Chronic Fatigue Syndrome prevalence is grossly overestimated using Oxford criteria compared to Centers for Disease Control (Fukuda) criteria in a U.S. population study

Abstract

Background.

Results from treatment studies using the low threshold Oxford criteria for recruitment may have been overgeneralized to patients diagnosed by more stringent CFS criteria.

Purpose.

To compare the selectivity of Oxford and Fukuda criteria in a U.S. population.

Methods.

Fukuda (Center for Disease Control (CDC)) criteria, as operationalized with the CFS Severity Questionnaire (CFSQ), were included in the nationwide rc2004 HealthStyles survey mailed to 6,175 participants who were representative of the US 2003 Census population. The 9 questionnaire items (CFS symptoms) were crafted into proxies for Oxford criteria (mild fatigue, minimal exclusions) and Fukuda criteria (fatigue plus ≥4 of 8 ancillary criteria at moderate or severe levels with exclusions). The comparative prevalence estimates of CFS were then determined. Severity scores for fatigue were plotted against the sum of severities for the 8 ancillary criteria. The 4 quadrants of scatter diagrams assessed putative healthy controls, CFS, chronic idiopathic fatigue, and CFS-like with insufficient fatigue subjects.

Results.

The Oxford criteria designated CFS in 25.5% of 2,004 males and 19.9% of 1,954 females. Based on quadrant analysis, 85% of Oxford-defined cases were inappropriately classified as CFS. Fukuda criteria identified CFS in 2.3% of males and 1.8% of females.

Discussion.

CFS prevalence using Fukuda criteria and quadrant analysis were near the upper limits of previous epidemiology studies. The CFSQ may have utility for on-line and outpatient screening. The Oxford criteria were untenable because they inappropriately selected healthy subjects with mild fatigue and chronic idiopathic fatigue and mislabeled them as CFS.

Introduction

At least 20 Chronic Fatigue Syndrome (CFS) / myalgic encephalomyelitis (ME) definitions have been proposed that differ in the severity of fatigue, presence of post-exertional malaise, and ancillary symptoms. [1–3]. These definitions stratify patients in different ways, leading to heterogeneous clinical groupings [4] and clinical conclusions that cannot be generalized between studies or patient populations. We argue that stratification of symptom severity is necessary for correct case designation, appropriate exclusion of subjects, comparisons between criteria, and for understanding the differential diagnosis of fatiguing illnesses.

The Oxford definitional criteria [5] for CFS are inappropriate (in our view) because they require only mild fatigue severity in contrast to the moderate and severe fatigue required in the Fukuda (Centers for Disease Control) definition [6,7] and other definitions, e.g., Carruthers’ Canadian ME/CFS criteria [8,9]. Even though it is highly unlikely that subjects with mild fatigue are representative of those with severe symptoms, studies using the low threshold Oxford criteria [10] that recruit mildly fatigued subjects continue to be published and their treatment recommendations generalized [11,12] to patients with severe fatigue, post-exertional malaise, pain and other CFS/ME characteristics. The consequences of this type of over-inclusiveness were explored in this study by determining the proportion of a U.S. population who met Oxford criteria as compared to Fukuda criteria for CFS.

Illustrations of Overlapping Criteria in CFS/ME

Criteria overlap was evident in an Australian CFS/ME patient database where 30% of presumed patients met Fukuda criteria, 32% Fukuda plus Carruthers criteria, 15% had chronic fatigue that did not meet CFS/ME criteria, and 23% had fatigue with other exclusionary causes [13]. In other reports, lack of specificity was evident by the designation of CFS in 15% of the healthy adults using the Fukuda criteria [6,14] and 38% of major depressive disorder patients using the Empiric CDC criteria [7,15]. Another example: An epidemiological study [17] (n = 183,841) identified 684 potential CFS cases, of which only 11% of these met Fukuda criteria for CFS. Seventy percent had exclusionary medical conditions, 11% had fatigue for less than 6 months or fewer than 4/8 ancillary criteria, and 7% were a combination of chronic idiopathic fatigue (CIF) [16] and CFS-like illness with insufficient fatigue syndrome (CFS-Like) [7,17]. CFS-Like subjects had no or minimal fatigue but substantial complaints on CFS ancillary criteria [17]. CIF and CFS-Like individuals have been lumped together as a category of “unwell subjects with insufficient symptoms for CFS,” [18,19] but this does not facilitate investigations of those with fatigue alone compared to those with increased symptoms without fatigue.

