Prognostic indicators in a Danish hospital-based population of colic horses.

Abstract

A prospective survey of 528 colic horses, referred to the Large Animal Hospital at the Royal Veterinary and Agricultural University of Copenhagen, Denmark, during the period August 1994 to December 1998, was undertaken to develop a predictive model for application in the clinical assessment of prognosis. In the multivariate logistic regression analysis, 357 colic cases were used in the elaboration of a simple clinical-practical model consisting of degree of pain, packed cell volume, capillary refill time and rectal temperature. The relationship between rectal temperature and outcome (survival/death) has been regarded as linear. It has also been reported to be nonsignificant. The present study suggests a strong U-shaped relation, which is easily transformed into a linear association and readily interpretable in the clinical situation if treated as a deviation from 38 degrees C. Several other clinical and laboratory variables were strongly related to outcome in the bivariate analysis. The changes in sensitivity and specificity of the multivariate model, when applied as a prognostic test, were presented with changing cut-off values. The cut-off value is the level, of predicted probability of death, at which the clinical decision to treat or subject to euthanasia is taken. The predictive performance of the model was further illustrated using a mortality of 19%. The optimal accurate classification for both survival and death was 87%, which was attained when a cut-off value of 86% was selected. To minimise the number of horses unnecessarily subjected to euthanasia, the cut-off value was increased. However, this simultaneously increased the number of misclassified survivors, i.e. the number of horses that would die despite treatment. Outcome was especially poorly predicted in 4 horses, as was indicated by extreme deviance residuals. In 2 of these horses the large residuals were attributable to sudden and severe changes in the course of disease. All 4 horses had changes in variable values towards abnormality prior to death. Repeated measurements are therefore suggested in order to increase the test performance in general. The prerequisites of clinical application of prognostic models are critically discussed. The main findings of the present study indicate that degree of pain, packed cell volume, capillary refill time and temperature deviation from 38 degrees C, used in a logistic regression model, offer a strong model for clinical assessment of prognosis.