis of an on-line system over time results in repeated 1132
OUR INDUSTRY TODAY
measurements on the same cows and quarters. In addition, for herds with good udder health and a low prevalence of subclinical mastitis, the discrepancy in observations between healthy and mastitic quarters is large. To collect observations efficiently, our study used an alternative sampling design, i.e., sampling that was conditional on the signals of the on-line EC system. The objective of this study was to evaluate the efficacy of detecting severe subclinical mastitis using a prototype on-line quarter milk EC (QMEC) measurement system (14). MATERIALS AND METHODS On-Line System
The on-line prototype system (sensors by Nedap, Groenlo, The Netherlands, and software by IMAG-DLO, Wageningen, The Netherlands, and IVO-DLO, Zeist, The Netherlands) for measurement of QMEC, milk temperature, and milk yield was available in a six-point open tandem milking parlor at the experimental farm "de Bunzing" of the Institute for Animal Production "Schoonoord" (Zeist, The Netherlands) (8, 14). Three stalls were equipped with sensors in the short milk tube, and three stalls were equipped with sensors in the milking claw. During the study period, but independent of the study, the three milking claw sensors were slightly adapted to minimize mixing of quarter milk from the different quarters at quarter sensor leveL Every 5 s during milking, EC was measured for each quarter. The average of the 20 highest EC per quarter was defined as QMEC (14). The QMEC of all milkings since calving were used for the calculation of the running average (RA) QMEC, with a slight modification of the original rules (14); RA QMEC
= .33
x RA QMEC (actual milking) + .67 x RA QMEC (previous milking).
Subsequently, the quarter with the lowest RA QMEC of a cow was defined as the reference quarter (14). A quarter was signaled by the system (8) if 1) the actual RA QMEC deviated by >15% from the actual RA QMEC of the reference quarter and the actual QMEC
1133
was 20% over the RA QMEC of the reference quarter in both previous milkings or 2) the actual QMEC deviated by >15% from the RA QMEC of the reference quarter in the previous milking and the cow was signaled by the system for an increased milk temperature. Study Design
For our study, a quarter was defined as EC positive if it was signaled by the system and as EC negative if the quarter was not signaled. All EC-positive quarters were sampled for bacteria and SCC at the next milking. For practical reasons and for cost reduction, only 10 randomly selected EC-negative quarters were sampled for bacteria and SCC at the same milking. Based on assumptions of expected sensitivity, specificity, and prevalence, 10 ECnegative quarters per EC-positive quarter seemed to be necessary for evaluation. The EC-positive quarters were omitted from the study population for 2 wk (28 milkings) after the first EC signaL During this period, the quarter could not be sampled as an EC-positive or EC-negative quarter. In this 14-d period, a second sample of the EC-positive quarter was taken 1 wk after the signal to observe the dynamics of the inflammation in the quarter. Additional EC signals that appeared within the 2-wk waiting period were recorded. The EC-negative quarters that were defined subclinical in the analysis were also omitted from the study population for 2 wk after that sampling. Milk Samples
Data for this study were obtained from March 21 to July 2, 1991. The number of cows milked in this period varied between 46 and 50 cows per milking. Quarter samples were taken aseptically according to the procedures recommended by the International Dairy Federation (9). Udder and teats were cleaned with paper towels, the first stripping was milked, and teat ends were disinfected with 70% alcohol in a cotton swab. Duplicate foremilk samples were obtained aseptically from the quarter for bacteriological culturing. An additional sample was taken for SCC determination in a vial with sodium azide for preservation of samples. Somatic cells were Journal of Dairy Science Vol. 77. No.4. 1994
1134
LANSBERGEN ET AL.
TABLE I. Sampling conditional on test status. Subclinical mastitis System
Positive
Negative
Total
Signaled Nonsignaled l Total extrapolated 2
a c' a + c
b d' b + d
a + b c' + d' a + b + c + d
'With every signaled quarter (a or b), 10 nonsignaled quarters (c' or d') were randomly sampled. 2The total number of nonsignaled quarters (c + d) was extrapolated by dividing the sampled quarters (c' + d') by the sampling fraction.
counted using a Fossomatic cell counter (Foss Electric, Hiller0d, Denmark). Other samples were stored at -20·C until bacteriological culturing. The time between collection of the milk samples and bacteriological culturing varied between 31 and 325 d. The median storage length was 202 d. Bacteriological procedures adhered to National Mastitis Council (19) standards: .1 ml from each sample was spread over an agar plate (Oxoid, Basingstoke, England) using a Drigalsky spatula and cultured on 6% bovine blood agar (aerobically and anaerobically), MacConkey number 3 agar, and thallium crystal violet toxin agar. Gram-positive cultures were differentiated with the catalase test. Streptococci were identified by means of CAMP reaction, ox bile, esculin, and sodium hippurate. Staphylococci were identified by coagulase test and were differentiated from micrococci by anaerobic culture. Gramnegative microorganisms were determined using cytochrome oxidase, triple sugar iron agar, ureum, indol, ornithine decarboxylase. nitrate reduction, and gelatine liquefaction. The number of colony-forming units of each type of bacteria was counted. Definitions
Prior to statistical analysis, the cultured bacterial species were grouped into two categories. Staphylococcus aureus, Streptococcus dysgalactiae, and Streptococcus uberis were classified as major pathogens. Escherichia coli and other Gram-negative bacteria in combination with clinical symptoms were also in the first category. The second category consisted of coagulase-negative staphylococci, corynebacteria, micrococci, Bacillus spp., strepJournal of Dairy Science Vol. 77, No.4, 1994
tococci group D, Enterobacter spp., remaining coliform bacteria, Aspergillus spp., and Acinetobacter spp. A milk culture was considered to be bacteriologically positive if >500 cfuJml of Grampositive bacteria or > 100 cfuJml of Gramnegative bacteria from the category of major pathogens were cultured from both milk samples (22, 23). For this study, a quarter was considered to be subclinical if 1) the quarter milk sample was cultured positive as defined and had a SCC >500 x 103cells/ml or 2) the quarter sample had a SCC > 1000 x 1
The system was evaluated using standard test performance characteristics. Analysis was performed on quarters because the system signaled by quarters. Sensitivity, specificity, and predictive values of positive and negative results of the system were computed (16). Variances of the point estimators for sensitivity and specificity were calculated using a Taylor expansion for ratio estimators (18). Simple correlations between EC, In SCC, and log number of major pathogens were calculated. The number of major pathogens was calculated if at least one of the duplicate samples contained > I 0 cfuJml and was the mean of the duplicate samples. Data were analyzed using PROC FREQ of SAS (25). To calculate sensitivity and specificity, the total number of EC-negative quarters at sampling milkings was extrapolated for subclinical and healthy groups (Table 1). To test the impact of repeated measurements on the estimation of the test characteristics, an analysis was performed using every
1135
OUR INDUSTRY TODAY
TABLE 2. Quarters with and without an electrical conductivity (EC) signal grouped by major pathogens (MP) in the quarters. EC Positive Negative
Escherichia coli
Staphylococcus aureus
Streptococcus uberis
Streptococcus dysgalactiae
No MP
Total
2
4 4
0 1
o
48 517
54 530
o
quarter only once per EC category. Using a random number generator, one sampling of each EC-positive quarter was selected. The same procedure was performed for the ECnegative quarters. Subsequently, the selected quarters were assigned to subclinical or healthy groups. In a second analysis, the efficacy of repeated EC signals for the detection of subclinical mastitis was evaluated. In this analysis, a quarter was EC positive when, within 14 d after the first signal, the quarter was again signaled one or more times. The group of ECnegative quarters consisted of quarters with only one EC signal in the 14-d period and quarters without an EC signal. The possible effect of sensor adaptation was assessed by stratification of the data into two periods. RESULTS Descriptive Statistics
Fifty-four EC-positive and 530 EC-negative quarter samples were sampled at 32 milkings. Six EC-positive quarters and 10 EC-negative quarters were not sampled for logistical reasons. In Table 2, the numbers of EC-positive and EC-negative quarters were grouped by results of bacteriological culturing. The number of
8
EC-positive and EC-negative quarters were grouped in two SCC classes in Table 3. Mean In SCC was 3.8 (SD = 1.3) for the EC-negative samples and 5.8 (SD = 1.8) for the EC-positive samples. Simple correlations between EC (millisiemens per centimeter), In SCC, and log10 number of major pathogens are shown in Table 4. The correlation between EC and In SCC in all sampled quarters (n = 611) was .43. The correlation between EC and In SCC in quarters containing major pathogens (n = 42) was .63. Four cases of clinical mastitis occurred; all were EC positive in the clinical period and were analyzed within the EC-positive category. Analysis of Diagnostic Performance
Because of the retrospectively defined waiting period for EC-negative quarters that were defined subclinical, 6 quarter samples were omitted from the analysis. Of 52 EC-positive quarters, 18 were classified as subclinical; of the 526 EC-negative quarters, 5 were classified as subclinical (Table 5). The predictive value was 35% when a signal was given and 99% when no EC signal was given. Predictive value of a positive test
TABLE 4. Simple correlation coefficients between electrical conductivity (EC), m(SCC), and log\O of major pathogens (MP) in quarters with Staphylococcus aureus (n = 32) or Streptococcus dysgalactiae (n = 10). m(SCC)
TABLE 3. Quarters with and without an electrical conductivity (EC) signal grouped into two SCC classes.