The range of severities of fatigue, pain, sleep disturbance and other symptoms in these various illness subcategories may help to identify the level of severity that is clearly dysfunctional or in need of treatment. In addition, it is of value to determine if fatigue and other symptoms have continuous, normal distributions in the general population because “illness” may be assigned to the tail of the normal curve distal to an arbitrary threshold, or may represent a separate and distinct pathophysiological entity with a bimodal distribution in health versus disease [20,21].

CFS Severity Questionnaire

The Fukuda criteria [6], operationalized as the CFS Severity Questionnaire (CFSQ) [22], represents an attempt to quantify a dysfunctional level of disturbance in the illness. The questionnaire data provide a mathematical framework to study the spectrum of nociceptive, interoceptive and fatiguing illnesses by allowing fatigue severity to be plotted on one axis of a scatter diagram and the sum of severities of the other 8 ancillary complaints (Sum8) in the Fukuda definition to be plotted on the perpendicular axis [22]. Thresholds were defined for problematic fatigue and ancillary symptoms. These divided the scatter plots into 4 quadrants: (i) healthy control state with minimal fatigue and ancillary symptoms, (ii) problematic fatigue (moderate or severe) with no ancillary symptoms (chronic idiopathic fatigue, CIF), (iii) problematic fatigue plus sufficient ancillary symptoms to meet CFS criteria, and (iv) sufficient ancillary symptoms but insufficient fatigue (CFS-Like). The characteristics of CIF and CFS-Like are not well understood but can be examined using this framework.

The CFSQ was included in the U.S. nationwide HealthStyles postal survey [23]. HealthStyles is a proprietary database product of Porter Novelli, Inc. [24], and is licensed by the CDC for audience analysis in health communication planning [25]. The mail-in survey has been conducted annually since 1995 [23]. It was designed to assess attitudes and beliefs of the U.S. population about chronic and infectious diseases and behaviors, exposure to health information and communication campaigns, and self-reported symptoms, diseases and disorders. HealthStyles data provide estimates of self-reported risk factors and conditions comparable to random-sampling methods [26]. The advantage of the national survey data was the unbiased selection of participants stratified to 2003 U.S. Census data.

Obvious limitations of relying solely on self-report of the nine Fukuda criterion items in these postal responses were the absence of clinical evaluations to confirm the presence or absence of each diagnostic criterion, inability to assess the full spectrum of symptoms from other definitions e.g., Canadian criteria [8,9], and the inability to investigate the differential diagnosis of CFS and exclude inflammatory, psychiatric and other medical conditions [7,27,28]. The HealthStyles survey included these definitional exclusions: arthritis-related symptoms, daily exercise for more than 30 minutes, and significant cardiovascular disease with transient ischemic attacks. It was not possible to assess the many other potential exclusions inherent in the differential diagnosis of CFS.

The objective of this investigation was to estimate the comparative prevalence of CFS based on the Oxford and Fukuda criteria in female and male subjects drawn from a sample representative of the U.S. population. Quadrant analysis was used as a second step to assess differences based on symptom severities.

Methods

HealthStyles survey.

The Porter Novelli HealthStyles survey [16,17] was conducted in 2004 with CDC support by Synovate, Inc., a sampling and data collection firm [29]. Synovate purchased a mailing list of approximately 600,000 community-dwelling adults with utility, telephone, credit card, loan, subscription or similar consumer products registered in their name. Potential participants completed a four-page recruitment survey to facilitate selection of an unbiased and nationally representative panel. Participants gave informed consent for use of their anonymous data. The 2004 HealthStyles demographics were reported to be comparable to the 2003 U.S. Census estimates for age, race, sex, household size, and household income except for underrepresentation of the lowest education level (not a high school graduate) and overrepresentation of the lowest income (<$10,000) [30]. Stratified random sampling generated a list of 10,000 participants for the 1^st wave ConsumerStyles survey mailing in May and June 2004. There were 6,207 respondents.

A subset (n=6,175) were then sent the 2^nd wave rc2004 HealthStyles survey that included the CFSQ during July and August 2004. Survey respondents received a 20-min calling card and sweepstake ticket with 1^st place prize of $1,000 and five 2^nd place prizes of $50. The final 4,345 anonymous survey datasets were purchased through Maibach Data Group, the responses screened, and 387 excluded because of incomplete CFSQ with no efforts to impute missing data. Data from 2,004 males and 1,954 females were stratified according to the Oxford and CDC criteria. Analysis of this totally anonymous database did not require IRB approval.

