Comparison of artificial neural network and logistic regression models for prediction of diabetes type II with complications

Type II diabetes mellitus (T2DM) is a growing health concern in the United States, affecting almost 30 million individuals, and currently ranking as the 7th leading cause of mortality. In addition, T2DM is associated with multi-systemic complications that contribute to both early mortality and decreased quality of life. Individuals with T2DM can be diagnosed by blood glucose tests, and previous studies have demonstrated increased risk factors for T2DM development, including obesity, particular ethnicities, personal history of polycystic ovary disease, or a family history of T2DM. The present study aimed to find the connection between T2DM complications including ketoacidosis, hyperosmolarity, renal manifestations, ophthalmic manifestations, neurological manifestations, and peripheral circulatory diseases with the most widespread risk factors, including gender, race, family history of diabetes, obesity, smoking, alcohol-related disorders, hyperlipidemia, hypertension, hypercholesterolemia, asthma, Vitamin D deficiency, and age. The strongest association was found between increasing age and peripheral circulatory disorders, with those over 65 years showing the highest correlation (OR=22.081). Strong connections were also found between Asian/Pacific Islanders and age >65 with renal complications, as well as between alcohol abuse and hyperglyceridemia with ketoacidosis (OR=3.303 and 2.992 respectively). This study also tested two predictive models, Logistic Regression and Neural Network (ANN), in modeling T2DM with complications. Classification methods tests showed that three complications – renal manifestations, neurological manifestations, and ketoacidosis – were better predicted by these models than the other complications, and that both models performed very similarly in both sensitivity and specificity. This study demonstrates that specific combinations of risk factors can predict increased probabilities of specific complications in T2DM patients, and that a neural network analysis model can predict these relationships as accurately and with the same sensitivity as a standard linear regression model.Ph.D.Includes bibliographical referencesby Muteb Hamed Saleh Alshammar