Machine Learning Approaches on Diagnostic Term Encoding With the ICD for Clinical Documentation.

Area of Science:

• Clinical Informatics
• Data Mining
• Machine Learning

Background:

• Electronic health records (EHRs) contain valuable clinical information.
• Manual encoding of diagnostic terms (DTs) using International Classification of Diseases (ICD) is time-consuming and expert-dependent.
• Automating DT encoding can enhance data retrieval and analysis in healthcare.

Purpose of the Study:

• To explore the application of text mining for aiding the encoding of diagnostic terms (DTs).
• To develop a robust machine learning (ML) model for high-dimensional ICD code classification.
• To improve the efficiency and accuracy of clinical documentation.

Main Methods:

• Utilized data mining and machine learning techniques for DT classification.
• Developed a robust instance representation method to enhance ML model performance.
• Trained and evaluated a classification system for over 1500 ICD codes.

Main Results:

• Achieved 92% precision for primary class (main disease) ICD code identification.
• Reached 88% precision for fully specified class (main disease with modifiers) ICD code identification.
• Reported 91.2% accuracy on a test set, validated by clinical experts.

Conclusions:

• The proposed text mining system significantly aids in the accurate encoding of diagnostic terms.
• The system is simple, portable, and highly useful for documentation and pharmacosurveillance.
• Publicly released software aims to support the clinical and research communities.