Cohort 1 (n = 411) was younger and had a higher proportion of women than Cohort 2 (n = 189; p < 0.001 by Fisher’s Exact test) (). However, when fatigue severity was compared between the Not Fatigued and Fatigued groups within Cohort 1 and Cohort 2 by multivariate analysis, then age, gender, and age x gender cross-product were not significantly different. Fatigue, the 8 ancillary criteria, Sum8 and the number of responses with severity scores of 2, 3 or 4 were all significantly higher in Fatigued than Not Fatigued groups (p < 10-10 by Tukey’s tests after significant ANOVAs). As a result, data from Cohorts 1 and 2 were combined into Cohort 1&2 for the rest of the analysis.
Table 1 Comparisons between Not Fatigued and Fatigued groups with Cohort 1, 2 and 1&2. Data were shown for age, gender, each of the CFS Severity Scores, Sum8, and the number of ancillary symptoms scored at severities of 2, 3 or 4 (mean [95% confidence (more ...)
Cohort 1&2 subject classification by fatigue severity and Sum8
The Not Fatigued group was defined by fatigue levels of 0 (none), 1 (trivial) or 2 (mild) (n = 289) (). The most frequent score was 0 for fatigue and 0 for Sum8 (13.1% of Cohort 1&2). The distribution of Sum8 was unimodal but highly skewed (kurtosis 18.4; skewness 3.02), and it extended as high as 25 out of 32. There was a spline suggesting a trend that correlated fatigue with Sum8 scores.
Figure 1 Distribution of Sum8 for each Fatigue Score in Cohort 1 & 2. The study population was divided into not fatigued (HC), chronic idiopathic fatigue (CIF), and Chronic Fatigue Syndrome (CFS) groups. The nonfatigued group was defined by fatigue levels (more ...)
The Fatigued group (n = 311) was divided into CIF and CFS. The CIF group had fatigue scores of 3 (moderate) or 4 (severe), and ≤ 3 ancillary criteria with severity scores of 2, 3 or 4. The highest Sum8 score in the CIF group was 15. The CIF group was small because the original recruitment strategy targeted healthy control and CFS subjects. The Sum8 distribution for CIF was unimodal, and overlapped with the lower end of the preliminary CFS distribution.
At this initial stage of the analysis, CFS was defined by moderate (3) or severe (4) fatigue plus at least 4 ancillary criteria with severities of 2, 3 or 4. The importance of 3 versus 4 positive ancillary criteria was demonstrated by the difference in the Sum8 distributions for CIF and CFS groups (). CFS had much higher Sum8 scores. The distribution of Sum8 scores in CFS was close to symmetrical (kurtosis -1.20; skewness 0.062).
The Sum8 distributions of CFS and HC subjects overlapped. Receiver operator analysis of Sum8 set the threshold between HC and CFS at 14 (blue arrow, and ). This threshold had specificity of 0.934 and sensitivity of 0.928 (area under the curve = 0.98) in this unique population (). The threshold was applied to the HC subjects (). The distal tails of the distribution curves with Sum8 ≥ 14 was defined as CFS – like with insufficient fatigue syndrome (CFSLWIFS).
Receiver – operator curve (ROC) for Sum8.
In essence, fatigue scores of ≥ 3 and Sum8 ≥ 14 divided the Cohort 1&2 population into 4 quadrants (). The left side with Fatigue < 3 showed HC plus the CFSLWIFS group and their higher Sum8 scores. The right lower quadrant represented CIF with Fatigue of 3 or 4 and Sum8 scores < 14. The CFS group (upper right quadrant) had at least 4 significantly problematic symptoms and a Sum8 score ≥ 14.
Figure 3 Sum of 8 ancillary symptom scores (Sum8). Sum8 scores for HC (n=259, yellow squares) and CFS (n=276, black diamonds) were plotted as a function of Fatigue Severity Score (means with 95% confidence intervals). The explained variance for Fatigue and Sum8 (more ...)
Sum8 as a proxy of fatigue
The mean Sum8 was highly correlated with Fatigue Severity Scores for the combined HC and CFS quadrants and groups (R2 = 0 .977) (). This indicated that Sum8 could be an effective proxy for Fatigue when subjects were strictly defined as HC and CFS. However, the CFSLWIFS subjects with high Sum8 and Fatigue Scores of 0 to 2, and the CIF subjects with Fatigue Scores of 3 or 4 but only 0 to 3 ancillary complaints, were outlier groups. Sum8 was not correlated with Fatigue for these two groups or in their quadrants (Figures 1 and ).
