Every year life span is increasing and simultaneously the proportion of people with one or more chronic diseases. This paper presents an implementation of a prototype with a decision tree to detect dangerous health conditions for Diabetes Type 1 and Diabetes Type 2. With the information we collect from Personal Health Devices and data from the Active-Assisted-Living environment, we are in the position to customize thresholds and to get individual results. With the help of a modified Glucose-Insulin Model (based on the minimal model of Stolwijk & Hardy) we predicted the future glucose concentration of the patient. We validated our model with an intention-to-treat pilot study including 8 subjects and obtained a significantly better (p < 2.2⁻¹⁶) result than the original model.