sec EC Positive Negative
17
37
4
526
IX
103 ce ll s/ml.
(cfulml)
Staph. aureus
>1000 1
;5;1000
Total 54 530
loglO(MP)
Strep. dys.
EC m(SCC) EC m(SCC)
.63** .63*
.31 .64** .68* .93**
*p < .05. **p < .001. Journal of Dairy Science Vol. 77. No.4, 1994
1136
LANSBERGEN ET AL.
TABLE 5. Subclinical mastitis status, as defined by results of bacteriological culturing and SCC of quarters with and without an electrical conductivity (EC) signal.
Total study period EC Positive EC Negative Before sensor adaptation EC Positive EC Negative After sensor adaptation EC Positive EC Negative
Subclinical Healthy
Total
18 5
34 521
52 526
5 4
21 268
26 272
13
13 253
26 254
I
was 19% before sensor adaptation and 50% after. Predictive value of a negative test was 99% in both periods. Extrapolation of EC-negative quarters gave the following results. The EC signal identified correctly 18 out of 71 subclinical and 5924 out of 5958 healthy quarters (Table 6). The sensitivity of the system was 25% (95% confidence interval; 6 to 45%). The specificity was 99.4% (99.2 to 99.6%). Sensitivity was 13% (0 to 28%) before sensor adaptation and 42% (0 to 85%) after adaptation. Specificity was 99.3 and 99.6% before and after, respectively. The estimated quarter prevalence was 1%. The 34 false-positive quarters were studied in detail. Two cows that were in estrus were signaled in all four quarters at the same time, causing 4 and 2 false-positive quarters, respectively, per cow. Two cows with clinical mastitis were also signaled in all 4 quarters, resulting in 2 and 3 false-positive quarters per cow. One false-positive quarter was from a cow with clinical mastitis in another quarter the day before. One cow with a hard nodule in a
TABLE 6. Subclinical mastitis status, as defined by results of bacteriological culturing and SCc. The quarters without an electrical conductivity (EC) signal were extrapolated to total numbers present at the milkings.
EC Positive EC Negative Extrapolated total
Subclinical
Healthy
Total
18 53
34 5924
52 5977
71
5958
6029
Journal of Dairy Science Vol. 77. No.4, 1994
quarter was signaled repeatedly, causing 4 false positives and 1 true positive. One cow caused 6 false positive signals for 1 quarter over the entire study, but the quarter seemed clearly healthy. The 5 false-negative quarters were also studied in more detail; 2 were caused by 1 consistently infected quarter (Strep. dysgalactiae), which would have caused 6 falsenegative signals without the 2-wk waiting period. The 3 other false-negative quarters were from 3 different cows. After random selection of only 1 measurement per quarter, II out of 31 EC-positive quarters had subclinical mastitis, and 198 out of 202 EC-negative quarters were healthy. Predictive values hardly changed and were 35 and 98%, respectively. Selection of 1 measurement per quarter in the period before sensor adaptation resulted in a predictive value of a positive test of 19% and of a test of 97%. After sensor adaptation, these values were 50 and 99%, respectively. Sensitivity and specificity could not be calculated because the total number of EC-negative quarters could not be extrapolated. In the analysis with repeated EC signals that occurred in the 2-wk waiting period, the number of false-positive quarters decreased in favor of the number of true-negative quarters. Of the 29 repeatedly signaled EC-positive quarters, 14 had subclinical mastitis. Of the 549 EC-negative quarters, 540 were healthy (Table 7). The predictive value of an EC signal (48%) improved, and the predictive value of no EC signal (98%) remained high. Before and after sensor adaptation, predictive values of a positive test was 36 and 56%, and predictive value of a negative test was 98% for both periods. After extrapolation, sensitivity and specificity for the total period were 13% (3 to
TABLE 7. Subclinical mastitis status, as defined by results of bacteriological culturing and SCc. A quarter was positive for electrical conductivity (EC) if more than one signal was given within a 14·d period and EC negative with only one or no EC signal. Subclinical Healthy EC Positive, repeated signals EC Negative
14 9
15 540
Total 29 549
1137
OUR INDUSTRY TODAY
24%) and 99.7% (99.6 to 99.9%), respectively. Sensitivity was 10% (0 to 24%) before sensor adaptation and 16% (0 to 32%) after, and specificity was 99.7% in both periods.
probability of a quarter being negative in this population.
DISCUSSION
The EC signals of the prototype system were developed to detect severe subclinical as well as clinical mastitis cases (14). Because of the original development goal, high threshold values for SCC and culturing were chosen in our subclinical mastitis definition. The high number of false positives could be a direct effect of the high thresholds in the subclinical mastitis definition. Reclassification of quarters with a SCC >200 x lcPcells/ml as subclinical reduced the number of false positives and increased predictive value of a positive test (from 35 to 60%). However, predictive value of a negative test decreased (from 99 to 87%) because of lowered sensitivity and increased prevalence. Because of the high thresholds, 5 ECnegative quarters were defined as healthy, although major pathogens were present and SCC was elevated in 3 quarters (Table 8). I n the analysis, these 5 quarters appeared to be true negatives, but false negatives would have been more accurate. An expected effect of the high thresholds was a relatively high sensitivity, because the subclinical defined cases were clearly abnormal. Sensitivity was not high in our study, possibly partly because of our study design. Grouping of the sampled quarters into subclinical and healthy quarters based on SCC only (Table 3) resulted in fewer false-positive
Use for Subclinical Mastitis Detection
The predictive values of the system are the parameters that a future user needs to know to manage a herd with the system. Predictive values are affected by sensitivity, specificity, and prevalence (15). The prevalence of subclinical mastitis in this herd was estimated by the number of subclinical quarters as a fraction of the total number of quarters. The estimated quarter prevalence was about 1%. As a result of high thresholds for bacteriological culturing and SCC in our subclinical mastitis definition, the estimated quarter prevalence was very low. However, the bulk tank SCC of this herd and the SCC per cow were low in comparison with the Dutch average values (29). The low prevalence was also supported by other samplings of the herd during the study. With the high specificity, the system could be used diagnostically. The relative importance of a high sensitivity or high predictive value of a positive test depends on the importance of missing cases or the actions taken on a positive test result. If the system is to be used as a tool for decisions on treatment of quarters, the ratio of cost to benefit of treatment of subclinical mastitis should be taken into account. Although a predictive value of a positive test of 48% (repeated signals) is not low in a population with a prevalence of 1%, the chance that subclinical mastitis is present in repeatedly EC-positive quarters is only about 50%. Additionally, subclinical quarters would not be treated as a result of the low sensitivity. Because of the low predictive value of a positive test in this herd, the system should not be used as a tool for decisions on culling. To use the system as a screening test, sensitivity must be high so that the number of false negatives is low. Even with a high predictive value of a negative test (99%), false-negative quarters remained as a result of the low sensitivity. The high predictive value of a negative test was only slightly higher than the prior
Effects of Subclinical Mastitis Definition on Evaluation
TABLE 8. Bacteriological results of 5 nonsignaled quartees that had see <500 x l03cellslmJ. The 5 quarters were classified as not being subclinical and appeared as true negatives in the study.
see l
Mp2
484 416
Streptococcus dysgalactiae Staphylococcus aureus Staph. aureus Staph. aureus Staph. aureus
Duplicate samples - - (cfulmJ) - -
353 78
35 IX
1200 1400 250 1500 300
800 1250 300 125 200
l03cellslmJ.