Severity scores for groups in Cohort 1&2
The HC group was younger than the CIF (p = 0.0001) and CFS (p < 10-10) groups (). However, age, gender, and age x gender crossproduct were not related to either fatigue or Sum8 by multivariate linear regression.
CFS Severity Scores and demographics for the combined Cohort 1&2. Data for HC, CFSLWIFS, CIF and CFS groups were shown as means with [95% confidence intervals]
Fatigue (p < 10-10), cognitive function (p < 0.05), and myalgia (p < 0.02) scores were significantly different between the 4 groups. HC and CIF had equivalent scores (p > 0.05) for sore throat and lymph nodes. CFSLWIFS and CFS had equivalent scores for sore throat, lymph nodes, arthralgia, headache and sleep. Exertional exhaustion was equivalent between CFSLWIFS and CIF. Sum8 was ranked CFS > CIF > CFSLWIFS > HC with all groups being significantly different from each other (p < 0.005). The number of positive ancillary criteria was equivalent for CFSLWIFS and CFS (~6 symptoms), while these numbers for HC and CIF were lower and significantly different from the other 3 groups.
Cronbach’s alpha was 0.924. Exclusion of any single item maintained alpha between 0.907 and 0.925 indicating internal consistency and reliability.
Fourteen CFS subjects completed the questionnaire after 0, 6 and 12 weeks. Coefficients of variation ranged from 0.017 for memory and concentration to 0.81 for sore lymph nodes (). Symptoms for individuals did not change significantly over this 3 month longitudinal period. Gulf War Illness subjects had an almost identical pattern of scores as the CFS subjects suggesting overlap with regard to case designation criteria but without any inferences about comparable inciting or on-going pathogenesis. In contrast, 26.5% of Type II diabetics met CFS criteria. The co-existence of diabetes and CFS highlighted the primacy of the physician examination for determining the chronology of symptoms and syndrome onset, and influence of potential exclusionary disorders on the diagnosis of CFS.
Longitudinal CFS, Gulf War Illness, and diabetic group demographics and questionnaire scores. Data are shown as mean [95% confidence interval], with (coefficient of variation) for the longitudinal CFS group
The validity of this approach and coherence of the CFS, CIF, CFSLWIFS and HC categories were corroborated by independent comparisons of fatigue, disability, pain and irritant sensitivity in Cohort 1&2.
Multidimensional Fatigue Inventory (MFI) 
General Fatigue scores were ranked CFS > CIF = CFSLWIFS > HC (). HC and CFS scores were significantly different from each of the other groups (p < 0.0005). Physical Fatigue and Reduced Activity had the same trend. Reduced Motivation scores were equivalent between HC, CFSLWIFS and CIF. The CFS and CFSLWIFS groups were equivalent due to the small size (n=16) and large variance of the latter group. Mental Fatigue was significantly lower for HC than the other 3 groups (p < 0.05). CFSLWIFS and CFS had equivalent scores. Although the MFI domain scores for CFSLWIFS and CIF were similar, the Fatigue Severity score distinguished between these 2 groups. MFI domains were positively correlated with both Fatigue Severity and Sum8; General Fatigue had the highest explained variances. Reduced Motivation was poorly correlated with both Fatigue and Sum8.
Multidimensional Fatigue Inventory Domains (mean [95% C.I.]) and correlations (R2). Some Cohort 1 subjects did not complete both this and the CFS Severity Questionnaire and so the group sizes were smaller than in
Disability and quality of life
SF-36 Domain Scores [33
] were generally similar for HC, CFSLWIFS and CIF groups, and significantly different from CFS scores (). Physical Functioning, Role Physical, Bodily Pain, General Health, Vitality and Social Functioning had similar patterns of responses. CFS had significantly worse (lower) scores for these compared to the other 3 groups, while CFSLWIFS and CIF had equivalent scores. The latter 2 groups were also equivalent for Role Physical, General Health, Vitality, and Social Functioning. These two group’s scores for Role Physical, General Health and Vitality were significantly lower than the HC group. HC and CFSLWIFS scores for Physical Functioning and Bodily Pain were equivalent. The general pattern of HC > CFSLWIFS = CIF > CFS was reinforced by the high explained variances for these 6 domains with Fatigue Severity and Sum8 (R2 > 0.527).
SF-36 Domain scores for each Group (mean [95% confidence intervals])
In contrast, Role Emotional and Mental Health were relatively poorly correlated with fatigue and Sum8 (R2 between 0.220 and 0.244) suggesting that these domains were inefficient for separating Fatigued from Not Fatigued subgroups. Mental Health scores were equivalent for HC, CFSLWIFS and CIF. HC had higher scores than CFS for Role Emotional and Mental Health.
