JMIR Publications

ABSTRACT

Background:

Nowadays, International Classification of Diseases (ICD) code is widely used as the reference on medical system and billing purposes. However, classifying diseases into ICD codes still mainly relies on humans reading a large amount of written materials as the basis for coding. Coding is both laborious and time consuming. Since the conversion of ICD-9 to ICD-10, the coding task became much more complicated and deep learning and Natural Language Processing related approaches were studied to assist disease coders.

Objective:

This paper aims at constructing a deep learning model for ICD-10 coding, where the model is to automatically determine the corresponding diagnosis and procedure codes based solely on free-text medical notes to improve accuracy and reduce human effort.

Methods:

We use diagnosis records of the National Taiwan University Hospital (NTUH) as the resources and apply natural language processing techniques, including global vectors, word to vectors, embeddings from language models, bidirectional encoder representations from transformers (BERT), and single head attention recurrent neural network, on the deep neural network architecture to implement ICD-10 auto-coding. Besides, we introduce the attention mechanism into the classification model to extract the keywords from diagnoses and visualize the coding reference for training freshmen in ICD-10. Sixty discharge notes were randomly selected to exam the change on the F1-score and the coding time by coders before and after using our model.

Results:

In experiments on the medical dataset of NTUH, our predicting result could achieve F1-score of 0.715 and 0.618 on ICD-10 Clinical Modification code and Procedure Coding System code, respectively, with BERT embedding approach on Gated Recurrent Unit classification model. The well-trained models were applied on the ICD-10 web service for coding and training to ICD-10 users. With this service, coders can code with the F1-score increased from median of 0.832 to 0.922 (P < 0.05), but not decreased time consumed.

Conclusions:

The proposed model significantly improved the F1-score but not decreased time in coding by disease coders.