CCS Coding of Discharge Diagnoses via Deep Neural Networks
Citations Over TimeTop 18% of 2017 papers
Abstract
A standard procedure in the medical domain is to code discharge diagnoses into a set of manageable categories known as the CCS codes. This is typically done by first manually coding the discharge diagnoses into the standard ICD codes and then using a one-to-one mapping between ICD and CCS codes. In this paper, we study the applicability of deep learning to perform automatic coding of discharge diagnoses into CCS codes. In particular, we build an LSTM network combined with a dense neural network that uses medically-trained word embeddings to code discharge diagnoses into single-level CCS codes. We also investigate the advantage of mapping discharge diagnoses into UMLS concepts before coding is carried out. Experimental results based on a large dataset of manually coded discharge diagnoses show that our deep-learning model outperforms the state-of-the-art automatic coding approaches and that the mapping to UMLS concepts consistently results in significant improvement in the coding accuracy.
Related Papers
- → The Tyranny of the Diagnosis Code1(2005)10 cited
- Hospital discharge diagnoses: how accurate are they and their international classification of diseases (ICD) codes?(1989)
- Computer-prompted diagnostic codes.(1995)
- Risk of Duplicate ICD Codes for Orthopedic and Injury Related Research.(2023)
- In Case of Emergency - Are ICD-10 Codes Enough?(2015)