Automatic construction of rule-based ICD-9-CM coding systems.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Manual assignment of ICD-9-CM codes for radiology reports is costly, time-consuming, and error-prone due to the complexity and volume of codes.
• Existing automated systems often rely on hand-crafted rules, which are difficult to scale for thousands of codes.

Purpose of the Study:

• To investigate the automatic construction of rule sets for ICD-9-CM coding systems.
• To develop hybrid systems that combine machine learning with rule-based approaches for improved efficiency and accuracy.

Main Methods:

• Utilized datasets from the International Challenge on Classifying Clinical Free Text Using Natural Language Processing.
• Developed and evaluated machine learning models to replace laborious steps in constructing rule-based coding systems.

Main Results:

• Achieved comparable performance to purely hand-crafted ICD-9-CM classifiers.
• The best model obtained a 90.26% F-measure on the training set and 88.93% on the test set.
• This performance would have ranked second in the challenge.

Conclusions:

• Machine learning models can effectively reproduce the performance of successful hand-crafted ICD-9-CM coding systems.
• Hybrid systems offer a rapid development pathway with reduced human effort.
• This approach is feasible for larger label sets and more data.