Factors associated with incidence of type II diabetes in pre-diabetic women using Bayesian Model Averaging

Diabetes is a chronic disease which usually begins with impaired glucose tolerance. This step is known as pre-diabetes. People with pre-diabetes are at greater risk for diabetes. Typically for the variable selection, stepwise approach is used which does not take into account model uncertainties. In this study, Bayesian Model Averaging (BMA) method was used to sort out the above shortcoming.
The study population was 734 pre-diabetic women with 20 years and older participated in Tehran Lipid and Glucose Study (TLGS). In this study, the stepwise and BMA variable selection methods were employed in logistic regression. Then area under curve (AUC) for both methods was computed and compared with Delong test. All analyses was done using R version 3.1.3.
BMA selected the fasting plasma glucose, 2 hours’ blood glucose, and family history of diabetes, body mass index and aspirin use at baseline as risk factors for diabetic. In addition to these factors, stepwise method selected diastolic blood pressure, history of past 3 months’ hospitalization, thyroid drug use and education. Although the number of variables selected by BMA (5 variables) was less than that of stepwise (9 variables), AUC for the two methods was not significant.
It seems that the BMA provide better model for screening of diabetes because with selecting fewer variables, prediction ability of the model is preserved