Cancer symptom clusters: an exploratory analysis of eight statistical techniques

Area of Science:

• Oncology
• Biostatistics
• Symptom Science

Background:

• Identifying symptom clusters (SC) is crucial in oncology, but varied statistical methods have led to inconsistent results.
• The optimal statistical approach for SC identification remains undetermined.

Purpose of the Study:

• To compare SC identified using eight different statistical techniques on a single dataset.
• To assess the reproducibility of SC found in prior research.

Main Methods:

• Reanalyzed a dataset of 1000 advanced cancer patients' symptom prevalence and severity.
• Conducted eight distinct cluster analyses, including hierarchical clustering and K-means validation.
• Utilized Spearman correlation, Kendall tau-b correlation, and Cohen kappa for similarity measures.

Main Results:

• Hierarchical clustering consistently identified similar symptom configurations across different thresholds (r=0.6, 0.5, 0.4).
• Seven symptom clusters were consistently identified and validated by K-means analysis.
• Four core symptom clusters (aerodigestive, fatigue/anorexia-cachexia, nausea/vomiting, upper GI) were robust across all analyses.

Conclusions:

• Symptom cluster identification is consistent regardless of the statistical method employed.
• Minor variations in cluster number and composition were observed between techniques.
• The consistent identification of seven SC supports their clinical significance and the validity of the SC concept.