CFSQ and Quadrant Method.

The Fukuda CDC criteria [6] were scaled so subjects could grade their symptoms over the previous 6 month period as none (score = 0), trivial (score = 1), mild (score = 2), moderate (score = 3) or severe (score = 4) [22]. This scoring system allowed investigation of the levels of fatigue and each of the ancillary criteria that can distinguish control from CFS subjects. The sum of severity scores for the 8 ancillary criteria was calculated as Sum8 (range 0 to 32). A significant threshold of Sum8 ≥ 14 was defined by receiver operating characteristics in 600 subjects from rhinitis, sinusitis, asthma, CFS and other clinical trials (specificity = 0.93; sensitivity = 0.93; Cronbach’s alpha = 0.92) [22].

Fatigue severity and Sum8 were treated as 2 independent dimensions and plotted against each other. The fatigue axis was divided into < and ≥ moderate severity, and Sum8 axis into <14 and ≥14 to create 4 quadrants: (i) healthy controls (HC) with none, trivial or mild fatigue and Sum8 < 14; (ii) CIF with moderate or severe fatigue and Sum8 < 14; (iii) CFS-like with insufficient fatigue syndrome (CFS-Like) with high symptom severities indicated by Sum8 ≥ 14 but fatigue that was absent, trivial or mild; and (iv) CFS with moderate or severe fatigue and Sum8 ≥ 14. Although this quadrant system was not intended for CFS diagnosis, it did provide a logical research framework for assessing CFS, CIF and CFS-Like categories of fatiguing illnesses in comparison to health.

Oxford criteria.

The Oxford criteria (“Oxford”) define a case of CFS if mild to severe symptoms of fatigue, sleep disturbance and myalgia are present [5]. The Oxford proxy was designated when all 3 items were present with mild or greater severities. No weight was given to the presence or absence of other complaints. Few medical conditions were articulated as exclusionary illnesses. Instead, there has been a recommendation that other medical conditions be prospectively listed in study publications. Depression, anxiety and chronic idiopathic fatigue (CIF) were permitted by the Oxford criteria.

Fukuda (Center for Disease Control; CDC) criteria.

For Fukuda criteria, CFS was defined by significant fatigue plus at least 4 of 8 ancillary criteria [6]. A long-standing ambiguity in the Fukuda criteria has been the level of severity required for positive identification of each criterion [7,15,27]. Attempts to quantify fatigue were added [7,31] but have not been widely used. The proxy used here required moderate or severe levels for fatigue and at least 4 ancillary criteria. Exclusionary conditions were applied for the CDC criteria (see below) [7,27,28].

Exclusions.

The Fukuda criteria and other more specified case definitions [6–9,27,28] have well described directives for excluding medical illnesses that may explain fatigue and the remainder of the symptom cluster. Based on our questionnaire data, we used these three exclusions: (1) HealthStyles arthritis scores plus CFSQ arthralgia severity scores that reached moderate or severe levels, given the medical implications of painful, red, hot, swollen joints with decreased range of motion and function as exclusionary of a CFS diagnosis; (2) Trans-ischemic attacks which imply significant cardiovascular disease and strokes; (3) Agreement or strong agreement with the statement “I exercise 30 minutes every day” in the HealthStyles questionnaire, which was considered exclusionary for reduced physical exertion and exertional exhaustion in CFS. Other known CFS exclusions were not part of the HealthStyles survey.

After data were cleaned to exclude these conditions and any incomplete questionnaire results, the male and female samples were found not to be comparable to each other, and may not have been representative of the U.S. population. Data were analyzed in SPSS.

Results

Quadrant analysis.

The female and male samples were divided into 4 quadrants (Table 1) [22]. CFS was initially designated in 8.8% of females (131/1,495), but 61.8% (81/131) of these individuals had exclusionary conditions. As a result, CFS was estimated to occur in 3.3% of this female population. The rates of CIF and CFS-Like in females (8.5% and 8.0%, respectively) and males (10.3% and 8.6%, respectively) were higher than previously estimated [17,27,32]. After exclusions, the rate of CFS in males was 4.7%.

Table 1.

Quadrant analysis. Females and males were assigned to the 4 quadrants. The numbers of subjects who met CFS criteria but were excluded using the HealthStyles exclusionary conditions were in the right hand column.