McGill Affective, Sensory and Total Pain Scores [35
] were equivalent for HC, CFSLWIFS and CIF groups, and significantly lower than CFS (p < 0.005) (). The Sensory scores were more highly correlated with Sum8 (R2
= 0.573) than Fatigue (R2
= 0.454). This was consistent with the painful ancillary symptoms included in Sum8. These data indicate that pain is a significant discriminator between CFS and the other 3 groups. This conclusion was supported by other findings of systemic hyperalgesia, allodynia and central sensitization in CFS [36
McGill short form Pain Scores. Scores are shown for HC, CFSLWIFS, CIF and CFS subjects who also completed the CFS Severity Questionnaire (mean [95% C.I.])
The Irritant Rhinitis Questionnaire showed that rhinorrhea scores were equivalent for CFSLWIFS, CIF and CFS subjects (). Congestion was equivalent in the CFSLWIFS and CFS groups, and significantly higher than HC. Congestion may be due to nasal vascular dilation, mucous hypersecretion, or be a trigeminal nociceptive perception of decreased nasal airflow [26
]. Fatigue and Sum8 were poorly correlated with congestion and rhinorrhea scores suggesting that mechanisms responsible for mucosal irritant sensitivity, fatigue, and the pain - and cognition - related Sum8 were distinct.
Irritant Rhinitis scores. Scores are shown for HC, CFSLWIFS, CIF and CFS subjects who also completed the CFS Severity Questionnaire (mean [95% C.I.])
Unsupervised hierachical clustering
Clusters of subjects were defined within the HC, CIF and CFS groups based on their CFS symptom severity scores. The CFSLWIFS group was too small for cluster analysis. Their mean scores were above 2 for all symptoms except fatigue, sore throat and tender lymph nodes (). This pattern was distinct from CFS and HC clusters. Headache and arthralgia scores were significantly higher for CFSLWIFS than for any of the HC clusters.
Figure 4 Clusters based on CFS Severity Score components. Clusters within A. CFSLWIFS, B. HC, C. CFS and D. CIF are shown. The single clade with severity scores that were significantly different from all of the other clades in the group were indicated on each (more ...)
HC had 4 clusters. The largest cluster (“A”, n=130) had scores of 0.2 or less for all symptoms (). HC cluster “A” symptom scores were significantly different from the other 3 clusters except for exertional exhaustion (equivalent to cluster “C”). Cluster “C” was the next largest (n=69), and generally had scores that were intermediate between “A” and the other HC clades. Clades “B” (n=37) and “D” (n=33) had similar, relatively elevated scores except for significantly higher scores for headache and myalgia in “D”. In preliminary examinations, a 3 cluster solution identified clades “A” and “C”, but combined clades “B” and “D”. When HC and CFSLWIFS were combined, a 5 cluster solution was obtained where the 4 HC clades were generally distinct from CFSLWIFS.
CFS subjects satisfied a 4 cluster solution (). CFS clade “A” had high scores for all symptoms. Clade “A” scores for sore throat and lymph nodes were significantly higher than the other 3 clades. This suggested that 40% of this CFS study population had chronic pharyngitis-like symptoms. Clade “D” (n=52) was similar to “A” except for significantly lower sore throat and lymph node scores. Clade “B” (n=53) was similar to “D” but had significantly lower myalgia scores compared to “A” and “D”, and arthralgia versus the 3 other clades. Clade “C” (n = 54) was unique in having significantly lower headache, cognition, and exertional exhaustion scores than the other 3 CFS clades.
The CIF group was optimally divided into 3 distinct clades. By definition, all had fatigue ≥ 3. CIF clade “A” (n = 14) had significant exertional exhaustion (p < 10-7 vs. the other 2 clades). Clade “C” (n = 22) had significant headaches compared to the other 2 clades (p < 10-5). Clade “B” had 15 subjects with low headache and exertional exhaustion scores.
Clustering for proteomic analysis
To demonstrate utility, CFS questionnaire responses were clustered for 56 CFS subjects who had lumbar punctures for cerebrospinal fluid proteomics [19
]. Questionnaire outcomes and ion peak signal intensities were assessed by 2-dimensional unsupervised hierarchical clustering. A heat map was generated showing synchrony between ion peaks and 4 clades of CFS subjects. This suggested that the ordinal scores were in some way indicative of the mechanisms leading to differences in cerebrospinal fluid constituents between the 4 clades. These relationships and proteomic results will be discussed in a separate publication.