Building a semantically annotated corpus for chronic disease complications using two document types

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Public Health

Background:

• Electronic health records (EHRs) and social media (Twitter) contain valuable patient health information, including disease complications and risk factors.
• Understanding disease risk factors is crucial for prevention and managing complications, particularly for conditions like hypertension and diabetes.
• Developing effective text-mining tools for health information extraction requires high-quality annotated data.

Purpose of the Study:

• To develop the PrevComp corpus, an annotated dataset for identifying disease complications, risk factors, and prevention measures.
• To focus on the interaction between hypertension and diabetes within the biomedical domain.
• To facilitate the creation of advanced text-mining tools for extracting critical health insights.

Main Methods:

• Integrated narrative text from electronic health records (EHRs) and tweets from Twitter.
• Developed a unique annotation scheme for disease complications, risk factors, and prevention measures, guided by domain experts.
• Conducted expert-driven annotation to ensure high-quality data, achieving F-scores of 0.60 for EHRs and 0.75 for tweets.

Main Results:

• Successfully created the PrevComp corpus, a novel resource combining EHR and Twitter data specific to hypertension and diabetes.
• The corpus is annotated for disease complications, risk factors, and prevention strategies.
• Achieved high inter-annotator agreement, indicating reliable and high-quality annotations.

Conclusions:

• The PrevComp corpus is a valuable, unique resource for advancing text-mining in the biomedical domain, specifically for hypertension and diabetes research.
• This annotated corpus will support the development of tools to extract vital information for disease risk factor management and prevention.
• The integration of EHR and Twitter data offers a comprehensive approach to capturing diverse health-related narratives.