An empirical evaluation of deep learning for ICD-9 code assignment using MIMIC-III clinical notes.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Documentation Improvement

Background:

• Accurate medical coding is crucial for hospital operations, including billing and patient history.
• Manual coding is labor-intensive, subjective, and requires specialized expertise.
• Automating medical code assignment from clinical notes presents a significant challenge.

Purpose of the Study:

• To evaluate the performance of deep learning systems for automated mapping of clinical notes to International Classification of Diseases, Ninth Revision (ICD-9) codes.
• To compare deep learning approaches against traditional machine learning methods for this task.
• To establish a baseline for future research in automated medical coding.

Main Methods:

• Utilized end-to-end deep learning models, including Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs).
• Applied algorithms to the Medical Information Mart for Intensive Care (MIMIC-III) dataset.
• Conducted extensive experiments across various algorithmic settings.

Main Results:

• Deep learning methods demonstrated superior performance compared to conventional machine learning algorithms.
• The best models achieved a 0.6957 F1-score and 0.8967 accuracy for predicting top 10 ICD-9 codes.
• Models estimated top 10 ICD-9 categories with a 0.7233 F1-score and 0.8588 accuracy, outperforming existing benchmarks.

Conclusions:

• Standard metrics were employed to assess ICD-9 code assignment performance on the MIMIC-III dataset.
• Developed evaluation tools and resources are publicly available for research.
• Deep learning offers a promising avenue for enhancing the accuracy and efficiency of medical coding.