SymptomGraph: Identifying Symptom Clusters from Narrative Clinical Notes using Graph Clustering

Area of Science:

• Computational biology
• Medical informatics
• Artificial intelligence in healthcare

Background:

• Patients with chronic diseases experience diverse symptoms impacting quality of life.
• Traditional symptom analysis relies on surveys and statistical methods, often missing nuances in clinical narratives.
• Electronic health records (EHR) contain rich, unstructured data on patient symptoms.

Purpose of the Study:

• To develop and validate SymptomGraph, a computational system for identifying symptom clusters from clinical notes.
• To analyze symptom patterns in colorectal cancer patients post-chemotherapy.
• To explore symptom variations based on comorbidities (e.g., diabetes) and patient demographics.

Main Methods:

• Utilized natural language processing (NLP) and artificial intelligence (AI) to extract symptoms from EHR clinical notes.
• Employed semantic symptom expression clustering to identify typical symptoms.
• Constructed a symptom graph based on symptom co-occurrences and applied graph clustering algorithms.

Main Results:

• SymptomGraph successfully identified distinct symptom clusters in colorectal cancer patients one year post-chemotherapy.
• Colorectal cancer patients with diabetes exhibited increased peripheral neuropathy symptoms.
• Later-stage cancer patients showed more memory loss, while younger patients had mental dysfunction related to substance abuse.

Conclusions:

• SymptomGraph is an effective tool for uncovering complex symptom clusters from narrative clinical data.
• The system can reveal disease-specific and patient-specific symptom profiles, aiding personalized medicine.
• SymptomGraph has broad applicability for analyzing symptom data in various acute and chronic diseases.