2Major pathogens. Journal of Dairy Science Vol. 77, No.4, 1994
1138
LANSBERGEN ET AL.
and false-negative quarters than grouping based on culturing only (Table 2). This result contrasts with findings of Nielen et al. (20), in which subclinical mastitis as defined by bacteriological culturing resulted in higher sensitivity than with subclinical mastitis defined by SCC. However, numbers of cultured positive quarters were very low in the current study. The correlations between EC and In SCC confirmed findings of previous studies (3, 4, 13, 28). Effects of Study Design on Evaluation
Repeated Measurements. The impact of repeated measurements on the predictive values was small, because predictive values only slightly changed when single observations per quarter were used. Therefore, quarters were consistently classified within EC categories. However, for sensitivity and specificity estimation, independent observations were assumed, and repeated measurements could have created a bias. Because false negatives remained present when only single observations were used, the low sensitivity with repeated measurements was not expected to change to a very high sensitivity. Comparison at a Given Moment. The shedding of the bacteria, the rise in SCC, and the rise in EC might be processes with different time frames, making a comparison at a given moment difficult. In addition, subclinical mastitis was often present for several weeks (7).
Sears et al. (27) found that the shedding of Staph. aureus from the mammary gland was not constant. For accurate diagnosis of subclinical quarters with bacteriological culturing, use of consecutive samples is advisable. The samples in our study that were positive for EC could have been taken at the very moment that minimal or no pathogens were shedded by the quarter, causing false positives. To consider the dynamics of the inflammation, EC-positive quarters were sampled again 1 wk later. Of the 34 resampled, 7 changed status, mostly because of a change in SCc. Three of 4 resampled quarters infected with Staph. aureus changed between bacteriologically negative and positive. Bacteriological results were also reversed in a quarter with a Strep. uberis infection. This Journal of Dairy Science Vol. 77, No.4, 1994
quarter was sampled as an EC-negative quarter and classified as a false negative. When the quarter was sampled as EC positive 24 h later, Strep. uberis was still present in both duplicate milk samples, but below the threshold of 500 cfu/ml. This second sample was not included in the study population because of the retrospectively defined 2-wk waiting period for EC-negatives that were subclinical but would have been false positive. An example of long duration of subclinical infection was a quarter that was consistently infected with Strep. dysgalactiae. This quarter was infected during the entire study and was only signaled twice. The quarter was erroneously not sampled as EC positive but was sampled as EC negative seven times. Extrapolation. In our study, the predictive value of a positive test was not influenced by the extrapolation procedure. The predictive value of a negative test was based on a random sample but was not influenced by the extrapolation. Because of the low numbers, the effect of the extrapolation on the precision of sensitivity estimation was large. This difference was reflected in the large confidence intervals. The effect on specificity estimation was smaller because of the much larger numbers of healthy quarters. Effect of Sensor Adaptation on Evaluation
The adaptation of the 3 milking claw sensors was to reduce further the mixing of milk from different quarters at sensor level. When milk from a high EC quarter mixes with milk of other quarters during measuring, the difference in EC between the quarters is underestimated. Two effects on test performance can be the result: 1) false signals in healthy quarters when a large rise of EC occurs in a high EC quarter or 2) no signal in the high EC quarter because the difference with the reference quarter is diminished by the mixing. Test characteristics improved after sensor adaptation, thereby retrospectively justifying the adaptation. However, sensitivity remained low. The 6 false positive signals for cows with a clinical mastitis quarter occurred before sensor adaptation. When the 6 false-positive signals were deleted from the data, predictive value of a positive test changed from 19 to
1139
OUR INDUSTRY TODAY
25% before sensor adaptation and from 35 to 39% over the entire study. Prior Evaluation
During development, the system was evaluated by comparison of EC signals with SCC and bacteriological culturing (14). Every 2 mo during I yr, quarter samples of all lactating cows were taken. Milk samples were grouped into health categories according to International Dairy Federation (9) definitions. If latent mastitis and secretion disturbance were not taken into account, 14 of 26 subclinical and 192 of 200 healthy cows were identified correctly by the system. Sensitivity was 53%, and specificity was 96%. Those results cannot be compared easily with those of the present study. Definitions of subclinical mastitis differed. Furthermore, the conditions of an EC signal were not completely comparable. Subclinical mastitis and EC signals were not compared at the same moment. Prevalence of subclinical mastitis was higher because data were analyzed by cow. However, both studies agree on high specificity, which coincides with other reports (20). The difference in sensitivity estimation was probably caused by the analysis by cow, combined with a longer period around sampling during which a positive EC signal was considered to be a true positive. In the present study, bias from repeated measurements might have underestimated sensitivity. Future Developments
Importance of Attention. Different signals for the user should be considered for cows suspected of having subclinical versus clinical mastitis. For subclinical cases, normally no direct action is taken under present management circumstances. For clinical cases, early and rapid detection is preferable because treatment should not be delayed. In the studied prototype, the two conditions causing an EC signal were developed for subclinical and clinical cases, respectively, and this information should be available to the user. Cows in this study that were signaled in all 4 quarters were in estrus or had a clinically mastitic quarter. Prevention of the mixing of milk from different quarters at sensor level
should remove influence of clinical quarters on other quarter readings. Effects of estrus were not based on mixing of quarter milk. The signals were caused by the second condition and should be expected to occur incidentally, especially when a cow has a raised body temperature during estrus. Subclinical Detection. Based on our results, repeated QMEC signals may offer good prospects for detection of subclinical mastitis (fable 7). Repeated signals in the same quarter within a certain time frame might be the best EC predictor for a persistent subclinical infection. However, the conditions in the studied prototype would not signal a quarter with a steadily high QMEC if the actual QMEC were <20% over the reference quarter during two milkings. Only relative large fluctuations in QMEC compared with those of the reference quarter would cause a signal. Signaling of quarters with steadily high QMEC as subclinical quarters might be advisable. Study Design. For detection of subclinical mastitis, a comparison of EC, SCC, and bacteriological culturing at the same moment might not be the most accurate comparison. To obtain sufficient data, samples based on EC signals could be taken during short periods in different herds. To consider the prior and future subclinical mastitis status of the quarters, frequent sampling of quarters before and after the EC testing period is important. Records for herds with high prevalence of subclinical mastitis should be included to improve sensitivity estimation. Sampling all quarters of a cow to obtain interquarter ratios for SCC may be advisable (10). CONCLUSIONS
The overall conclusion based on the study in this herd is that the efficacy was low for the prototype on-line EC system as a screening or diagnostic test for subclinical mastitis. The low prevalence of subclinical mastitis cases, in combination with repeated measurements and the extrapolation procedure, might have caused a low sensitivity estimation. ACKNOWLEDGMENTS
The authors gratefully acknowledge the help of the following persons: l.P.A. Peek, Journal of Dairy Science Vol. 77. No.4. 1994
1140
LANSBERGEN ET AL.
