Identifying risk factors for heart disease over time: Overview of 2014 i2b2/UTHealth shared task Track 2

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Analysis

Background:

• Diabetic patients have multiple risk factors for Coronary Artery Disease (CAD).
• Longitudinal medical records contain valuable information on disease risk factors and progression.
• Accurate identification of these factors is crucial for patient management.

Purpose of the Study:

• To evaluate automated systems for identifying CAD risk factors in diabetic patients' longitudinal records.
• To assess the capability of systems in tracking the presence and progression of these risk factors over time.
• To benchmark natural language processing (NLP) approaches in clinical risk factor identification.

Main Methods:

• Utilized the 2014 i2b2/UTHealth NLP shared task dataset.
• Focused on identifying risk factors like hypertension, hyperlipidemia, obesity, smoking, and family history.
• Employed various NLP techniques including lexicons, rules, and machine learning (Support Vector Machines).

Main Results:

• Twenty teams participated, submitting 49 system runs.
• Top systems achieved high F1 scores, with six exceeding 0.90.
• Successful systems combined multiple NLP strategies for optimal performance.

Conclusions:

• Automated NLP systems demonstrate high accuracy in identifying CAD risk factors.
• Tracking the progression of these risk factors over time is feasible with current technology.
• The findings support the integration of NLP for enhanced clinical risk assessment and monitoring.