Predicting the level of metabolic control using collaborative filtering

Abstract

Collaborative Filtering is a well-approved method for prediction of consumer behaviour in marketing strategies. We adapt and evaluate this method for prognosis of the longterm metabolic control level in diabetic patients. The underlying data for the prediction were extracted from a central diabetic data pool (DPVSCIENT) [1],[2], containing longtime documentation of about 60% of all young patients with type-1 diabetes in Germany. Prediction results were successfully checked against random values and evaluated calculating sensitivity, specifity and total performance of a prognosis test. Best results were: sensitivity = 76%, specifity = 92% and total performance = 84%. This novel approach in diabetology demonstrates tracking for metabolic control and allows to predict favorable or unfavorable results, providing an objective basis to target intervention in individual patients.