Machine Reading of Clinical Notes for Automated ICD Coding

The International Classification of Diseases (ICD) is a healthcare classification system maintained by the World Health Organization. It provides a set of diagnostic codes for classifying diseases (including a wide range of signs, symptoms and external causes of injury or disease) and surgical procedures. ICD codes are used for a number of different tasks, such as reporting health conditions and carrying out medical billing. The standard coding procedure consists in assigning one or more ICD codes to a patient's hospital visit: this operation is performed by medical coders, who read and review the clinical notes written by physicians and then assign the appropriate ICD codes according to the coding guidelines. This process can be time consuming and error-prone. With the rising popularity of electronic health records (EHRs) systems for the automated reading of the clinical notes and codes assignment have been proposed in the recent years. In the present work we review different supervised learning approaches for automated ICD coding and we propose our model, based on a convolutional neural network with attention layers, which achieves state of the art results. The experiments are performed on the publicly available MIMIC-III dataset, which contains de-identified EHRs of 58,976 patient visits at the Beth Israel Deaconess Medical Center from 2001 to 2012.