Supervised ICD Code Assignment to Short Clinical Problem List Entries.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Data Management

Background:

• Clinical information systems frequently utilize free-text entries across diverse applications.
• Accurate and efficient coding of patient problems is crucial for healthcare data analysis and management.

Purpose of the Study:

• To evaluate the feasibility of automatically assigning International Classification of Diseases, 10th Revision (ICD-10) diagnosis codes to brief patient problem list entries (≤50 characters).
• To compare the performance of different machine learning classifiers for this automated coding task.

Main Methods:

• Utilized patient-based short problem list entries (≤50 characters).
• Employed machine learning classifiers, specifically Random Forest and AdaBoost algorithms.
• Evaluated classifier performance on both unbalanced and balanced datasets using F-measure metrics.

Main Results:

• Random Forest achieved an F-measure of 0.87 on unbalanced data and 0.88 on balanced data.
• AdaBoost achieved an F-measure of 0.85 on unbalanced data and 0.94 on balanced data.
• Both classifiers demonstrated strong performance, particularly AdaBoost on the balanced dataset.

Conclusions:

• Automated post-assignment of ICD-10 codes to short patient problem lists is feasible with high accuracy.
• Machine learning, especially AdaBoost, shows significant promise for improving the efficiency and accuracy of clinical coding processes.
• This approach can enhance the utility of clinical information systems for various healthcare applications.