Modeling effects of diabetes risk factors using modified logistics regression

Abstract

This paper model the effects diabetes risk factors using modified logistic regression model (MLRM) since ordinary least square model (OLSM) is deficient, especially when dealing with irreversible variables. A real data was sampled from medical out-patient department (MOPD) unit, Federal Medical Centre Azare using systematic sampling tech nique. A total sampled of 407 patients’ data were used to test and compare the MLRM and the other competitive model. Gnu Regression, Econometrics and Time-series Library (GRETL) and Statistical Package for Social & Sciences (SPSS) programming bundles were utilized during the information analysis. The result reveals that MLRM is more efficient than LRM. Chi-square was employed to test for dependency between diabetes status (DBStatus) and the other suspected variables; it was shown that there is association between DBStatus and virtually all the variables incorporated in the MLRM model. The MLRM model can be used to predict the chance of an individual to be diabetic given data as input variables. Based on this model, it was observed that the reduced MLRM Model M2
 is the suitable model than M1
 that is the Model with all the variables. This paper also 
revealed that blood pressure (BP) systolic, weight, height, BP diastolic, age, and sugar 
level are statistically significant, while the other remaining variables are not statistically 
significant. We recommend for further investigation using the machine learning approach.