Prediction of the Insulin Sensitivity Index using Bayesian Networks

The insulin sensitivity index (SI) can be used in assessing the risk of developing type 2 diabetes. An intravenous study is used to determine SI using Bergmans minimal model. However, an intravenous study is time consuming and expensive and therefore not suitable for large scale epidemiological studies. In this paper we learn the parameters and structure of several Bayesian networks relating measurements from an oral glucose tolerance test to the insulin sensitivity index determined from an intravenous study on the same individuals. The networks can then be used in prediction of SI from an oral glucose tolerance test instead of an intravenous study. The methodology is applied to a dataset with 187 patients. We find that the SI values from this study are highly correlated to the SI values determined from the intravenous study.