A logistic regression modeling on the prevalence of diabetes mellitus in the North Western Part of Nigeria

Abstract

The advances in Health sciences have led to a significant production of data such as clinical information, generated from large Electronic Health Record (EHRs). Extensive research in all aspects of has led to the generation of huge amount of data. This paper focuses on the prediction of diabetes mellitus model, using a generalized linear model a special case of a logistic regression model, from the dataset, collected from the northwestern states of Nigeria. The methods used in the study are classification, logistic
regression, and confusion matrix. The datasets were partitioned into training and testing sets, the training part used in building the model while testing part used in validating the model. The best-fitted model algorithm converged with null deviance: 226.568, residual deviance: 30.287 and AIC: 42.287, variables AGE, GLU, BMI and WGT were significance at 0.05, 0.001, 0.025 and 0.025 P-values respectively. The model was predicted as follows: Forty-three (43) times the patient was actually diabetic; the model also predicted as diabetic. Fifty (50) times the patient was non-diabetic; the model also predicted as the non-diabetic. One (1) times the model was actually non-diabetic; the model predicted diabetic (Type I) error. Zero (0) time the patient was actually diabetic; the model predicted non-diabetic (Type II) error. This accounted for the model accuracy of about 98.9% at α = 0.05.