CORRELATION USING GENERALISED ESTIMATING EQUATIONS. LESLIE, K.E.a, BATEMAN, K.a, BARNUM, D.a, SCHUKKEN, Y.H.ab. Elimination of S.aureus ...
The Kenya Veterinarian 18(2) (1994) 149-151 - 4.3 149
EFFECT ESTIMATION OF REPEATED VERSUS SINGLE DRY COW TREATMENT, CORRECTING FOR INTRACOW AND INTRAHERD CORRELATION USING GENERALISED ESTIMATING EQUATIONS.
LESLIE, K.E. a, BATEMAN, K. a , BARNUM, D. a, SCHUKKEN, Y.H.ab
Elimination of S.aureus is most frequently attempted during the dry period. Treatment in lactation is considered less efficacious and economically not feasible. Treatment success at dry off has been reported around 60-70% (Sol et al., 1992, Pankey et a1.,1982). Treatment success is effected by herd effects, the age of the cow (Sol et al., 1992), the SCC at dry off (Sol et al., 1992), and dosage and treatment duration (Osteras et al., 1991) of the antibiotic. Hence, considerable intraherd and intracow correlations may be present in such data. Currently, few data exist that evaluate double dry cow treatment for its effect on enhancing the cure, and preventing new infections. The purpose of this study was to evaluate the effect of a double application of dry cow antibiotics in a randomised field trial. Special attention was paid to the correction for intraclass correlation when estimating the effect of treatment.
MATERIALS AND METHODS
The study was designed as a randomised field study. All 123 animals in 23 herds that qualified for the study had an equal probability of obtaining either a single antibiotic treatment (at dry off) or a double treatment (at dry off and 30 days before expected calving). All treatments were intramammary. The unit of randomisation was the individual cow. To account for possible confounding factors, the age, lactation month, dry period length and farm were recorded for each cow. The herds were surveyed and all S.aureus positive animals were eligible for entry into the study. At the time of dry-off, a second sample was taken. After calving, duplicate samples were obtained with approximately five days in between. Identification of microorganisms were as described in National Mastitis Council (1989). An infection was defined as present when the two subsequent samples, either before dry off or after calving, were positive for the same microorganism. In addition, a quarter was also defined positive for S.aureus when one of two samples showed growth of this microorganism, and the mean SCC was greater than 250,000 cells/ml. The probabilities of cure and of new infection were modelled using logistic regression analysis. The basic observational unit was the quarter of a cow. Subsequently, the analysis was corrected for intracow correlation using the Generalised Estimating Equations procedure according to Zeger et al. (1988). All models included fixed herd parameters to account for intraherd correlation. The initial regression model included the effect of treatment (single vs. double), SCC, age in years, lactation month at dry-off, dry period length, and a quarter code. Intraherd and intracow correlation coefficients were calculated using an ANOVA procedure, with herd and cow respectively as the classification variable. All analysis were done in SAS (1985), using Proc IML (with a special program written by M. Rezaul Karim) and Proc ANOVA.
RESULTS
Of the 492 quarters, 222 were infected before dry off; these quarters were eligible for cure during the dry period. Of the 222 quarters, 198 were infected with S. aureus, 18 with Staph spp., 4 with streptococci
a Department of Population Medicine, University of Guelph, Ontario, Canada b Department of Herd Health and Reproduction, University of Utrecht, Utrecht, The Netherlands
150 (non- agalactia) and 2 with gram negative microorganisms. The overall percentage cure was 78%, individually 76% for S. aureus, 89% for staph spp. Of the 492 quarters, 44 (8.9%) experienced a new infection, 30 with S.aureus, 6 with staph spp., and 8 with other pathogens. The intraclass correlation coefficients for cures were 0.056 and 0.29 for herds and cows, respectively. The intraclass correlation coefficients for new infections were 0.06 and 0.17 for herds and cows, respectively. This indicated a substantial within-cow correlation and a small within-herd correlation. The results of final GEE and ordinary logistic regression models are shown in Table 1. The model included an interaction term between treatment and SCC. This interaction term indicated that a double treatment was beneficial at SCC levels below approximately one million cells per ml., but showed a lower cure compared to a single treatment above this level. The probability of cure was also affected by age (higher age, lower cure), and SCC (higher SCC, lower cure).
Table 1.
Results of the final logistic regression analysis on the probability of cure. Comparison of GEE model and ordinary logistic regression Ordinary LR
GEE LR Parameter
Estimate
SE
Estimate
SE
Intercept Age (yr) SCC at dry off Treatment Treatment*SCC
2.76 -.17 -.000104 .80 -.0015
.64 .073 .00014 .68 .00055
2.73 -.17 -.000063 .82 -.0015
.56 .067 .00016 59 .00050
The results of the final GEE logistic regression model showed that a double treatment resulted in'a nonsignificant 50% reduced risk of a new infection (odds ratio 0.49, 95% CI 0.22 - 1.07). The same ordinary logistic regression model resulted in an odds ratio of 052 and a 95% CI of 0.27 - 1.00. There was no significant effect of age, SCC level, month of lactation or the dry period length on risk of a new infection.
DISCUSSION The cure rate observed in this study (76% for S.aureus) is very similar to the results from other studies using the same antibiotic treatment (Newbould, 1979). Also other intramammary dry cow products seem to have a similar efficacy (Sol et al., 1992, Pankey et al., 1982). Important intraclass correlations were present in the data. This leads in general to an underestimation of variance, and an inflation of statistical significance. An approximate 15% increase in standard deviation resulted from correction for intraclass correlation. This resulted, in some instances, in a change in statistical significance from under to over the 5% probability level. However, there was no change in the values of the estimates (see Table 1). From our data it appears that correction for intraclass correlation is rather important. Double treatment had a somewhat puzzling effect on cure rate. It appeared to perform better than a single treatment in cows with an SCC below 1 x10 6 , but performed worse in cows with a higher SCC. The surprising observation was the consistently high cure among all levels of SCC in the single treatment group. Other studies using single dry cow treatments have not observed this, and reported instead a decline in cure rate with increasing SCC (Sol et al. 1992). This reported decline with increasing SCC, and also with increasing age, was similar to the decline observed in the double treatment group in this study.
151 The rate of occurrence of new infections in this study was fairly similar to data from other studies. ,Pankey et al. (1982) observed new infections in 10.9% of quarters. The occurrence of new infections was halved in the double treated group in our study. This may be a significant finding for herds with a naturally high rate of incidence of new infections during the dry period. Such herds might benefit from a double dry cow treatment.
REFERENCES National Mastitis Council., 1989: Microbiological Procedures for Use in the Diagnosis of Bovine Mastitis. 3rd edition. Ames, Iowa Carter Press Inc. Newbould, F.H.S., 1979. The use of induced mammary infections for evaluating dry cow treatment products. Can. J. Comp. Med., 43: 430-433. Osteras, 0., Sandvik, L., Aursjo, J., Gjul, G.G., and Jorstad, A. 1991. Assessment of strategy on selective dry cow therapy for mastitis control. J. Vet. Med. B, 38: 513-522. Pankey, LW., Barker, R.M., Twomey, A., and Duirs, G., 1982. Comparative efficacy of dry-cow treatment regimes against S.aureus. N. Z. Vet. J., 30: 13-15. Sol, J. Hat-ink, J., Snoep, JJ. Schukken, Y.H., 1992. Factors effecting the result of dry cow treatment. In Proc World Cattle Congress, Minneapolis, MN, USA Zeger, S.L., Liang, K.Y. and Albert, P.S., 1988. Models for longitudinal data: a generalized estimation equation approach. Biometrics, 44: 1049-1060.