Pervasive ensemble data mining models to predict organ failure and patient outcome in intensive medicine

Abstract

The number of patients admitted in Intensive care units (ICU) with organ failure is significant. This type of situation is common in Intensive Medicine. Intensive medicine is a specific area of medicine with the objective to avoid organ failure and recover patients in seriously ill conditions. This type of problems can culminate in the death of patient. In order to help the intensive medicine professionals in the moment of they take the decision, a Pervasive Intelligent Decision Support System called INTCare were developed. INTCare uses ensemble data mining to predict the probability of occurring an organ failure or patient die in the next hour. With the purpose to obtain always the better results, a measure was implemented to assess the models quality. All the transforming process and model induction are performed automatically and in real-time. The ensembles use online-learning to improve their models. In this paper the ensemble approach was explored and the results were compared at level of sensitivity, specificity, accuracy and total error. After the analysis was possible conclude that the ensembles are a too valid option to help the decision process in intensive Medicine.