A decision support system for ICU readmissions prevention

Abstract

This paper presents a proposal for a decision support system for the prevention of Intensive Care Unit readmissions, that is being built based on data from the MIMIC II database. The proposed system fuses the results of two distinct classification approaches, one based on available numerical data, and the other based on medical text annotations. The former uses neural fuzzy models and fuzzy modeling in order to provide interpretable classification results. The latter is based on the creation of top-k edited word lists and fuzzy fingerprints. Model fusion is accomplished using a two-tiered fuzzy rule based architecture.