Using natural language processing to identify symptoms in systemic mastocytosis

Area of Science:

• Medical Informatics
• Natural Language Processing
• Rare Diseases

Background:

• Systemic mastocytosis (SM) is a rare, multisystem disease with varied symptoms, often leading to delayed diagnosis.
• Unstructured clinical notes in electronic health records (EHRs) contain valuable symptom data but lack scalable extraction methods.
• Earlier recognition of SM is hindered by challenges in systematically documenting and analyzing patient symptoms.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for identifying 23 potential SM-related symptoms.
• To assess the algorithm's effectiveness in extracting symptoms from unstructured EHR data for SM and comparison groups.
• To evaluate the potential of NLP in improving the early detection of rare diseases like SM.

Main Methods:

• A retrospective study utilized EHR data from Kaiser Permanente Southern California (2008-2023).
• A rule-based NLP algorithm was developed using annotated training data and validated on double-annotated notes.
• The algorithm was applied to a large cohort of electronic health records (EHRs) from SM, chronic spontaneous urticaria (CSU), and control patients.

Main Results:

• The NLP algorithm demonstrated high precision and recall (over 90%) for most SM-related symptoms.
• Lower precision was noted for "epigastric or abdominal bloating" and "swelling" due to ambiguous references.
• Symptom documentation varied across groups, with SM patients showing more gastrointestinal and systemic symptoms, and CSU patients more cutaneous symptoms.

Conclusions:

• Rule-based NLP is feasible for identifying SM symptoms from unstructured EHR narratives with strong performance.
• The algorithm successfully extracted meaningful patterns in symptom documentation, supporting its utility for rare disease research.
• NLP applications can enhance early recognition of rare diseases and inform data-driven healthcare strategies.