Comparison of artificial neural network predictive power with multiple logistic regressions to determine patients with and without diabetic retinopathy

Diabetes mellitus is a high prevalent disease among the population, and if not controlled, it causes complications and irreparable damage to the eye and cause blindness. This study goal is to investigate the predictive power of multiple logistic regression model and the Artificial Neural Network Multi-layer Perceptron (MLP) in determining patients with and without diabetic retinopathy. Of 16,000 diabetic cases from Kermanshah diabetic center a sample including 150 cases and 150 controls were enrolled. Demographic data, BMI, FBS, Hba1c, blood pressure, cholesterol (TC) and duration of disease, smoking status, and age of patient, and health records were collected into two separate checklists. For identifying risk factors, and artificial neural network models multiple logistic regression was fitted to the data and the Rock charts was used to compare the predictive power of the models. Also sensitivity and specificity were analyzed together with the standards of both models (ROC curve, sensitivity and specificity) and superior model was introduced. The predictive power of logistic regression and MLP were 73.0 and 83.0, respectively. The MLP model features (80%) and sensitivity (85%) were higher. Variables of, BMI (p<0.0001), age (p<0.0001) duration of diabetes (p<0.0001) in the logistic regression model, the variables of age, FBS, duration of diabetes, BMI, smoking status, TC according to the Wrapper, the predictive power of 83% in MLP were significant. In this study, the MLP model showed more power to identify diabetic retinopathy patients from those without retinopathy. Thus, in communities that case and control groups have high affinity (like this study), discovering the difference needs a more powerful method such as artificial neural network MLP. This method is recommended for medical research.