Development of a Multivariate Statistical Model to Predict Congestive Heart Failure in Canine Mitral Valve Disease
Department of Clinical Studies, School of Veterinary Medicine, University of Pennsylvania
Philadelphia, PA, USA
The progression of canine mitral valve disease (MVD) is variable, and predicting the risk and time to onset of congestive heart failure (CHF) in asymptomatic dogs is challenging. Accurate prediction would permit intervention and ostensibly improve outcome. The purpose of this study was to retrospectively develop a logistic regression model that identifies which variables are likely to best distinguish dogs with and without CHF due to MVD, and to then evaluate the ability of these variables to predict risk of CHF in a cohort of asymptomatic dogs followed over time. To help identify critical variables, medical records of 82 dogs with MVD were retrospectively evaluated and physical examinations, thoracic radiographs, and echocardiograms were reviewed. Forty five dogs (55%) were diagnosed with CHF based on thoracic radiographs and 37 dogs (45%) were asymptomatic. The association between presence or absence of CHF and age, tricuspid regurgitation velocity, left ventricular dimension at end-diastole, left ventricular dimension at end-systole, left atrial size (LA:Ao), vertebral heart size, heart rate, and serum N-terminal pro-B-type natriuretic peptide (NT-proBNP) were examined by multiple logistic regression. NT-proBNP had the strongest association with the presence of CHF (p<0.001), and the other variables were individually added to determine their significance and effect on the model. The combination of NT-proBNP and LA:Ao yielded the best goodness-of-fit value (Hosmer-Lemeshaw GOF=0.39). Analysis of the receiver operating characteristic curve indicated that the combination of LA:Ao and NT-proBNP was highly accurate in differentiating dogs with CHF from those without CHF. The area under the curve for the ROC curve was 0.98. There was a 10% probability of CHF when LA:Ao is 1.7 and NT-proBNP is 541 pmol/L; a 45% probability when LA:Ao is 1.8 and NT-proBNP is 1086 pmol/L; and a 100% probability when LA:Ao is 2.8 and NT-proBNP is 2192 pmol/L. The developed statistical model identifies critical variables, namely NT-proBNP and LA:Ao, that should be prospectively and longitudinally examined in a cohort of dogs with asymptomatic MVD with the goal of predicting when any individual dog is going to experience the first episode of CHF.