Automated ICD-10 code assignment of nonstandard diagnoses via a two-stage framework

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Health Information Management

Background:

• Electronic medical records (EMRs) contain valuable clinical data but suffer from nonstandard diagnoses due to physician variability.
• Manual ICD-10 coding is inefficient, costly, and time-consuming, hindering medical study quality.
• The large number of ICD-10 subcategory codes (23,000) presents a significant challenge for automated assignment.

Purpose of the Study:

• To develop and evaluate a novel two-stage framework for automated ICD-10 code assignment from nonstandard EMR diagnoses.
• To address the challenge of training data sparsity in medical coding datasets.
• To improve the efficiency and accuracy of automated diagnostic coding.

Main Methods:

• A two-stage framework leveraging the hierarchical structure of ICD-10 codes was proposed.
• The framework first examines broad category codes (approx. 1900) and then searches relevant subcategory codes.
• Additional supervised information was introduced to mitigate training data sparsity.

Main Results:

• The proposed framework significantly reduces the number of codes to examine compared to methods searching all 23,000 subcategory codes.
• Experimental results demonstrate improved performance in automated ICD-10 code assignment.
• The approach effectively handles nonstandard diagnoses present in EMRs.

Conclusions:

• The developed two-stage framework offers a more efficient and effective solution for automated ICD-10 code assignment.
• This method enhances the quality and reliability of medical studies by standardizing diagnoses from EMRs.
• The framework provides a promising approach to overcome challenges in automated medical coding.