This is the first published study, of which we are aware, that screened defined populations for unwellness and then used standardized, validated instruments to define CFS, unwellness, and wellness based on functional impairment, characteristics of fatigue, and frequency/severity of the 8 case defining symptoms [7
]. Using this approach, we found 2.54% of the Georgia population to suffer from CFS, which was 10-fold higher than previous estimates in the population of Wichita (0.24%) [3
] and 6-fold higher than estimated in the Chicago population (0.42%) [5
]. We are aware of no other published population-based surveys of CFS. However, several studies have published estimates of CFS prevalence; although they cannot be directly compared to the 2 U.S. studies, their prevalence estimates serve to put the present study into a more complete perspective. A well-conducted survey of the Group Health Cooperative of Puget Sound estimated that between 0.75 and 0.27% of that HMO population had CFS [19
]. A survey of primary care patients in England, published in 1997, estimated that 2.6% of that population met criteria for CFS [20
]. Finally, analysis of data from the Australian National survey of Mental Health and Wellbeing estimated that 1.5% of the Australian adult population suffers from chronic neurasthenia (defined similarly to CFS) [21
In part, the increased prevalence we estimated in Georgia reflects a difference in screening criteria. The Georgia survey screened for unwell (the core symptoms of CFS), whereas previous studies have screened only for fatigue. Our less restrictive approach allowed the inclusion of potential cases whom, although noted as unwell-not fatigued by a household informant, endorsed chronic fatigue upon detailed in-person interviewing. Sixty-nine people whom the household informant identified during the screening telephone interview as well or unwell without fatigue were classified as CFS-like based on their responses during detailed telephone interview (7.6% of all CFS-like). Further, 13 clinic participants classified as well or unwell without fatigue, based on their detailed telephone interview were diagnosed as CFS when evaluated clinically. In other words, 11.5% of subjects with CFS would not have been detected in previous studies that queried participants only for fatigue.
The 6- to 10-fold greater prevalence estimates also reflect application of more sensitive and specific measures of the CFS diagnostic parameters specified by the 1994 case definition. Previous prevalence estimates from population surveys and those based on patients attending clinics did not use validated standardized instruments to define CFS; rather they simply queried as to the presence or absence of fatigue, accompanying symptoms, and impairment. In 2003, the International Chronic Fatigue Syndrome Study Group
published recommendations concerning application of the case definition [7
]. They recommended the use of validated instruments to obtain standardized measures of the major symptom domains of the illness, and this study implemented those recommendations. The Study Group
specifically recommended: 1) the SF-36, to measure functional impairment; 2) the Checklist Individual Strength or MFI, to obtain reproducible quantifiable measures of fatigue; 3) and the CDC Symptom Inventory to document toe occurrence, duration and severity of the symptom complex. The manner in which we applied the case definition in Georgia has been shown to detect about 3 times the number of CFS cases as verbatim application of the 1994 definition [8
]. Of course, we cannot exclude the possibility that CFS prevalence may be higher in Georgia than Wichita and Chicago.
However, the manner in which we chose and applied subscales and their cutoffs from the SF-36 and MFI can be debated. We used the SF-36 physical function, role physical, social function and role emotional subscales to define an illness severe enough to "result in substantial reduction in previous levels of occupational, educational, social, or personal activities." [11
]. In particular, we included the role emotional subscale to capture the relation between functional emotional impairment and reduced social and personal activities. We ascertained the onset and duration of fatigue during interview (the case definition requires => 6 months of fatigue that is of new or definite onset) and utilized the MFI general fatigue and reduced activity scales to define severe fatigue. We used stringent (i.e., =< 25th
percentile population norms on any of the 4 SF-36 scales) to define severe functional impairment. Numerous publications tabulate slightly different population norms and we chose those published by Quality Metric [22
]. We are not aware of published population norms for the MFI, so we used the cut-offs established in a previous CDC study (=> than the median determined in Wichita) [8
], which is more sensitive and less specific than the 25th
percentile SF-36 cutoff. There are no population norms for the Symptom Inventory, so we also used cutoffs applied in the previous study. Finally, population norms are not cast in iron and one might consider defining cut-offs specific to each population studied. In the end, we decided to use these cutoffs because we believe they make sense; because we used them in the Wichita study and can compare findings in similarly ill individuals; and, because others can replicate the findings if they use the same cutoffs and stratify their populations based on variations of the cutoffs.