		Fatigue		Excluded CFS subjects
		None, Trivial, Mild	Moderate, Severe
Female (n = 1,495)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 74.7% (1,117)	CIF 8.5% (127)
	14 to 32	CFS-Like 8.0% (120)	CFS 3.3% (50)	5.4% (81)
Male (n = 2,004)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 67.9% (1,361)	CIF 10.3% (206)
	14 to 32	CFS-Like 8.6% (173)	CFS 4.7% (95)	8.4% (169)

Plotting fatigue severity against the sum of the 8 ancillary severity scores showed the general change in these parameters with increasing severity (Figure 1). The most frequent pair of responses was no fatigue plus no ancillary symptoms, found in 9.5% of females and 7.6% of males. The frequency distribution for fatigue severity was shifted to the left in women compared to men indicating that women had lower levels of fatigue in this population. Fatigue and Sum8 scores decreased in coordinated, exponential fashion so that a spline was evident in their distributions. This fan-shaped distribution suggested that fatigue and the other symptoms had continuous distributions in this population without obvious breakpoints. This emphasized the need for objective markers to reinforce the clinical diagnosis of moderate to severe fatigue plus ≥4 other complaints.

Figure 1.

Quadrant method for the HealthStyles survey population. Sum8 (range 0 to 32) was binned and then plotted versus Fatigue Severity for the (a) 1,945 females and (b) 2,004 males who remained after removing subjects with incomplete questionnaire responses. Bar graphs were rotated to provide 2 orthogonal perspectives that showed the progressive increases in Fatigue Severity and Sum8 along the diagonal splines. The open row and column on each axis indicate the boundaries for significant fatigue and Sum8 ≥ 14. These boundaries defined HC, CFS-Like, CIF and CFS quadrants. Frequency distributions projected the Sum8 scores for each level of Fatigue Severity in (c) females and (d) males. A larger proportion of females had Fatigue Severity of “none” compared to males and accounted for the relatively lower proportion of CFS in females than males in this survey population.

Oxford criteria.

Using Oxford criteria, CFS was designated in 19.9% of women (n = 297) (Table 2). These women were then divided into 5 groups by quadrant analysis and the HealthStyles exclusionary conditions. Astonishingly, one quarter (25.6%) of the Oxford CFS women fell into the healthy control quadrant. CFS-Like was present in 22.2% and CIF in 13.5% of the female sample. The predominance of mild fatigue females selected by the Oxford criteria was evident from the frequency analysis (Figure 2). 23.6% (n = 70) of Oxford subjects had exclusionary conditions leaving only 15.2% (n = 45) in the CFS quadrant. Removing the female subjects with exclusions from the CFS quadrant would have dramatically reduced the number and frequency of CFS subjects but the Oxford criteria do not specifically exclude these subjects as is customary with Fukuda and other more rigorous definitions.

Table 2.

Quadrant analysis of subjects meeting Oxford criteria for CFS. Females and males were assigned to the 4 quadrants. The subset of Oxford CFS subjects who met HealthStyles exclusionary conditions were in the right hand column.

		Fatigue		CFS subjects with exclusions
		None, Trivial, Mild	Moderate, Severe
Female CFS 19.9% (n = 297 / 1,495)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 25.6% (76)	CIF 13.5% (40)
	14 to 32	CFS-Like 22.2% (66)	CFS 15.2% (45)	23.6% (70)
Male CFS 25.5% (n = 511 / 2,004)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 24.5% (125)	CIF 12.5% (64)
	14 to 32	CFS-Like 19.6% (100)	CFS 14.9% (76)	28.6% (146)

Figure 2.

Quadrant analysis of females (n = 1,495). (a) The Oxford criteria permitted mild fatigue (green bars) which skewed the putative CFS subjects into healthy control and CFS-Like quadrants. (b) The rigorous CDC criteria limited subject selection to the CFS quadrant with moderate or severe fatigue and the sum of the 8 ancillary criteria scores (Sum8) from 14 to 32.

The Oxford criteria determined that 25.5% of the males (n = 511) were CFS cases (Table 3). Within this male group, 28.6% (n = 146) had exclusionary conditions, with only 14.9% (n = 76) remaining in the CFS quadrant. Like the females, one quarter of the male CFS subjects selected by the Oxford criteria fell into the healthy control quadrant (24.5%) (Figure 3). CFS-Like cases accounted for 19.6%, and CIF for 12.5% of Oxford males. Thus allowing mild fatigue under the Oxford criteria grossly overestimated the prevalence of CFS in this U.S. population survey.

