bivariate logistic regression

When it is desired to examine occurrence of two events simultaneously, it is common to use bivariate statistical models such as bivariate logistic regression. Due to the limitations of classical methods in real situations, other methods such as artificial neural networks (ANN) are concerned. The aim of this study was comparing the predictive accuracy of bivariate logistic regression and artificial neural network models in diagnosis of death occurrence and heart block in myocardial infarction patients.
Material and In this study, data was taken from a census in a cross-sectional study in which 263 patients with myocardial infarction cases who admitted to Hajar hospital heart care in 2013 to 2014. Gender, type of stroke, history of diabetes, previous history of hypertension, lipid disorders, history of heart disease, cardiac output fraction, systolic blood pressure, diastolic blood pressure, fasting and non-fasting blood sugar, cholesterol, triglycerides, low-density cholesterol, smoking, type of treatment, the troponin enzymes and insurant type were considered as explanatory variables and occurrence of death and heart block were used as dependent variables. Bivariate logistic regression and neural network model was fitted. Both models were predicted and the accuracy of them were compared. Models were fitted by MATLAB2013a and Zelig in R3.2.2.
Predictive accuracy of bivariate logistic regression model was 77.7% for the training and 78.48% for the test data. In ANN model, LM and OSS algorithms had best performance with 83.69% and 83.15% predictive accuracy for training data and 84.81% and 83.54% for testing data, respectively.
This research showed that the neural network method is more accurate than bivariate logistic regression to joint predicting the occurrence of death and heart block in patients with myocardial infarction.