The CDC Institutional Review Board, as required by Department of Health and Human Services regulations, approved the study. All participants were volunteers who gave informed consent. The study was conducted in English. Non-English speaking respondents were not included.
Metropolitan, urban, rural
The definitions of metropolitan, urban and rural geographic strata are complex. The U.S. Office of Management and Budget defines several categories of metropolitan statistical areas according to specific standards. In general, metropolitan areas contain at least a million residents living in a core area (i.e., central city), together with adjacent communities that have a high degree of economic and social integration with that core. Atlanta, with approximately 4 million residents, is a metropolitan city. The Census Bureau defines urban and rural areas independently of OMB's classification. Urban and rural can occur inside of and outside of metropolitan areas. Typically, settled areas of 2,500 or more are considered to be urban and the remainder rural. Based on these definitions, we determined that the cities of Macon and Warner Robins (with populations of 300,000 and 48,000, respectively) were urban. For this study, we considered the counties surrounding Macon and Warner Robins to be rural.
Survey in general
The survey was conducted between September 2004 and July 2005. It included residents of three areas of Georgia: metropolitan (Atlanta – Fulton and DeKalb counties), urban (Macon – Bibb County and Warner Robins in adjacent Houston County), and rural (10 counties surrounding Bibb County – Houston -excluding Warner Robins, Baldwin, Bleckley, Crawford, Jones, Macon, Monroe, Peach, Twiggs, and Wilkinson). The survey used the same strategy as previously reported CDC population surveys of CFS [7, 6]. We used list-assisted random-digit dialing  and an advance letter  to contact households containing persons aged 18–59 years in the three population strata.
Telephone screening interviews
In contrast to our previous studies, which screened households for fatigue, in this study we modified the initial screening interview to cover a broader range of CFS defining symptoms. In brief, the screening interview asked a household informant (=> 18 years) to report the age, sex, ethnicity and health status of each household member aged 18 and older and to identify unwell household members, who the informant noticed to have at least one of the CFS defining symptoms (fatigue, cognitive impairment, unrefreshing sleep, muscle pain, joint pain, sore throat, tender lymph nodes, or headache) for ≥ 1 month, and well residents, who had none of these problems for => 1 month.
Detailed telephone interviews
Household residents between 18 and 59 years of age who were identified by the informant as unwell with fatigue, randomly selected persons identified as unwell without fatigue (i.e., identified with cognitive impairment, unrefreshing sleep, muscle pain, joint pain, sore throat, tender lymph nodes, or headache), and a random sample of people identified as well were asked to complete a detailed telephone interview. The detailed interview covered fatigue status and duration, other symptoms, race ("What race do you consider yourself to be? Please note that you may choose more than one option. White, black of African American, Asian, American Indian or Alaskan Native, native Hawaiian or other Pacific islander"), self-identified Hispanic/Latino or Spanish origin or descent, other demographic characteristics, and medical history. Based on their responses to the detailed interview, respondents were classified as: 1) having a medical or psychiatric condition considered exclusionary for CFS ; 2) having CFS-like illness if they reported severe fatigue lasting 6 months or longer that was not alleviated by rest, that caused substantial reduction in occupational, educational, social or personal activities, and that was accompanied by at least 4 of the CFS case defining symptoms ; 3) being chronically unwell (reporting any of the CFS defining symptoms) with or without fatigue);, 4) or being well.
All respondents between 18 and 59 years who had no exclusionary conditions per interview, and were classified as having a CFS-like illness were invited for a one-day clinical examination to further investigate exclusionary medical and psychiatric conditions. We also invited a similar number of randomly selected participants identified with chronic unwellness(at least six months of unwellness with or without fatigue but not CFS-like). Finally, we invited well participants, matched to the CFS-like on geographic stratum, sex, race/ethnicity and age (within 3 years), for a one-day clinical examination. Those from the urban or rural areas attended a clinic in Macon, and those from Atlanta attended a similar clinic in Atlanta. No more than 4 participants attended a clinic on any day, and appointments were staggered for optimal flow. Clinic staff with responsibilities for examinations was not aware of participants' telephone interview responses or classification. The authors and CDC CFS Research Program staff attended clinics on a regular basis to assess operations.
