A Graph Enhanced Label Attention Model for ICD Coding from Clinical Text

Anonymous

Abstract: Medical code assignment from clinical texts is a crucial task in the healthcare industry. Clinical texts are typically very long sequences and the number of possible labels are large, making this task quite challenging. Recent work applies deep neural network models to encode the medical notes and assign medical codes to clinical documents. Some works use effective attention mechanisms to construct label-specific document representations and show promising results. In this paper, we propose a new attention mechanism, GE-LAAT (graph enhanced label attention), which utilizes code graphs to learn robust representation vectors for medical codes and improve upon the state of the art models. Experiments on the MIMIC-III dataset are conducted to show the effectiveness of our proposed model.