Th.A.M. Verkerk, and other personnel of "de Bunzing" for taking all of the milk samples and R. T. van Zonneveld for providing additional data for the analysis. This research has been funded by the Dutch Foundation for Knowledge-Based Systems (SKBS).
REFERENCES I Batra, T. R., and A. J. McAllister. 1984. A comparison of mastitis detection methods in dairy cattle. Can. J. Anim. Sci. 64:305. 2 Chamings, R. 1., G. Murray, and 1. M. Booth. 1984. Use of conductivity meter for the detection of subclinical mastitis. Vet. Rec. 114:243. 3 Emanuelson, U., T. Olsson, O. Holmberg, M. Hageltorn, T. Mattila, L. Nelson, and G. AstrOm. 1987. Comparison of some screening tests for detecting mastitis. J. Dairy Sci. 70:880. 4 Fernando, R. S., S. L. Spahr, and E. H. Jaster. 1985. Comparison of electrical conductivity of milk with other indirect methods for detection of subclinical mastitis. J. Dairy Sci. 68:449. 5 Gebre-Egziabher, A., H. C. Wood, J. D. Robar, and G. Blankenagel. 1979. Evaluation of automatic mastitis detection equipment. J. Dairy Sci. 62:1108. 6 Greatix, G. R., J. C. Quale, and R. A. Coombe. 1968. The detection of abnormal milk by electrical means. J. Dairy Sci. 35:213. 7 Grommers, F. J., D. van de Geer, and C.A.A. in 't Veen. 1985. Duration of bovine intramammary infections in commercial dairy herds. Vet. Rec. 116:581. 8 Hogewerf, P. H., K. Maatje, and W. Rossing. 1992. Computed aided systems for health and reproduction control in dairy cows. Page 483 in Proc. Symp. Prospects Automatic Milking, Eur. Assoc. Anim. Prod. Publ. 65. Pudoc Sci. Publ., Wageningen, Neth. 9 International Dairy Federation, 198 I. Laboratory methods for use in mastitis work. Int. Dairy Fed. Doc. 132., Int. Dairy Fed., Brussels, Belgium. 10 Jensen, N. E., and K. Knudsen. 1991. Interquarter comparison of markers of subclinical mastitis: somatic cell count, electrical conductivity, N-acetyl-j3glucosaminidase and antitrypsin. J. Dairy Res. 58:389. II Kitchen, B. J. 1981. Review of the progress of dairy science: bovine mastitis: milk compositional changes and related diagnostic tests. J. Dairy Res. 48:167. 12 Linzell, J. L., and M. Peaker. 1975. Efficacy of the measurement of the electrical conductivity of milk for the detection of subclinical mastitis in cows: detection of infected cows at a single visit. Br. Vet. 1. 131:447. 13 Little, T,W.A., C. N. Herbert, and D. Forbes. 1968. Electrical conductivity and the leucocyte count of bovine milk. Vet. Rec. 82:431. 14 Maatje, K., P.I.M. Huijsmans, W. Rossing, and P. H. Hogewerf. 1992. The efficacy of in-line measurement of quarter milk electrical conductivity, milk yield and
Journal of Dairy Science Vol. 77, No.4, 1994
milk temperature for the detection of clinical and subclinical mastitis. Livest. Prod. Sci. 30:239. 15 Martin, S. W. 1984. Estimating disease prevalence and the interpretation of screening test results. Prevo Vet. Med. 2:463. 16 Martin, S. W., A. H. Meek, and P. Willeberg. 1987. Veterinary Epidemiology: Principles and Methods. Iowa State Univ. Press, Ames. 17 Mijnen, E., FRJ. Jaartsveld, G.A.A. Albers, M'w.A. Verstegen, and M.J.M. Tielen. 1983. The value of cell count, lactose content, pH and conductivity of milk for mastitis detection in individual cows. Neth. Milk Dairy 1. 37:65. 18 Mood, M. A., F. A. Graybill, and D. C. Boes. 1974. Introduction to the Theory of Statistics. McGraw-Hili Book Co., New York, NY. 19 National Mastitis Council. 1987. Laboratory and Field Handbook on Bovine Mastitis. W. D. Hoard & Sons Co., Fort Atkinson, WI. 20 Nielen, M., H. Deluyker, Y. H. Schukken, and A. Brand. 1992. Electrical conductivity of milk: measurement, modifiers, and meta analysis of mastitis detection performance. 1. Dairy Sci. 75:606. 21 Okigbo, L. M., M. A. Sheliah, G. H. Richardson, C. A. Ernstrom, R. 1. Brown, and E. L. Tippetts. 1984. Portable conductivity meter for detecting abnormal milk. J. Dairy Sci. 67:1510. 22 Oliver, S. P., S. H. King, M. J. Lewis, P. M. Torre, K. R. Matthews, and H. H. Dowlen. 1990. Efficacy of chlorhexidine as a postmilking teat disinfectant for the prevention of bovine mastitis during lactation. 1. Dairy Sci. 73:2230. 23 Oliver, S. P., S. H. King, P. M. Torre, E. P. Schull, H. H. Dowlen, M. J. Lewis, and L. M. Sordillo. 1989. Prevention of bovine mastitis by Postmilking teat disinfectant containing chlorous acid and chlorine dioxide in soluble polymer gel. J. Dairy Sci. 72:3091. 24 Peaker, M. 1978. The electrical conductivity of milk for the detection of subclinical mastitis in cows: comparison of various methods of handling conductivity data with the use of cell count and bacteriological examination. Br. Vet. 1. 134:314. 25 SAS(/!) User's Guide: Statistics, Version 6.03 Edition. 1988. SAS Inst., Inc., Cary, NC. 26 Schakenraad, M.H'w., and A. A. Dijkhuizen. 1990. Economic losses due to bovine mastitis in Dutch dairy herds. Neth. J. Agric. Sci. 38:89. 27 Sears, P. M., B. S. Smith, P. B. English, P. S. Herer, and R. N. Gonzalez. 1990. Shedding pattern of Staphylococcus aureus from bovine intramammary infections. J. Dairy Sci. 73:2785. 28 Sheldrake, R. F., G. D. McGregor, and R.I.T. Hoare. 1983. Somatic cell count, electrical conductivity, and serum albumin concentration for detecting bovine mastitis. J. Dairy Sci. 66:548. 29 Vecht, U., H. 1. Wisselink, and P. R. Defize. 1987. Verslag landelijke steekproef mastitis 1985/1986. Rapport Centraal Diergeneeskundig Instituut, Lelystad, Neth.
