Extracting Angina Symptoms from Clinical Notes Using Pre-Trained Transformer Architectures

Area of Science:

• Cardiology
• Medical Informatics
• Natural Language Processing

Background:

• Anginal symptoms indicate heightened cardiac risk and necessitate adjustments in cardiovascular management.
• Automating the identification of these symptoms from clinical notes can improve efficiency and patient care.
• Current methods for symptom extraction from electronic health records are often manual and time-consuming.

Purpose of the Study:

• To evaluate the efficacy of pre-trained transformer architectures in automatically detecting and characterizing anginal symptoms.
• To assess the model's performance in identifying specific symptoms such as chest pain and shortness of breath from primary care physician notes.

Main Methods:

• Utilized a pre-trained transformer architecture on 459 primary care physician notes.
• Focused on the 'history of present illness' section for symptom extraction.
• Annotated notes for positive and negative mentions of chest pain, shortness of breath, and related characteristics.

Main Results:

• Achieved high sensitivity and specificity in detecting chest pain or discomfort, substernal chest pain, shortness of breath, and dyspnea on exertion.
• Demonstrated promising performance in identifying key anginal symptoms from clinical text.
• Limited performance was observed in extracting provocation and palliation factors due to small sample sizes.

Conclusions:

• Pre-trained transformer architectures show significant potential for automating the extraction of anginal symptoms from clinical documentation.
• This approach can aid in faster risk stratification and timely cardiovascular management adjustments.
• Further research with larger datasets is warranted to refine models for more nuanced symptom characterization.