Automatic extraction and aggregation of diseases from clinical notes

Abstract

Clinical notes provide medical information about the patient’s health. The automatic extraction of this information is relevant in order to analyze patterns for grouping patients with similar characteristics. In this paper, we used MetaMap to extract diseases present in 412 discharge summaries of obesity patients. The UMLS intra-source vocabulary relationships were used to make automatic aggregation of diseases. The results showed an average of 0.81 for recall, 0.92 for precision, and 0.84 for F-score. Finally, with the diseases extracted and aggregated three sub-graphs were identified; they correspond to patients with sleep apnea, those with heart diseases, and those with communicable diseases.