Evaluation of a Prototype On-Line Electrical Conductivity System for Detection of Subclinical Mastitis L.M.T.E. LANSBERGEN, M. NIELEN,1 T.J.G.M. LAM, A. PENGOY,2 and Y. H. SCHUKKEN Department of Herd Health and Reproduction Utrecht University Yalelaan 7 3584 CL Utrecht, The Netherlands K. MAATJE IYO-DLO Institute for Animal Production PO Box 501 3700 AM Zeist, The Netherlands ABSTRACT
A prototype on-line system for measurement of electrical conductivity of quarter milk was evaluated for accuracy in detection of subclinical mastitis compared with that of bacteriological culture and SCC of sampled quarters. Because of the low quarter prevalence of mastitis, quarters were sampled conditionally from the signals of the on-line system. All signaled quarters and a random selection of the nonsignaled quarters were sampled. To calculate sensitivity and specificity, the total number of nonsignaled quarters was extrapolated. The system identified correctly 18 out of 23 subclinical quarters and 521 out of 555 healthy quarters. Quarter prevalence was about 1%. Predictive value of a positive test (35%) and the predictive value of a negative test (99%) were calculated, as well as sensitivity (25%) and specificity (99%), after extrapolation of the total number of nonsignaled quarters. Because of repeated measurements, sensitivity may be underestimated. When signaled quarters were defined by repeated signals within 14 d, predictive value positive increased to 48%. The prototype on-line system did not detect subclinical mastitis very ac-
Received August 30. 1993. Accepted November 3, 1993. lCorresponding author. 2Yeterinary Microbiological Diagnostic Centre. 1994 J Dairy Sci 77:1132-1140
curately because of suboptimal test characteristics. (Key words: on-line electrical conductivity, subclinical mastitis detection) Abbreviation key: EC = electrical conductivity, QMEC = quarter milk EC, RA = running average. INTRODUCTION
Modem, large dairy herds often use automated management systems. The costs of mastitis (26), combined with stricter legislation for bulk SCC, indicate a potential use for an efficient and direct (on-line) method for the detection of subclinical and clinical mastitis. At present, on-line mastitis detection systems based on electrical conductivity (EC) are becoming commercially available. According to Kitchen (11), mastitis increased the EC of milk because of changes in ionic concentrations. As a result of the damage of udder tissue, concentrations of lactose and K+ decreased, and concentrations of Na+ and Cl- increased. Several studies have evaluated measurement of milk EC as a method for detecting subclinical or clinical mastitis (20). Most investigators (4, 5, 6, 12, 14, 21, 24) considered milk EC to be accurate for the detection of mastitis; others (I, 2, 17) considered it to be less accurate. In the future, more on-line mastitis detection systems probably will be developed that need evaluation before commercial use. One of the problems in evaluation is the definition of subclinical mastitis (20). Also, analysis of an on-line system over time results in repeated 1132
OUR INDUSTRY TODAY
measurements on the same cows and quarters. In addition, for herds with good udder health and a low prevalence of subclinical mastitis, the discrepancy in observations between healthy and mastitic quarters is large. To collect observations efficiently, our study used an alternative sampling design, i.e., sampling that was conditional on the signals of the on-line EC system. The objective of this study was to evaluate the efficacy of detecting severe subclinical mastitis using a prototype on-line quarter milk EC (QMEC) measurement system (14). MATERIALS AND METHODS On-Line System
The on-line prototype system (sensors by Nedap, Groenlo, The Netherlands, and software by IMAG-DLO, Wageningen, The Netherlands, and IVO-DLO, Zeist, The Netherlands) for measurement of QMEC, milk temperature, and milk yield was available in a six-point open tandem milking parlor at the experimental farm "de Bunzing" of the Institute for Animal Production "Schoonoord" (Zeist, The Netherlands) (8, 14). Three stalls were equipped with sensors in the short milk tube, and three stalls were equipped with sensors in the milking claw. During the study period, but independent of the study, the three milking claw sensors were slightly adapted to minimize mixing of quarter milk from the different quarters at quarter sensor leveL Every 5 s during milking, EC was measured for each quarter. The average of the 20 highest EC per quarter was defined as QMEC (14). The QMEC of all milkings since calving were used for the calculation of the running average (RA) QMEC, with a slight modification of the original rules (14); RA QMEC
= .33
x RA QMEC (actual milking) + .67 x RA QMEC (previous milking).
Subsequently, the quarter with the lowest RA QMEC of a cow was defined as the reference quarter (14). A quarter was signaled by the system (8) if 1) the actual RA QMEC deviated by >15% from the actual RA QMEC of the reference quarter and the actual QMEC
1133
was 20% over the RA QMEC of the reference quarter in both previous milkings or 2) the actual QMEC deviated by >15% from the RA QMEC of the reference quarter in the previous milking and the cow was signaled by the system for an increased milk temperature. Study Design
For our study, a quarter was defined as EC positive if it was signaled by the system and as EC negative if the quarter was not signaled. All EC-positive quarters were sampled for bacteria and SCC at the next milking. For practical reasons and for cost reduction, only 10 randomly selected EC-negative quarters were sampled for bacteria and SCC at the same milking. Based on assumptions of expected sensitivity, specificity, and prevalence, 10 ECnegative quarters per EC-positive quarter seemed to be necessary for evaluation. The EC-positive quarters were omitted from the study population for 2 wk (28 milkings) after the first EC signaL During this period, the quarter could not be sampled as an EC-positive or EC-negative quarter. In this 14-d period, a second sample of the EC-positive quarter was taken 1 wk after the signal to observe the dynamics of the inflammation in the quarter. Additional EC signals that appeared within the 2-wk waiting period were recorded. The EC-negative quarters that were defined subclinical in the analysis were also omitted from the study population for 2 wk after that sampling. Milk Samples
Data for this study were obtained from March 21 to July 2, 1991. The number of cows milked in this period varied between 46 and 50 cows per milking. Quarter samples were taken aseptically according to the procedures recommended by the International Dairy Federation (9). Udder and teats were cleaned with paper towels, the first stripping was milked, and teat ends were disinfected with 70% alcohol in a cotton swab. Duplicate foremilk samples were obtained aseptically from the quarter for bacteriological culturing. An additional sample was taken for SCC determination in a vial with sodium azide for preservation of samples. Somatic cells were Journal of Dairy Science Vol. 77. No.4. 1994
1134
LANSBERGEN ET AL.
TABLE I. Sampling conditional on test status. Subclinical mastitis System
Positive
Negative
Total
Signaled Nonsignaled l Total extrapolated 2
a c' a + c
b d' b + d
a + b c' + d' a + b + c + d
'With every signaled quarter (a or b), 10 nonsignaled quarters (c' or d') were randomly sampled. 2The total number of nonsignaled quarters (c + d) was extrapolated by dividing the sampled quarters (c' + d') by the sampling fraction.
counted using a Fossomatic cell counter (Foss Electric, Hiller0d, Denmark). Other samples were stored at -20·C until bacteriological culturing. The time between collection of the milk samples and bacteriological culturing varied between 31 and 325 d. The median storage length was 202 d. Bacteriological procedures adhered to National Mastitis Council (19) standards: .1 ml from each sample was spread over an agar plate (Oxoid, Basingstoke, England) using a Drigalsky spatula and cultured on 6% bovine blood agar (aerobically and anaerobically), MacConkey number 3 agar, and thallium crystal violet toxin agar. Gram-positive cultures were differentiated with the catalase test. Streptococci were identified by means of CAMP reaction, ox bile, esculin, and sodium hippurate. Staphylococci were identified by coagulase test and were differentiated from micrococci by anaerobic culture. Gramnegative microorganisms were determined using cytochrome oxidase, triple sugar iron agar, ureum, indol, ornithine decarboxylase. nitrate reduction, and gelatine liquefaction. The number of colony-forming units of each type of bacteria was counted. Definitions
