Abstract
BACKGROUND: The use of different statistical methods and inclusion criteria when deriving symptom clusters in cancer patients are contributing factors in cluster inconsistencies across studies. Primary objective was to extract symptom clusters in a subgroup of patients reporting non-zero Brief Pain Inventory (BPI) scores at baseline, and to compare clusters with those identified in the total patient sample. METHODS: Principal Component Analysis (PCA), Hierarchical Cluster Analysis (HCA) and Exploratory Factor Analysis (EFA) were performed on the non-zero subgroup and total patient sample to identify symptom clusters at baseline and 1, 2 and 3 months following radiotherapy. RESULTS: At baseline, different symptom clusters were derived from the non-zero subgroup and the total patient population. Only PCA identified identical clusters. Over time, clusters extracted using the three statistical methods varied, with a few exceptions where the same clusters were extracted using two different methods at a specific time point. A complete consensus between all three methods was not noted at any time. The BPI, which is a short assessment tool, may lead to the extraction of oversimplified clusters. In addition, since this study analyzed results in the non-zero subgroup, clusters derived may be reflective of patients with poorer prognosis as these patients experienced all symptoms. CONCLUSION: Analyzing data compiled from all eligible consenting patients may not provide clinically relevant clustering among all symptoms in the assessment tool. The composition of symptom clusters varied with the inclusion of patients with zero symptom severity scores and with the statistical method employed.