Taxonomy-based dissimilarity measures for profile identification in medical data

Abstract

The lists of diagnostic codes which are usually recorded in the hospitals for health management and/or costs reimbursement purposes can represent a useful source of information in the analysis of the (dis)similarity between different patients, as long as appropriate measures exist to estimate this (dis)similarity. The aim of this paper is to analyze various measures obtained by using different ways of computing the information content corresponding to entities in a taxonomy and by aggregating different types of measures. The discriminative power of these measures is evaluated by analyzing their ability to explain existing groups in data. A case study based on medical records containing lists of ICD (International Classification of Diseases) codes is presented and the proposed dissimilarity measures are used to identify prototypes in groups of patients.