Prior to statistical analysis, the cultured bacterial species were grouped into two categories. Staphylococcus aureus, Streptococcus dysgalactiae, and Streptococcus uberis were classified as major pathogens. Escherichia coli and other Gram-negative bacteria in combination with clinical symptoms were also in the first category. The second category consisted of coagulase-negative staphylococci, corynebacteria, micrococci, Bacillus spp., strepJournal of Dairy Science Vol. 77, No.4, 1994
tococci group D, Enterobacter spp., remaining coliform bacteria, Aspergillus spp., and Acinetobacter spp. A milk culture was considered to be bacteriologically positive if >500 cfuJml of Grampositive bacteria or > 100 cfuJml of Gramnegative bacteria from the category of major pathogens were cultured from both milk samples (22, 23). For this study, a quarter was considered to be subclinical if 1) the quarter milk sample was cultured positive as defined and had a SCC >500 x 103cells/ml or 2) the quarter sample had a SCC > 1000 x 1
The system was evaluated using standard test performance characteristics. Analysis was performed on quarters because the system signaled by quarters. Sensitivity, specificity, and predictive values of positive and negative results of the system were computed (16). Variances of the point estimators for sensitivity and specificity were calculated using a Taylor expansion for ratio estimators (18). Simple correlations between EC, In SCC, and log number of major pathogens were calculated. The number of major pathogens was calculated if at least one of the duplicate samples contained > I 0 cfuJml and was the mean of the duplicate samples. Data were analyzed using PROC FREQ of SAS (25). To calculate sensitivity and specificity, the total number of EC-negative quarters at sampling milkings was extrapolated for subclinical and healthy groups (Table 1). To test the impact of repeated measurements on the estimation of the test characteristics, an analysis was performed using every
1135
OUR INDUSTRY TODAY
TABLE 2. Quarters with and without an electrical conductivity (EC) signal grouped by major pathogens (MP) in the quarters. EC Positive Negative
Escherichia coli
Staphylococcus aureus
Streptococcus uberis
Streptococcus dysgalactiae
No MP
Total
2
4 4
0 1
o
48 517
54 530
o
quarter only once per EC category. Using a random number generator, one sampling of each EC-positive quarter was selected. The same procedure was performed for the ECnegative quarters. Subsequently, the selected quarters were assigned to subclinical or healthy groups. In a second analysis, the efficacy of repeated EC signals for the detection of subclinical mastitis was evaluated. In this analysis, a quarter was EC positive when, within 14 d after the first signal, the quarter was again signaled one or more times. The group of ECnegative quarters consisted of quarters with only one EC signal in the 14-d period and quarters without an EC signal. The possible effect of sensor adaptation was assessed by stratification of the data into two periods. RESULTS Descriptive Statistics
Fifty-four EC-positive and 530 EC-negative quarter samples were sampled at 32 milkings. Six EC-positive quarters and 10 EC-negative quarters were not sampled for logistical reasons. In Table 2, the numbers of EC-positive and EC-negative quarters were grouped by results of bacteriological culturing. The number of
8
EC-positive and EC-negative quarters were grouped in two SCC classes in Table 3. Mean In SCC was 3.8 (SD = 1.3) for the EC-negative samples and 5.8 (SD = 1.8) for the EC-positive samples. Simple correlations between EC (millisiemens per centimeter), In SCC, and log10 number of major pathogens are shown in Table 4. The correlation between EC and In SCC in all sampled quarters (n = 611) was .43. The correlation between EC and In SCC in quarters containing major pathogens (n = 42) was .63. Four cases of clinical mastitis occurred; all were EC positive in the clinical period and were analyzed within the EC-positive category. Analysis of Diagnostic Performance
Because of the retrospectively defined waiting period for EC-negative quarters that were defined subclinical, 6 quarter samples were omitted from the analysis. Of 52 EC-positive quarters, 18 were classified as subclinical; of the 526 EC-negative quarters, 5 were classified as subclinical (Table 5). The predictive value was 35% when a signal was given and 99% when no EC signal was given. Predictive value of a positive test
TABLE 4. Simple correlation coefficients between electrical conductivity (EC), m(SCC), and log\O of major pathogens (MP) in quarters with Staphylococcus aureus (n = 32) or Streptococcus dysgalactiae (n = 10). m(SCC)
TABLE 3. Quarters with and without an electrical conductivity (EC) signal grouped into two SCC classes.
sec EC Positive Negative
17
37
4
526
IX
103 ce ll s/ml.
(cfulml)
Staph. aureus
>1000 1
;5;1000
Total 54 530
loglO(MP)
Strep. dys.
EC m(SCC) EC m(SCC)
.63** .63*
.31 .64** .68* .93**
*p < .05. **p < .001. Journal of Dairy Science Vol. 77. No.4, 1994
1136
LANSBERGEN ET AL.
TABLE 5. Subclinical mastitis status, as defined by results of bacteriological culturing and SCC of quarters with and without an electrical conductivity (EC) signal.
Total study period EC Positive EC Negative Before sensor adaptation EC Positive EC Negative After sensor adaptation EC Positive EC Negative
Subclinical Healthy
Total
18 5
34 521
52 526
5 4
21 268
26 272
13
13 253
26 254
I
was 19% before sensor adaptation and 50% after. Predictive value of a negative test was 99% in both periods. Extrapolation of EC-negative quarters gave the following results. The EC signal identified correctly 18 out of 71 subclinical and 5924 out of 5958 healthy quarters (Table 6). The sensitivity of the system was 25% (95% confidence interval; 6 to 45%). The specificity was 99.4% (99.2 to 99.6%). Sensitivity was 13% (0 to 28%) before sensor adaptation and 42% (0 to 85%) after adaptation. Specificity was 99.3 and 99.6% before and after, respectively. The estimated quarter prevalence was 1%. The 34 false-positive quarters were studied in detail. Two cows that were in estrus were signaled in all four quarters at the same time, causing 4 and 2 false-positive quarters, respectively, per cow. Two cows with clinical mastitis were also signaled in all 4 quarters, resulting in 2 and 3 false-positive quarters per cow. One false-positive quarter was from a cow with clinical mastitis in another quarter the day before. One cow with a hard nodule in a
TABLE 6. Subclinical mastitis status, as defined by results of bacteriological culturing and SCc. The quarters without an electrical conductivity (EC) signal were extrapolated to total numbers present at the milkings.
EC Positive EC Negative Extrapolated total
Subclinical
Healthy
Total
18 53
34 5924
52 5977
71
5958
6029
Journal of Dairy Science Vol. 77. No.4, 1994
quarter was signaled repeatedly, causing 4 false positives and 1 true positive. One cow caused 6 false positive signals for 1 quarter over the entire study, but the quarter seemed clearly healthy. The 5 false-negative quarters were also studied in more detail; 2 were caused by 1 consistently infected quarter (Strep. dysgalactiae), which would have caused 6 falsenegative signals without the 2-wk waiting period. The 3 other false-negative quarters were from 3 different cows. After random selection of only 1 measurement per quarter, II out of 31 EC-positive quarters had subclinical mastitis, and 198 out of 202 EC-negative quarters were healthy. Predictive values hardly changed and were 35 and 98%, respectively. Selection of 1 measurement per quarter in the period before sensor adaptation resulted in a predictive value of a positive test of 19% and of a test of 97%. After sensor adaptation, these values were 50 and 99%, respectively. Sensitivity and specificity could not be calculated because the total number of EC-negative quarters could not be extrapolated. In the analysis with repeated EC signals that occurred in the 2-wk waiting period, the number of false-positive quarters decreased in favor of the number of true-negative quarters. Of the 29 repeatedly signaled EC-positive quarters, 14 had subclinical mastitis. Of the 549 EC-negative quarters, 540 were healthy (Table 7). The predictive value of an EC signal (48%) improved, and the predictive value of no EC signal (98%) remained high. Before and after sensor adaptation, predictive values of a positive test was 36 and 56%, and predictive value of a negative test was 98% for both periods. After extrapolation, sensitivity and specificity for the total period were 13% (3 to
TABLE 7. Subclinical mastitis status, as defined by results of bacteriological culturing and SCc. A quarter was positive for electrical conductivity (EC) if more than one signal was given within a 14·d period and EC negative with only one or no EC signal. Subclinical Healthy EC Positive, repeated signals EC Negative
14 9
15 540
Total 29 549
1137
OUR INDUSTRY TODAY
24%) and 99.7% (99.6 to 99.9%), respectively. Sensitivity was 10% (0 to 24%) before sensor adaptation and 16% (0 to 32%) after, and specificity was 99.7% in both periods.
probability of a quarter being negative in this population.
