Coronary artery disease risk assessment from unstructured electronic health records using text mining

Area of Science:

• Medical Informatics
• Cardiology
• Public Health

Background:

• Coronary artery disease (CAD) poses a significant health risk, often necessitating early intervention.
• Predicting CAD risk relies on identifying key factors, frequently embedded within unstructured clinical narratives.
• Diabetic patients represent a high-risk group for CAD, underscoring the need for precise risk assessment.

Purpose of the Study:

• To develop and evaluate methods for extracting Framingham risk factors from unstructured electronic health records (EHRs) using clinical text mining.
• To calculate 10-year CAD risk scores for a cohort of diabetic patients.
• To address challenges posed by missing data in EHRs for risk score calculation.

Main Methods:

• A rule-based system was developed to extract specific risk factors (age, gender, cholesterol levels, blood pressure, diabetes, smoking history) from clinical notes.
• Clinical text mining techniques were applied to unstructured EHR data.
• Imputation strategies were implemented to handle missing data points crucial for risk score calculation.

Main Results:

• The text mining system demonstrated reliable extraction of CAD risk factors.
• A significant amount of data required for Framingham risk score calculation was found to be missing.
• After imputation, analysis revealed that the majority of diabetic patients in the cohort are at moderate risk of developing CAD within 10 years.

Conclusions:

• Clinical text mining is a viable approach for extracting essential CAD risk factors from unstructured EHRs.
• Addressing missing data through systematic analysis and imputation is critical for accurate risk prediction in clinical populations.
• Diabetic patients, as a cohort, exhibit a substantial burden of moderate 10-year CAD risk, highlighting the importance of proactive management and prevention strategies.