A 4-cluster solution proved to be the best fit for these data because it generated the most interpretable profiles and also achieved a smaller error sum of squares (4.35%) compared to the 3- and 2-cluster solutions (5.34% and 13.07%, respectively). Expanding from a 4- to a 5-cluster solution did not substantially reduce the error sum of squares (which would indicate a reduction in unexplained variance). A statistical Relocate procedure was used to transfer individuals between clusters to achieve a better fit and more reliable cluster solution. This last refinement resulted in a reassignment of 28 individuals to attain the most homogenous clusters and a further reduction in the error sum of squares (3.86%). The final solution contained 4 relatively homogenous clusters with homogeneity coefficients ranging from 1.37 to 1.86 (illustrated in ).
Fig. 1 Psychological and biomedical profiles of FM subgroups. Cluster I (N = 19) was named Maltreated; Cluster II (N = 20), Dysregulated Biology; Cluster III (N = 36), Normal Biology; Cluster IV (N = 18), Positive Outlook. CTQ = Childhood Maltreatment Questionnaire; (more ...)
A series of univariate ANOVA’s confirmed that the 4 subgroups differed significantly from each other on 13 out of 14 cluster variables, P’s<.03, with the single exception being HA1c (). Differences between the 4 clusters were not attributable to demographic characteristics (). The only significant demographic difference was that individuals in Cluster III were slightly younger (by an average of 5 years, t = 2.97, P<.005).
Mean values (standard deviation) of untransformed cluster variables by subgroup.
As can be seen in , individuals in Cluster I were most likely to have reported childhood maltreatment (t = 10.54, P < .001), had the highest perceived stress (t = 5.61, P < .001), and lowest ratio of positive-to-negative affect (t= 4.75, P < .001). They also had lower testosterone (t = 1.98, P =.05) and low ANA titers (t = 2.27, P < .03). These patients were designated as the Maltreated subgroup.
Women in Cluster II had the highest ANA titers (t = 4.06, P = .001) and total cholesterol levels (t = 3.96, p < .001), larger BMI values (t = 2.21, P < .04.), with trends toward the highest ESR (t = 2.02, P = .056) and lowest creatinine clearance (t = 1.85, P =.067). These patients also had the lowest NK cell numbers (t=3.95, P < .001), cortisol (t = 2.78, P < .007), growth hormone (t = 3.20, P < .002), and testosterone levels (t = 3.80, P < .001). Given that they were distinctive on nearly every biological index measured, Cluster II was designated as having Dysregulated Biology. This group reported the highest levels of anxiety (t = 2.92, P = .005) with high perceived stress (t = 2.91, P < .007), and negative-to-positive affect (t = 3.28, P < .003).
In contrast, Cluster III was distinguished from the others by normal BMI (t = 4.70, P < .001), cholesterol (t= 4.78, P < .001), ESR (t = 3.86, P < .001), and HA1c levels (t = 2.33, P = .02). These women had the highest GH (t = 2.68, P < .009) and testosterone levels (t = 4.18, P < .001) with a trend for higher cortisol (t = 1.91, P = .06). Because their metabolic and neuroendocrine measures appeared to be better regulated without indications of inflammation, this cluster was named Normal Biology. Normal Biology patients had experienced little childhood maltreatment (t = 4.34, P < .001), yet reported high anxiety levels (t = 2.01, P = .047).
Cluster IV reported the most positive relative to negative affect (t= 7.40, P < .001) along with the lowest levels of perceived stress (t = 8.95, P < .001) and anxiety (t = 11.01, P < .001). Because of their more optimal psychological profile, this group was labeled Positive Outlook. They also tended to excrete the highest levels of cortisol, without evidence of hypocortisolemia (t = 1.90, P = .06).
Univariate ANOVA indicated that there were no differences in TP counts between the 4 clusters (F(3,89)=.21, NS). All women met ACR diagnostic criteria of having 11 or more TP. Nonetheless, discerning these four subgroups proved to have statistically predictive value for understanding variation in patients’ subjective experience of pain and ability to function in daily life. There were significant differences in self-reported pain intensity on the VAS across the four subgroups, (F(3,89) = 3.23, P = .026). As illustrated in , the Maltreated group reported the most pain (t = 2.01, P < .05), while the Positive Outlook cluster reported the least (t = 2.78, P < .007). There were also group differences in GAF (F(3,87) = 3.29, P = .024), with the Normal Biology cluster reporting relatively high global functioning (t = 2.10, P = .04). The Positive Outlook cluster functioned at a similarly high level, but because of the smaller number of subjects, the difference was significant only when compared directly with the Maltreated (P < .05) or Dysregulated Biology groups (P < .05). There were also tendencies for the clusters to differ in fatigue and ability to complete tasks. While an overall ANOVA did not reach significance across the four subgroups, when the Dysregulated Biology cluster was specifically targeted in post hoc analyses, these women were the most fatigued (t = 2.34, P = .024) and the least able to complete tasks (t = 2.07, P = .042).
Subgroup differences in the FM symptoms of subjective pain, measured by the Visual Analogue Scale, and Global Assessment of Functioning, assessed with a Structured Clinical Interview for DSM-IV.