DISCUSSION
The EC signals of the prototype system were developed to detect severe subclinical as well as clinical mastitis cases (14). Because of the original development goal, high threshold values for SCC and culturing were chosen in our subclinical mastitis definition. The high number of false positives could be a direct effect of the high thresholds in the subclinical mastitis definition. Reclassification of quarters with a SCC >200 x lcPcells/ml as subclinical reduced the number of false positives and increased predictive value of a positive test (from 35 to 60%). However, predictive value of a negative test decreased (from 99 to 87%) because of lowered sensitivity and increased prevalence. Because of the high thresholds, 5 ECnegative quarters were defined as healthy, although major pathogens were present and SCC was elevated in 3 quarters (Table 8). I n the analysis, these 5 quarters appeared to be true negatives, but false negatives would have been more accurate. An expected effect of the high thresholds was a relatively high sensitivity, because the subclinical defined cases were clearly abnormal. Sensitivity was not high in our study, possibly partly because of our study design. Grouping of the sampled quarters into subclinical and healthy quarters based on SCC only (Table 3) resulted in fewer false-positive
Use for Subclinical Mastitis Detection
The predictive values of the system are the parameters that a future user needs to know to manage a herd with the system. Predictive values are affected by sensitivity, specificity, and prevalence (15). The prevalence of subclinical mastitis in this herd was estimated by the number of subclinical quarters as a fraction of the total number of quarters. The estimated quarter prevalence was about 1%. As a result of high thresholds for bacteriological culturing and SCC in our subclinical mastitis definition, the estimated quarter prevalence was very low. However, the bulk tank SCC of this herd and the SCC per cow were low in comparison with the Dutch average values (29). The low prevalence was also supported by other samplings of the herd during the study. With the high specificity, the system could be used diagnostically. The relative importance of a high sensitivity or high predictive value of a positive test depends on the importance of missing cases or the actions taken on a positive test result. If the system is to be used as a tool for decisions on treatment of quarters, the ratio of cost to benefit of treatment of subclinical mastitis should be taken into account. Although a predictive value of a positive test of 48% (repeated signals) is not low in a population with a prevalence of 1%, the chance that subclinical mastitis is present in repeatedly EC-positive quarters is only about 50%. Additionally, subclinical quarters would not be treated as a result of the low sensitivity. Because of the low predictive value of a positive test in this herd, the system should not be used as a tool for decisions on culling. To use the system as a screening test, sensitivity must be high so that the number of false negatives is low. Even with a high predictive value of a negative test (99%), false-negative quarters remained as a result of the low sensitivity. The high predictive value of a negative test was only slightly higher than the prior
Effects of Subclinical Mastitis Definition on Evaluation
TABLE 8. Bacteriological results of 5 nonsignaled quartees that had see <500 x l03cellslmJ. The 5 quarters were classified as not being subclinical and appeared as true negatives in the study.
see l
Mp2
484 416
Streptococcus dysgalactiae Staphylococcus aureus Staph. aureus Staph. aureus Staph. aureus
Duplicate samples - - (cfulmJ) - -
353 78
35 IX
1200 1400 250 1500 300
800 1250 300 125 200
l03cellslmJ.
2Major pathogens. Journal of Dairy Science Vol. 77, No.4, 1994
1138
LANSBERGEN ET AL.
and false-negative quarters than grouping based on culturing only (Table 2). This result contrasts with findings of Nielen et al. (20), in which subclinical mastitis as defined by bacteriological culturing resulted in higher sensitivity than with subclinical mastitis defined by SCC. However, numbers of cultured positive quarters were very low in the current study. The correlations between EC and In SCC confirmed findings of previous studies (3, 4, 13, 28). Effects of Study Design on Evaluation
Repeated Measurements. The impact of repeated measurements on the predictive values was small, because predictive values only slightly changed when single observations per quarter were used. Therefore, quarters were consistently classified within EC categories. However, for sensitivity and specificity estimation, independent observations were assumed, and repeated measurements could have created a bias. Because false negatives remained present when only single observations were used, the low sensitivity with repeated measurements was not expected to change to a very high sensitivity. Comparison at a Given Moment. The shedding of the bacteria, the rise in SCC, and the rise in EC might be processes with different time frames, making a comparison at a given moment difficult. In addition, subclinical mastitis was often present for several weeks (7).
Sears et al. (27) found that the shedding of Staph. aureus from the mammary gland was not constant. For accurate diagnosis of subclinical quarters with bacteriological culturing, use of consecutive samples is advisable. The samples in our study that were positive for EC could have been taken at the very moment that minimal or no pathogens were shedded by the quarter, causing false positives. To consider the dynamics of the inflammation, EC-positive quarters were sampled again 1 wk later. Of the 34 resampled, 7 changed status, mostly because of a change in SCc. Three of 4 resampled quarters infected with Staph. aureus changed between bacteriologically negative and positive. Bacteriological results were also reversed in a quarter with a Strep. uberis infection. This Journal of Dairy Science Vol. 77, No.4, 1994
quarter was sampled as an EC-negative quarter and classified as a false negative. When the quarter was sampled as EC positive 24 h later, Strep. uberis was still present in both duplicate milk samples, but below the threshold of 500 cfu/ml. This second sample was not included in the study population because of the retrospectively defined 2-wk waiting period for EC-negatives that were subclinical but would have been false positive. An example of long duration of subclinical infection was a quarter that was consistently infected with Strep. dysgalactiae. This quarter was infected during the entire study and was only signaled twice. The quarter was erroneously not sampled as EC positive but was sampled as EC negative seven times. Extrapolation. In our study, the predictive value of a positive test was not influenced by the extrapolation procedure. The predictive value of a negative test was based on a random sample but was not influenced by the extrapolation. Because of the low numbers, the effect of the extrapolation on the precision of sensitivity estimation was large. This difference was reflected in the large confidence intervals. The effect on specificity estimation was smaller because of the much larger numbers of healthy quarters. Effect of Sensor Adaptation on Evaluation
The adaptation of the 3 milking claw sensors was to reduce further the mixing of milk from different quarters at sensor level. When milk from a high EC quarter mixes with milk of other quarters during measuring, the difference in EC between the quarters is underestimated. Two effects on test performance can be the result: 1) false signals in healthy quarters when a large rise of EC occurs in a high EC quarter or 2) no signal in the high EC quarter because the difference with the reference quarter is diminished by the mixing. Test characteristics improved after sensor adaptation, thereby retrospectively justifying the adaptation. However, sensitivity remained low. The 6 false positive signals for cows with a clinical mastitis quarter occurred before sensor adaptation. When the 6 false-positive signals were deleted from the data, predictive value of a positive test changed from 19 to
1139
OUR INDUSTRY TODAY
25% before sensor adaptation and from 35 to 39% over the entire study. Prior Evaluation
During development, the system was evaluated by comparison of EC signals with SCC and bacteriological culturing (14). Every 2 mo during I yr, quarter samples of all lactating cows were taken. Milk samples were grouped into health categories according to International Dairy Federation (9) definitions. If latent mastitis and secretion disturbance were not taken into account, 14 of 26 subclinical and 192 of 200 healthy cows were identified correctly by the system. Sensitivity was 53%, and specificity was 96%. Those results cannot be compared easily with those of the present study. Definitions of subclinical mastitis differed. Furthermore, the conditions of an EC signal were not completely comparable. Subclinical mastitis and EC signals were not compared at the same moment. Prevalence of subclinical mastitis was higher because data were analyzed by cow. However, both studies agree on high specificity, which coincides with other reports (20). The difference in sensitivity estimation was probably caused by the analysis by cow, combined with a longer period around sampling during which a positive EC signal was considered to be a true positive. In the present study, bias from repeated measurements might have underestimated sensitivity. Future Developments
Importance of Attention. Different signals for the user should be considered for cows suspected of having subclinical versus clinical mastitis. For subclinical cases, normally no direct action is taken under present management circumstances. For clinical cases, early and rapid detection is preferable because treatment should not be delayed. In the studied prototype, the two conditions causing an EC signal were developed for subclinical and clinical cases, respectively, and this information should be available to the user. Cows in this study that were signaled in all 4 quarters were in estrus or had a clinically mastitic quarter. Prevention of the mixing of milk from different quarters at sensor level
should remove influence of clinical quarters on other quarter readings. Effects of estrus were not based on mixing of quarter milk. The signals were caused by the second condition and should be expected to occur incidentally, especially when a cow has a raised body temperature during estrus. Subclinical Detection. Based on our results, repeated QMEC signals may offer good prospects for detection of subclinical mastitis (fable 7). Repeated signals in the same quarter within a certain time frame might be the best EC predictor for a persistent subclinical infection. However, the conditions in the studied prototype would not signal a quarter with a steadily high QMEC if the actual QMEC were <20% over the reference quarter during two milkings. Only relative large fluctuations in QMEC compared with those of the reference quarter would cause a signal. Signaling of quarters with steadily high QMEC as subclinical quarters might be advisable. Study Design. For detection of subclinical mastitis, a comparison of EC, SCC, and bacteriological culturing at the same moment might not be the most accurate comparison. To obtain sufficient data, samples based on EC signals could be taken during short periods in different herds. To consider the prior and future subclinical mastitis status of the quarters, frequent sampling of quarters before and after the EC testing period is important. Records for herds with high prevalence of subclinical mastitis should be included to improve sensitivity estimation. Sampling all quarters of a cow to obtain interquarter ratios for SCC may be advisable (10). CONCLUSIONS
The overall conclusion based on the study in this herd is that the efficacy was low for the prototype on-line EC system as a screening or diagnostic test for subclinical mastitis. The low prevalence of subclinical mastitis cases, in combination with repeated measurements and the extrapolation procedure, might have caused a low sensitivity estimation. ACKNOWLEDGMENTS
The authors gratefully acknowledge the help of the following persons: l.P.A. Peek, Journal of Dairy Science Vol. 77. No.4. 1994
1140
LANSBERGEN ET AL.
