Clustering Barotrauma Patients in ICU-A Data Mining Based Approach Using Ventilator Variables

Predicting barotrauma occurrence in intensive care patients is a difficult task. Data Mining modelling can contribute significantly to the identification of patients who will suffer barotrauma. This can be achieved by grouping patient data, considering a set of variables collected from ventilators directly related with barotrauma, and identifying similarities among them. For clustering have been considered k-means and k-medoids algortihms (Partitioning Around Medoids). The best model induced presented a Davies-Bouldin Index of 0.64. This model identifies the variables that have more similarity among the variables monitored by the ventilators and the occurrence of barotrauma.