Table 3.

Quadrant analysis of subjects meeting CDC criteria for CFS. Females and males were assigned to the 4 quadrants. The numbers of subjects who met CFS criteria but were excluded using the HealthStyles exclusionary conditions were in the right hand column.

		Fatigue		CFS subjects with exclusions
		None, Trivial, Mild	Moderate, Severe
Female CFS 1.8% (n = 27 / 1,495)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 0 (0%)	CIF 11.1% (3)
	14 to 32	CFS-Like 0 (0%)	CFS 88.9% (24)	0 (0%)
Male CFS 2.3% (n = 46 / 2,004)
Sum of 8 ancillary symptom scores (Sum8)	0 to 13	HC 0% (0)	CIF 4.3% (2)
	14 to 32	CFS-Like 0% (0)	CFS 95.7% (44)	0% (0)

Figure 3.

Quadrant analysis of males (n = 2,004). (a) The Oxford criteria permitted mild fatigue (green bars) which skewed the putative CFS subjects into healthy control and CFS-Like quadrants. (b) The rigorous CDC criteria limited subject selection to the CFS quadrant with moderate or severe fatigue and the sum of the 8 ancillary criteria scores (Sum8) from 14 to 32.

Fukuda criteria.

The more stringent Fukuda criteria designated CFS in 1.8% of females (Table 2) and 2.3% of males (Table 3). None of these CFS subjects met any of the 3 exclusionary conditions available in the HealthStyles survey. The Fukuda CFS subjects were also assessed by the quadrant process (Figures 2 and 3). The Fukuda requirement for moderate or severe fatigue meant that none of the 27 female and 46 male CFS subjects could be categorized into the control or CFS-Like quadrants (Tables 2 and 3). Only 3 females and 2 males fell into the CIF quadrant. The frequency analysis plots of fatigue severity against the sum of the 8 ancillary symptom severities demonstrated that these subjects were skewed into the CFS quadrant using this multi-step evaluation (Figures 2 and 3).

Comparing Oxford and Fukuda criteria.

The Oxford criteria identified 297 females and 511 males as CFS, but only 7.7% of females and 8.4% of males also met the Fukuda criteria (Table 4). Thus the Oxford criteria designated about 10-fold more females and males as CFS compared to the Fukuda criteria. This stark difference suggests that studies using the Oxford criteria have investigated largely mild fatigue and recruited fewer than 10% of “true” CFS subjects. Over 90% of Oxford CFS subjects were false positives when subsequently assessed using the more rigorous quadrant and Fukuda criteria.

Table 4.

Overlap of CFS designation between criteria. The left hand column indicated the criteria used to designate CFS status. The 2^nd column was the size of that group. The rest of each row showed the proportions also meeting CFS designation by other criteria.

CFS criteria	N	Oxford	Quadrant	CDC
Females (n = 1,495)
Oxford	297	100% (297)	15.2% (45)	7.7% (23)
Quadrant	50	90.0% (45)	100% (50)	48.0% (24)
CDC	27	85.2% (23)	88.9% (24)	100% (27)
Males (n = 2,004)
Oxford	511	100% (511)	18.6% (95)	8.4% (43)
Quadrant	95	80.0% (76)	100% (95)	46.3% (44)
CDC	46	93.5% (43)	95.7% (44)	100% (46)

Exclusions.

The rates for each HealthStyles exclusion condition were similar in females and males in each of the 4 quadrants (Table 5). Exclusionary daily exercise was endorsed by about 30% of all subjects regardless of complaints. Arthritic complaints, considered an exclusion because of the potential for inflammatory arthritis, were more common in CFS and CFS-Like (38.3% to 53.0%) than CIF and controls subjects (7.3 to 17.3%). Finally, transient ischemic attacks were endorsed by more CFS subjects (about 9%) than controls (3%) which may suggest cardiovascular disease.

Table 5.

HealthStyles exclusionary condition endorsements in each group defined by the quadrant method from female (n=1,495) and male (n=2,004) subjects.

