Quantifying the effect of statin use in pre-diabetic phenotypes discovered through association rule mining.

Prediabetes is the most important risk factor for developing type-2 diabetes mellitus, an important and growing epidemic. Prediabetes is often associated with comorbidities including hypercholesterolemia. While statin drugs are indicated to treat hypercholesterolemia, recent reports suggest a possible increased risk of developing overt diabetes associated with the use of statins. Association rule mining is a data mining technique capable of identifying interesting relationships between risks and treatments. However, it is limited in its ability to accurately calculate the effect of a treatment, as it does not appropriately account for bias and confounding. We propose a novel combination of propensity score matching and association rule mining to account for this bias, and find meaningful associations between a treatment and outcome for various subpopulations. We demonstrate this technique on a real diabetes data set examining the relationship between statin use and diabetes, and identify risk and protective factors previously not clearly defined.