COVID-19 SignSym

COVID-19 SignSym extracts and normalizes signs and symptoms of COVID-19 from electronic health records (EHRs) to produce structured, OMOP-compatible clinical concepts for research.

Key Features:

• Extraction of Clinical Concepts: Extracts signs and symptoms and eight associated attributes (Body location; Severity; Temporal expression; Subject; Condition; Uncertainty; Negation; Course) from unstructured clinical text.
• CLAMP Adaptation: Adapts the CLAMP clinical NLP framework to identify and normalize COVID-19-related clinical concepts within EHR text.
• Mapping to Standard Concepts: Maps extracted clinical concepts and attributes to standard concepts in the Observational Medical Outcomes Partnership (OMOP) common data model.
• Hybrid Methodology: Combines deep learning-based models with curated lexicons and pattern-based rules.
• Performance Evaluation: Demonstrated high performance in extracting relevant data across three external sites containing clinical notes of COVID-19 patients and on online medical dialogues related to COVID-19.
• Generalizable Workflow: Provides a workflow that can be generalized to adapt existing NLP tools to other clinical information needs rapidly.

Scientific Applications:

• Clinical study support: Provides OMOP-compatible structured data from unstructured EHRs to support clinical studies of COVID-19.
• Disease and symptom analysis: Enables analysis of disease patterns, symptomatology, and patient outcomes by supplying normalized signs and symptoms.
• Multi-institutional research: Has been applied by 16 healthcare organizations to enhance COVID-19 clinical research efforts.

Methodology:

Adapted the CLAMP framework and employed deep learning-based models augmented by curated lexicons and pattern-based rules to extract signs and symptoms and their eight attributes and map them to OMOP concepts.