Exclusions	CFS	CFS-Like	CIF	HC	Total
Agree or strongly agree that exercise for 30 minutes per day is important
Female	29.8% (39)	22.5% (27)	29.1% (37)	34.6% (386)	32.7% (489)
Male	20.5% (54)	31.2% (54)	22.3% (46)	32.3% (440)	29.6% (594)
Arthritis plus moderate or severe Arthralgia Score
Female	41.2% (54)	38.3% (46)	17.3% (22)	7.3% ( 82)	13.6% (204)
Male	53.0% (140)	49.1% (85)	17.0% (35)	8.3% (113)	18.6% (373)
Transient ischemic attack (TIA)
Female	9.2% (12)	6.7% (8)	4.7% (6)	3.0% (33)	3.9% (59)
Male	9.1% (22)	6.1% (10)	4.0% (8)	3.3% (44)	4.2% (84)

Subjects who met the Oxford, Fukuda and quadrant criteria for CFS were then compared for HealthStyles exclusion conditions. Rates of exclusion were similar between Oxford and quadrant methods for exercise (about 25%), arthritis (31.6% to 53.0%), and transient ischemic attacks (about 8%) (Table 6). Exclusions for females and males were also comparable. None of the CFS subjects identified by Fukuda criteria were excluded.

Table 6.

Reasons for exclusion from CFS diagnosis. Female (n=1,495) and male (n=2,004) groups were assessed for CFS case designation criteria. The percentages (numbers) of putative CFS cases who met any of the HealthStyles survey exclusionary conditions were shown. The Oxford criteria do not mandate that subjects be excluded for these reasons.

Exclusion	N	Oxford	Quadrant	CDC
Agree or strongly agree that exercise for 30 minutes per day is important
Female	489	27.2% (81)	29.8% (39)	0% (0)
Male	594	25.8% (132)	20.5% (54)	0% (0)
Arthritis plus moderate or severe Arthralgia Score
Female	204	31.6% (94)	41.2% (54)	0% (0)
Male	373	40.5% (207)	53.0% (140)	0% (0)
Transient ischemic attack (TIA)
Female	59	6.4% (19)	9.2% (12)	0% (0)
Male	84	7.2% (37)	8.3% (22)	0% (0)
Percentage (number) excluded from CFS diagnosis
Female		52.2% (155)	61.4% (81)	0% (0)
Male		58.7% (300)	64.0% (169)	0% (0)

Sensitivity.

When the quadrant method was used as the reference to define CFS prevalence in females, the sensitivities for CFS detection were 15.2% for the Oxford criteria and 88.9% for the CFS criteria. In males, the sensitivities for detecting CFS were 14.9% for the Oxford and 95.7% for Fukuda criteria.

Age.

Women in the CFS group selected by quadrant analysis who had HealthStyles exclusions (53.4±12.2 years, n=81 [mean±SD]) were significantly older than quadrant CFS (45.5±13.5, n=50) and control (48.4±14.0, n=1,117) groups (ANOVA and Tukey’s HSD p<0.05). In addition, CFS males selected by the quadrant system (42.8±13.8, n=95) and Fukuda criteria (38.7±10.3, n=46) prior to exclusions were significantly younger than those with Oxford CFS (49.2±15.1, n=511) and quadrant system CFS who had HealthStyles exclusions (53.0±14.8, n=169). Quadrant CFS males with exclusions were also older than the healthy control group (47.5±14.4, n=1,362). Ages were similar for Oxford CFS and controls, and the other quadrant system groups. By comparison, CFS males selected by Fukuda criteria and quadrant systems were significantly younger than healthy controls suggesting younger onset of their disease. Both female and male CFS subjects with HealthStyles exclusions were significantly older suggesting that they had older age onset of illnesses (e.g., atherosclerosis, arthritis) associated with chronic fatigue, pain, sleep and cognitive problems.

Discussion

The proxy definition used for Oxford CFS criteria selected 19.9% of females (297/1,495) and 25.5% of males (511/2,004) in this U.S. population sample. Factoring in the 3 exclusions from the HealthStyles survey reduced these rates to 10.5% and 9.5%, respectively. By comparison, proxy Fukuda criteria, which required moderate or severe fatigue plus 4 of 8 additional symptoms yielded a CFS prevalence of 2.3% of males and 1.8% of females, an approximate 10-fold reduction compared to the Oxford criteria.

