Accurate and Well-Calibrated ICD Code Assignment with a Chunk-Based Classifier Attending over Diverse Label Embeddings

Anonymous

Abstract: Although the International Classification of Diseases (ICD) has been adopted worldwide, manually assigning ICD codes to clinical text is time-consuming, error-prone, and expensive, motivating the development of automated approaches. This paper describes a novel deep learning approach for ICD coding, combining several ideas from previous related work. In particular, we split long clinical documents into chunks, and use a strong Transformer-based model for processing each of the chunks independently. The resulting representations are processed with a max-pooling operation, and combined with a label embedding mechanism that explores diverse ICD code synonyms. Experiments with different splits of the MIMIC-III dataset show that the proposed approach outperforms the current state-of-the-art models in ICD coding, while also leading to properly calibrated results that can effectively inform downstream tasks such as text quantification.

Paper Type: long

Research Area: NLP Applications

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.