Th.A.M. Verkerk, and other personnel of "de Bunzing" for taking all of the milk samples and R. T. van Zonneveld for providing additional data for the analysis. This research has been funded by the Dutch Foundation for Knowledge-Based Systems (SKBS).
REFERENCES I Batra, T. R., and A. J. McAllister. 1984. A comparison of mastitis detection methods in dairy cattle. Can. J. Anim. Sci. 64:305. 2 Chamings, R. 1., G. Murray, and 1. M. Booth. 1984. Use of conductivity meter for the detection of subclinical mastitis. Vet. Rec. 114:243. 3 Emanuelson, U., T. Olsson, O. Holmberg, M. Hageltorn, T. Mattila, L. Nelson, and G. AstrOm. 1987. Comparison of some screening tests for detecting mastitis. J. Dairy Sci. 70:880. 4 Fernando, R. S., S. L. Spahr, and E. H. Jaster. 1985. Comparison of electrical conductivity of milk with other indirect methods for detection of subclinical mastitis. J. Dairy Sci. 68:449. 5 Gebre-Egziabher, A., H. C. Wood, J. D. Robar, and G. Blankenagel. 1979. Evaluation of automatic mastitis detection equipment. J. Dairy Sci. 62:1108. 6 Greatix, G. R., J. C. Quale, and R. A. Coombe. 1968. The detection of abnormal milk by electrical means. J. Dairy Sci. 35:213. 7 Grommers, F. J., D. van de Geer, and C.A.A. in 't Veen. 1985. Duration of bovine intramammary infections in commercial dairy herds. Vet. Rec. 116:581. 8 Hogewerf, P. H., K. Maatje, and W. Rossing. 1992. Computed aided systems for health and reproduction control in dairy cows. Page 483 in Proc. Symp. Prospects Automatic Milking, Eur. Assoc. Anim. Prod. Publ. 65. Pudoc Sci. Publ., Wageningen, Neth. 9 International Dairy Federation, 198 I. Laboratory methods for use in mastitis work. Int. Dairy Fed. Doc. 132., Int. Dairy Fed., Brussels, Belgium. 10 Jensen, N. E., and K. Knudsen. 1991. Interquarter comparison of markers of subclinical mastitis: somatic cell count, electrical conductivity, N-acetyl-j3glucosaminidase and antitrypsin. J. Dairy Res. 58:389. II Kitchen, B. J. 1981. Review of the progress of dairy science: bovine mastitis: milk compositional changes and related diagnostic tests. J. Dairy Res. 48:167. 12 Linzell, J. L., and M. Peaker. 1975. Efficacy of the measurement of the electrical conductivity of milk for the detection of subclinical mastitis in cows: detection of infected cows at a single visit. Br. Vet. 1. 131:447. 13 Little, T,W.A., C. N. Herbert, and D. Forbes. 1968. Electrical conductivity and the leucocyte count of bovine milk. Vet. Rec. 82:431. 14 Maatje, K., P.I.M. Huijsmans, W. Rossing, and P. H. Hogewerf. 1992. The efficacy of in-line measurement of quarter milk electrical conductivity, milk yield and
Journal of Dairy Science Vol. 77, No.4, 1994
milk temperature for the detection of clinical and subclinical mastitis. Livest. Prod. Sci. 30:239. 15 Martin, S. W. 1984. Estimating disease prevalence and the interpretation of screening test results. Prevo Vet. Med. 2:463. 16 Martin, S. W., A. H. Meek, and P. Willeberg. 1987. Veterinary Epidemiology: Principles and Methods. Iowa State Univ. Press, Ames. 17 Mijnen, E., FRJ. Jaartsveld, G.A.A. Albers, M'w.A. Verstegen, and M.J.M. Tielen. 1983. The value of cell count, lactose content, pH and conductivity of milk for mastitis detection in individual cows. Neth. Milk Dairy 1. 37:65. 18 Mood, M. A., F. A. Graybill, and D. C. Boes. 1974. Introduction to the Theory of Statistics. McGraw-Hili Book Co., New York, NY. 19 National Mastitis Council. 1987. Laboratory and Field Handbook on Bovine Mastitis. W. D. Hoard & Sons Co., Fort Atkinson, WI. 20 Nielen, M., H. Deluyker, Y. H. Schukken, and A. Brand. 1992. Electrical conductivity of milk: measurement, modifiers, and meta analysis of mastitis detection performance. 1. Dairy Sci. 75:606. 21 Okigbo, L. M., M. A. Sheliah, G. H. Richardson, C. A. Ernstrom, R. 1. Brown, and E. L. Tippetts. 1984. Portable conductivity meter for detecting abnormal milk. J. Dairy Sci. 67:1510. 22 Oliver, S. P., S. H. King, M. J. Lewis, P. M. Torre, K. R. Matthews, and H. H. Dowlen. 1990. Efficacy of chlorhexidine as a postmilking teat disinfectant for the prevention of bovine mastitis during lactation. 1. Dairy Sci. 73:2230. 23 Oliver, S. P., S. H. King, P. M. Torre, E. P. Schull, H. H. Dowlen, M. J. Lewis, and L. M. Sordillo. 1989. Prevention of bovine mastitis by Postmilking teat disinfectant containing chlorous acid and chlorine dioxide in soluble polymer gel. J. Dairy Sci. 72:3091. 24 Peaker, M. 1978. The electrical conductivity of milk for the detection of subclinical mastitis in cows: comparison of various methods of handling conductivity data with the use of cell count and bacteriological examination. Br. Vet. 1. 134:314. 25 SAS(/!) User's Guide: Statistics, Version 6.03 Edition. 1988. SAS Inst., Inc., Cary, NC. 26 Schakenraad, M.H'w., and A. A. Dijkhuizen. 1990. Economic losses due to bovine mastitis in Dutch dairy herds. Neth. J. Agric. Sci. 38:89. 27 Sears, P. M., B. S. Smith, P. B. English, P. S. Herer, and R. N. Gonzalez. 1990. Shedding pattern of Staphylococcus aureus from bovine intramammary infections. J. Dairy Sci. 73:2785. 28 Sheldrake, R. F., G. D. McGregor, and R.I.T. Hoare. 1983. Somatic cell count, electrical conductivity, and serum albumin concentration for detecting bovine mastitis. J. Dairy Sci. 66:548. 29 Vecht, U., H. 1. Wisselink, and P. R. Defize. 1987. Verslag landelijke steekproef mastitis 1985/1986. Rapport Centraal Diergeneeskundig Instituut, Lelystad, Neth.