The other important new finding is that, despite differences in demographic factors, CFS prevalence was similar in the metropolitan, urban, and rural populations we surveyed (about 2.5%). However, differences in sex-specific prevalence among the strata must be evaluated in more detail. The striking differences between female and male rates in the 3 strata may indicate risk effects of gender (a social construct) in distinction to sex (a biologic attribute). In addition, the high prevalence among those of Hispanic ethnicity in the metropolitan area bears further investigation in a study designed to include Spanish-speaking persons.
The final major finding concerns the high proportion of study participants in whom the survey documented previously undiagnosed medical or psychiatric conditions. Overall 48% of persons recruited for clinical evaluation because CFS-like illness was identified on phone interview had exclusionary medical or psychiatric conditions, which is similar to the occurrence of such illness in other studies [2
]. Most of these exclusions are amenable to treatment if appropriately recognized. It is also important to note the relatively common occurrence of such conditions in people with other categories of unwellness identified during the phone interview. Indeed, 16% of respondents classified as well on the basis of interview had such exclusions identified. Further analyses will address whether the onset of psychiatric illness preceded or followed development of CFS-like illness.
Interpretation of the findings must consider obvious study weaknesses. First, like other telephone surveys, we faced challenges of nonresponse. Potential respondents can be lost in the processes of determining whether telephone numbers belong to households, completing screening interviews with household respondents, and maintaining contact with selected subjects to complete detailed interviews. In addition, some people are reluctant to schedule a full-day clinical examination. A random-digit-dialing survey that subsamples some types of subjects does not have a standard, single-number response rate. Taking into account the resolution of sampled telephone numbers (as residential or not), completion of screening interviews, and completion of detailed interviews, we calculated a response rate of 47.8% through the detailed interview. A further calculation, including the completion of clinical evaluations by subjects with CFS-like illness and chronically unwell subjects, produced a response rate of 27.5% through the clinic stage. Although nonresponse is always an appropriate subject for concern, detailed adjustments in the sampling weights often are able to mitigate its adverse effects. Thus, the adjustment for nonresponse on the detailed interview used a total of 195 cells, taking into account stratum, illness classification, sex, age, and race; post-stratification aligned respondents' weights with population totals; and the adjustment for nonresponse on the clinical evaluation used a total of 33 cells based on stratum, illness classification, sex, and age.
The second weakness concerns obtaining clinical information by telephone interview. Initial classification of subjects as well, unwell not fatigued, and unwell fatigued based on telephone interviews with household informants was confirmed by detailed interviews in 65%, 69% and 49% of subjects in those respective categories. Except for the unwell fatigued group, these levels of agreement between household informants and self-reports were similar to the 66% agreement detected in a study designed to examine concordance between self-reported health conditions and proxy information among adults => 65 years of age in the United Kingdom [23
]. This degree of agreement was considered reasonable. The lower level of agreement detected among the unwell fatigued may be due to the relative nonspecificity of fatigue, compared to other symptoms of unwellness, including problems with memory/concentration, unrefreshing sleep, or muscle/joint pain.
Third, the study was conducted in standard 8th
grade-level English, which may have led to an under-sampling of ethnic minority groups (e.g., Spanish speaking). Prevalence of CFS was unusually high among the metropolitan and rural Hispanic populations and unusually low among rural Hispanic residents. This may reflect important ethnic differences in risk [2
] and we weighted the sample to allow for ethnic differences. Most likely it reflects language-related misunderstanding of the questions and weighting cannot address this. It will be important to further evaluate the occurrence of and risk factors for CFS in English and non-English speaking metropolitan, urban, and rural Hispanic populations [24
Fourth, the study utilized Atlanta (Fulton and DeKalb counties) to represent metropolitan Georgia, Macon and Warner Robins to represent urban Georgia, and the counties surrounding the urban area to represent rural Georgia. We did this for logistic reasons, and the results cannot a-priori be generalized to other populations. Indeed, populations of the 10 rural county seats varied from 587 to 19,000 (median 2,000), so several exceeded the Census Bureau definition of rural.
Finally, in spite of a rigorous case definition, CFS has no diagnostic markers or characteristic physical signs. Thus, CFS diagnosis is based on self-reported symptoms, disability and exclusion of known diseases. Therefore, potential misclassification of study subjects remains a concern, as CFS is clinically heterogeneous and likely represents more than one entity [25