Clustering de pacientes en MIMIC-III para modelos de predicción de mortalidad hospitalaria y duración de la estancia en UCI

Abstract

[en] Healthcare data availability is growing recently due to the digitalization of clinical records. Therefore, this large amount of data is being used by researchers to improve decision-making process, resources allocation and to address several issues. The aim of this Bachelor’s thesis is to investigate if unsupervised clustering of patients could be helpful to improve predictive models performance for mortality and length of stay in the Intensive Care Unit. The data used belongs to the open source database called MIMIC-III (Medical Information Mart for Intensive Care III). Results shows that clustering prior to predictive models training improved accuracy for the most significant cluster.