JMIR Publications

ABSTRACT

Background:

Patients with chronic or severe illnesses often experience multiple inter-related symptoms that co-occur as symptom clusters (SCs). These clusters contribute to functional impairment, treatment non-adherence, and reduced quality of life. Electronic health records (EHRs), particularly unstructured data such as clinical notes, offer valuable opportunities to capture complex and dynamic symptom experiences including SCs. However, methods for extracting SCs from EHRs are poorly understood. This scoping review aimed to evaluate how previous research has identified SCs from EHR data.

Objective:

This scoping review aims to systematically identify how SCs were identified from EHR data and evaluate the methodological approaches used in this emerging area of research.

Methods:

Guided by PRISMA-ScR guidelines, a comprehensive search of five key databases was conducted for articles published from 2000–2025. Studies included patients over 18 years old and examined SCs from EHRs. Two reviewers independently screened titles, abstracts, and full-text articles for inclusion. Relevant study data were synthesized using narrative and mapping approaches.

Results:

Twenty studies were included for final review. Most were conducted in the US. EHR data studied commonly comprised structured data (e.g., diagnosis codes) and unstructured clinical notes. Natural language processing (NLP), including rule-based approaches, named entity recognition, concept mapping was primarily applied to unstructured clinical notes to extract symptom-related information from EHRs. SCs were identified mainly using exploratory factor analysis, cluster analysis, and network analysis. Most cases involved a two-step process involving the extraction of symptoms followed by symptom clustering analysis was performed. Across oncology, cardiovascular, COVID-19, and other chronic disease populations, commonly identified SCs included gastrointestinal, fatigue-related, psychoneurological, cardiopulmonary, cognitive, and pain-related patterns, many of which were associated with hospitalization, emergency department utilization, mortality, and disease progression.

Conclusions:

Real-world EHR data have potential for SC analysis. However, methodological heterogeneity, limited validation, and lack of clinical integration remain challenges. Future research should emphasize standardized workflows, longitudinal modeling, and advanced artificial intelligence methods, including large language models, to improve reproducibility and clinical translation. Clinical Trial: Not applicable