Development of an End-of-Life Aligned Symptom Expression Lexicon (EASEL) for Natural Language Processing in Hospice Care

Effective symptom monitoring in hospice care relies on a combination of both standardized assessments, such as the Edmonton Symptom Assessment System (ESAS), and narrative documentation completed by interprofessional care teams. However, the majority of hospice care occurs in home settings where familial caregivers describe patient symptoms through informal, observational language. Standardized medical terminologies, which are primarily optimized for acute care settings, would not capture these colloquial expressions. This gap in domain-specific lexical resources limits the ability to apply natural language processing (NLP) for symptom extraction and analysis in end-of-life care. In order to bridge this gap, this study aimed to develop a comprehensive symptom lexicon incorporating terminology from palliative care literature, a pre-existing lexicon derived from clinical notes, and at-home hospice care documentation. To capture terminology unique to home hospice care settings, this study obtained transcribed hospice staff notes and caregiver narratives from 204 patients who received care at a social model hospice residential care home. These transcribed records yielded 117,367 narrative entries, which were analyzed for symptom expressions. The systematic integration of these sources led to the development of the End-of-life Aligned Symptom Expression Lexicon (EASEL), a resource designed to bridge clinical terminology with caregiver language. EASEL contains 2,322 symptom terms organized into 30 symptom categories, including 867 colloquial expressions (37.3%) that represent caregiver language not typically found in standard clinical terminologies. EASEL could be a publicly available, ESAS-aligned symptom lexicon specifically designed for end-of-life care documentation. A hospice-specific lexicon capturing caregiver language could improve symptom identification and tracking in end-of-life care narratives, thereby supporting data-driven quality improvement and optimized symptom management at the end of life