The current data predict that the Oxford criteria will have poor sensitivity and specificity for CFS diagnosis when compared to the more stringent Fukuda [6,7], ME [8,9], and Systemic Exertion Intolerance Disease (SEID) criteria [20]. This is because the Oxford criteria for CFS [31] require only mild fatigue without specific requirements for additional symptoms of exertional exhaustion, pain, and/or cognitive dysfunction. In addition, Oxford does not specify any exclusionary conditions, such as major depression or autoimmune disorders. This is also important because exclusionary conditions were found in 23% - 70% of CFS cases sent for clinical referrals [13,32,34]. The impact of exclusionary conditions on the differential diagnosis of CFS was confirmed in our findings. As such, treatment recommendations based on studies that used the Oxford criteria [5,10–12] may largely apply to subjects with mild fatigue [16], and may not be clinically justified for more severe CFS/ME cases who meet Fukuda, Carruthers or SEID criteria.

The Fukuda criteria estimated prevalence in the current study is 1.8% in females and 2.3% in males. Published point prevalence rates using Fukuda criteria have ranged from 0.042%, [32], 0.3%, [35] to 2.54% [17] in mixed gender populations, and 0.37% for females and 0.083% for males [36]. Nacul et al. estimated a minimum prevalence rate of 0.2% in England with 0.19% meeting the Fukuda CDC definition and 0.11% the Carruthers definition [37]. Meta-analysis of 14 studies estimated a prevalence of 3.28% (95% CI: 2.24–4.33) for self-reported data, but only 0.76% (95% CI: 0.23–1.29) based on clinical assessment [13]. Thus, the Fukuda proxy analysis based on the CFSQ and Fukuda criteria matched the upper bounds of CFS prevalence estimates in epidemiological studies.

The Utility of the CFSQ and Quadrant Analysis

The CFSQ may be useful as a rapid screen to appraise potential cases for CFS in clinical practice and epidemiological surveys. It offers the additional advantage of assigning subjects to CIF and CFS-Like diagnostic categories. Prevalence rates for groups with insufficient fatigue (CFS-Like) and sufficient fatigue but insufficient ancillary findings (CIF) have not been delineated as they are generally pooled together into an “insufficient fatigue” group (prevalence ~0.74%) [32,18,19]. Diagnosis is challenging because almost half of CFS and “insufficient fatigue” patients have at least one non-exclusionary comorbid condition [32].

Quadrant analysis in this study offered an informative view of the frequency distribution of different levels of fatigue and the ancillary symptoms in this U.S. population. As shown previously [22], frequencies were highest at the origin (no fatigue and no other symptoms), then decreased smoothly to the upper limits of each scale [Figure 1]. There were no discontinuities or breakpoints in these curves that could be used as diagnostic thresholds. Instead, thresholds were defined by receiving operating characteristics [22]. As expected, the healthy control quadrant had the highest proportion of survey respondents. In clinical evaluations, the essential next step is to identify exclusionary conditions. This important task was underlined by the high rates of subjects who met CFS criteria plus at least 1 of the 3 exclusions available in the HealthStyles survey (Tables 2, 3, 5, 6). Apart from relatively straightforward exclusions, post-exertional malaise is a critical and discriminating aspect of CFS, but is more challenging to assess as a single questionnaire item on a mail-in or online survey than during personal interviews.

Apart from CFS itself, the CIF quadrant offers a method to evaluate fatigue as a solitary complaint. This is relevant to aging where fatigue has been investigated without assessing co-morbid pain, systemic hyperalgesia, sleep disturbances, cognitive dysfunction and post-exertional malaise [16]. Use of the CFSQ in healthy aging may provide insights into the natural progression of each of these symptoms and the potential for discovery of what could be an aging-related symptom complex similar to CFS.

The CFS-Like quadrant represented subjects with an extensive positive review of systems (sum of 8 ancillary symptoms ≥ 14) in the absence of fatigue. The most important symptoms, clinical characteristics, prevalence and pathologies in this segment of the population have not been evaluated. Instead, individual symptoms such as pain have been the focus.

The CFSQ quadrant method [22] allowed comparisons between sets of CFS/ME criteria (Table 4), but was not intended as a new diagnostic for CFS. Unlike other systems, it did provide a rationale for distinguishing CFS-Like and CIF groups from healthy controls and CFS. This method demonstrated increased selectivity based on the escalation of symptom severity during the progression from the low threshold Oxford to the more restrictive Fukuda criteria. The 9 Fukuda-based symptom questions can be used in large surveys and potentially in an outpatient setting as a rapid, high sensitivity but low specificity screening tool to identify potential CFS subjects before moving on to more thorough history and physical examinations in clinical practice.

