Predict hourly patient discharge probability in intensive care units using data mining

Abstract

The length of stay (LOS) is an important metric to manage hospital units since a correct prevision of the LOS can contribute to reduce costs and optimize resources. This metric become more fundamental in intensive care units (ICU) where controlling patient condition and predict clinical events is very di cult. A set of experiences was made using data mining techniques in order to predict something more ambitious than LOS. Using the data provided by INTCare system it was possible to induce models with a very good sensitivity (95%) in order to predict the probability of a patient be discharged in the next hour. The results achieved also allow for predicting the bed occupancy rate in ICU for the next hour. The work done represents a novelty in this area and contributes to improve the decision making process providing new knowledge in real time.