A Hybrid Clustering Prediction for Type 1 Diabetes Aid: Towards Decision Support Systems Based upon Scenario Profile Analysis

Type 1 diabetic patients present large variability reducing dramatically the ability to achieve adequate blood glucose control. Lifestyle and physiological factors highly impact their treatments which require some predictive capabilities to prevent as many adverse events as possible. The identification of the characteristic profiles of these patients would lead to an improvement in the accuracy of treatments in the specific scenarios that they face. This study presents the proof of concept of a clinical decision support system combining a classifier of glycemic profiles and a predictor of blood glucose levels. The system is aimed to identify data profiles according to a given scenario and to generate prediction models based of these scenarios to forecast blood glucose levels. The experiments were conducted in silico, by simulating different characteristics profiles of type 1 diabetic patients, in order to prove the feasibility of the approach. Clinical decision support systems based on this methodology could assist type 1 diabetic patients in their treatments according to patient's conditions and to the situations faced by patients.