Title: A Unified Review of Deep Learning for Automated Medical Coding

Abstract: Automated medical coding, an essential task for healthcare operation and delivery, makes unstructured data manageable by predicting medical codes from clinical documents. Recent advances in deep learning models in natural language processing have been widely applied to this task. However, it lacks a unified view of the design of neural network architectures for medical coding. This review proposes a unified framework to provide a general understanding of the building blocks of medical coding models and summarizes recent advanced models under the proposed framework. Our unified framework decomposes medical coding into four main components, i.e., encoder modules for text feature extraction, mechanisms for building deep encoder architectures, decoder modules for transforming hidden representations into medical codes, and the usage of auxiliary information. Finally, we discuss key research challenges and future directions.