Risk factors for subclinical intramammary infection in dairy goats in two longitudinal field studies evaluated by Bayesian logistic regression
Research output: Contribution to journal › Journal article › Research › peer-review
Standard
Risk factors for subclinical intramammary infection in dairy goats in two longitudinal field studies evaluated by Bayesian logistic regression. / Koop, Gerrit; Collar, Carol A.; Toft, Nils; Nielen, Mirjam; van Werven, Tine; Bacon, Debora; Gardner, Ian A.
In: Preventive Veterinary Medicine, Vol. 108, No. 4, 2013, p. 304-312.Research output: Contribution to journal › Journal article › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - JOUR
T1 - Risk factors for subclinical intramammary infection in dairy goats in two longitudinal field studies evaluated by Bayesian logistic regression
AU - Koop, Gerrit
AU - Collar, Carol A.
AU - Toft, Nils
AU - Nielen, Mirjam
AU - van Werven, Tine
AU - Bacon, Debora
AU - Gardner, Ian A.
PY - 2013
Y1 - 2013
N2 - Identification of risk factors for subclinical intramammary infections (IMI) in dairy goats should contribute to improved udder health. Intramammary infection may be diagnosed by bacteriological culture or by somatic cell count (SCC) of a milk sample. Both bacteriological culture and SCC are imperfect tests, particularly lacking sensitivity, which leads to misclassification and thus to biased estimates of odds ratios in risk factor studies. The objective of this study was to evaluate risk factors for the true (latent) IMI status of major pathogens in dairy goats. We used Bayesian logistic regression models that accounted for imperfect measurement of IMI by both culture and SCC. Udder half milk samples were collected from 530 Dutch and 438 California dairy goats in 10 herds on 3 occasions during lactation. Udder halves were classified as positive or negative for isolation of a major pathogen (mostly Staphylococcus aureus) on bacteriological culture and as positive or negative for SCC (cut-off of 2000 x 10(3) cells/mL). Potentially controllable risk factors (udder conformation, teat size, teat shape, teat placement, teat-end shape, teat-end callosity thickness, teat-end callosity roughness, caprine arthritis encephalitis-virus infection status, and kidding season), and uncontrollable risk factors (parity, lactation stage, milk yield, pregnancy status, and breed) were measured in the Dutch study, the Californian study or in both studies. Bayesian logistic regression models were constructed in which the true (but latent) infection status was linked to the joint test results, as functions of test sensitivity and specificity. The latent IMI status was the dependent variable in the logistic regression model with risk factors as independent variables and with random herd and goat effects. For the combined data from both studies, the culture-based estimate of apparent prevalence of major pathogens in udder halves was 2.6% (137/5220) and the estimate of the apparent prevalence of high SCC was 11.0% (581/5294). The model was able to estimate the performance characteristics of bacteriological culture and SCC together with the effect of risk factors on the true IMI status. Higher parity, late lactation and low milk yield were significantly related to higher odds of the latent IMI status. The only significant controllable risk factor was an udder base below the hocks. Lack of a perfect reference test is a common problem in veterinary epidemiology and may lead to biased estimates of odds ratios or other measures of association in risk factor studies. The approach described herein can be used to address these problems. (C) 2012 Elsevier B.V. All rights reserved.
AB - Identification of risk factors for subclinical intramammary infections (IMI) in dairy goats should contribute to improved udder health. Intramammary infection may be diagnosed by bacteriological culture or by somatic cell count (SCC) of a milk sample. Both bacteriological culture and SCC are imperfect tests, particularly lacking sensitivity, which leads to misclassification and thus to biased estimates of odds ratios in risk factor studies. The objective of this study was to evaluate risk factors for the true (latent) IMI status of major pathogens in dairy goats. We used Bayesian logistic regression models that accounted for imperfect measurement of IMI by both culture and SCC. Udder half milk samples were collected from 530 Dutch and 438 California dairy goats in 10 herds on 3 occasions during lactation. Udder halves were classified as positive or negative for isolation of a major pathogen (mostly Staphylococcus aureus) on bacteriological culture and as positive or negative for SCC (cut-off of 2000 x 10(3) cells/mL). Potentially controllable risk factors (udder conformation, teat size, teat shape, teat placement, teat-end shape, teat-end callosity thickness, teat-end callosity roughness, caprine arthritis encephalitis-virus infection status, and kidding season), and uncontrollable risk factors (parity, lactation stage, milk yield, pregnancy status, and breed) were measured in the Dutch study, the Californian study or in both studies. Bayesian logistic regression models were constructed in which the true (but latent) infection status was linked to the joint test results, as functions of test sensitivity and specificity. The latent IMI status was the dependent variable in the logistic regression model with risk factors as independent variables and with random herd and goat effects. For the combined data from both studies, the culture-based estimate of apparent prevalence of major pathogens in udder halves was 2.6% (137/5220) and the estimate of the apparent prevalence of high SCC was 11.0% (581/5294). The model was able to estimate the performance characteristics of bacteriological culture and SCC together with the effect of risk factors on the true IMI status. Higher parity, late lactation and low milk yield were significantly related to higher odds of the latent IMI status. The only significant controllable risk factor was an udder base below the hocks. Lack of a perfect reference test is a common problem in veterinary epidemiology and may lead to biased estimates of odds ratios or other measures of association in risk factor studies. The approach described herein can be used to address these problems. (C) 2012 Elsevier B.V. All rights reserved.
U2 - 10.1016/j.prevetmed.2012.11.007
DO - 10.1016/j.prevetmed.2012.11.007
M3 - Journal article
C2 - 23182030
VL - 108
SP - 304
EP - 312
JO - Preventive Veterinary Medicine
JF - Preventive Veterinary Medicine
SN - 0167-5877
IS - 4
ER -
ID: 119574799