Limitations

There are several limitations to this report. HealthStyles survey respondents may reflect a selection bias despite being sampled from a national database and being stratified to match the 2003 US census for age, sex, race, ethnicity, marital status, income and geographical region. This distribution may be more diverse than populations sampled from tertiary care and other medical settings [1,2,22,33–36]. In addition, completion of the HealthStyles survey required intact functional capacity, reading comprehension and English fluency. Subjects who could not complete the questionnaire in a competent fashion would have been excluded during the initial vetting process and 1^st wave ConsumerStyles survey. The health literacy of some participants may have limited their understanding of the CFSQ, self-reported health, and other questions that may contribute to a high number of incomplete CFSQ responses.

Another limitation is due to reproducibility issues. Respondents completed the CFSQ in both the 1^st and 2^nd waves. The data reported here were from the 2^nd wave suggesting subjects should have been familiar with the format. Outcomes from the 1^st wave could not be used to test intra-subject reproducibility because of a questionnaire formatting flaw for fatigue. Therefore, this is a cross-sectional rather than a longitudinal assessment. Furthermore, respondents were instructed to rate their symptom severities for the past 6 month period which may not allow for the influence of recall bias, particularly if a severe flare of symptoms had been recent. This was unlikely to have affected subjects with none or trivial complaints.

The data could not be re-stratified to US Census - based norms because different age groups may have been affected by exercise (younger) compared to arthritis plus trans-ischemic attacks (older). The prevalence of CFS was lower in women than men. However, this was because a higher percentage of women had zero fatigue so that their frequency distribution was shifted to the left compared to men (Figure 1). A sampling bias could explain this relatively higher rate of CFS in males if it was due to an artifact of oversampling low income or rural males who did not work and so had time to complete the survey. African-American men in rural Georgia also had a higher rate of CFS than women [17]. In addition, more women were excluded because of incomplete HealthStyles CFSQ responses and endorsing daily exercise of 30 minutes or longer.

Proxies were derived post hoc from the CFSQ and other HealthStyles data. The proxies were satisfactory for the Oxford criteria of mild to severe fatigue, sleep and myalgia, and Fukuda criteria of moderate or severe fatigue plus moderate or severe ancillary symptoms plus allowance for exclusionary conditions. Items for the more stringent Carruthers criteria [8,9] and orthostatic intolerance for the systemic exertion intolerance disease [20] were not included in the survey.

The large proportion of subjects excluded by the small number of HealthStyles exclusionary conditions suggested that even more respondents may have had false positive CFS status because of other medical or psychiatric comorbidities. The exclusion criteria were clearly incomplete for the conditions considered in the differential diagnosis of CFS. It was not feasible to inquire into all of the exclusions and differential diagnosis of CFS [27,28] in this on-line survey. Despite these approximations, the rates of CFS determined by the Fukuda criteria were in the upper range reported from epidemiological and other questionnaires studies. Of special note is that the Oxford criteria do not necessarily apply these exclusions to subject selection [10].

Implications

The Oxford criteria appeared not to be useful because they selected roughly equivalent numbers of healthy control, CFS-Like, CIF and CFS subjects. This observation suggests that treatment studies based on Oxford criteria for participant inclusion may be seriously flawed because they can potentially select a cross-section of the healthy general population, rather than a rigorously defined CFS group. In addition, the HealthStyles population data indicate that the Oxford criteria cannot discriminate between idiopathic fatigue and control subjects. Thus pathological mechanisms, biomarkers, and treatments identified from Fukuda-based studies are not likely to be confirmed if the Oxford criteria are used because healthy control and other non-CFS subjects will dilute the authentic CFS signal(s). The lack of sensitivity and specificity for designation of CFS suggests that the Oxford criteria should not be used. In addition, treatment studies based on the Oxford criteria must be viewed skeptically [1,20,33].

Abbreviations:

CFS

Chronic Fatigue Syndrome

Oxford

1991 Oxford criteria

CDC

Center for Disease Control CFS criteria requiring moderate or severe for fatigue and ancillary criteria

CIF

chronic idiopathic fatigue

CFS-Like

CFS-like with insufficient fatigue syndrome

CFSQ

CFS Symptom Severity Questionnaire

Sum8

sum of Severity Scores for the 8 ancillary criteria

SEID

Systemic Exertion Intolerance Disease