Multi-features-Based Automatic Clinical Coding for Chinese ICD-9-CM-3

ICD-9-CM Volume 3 (ICD-9-CM-3), as a subset of the ICD-9-CM, is a standard system used to classify operations and medical procedures for billing purposes. With the gradual maturity of the DRG system, the precise coding of ICD-9-CM-3 is increasingly important for both hospital revenue and patients’ health. In this paper, a method based on BERT and NER capturing Multi-Features for automatic Chinese ICD-9-CM-3 Coding (BNMF) is proposed to support doctors to make decisions so that traditional manual coding in hospitals can be replaced. The method is designed focusing on short text information from electronic medical records (EMR) and make decisions by combining the semantic features of the clinical text, the structured features of Chinese ICD-9-CM-3, and the axis knowledge of Chinese ICD-9-CM-3 text. Meanwhile, fusion of spatial and temporal features is made to better capture semantic features. The experiments demonstrate that the efficiency of our framework is higher than those of state-of-the-art methods in the field of classification of the short medical corpus.