Categorize readmitted patients in Intensive Medicine by means of Clustering Data Mining

Abstract

With a constant increasing in the health expenses and the aggravation of the global economic situation, managing costs and resources in healthcare is nowadays an essential point in the management of hospitals. The goal of this work is to apply clustering techniques to data collected in real-Time about readmitted patients in Intensive Care Units in order to know some possible features that affect readmissions in this area. By knowing the common characteristics of readmitted patients it will be possible helping to improve patient outcome, reduce costs and prevent future readmissions. In this study, it was followed the Stability and Workload Index for Transfer (SWIFT) combined with the results of clinical tests for substances like lactic acid, leucocytes, bilirubin, platelets and creatinine. Attributes like sex, age and identification if the patient came from the chirurgical block were also considered in the characterization of potential readmissions. In general, all the models presented very good results being the Davies-Bouldin index lower than 0.82, where the best index was 0.425.