Heart disease risk factors detection from electronic health records using advanced NLP and deep learning techniques

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Heart disease is a leading cause of death, necessitating improved risk factor identification for prevention and diagnosis.
• Current methods for detecting heart disease risk factors in clinical notes are often labor-intensive and incomplete.
• Automated detection of risk factors is crucial for disease progression modeling and clinical decision-making.

Purpose of the Study:

• To enhance the identification of heart disease risk factors, diagnoses, and medications in clinical notes.
• To improve upon existing methods by employing advanced stacked word embedding techniques.
• To address the limitations of previous approaches in the 2014 i2b2 (Informatics for Integrating Biology and Beyond) challenge.

Main Methods:

• Utilized stacked word embeddings, combining various embedding types for improved feature representation.
• Employed BERT and character embeddings (CHARACTER-BERT Embedding) stacking as an advanced technique.
• Applied these methods to the 2014 i2b2 heart disease risk factors challenge dataset.

Main Results:

• Achieved a significant F1 score of 93.66% using the proposed stacked embedding model.
• Demonstrated substantial improvement compared to all previously developed models for the 2014 i2b2 challenge.
• The CHARACTER-BERT Embedding stacking approach proved highly effective for risk factor detection.

Conclusions:

• Stacked word embeddings, particularly CHARACTER-BERT Embedding stacking, represent a significant advancement in detecting heart disease risk factors from clinical text.
• The developed model offers a more accurate and efficient approach to extracting critical health information from clinical narratives.
• This work contributes to better disease understanding and clinical decision support systems.