A network-based analysis of medical information extracted from electronic medical records

Clinical notes constitute a rich source of medical information that could be useful in identifying graphs of patients with similar characteristics. Network-based approaches permit to visualize associations between medical entities and to infer medical knowledge. This paper aims to apply such an approach to identify the graphs of obesity patients as well as relationships between diseases and treatments extracted from discharge summaries. Two experiments were designed. In the first experiment, a 412-node graph representing patients was constructed to identify patient groups. Graphs were obtained with the modularity function. In the second, some bipartite graphs were constructed to identify disease-treatment relationships from patient graphs. The results were congruent in both experiments. Patient graphs corresponding to obese patients with diseases derived from a metabolic problem were identified; some had infectious diseases, while others had diseases derived from a mechanical problem. Furthermore, graphs of diseases and treatments related to obesity could be observed. This work identified obesity-patient graphs and relationships between diseases and treatments based on a network approach, which took into account information extracted from clinical notes.