Use of Cluster Analysis to Delineate Symptom Profiles in an Ehlers-Danlos Syndrome Patient Population

Abstract

Context: The Ehlers-Danlos Syndromes (EDSs) are a set of rare heritable disorders of connective tissue, characterized by defects in the structure and synthesis of extracellular matrix elements that lead to a myriad of problems including joint hypermobility and skin abnormalities. Because EDS affects multiple organ systems, defining clear boundaries and recognizing overlapping clinical features shared by disease phenotypes is challenging.

Objectives: The objective of this study was to seek evidence of phenotypic subgroups of patients with distinctive symptom profiles and describe these resulting subgroups.

Methods: Data were extracted from a repository assembled 2001-2013 by the National Institute on Aging Intramural Research Program. Agglomerative hierarchical clustering was used to form distinct subgroups of patients with respect to the domains of pain, physical and mental fatigue, daytime sleepiness, and nighttime sleep. Domains were selected based on literature review, clinician expertise, and guidance from patient advisors.

Results: One hundred seventy-five patients met all inclusion criteria. Three subgroups were identified. The Pain Dominant subgroup (39 patients) had the highest mean pain values, but lowest mean values of other symptoms. The High Symptom Burden subgroup (71 patients) had high mean values in all domains. The Mental Fatigue subgroup (65 patients) had a high mean value for mental fatigue and daytime sleepiness, but a lower mean value for pain.

Conclusion: The subgroups aligned with clinical observation of the heterogeneous nature of EDS, with overlapping symptoms between subtypes and a wide divergence in degree of symptoms within subtypes. This exploratory study helps characterize the various phenotypes and comorbidities of patients with EDS.

Keywords: Ehlers-Danlos Syndrome; cluster analysis; symptom profile.

Fig. 1.

Scatter plot matrix for each pair of variables, with density estimates shown on diagonal and Spearman’s correlation statistics shown in upper corner. Points were slightly jittered (offset) within the scatter plots.

Fig. 2.

Final dendrogram based on agglomerative hierarchical clustering. The leaves on the very bottom of the plot indicate individual patients. Line segments at various heights (y-axis) indicate the distances at which clusters were combined.

Fig 3.

(A) Mean values by subgroup for each standardized variable. (B) Parallel coordinates plots showing standardized values across variables for all individuals classified within each subgroup.