The Use of Cluster Analysis by Partitioning around Medoids (PAM) to Examine the Heterogeneity of Patients with Low Back Pain within Subgroups of the Treatment Based Classification System

Current evidence in low back pain (LBP) focuses on population averages and traditional multivariate analyses to find the significant difference between variables. Such a focus actively obscured the heterogeneity and increased errors. Cluster analysis (CA) addresses the mentioned shortcomings by calculating the degree of similarity among the relevant variables of the different objects.  This study aims to evaluate the agreement between the treatment-based classification (TBC) system and the equivalent 3 cluster typology created by partitioning around medoids (PAM) analysis. In this cross-sectional study, a convenient sample of 90 patients with low back pain (50 males and 40 females) aged 20 to 65 years was included in the study. The patients were selected based on the 21 criteria of 2007 TBC system. An equivalent 3 cluster typology (C3) was applied using PAM method. Cohen’s Kappa was run to determine if there was agreement between the TBC system and the equivalent C3 typology.  PAM analysis revealed the evidence of clustering for a C3 cluster typology with average Silhouette widths of 0.12. Cohen’s Kappa revealed fair agreement between the TBC system and C3 cluster typology (Percent of agreement 61%, Kappa=0.36, P<0.001). Selected criteria by PAM analysis were different with original TBC system.   Higher probability of chance agreement was observed between two classification methods. Significant inhomogeneity was observed in subgroups of the 2007 TBC system.