Breast Cancer Symptom Clusters Derived from Social Media and Research Study Data Using Improved K-Medoid Clustering

Area of Science:

• Oncology
• Data Science
• Patient-Reported Outcomes

Background:

• Cancer patients, particularly those with breast cancer, often experience multiple concurrent symptoms during treatment.
• These co-occurring symptoms, or symptom clusters, can significantly impact patient functioning, mental health, and quality of life.
• Traditional research methods for identifying symptom clusters may not capture the full spectrum of patient experiences.

Purpose of the Study:

• To identify and characterize symptom clusters in breast cancer survivors.
• To compare symptom cluster patterns derived from social media data versus traditional research study data.
• To evaluate an improved K-Medoid clustering algorithm for analyzing these datasets.

Main Methods:

• Collected 50,426 social media messages from Medhelp.com and 653 questionnaires from a research study.
• Applied an improved K-Medoid clustering algorithm to both social media and clinical datasets.
• Analyzed symptom networks and clustering performance using average silhouette width (ASW).

Main Results:

• Identified consistent symptom clusters across social media and clinical data, including gastrointestinal, menopausal, mood-related, cognitive, and pain symptoms.
• Social media data provided a sparser network, facilitating easier partitioning compared to clinical data.
• The revised K-Medoid clustering improved performance by avoiding local optima and maintaining non-decreasing ASW.

Conclusions:

• An integrative approach using both social media and clinical data offers a comprehensive understanding of breast cancer symptom clusters.
• Social media data can contextualize clinical findings, while clinical data can enhance precision and recall.
• Findings support the development of targeted interventions addressing combined symptom effects.