Identification of hidden patterns in clinical database through data mining techniques for the study of diabetes pathophysiology

This study aimed to identify hidden patterns for the study of diabetes pathophysiology through data mining techniques in a database constituted by features that quantify specific metabolic processes in a population of women with a previous history of gestational diabetes and thus, particularly prone to progress toward type 2 diabetes (T2DM). Interest was reserved to insulin clearance process as relevant feature for the prediction of T2DM. In particular, a systematic review has been conducted to examine the epidemiological determinants, phenomena and processes categories affecting insulin clearance. The knowledge of epidemiological determinants as well as the determinants deriving from the phenomena and processes category that affected insulin clearance was fundamental to understand this complex, not completely known process which exerts an important role in diabetes. In particular, being insulin clearance involved in the top ten features in the logistic regression (at 8th and 9th position of the ranking for first-phase insulin clearance and second-phase insulin clearance, respectively) as well as in the decision tree (at the 3rd and 6th position of the ranking for second-phase insulin clearance and first-phase insulin clearance, respectively), it resulted a relevant feature for the prediction of type 2 diabetes in a population of women with a history of gestational diabetes. The information obtained from the resultant pattern could be of interest for the diabetes pathophysiology.