GPT-Enhanced Hierarchical Deep Learning Model for Automated ICD Coding

Abstract

In healthcare, accurate communication is critical, and medical coding, especially coding using the ICD (International Classification of Diseases) standards, plays a vital role in achieving this accuracy. Traditionally, ICD coding has been a time-consuming manual process performed by trained professionals, involving the assignment of codes to patient records, such as doctor's notes. In this paper, we present an automated ICD coding approach using deep learning models and demonstrate the feasibility and effectiveness of the approach across subsets of ICD codes. The proposed method employs a fine-grained approach that individually predicts the appropriate medical code for each diagnosis. In order to utilize sufficient evidence to enhance the classification capabilities of our deep leaning models, we integrate GPT-4 to extract semantically related sentences for each diagnosis from doctor's notes. Furthermore, we introduce a hierarchical classifier to handle the large label space and complex classification inherent in the ICD coding task. This hierarchical approach decomposes the ICD coding task into smaller, more manageable subclassification tasks, thereby improving tractability and addressing the challenges posed by the high number of unique labels associated with ICD coding.