Application of the log-linear and logistic regression models in the prediction of systemic lupus erythematosus in the dog

Abstract

This study sought to mathematically define canine systemic lupus erythematosus (SLE) by unifying diagnostic criteria proposed by others. Thirty-one cases of canine SLE were selected for modeling when 4 different published schemes agreed on the diagnosis, and 122 controls were selected when a patient's status met no scheme's criteria. The log-linear method showed an association between SLE and polyarthritis, hematologic abnormalities, renal damage, dermatologic disorders, and antinuclear antibody test response (positive). Logistic regression was then used to derive a predictive algorithm that could identify cases and controls with which all published criteria would be in accordance. The final equation correctly classified 93.5% of the affected dogs and 98.4% of the controls. It was concluded that the log-linear and logistic regression models are useful for the diagnosis of clinically similar, but distinguishable, disease states.