Deep Learning on Electronic Health Records to Improve Disease Coding Accuracy.

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Science

Background:

• Accurate patient phenotyping is crucial in biomedical informatics.
• International Classification of Diseases (ICD) codes are vital for population health and cohort discovery but exhibit coding variability.
• Limited clinical information necessitates reliable methods for ICD code assignment.

Purpose of the Study:

• To develop and validate a deep learning methodology for predicting ICD codes.
• To model the decision-making process of human coders.
• To enhance the accuracy of ICD code assignment using patient data.

Main Methods:

• Utilized deep learning models to predict ICD codes.
• Incorporated patient demographics, lab results, medications, and historical ICD codes as input features.
• Compared model predictions against human coder assignments and clinician assessments.

Main Results:

• The deep learning model achieved high accuracy in predicting ICD codes for diabetes, acute renal failure, and chronic kidney disease.
• The model demonstrated superior performance compared to human coders when assessed by a panel of clinicians.
• Analysis of discrepancies revealed the model's advantage over traditional coder-assigned ICD codes.

Conclusions:

• Deep learning offers a robust methodology for accurate ICD code prediction.
• This approach can significantly improve clinical phenotyping and cohort discovery, especially with limited data.
• The developed model outperforms human coders in assigning ICD codes, highlighting its potential clinical utility.