Study participants who underwent a detailed telephone interview and met criteria of the 1994 CFS case definition  were classified as CFS-like. In brief, criteria for classification as CFS-like on telephone included persistent or relapsing fatigue of at least 6 months' duration; the fatigue was not relieved by rest and caused substantial reduction in previous levels of occupational, educational, social, or personal activities. Exclusionary conditions included self-reported medical or psychiatric conditions that could cause the fatigue. Finally, the medically/psychiatrically unexplained fatigue must have been accompanied by at least 4 of the 8 CFS case defining symptoms : 1) unusual post-exertional malaise; 2) unrefreshing sleep; 3) impaired memory or concentration; 4) headaches; 5) muscle pain; 6) multi-joint paint without swelling or redness; 7) sore throat; 8) tender cervical/axillary lymph nodes. CFS-like subjects differ from CFS by not having been evaluated clinically in the study.
The objective of the clinical evaluation was to classify participants' clinical status and diagnose exclusionary medical and psychiatric conditions. As recommended by the International CFS Study Group , participants were classified as CFS, unexplained chronic illness not meeting criteria for CFS (termed ISF), or well by using standardized reproducible criteria for measuring specifics of the 1994 case definition . We used the Multidimensional Fatigue Inventory (MFI)  to assess fatigue status. For classification as CFS, those with a score => well-population medians on the general fatigue or reduced activity scales of the MFI were considered to meet fatigue criteria of the 1994 case definition. Functional impairment was assessed by the medical outcomes survey short form-36 (SF-36) . For classification as CFS, those with a score =< 25th percentile of population norms in the physical function or role physical, or social function, or role emotional subscales of the SF-36 were considered to have substantial reduction in activities as specified in the 1994 definition. Finally we used the CDC Symptom Inventory (SI)  to evaluate occurrence, frequency and severity of the 8 CFS-defining accompanying symptoms. The SF-36, MFI and SI domain scores require complete data for the subscales. We imputed a zero value in the case of one-item non-response for subscales contributing to the relevant domains. For classification as CFS, those reporting => 4 case defining symptoms and who scored > 25 on the SI concerning frequency and severity of the 8 case defining symptoms  were considered to meet accompanying symptom criteria of the 1994 case definition. Participants who fulfilled some, but not all of these criteria were classified as ISF. Those who met none of the criteria were considered to be well. The MFI, SF-36 and SI are self-administered paper and pencil forms. A trained clinic supervisor reviewed forms and helped subjects complete missing or misunderstood portions.
To screen for medical conditions considered exclusionary for CFS [11, 7], participants completed past medical history questionnaires and were requested to bring all their medications and supplements to the clinic. A licensed nurse practitioner or physician assistant reviewed subjects' past medical histories and medications to clarify omissions or discrepancies and also catalogued all medications. Relevant information was brought to the attention of the study physicians. A specifically trained licensed physician then performed a standardized physical examination . The examination was expanded if there were any concerns. The examiner recorded a differential diagnosis. Blood and urine specimens were obtained for laboratory screening tests to identify possible underlying or contributing medical conditions as stipulated by the case definition [11, 7]. Laboratory tests included a complete blood count with differential, c-reactive protein, alanine aminotransferase (ALT), SGPT, albumin, alkaline phosphatase, asparatate aminotransferase (AST), SGOT, total bilirubin, calcium, carbon dioxide, chloride, creatinine, glucose, potassium, total protein, sodium, urea nitrogen BUN, antinuclear antibodies, rheumatoid factor, TSH, free T4, and urinalysis.
To screen for psychiatric conditions considered exclusionary for CFS [11, 7], a trained and experienced licensed psychiatric social worker, clinical psychologist, psychiatric nurse practitioner or certified psychiatric research nurse administered the research version of the SCID . They underwent specific training for the SCID. Psychologists on the CDC CFS Research Program monitored their technique on a regular basis. The SCID included the screening module, mood episodes, psychotic symptoms, psychotic disorders, mood disorders, substance use disorders, anxiety disorders, somatoform disorders, eating disorders, and adjustment disorders.
A review committee of CDC and Emory University physicians and psychologists reviewed medical and psychiatric evaluations to determine the presence of medical and psychiatric conditions exclusionary for CFS. Members of the review committee were not aware of subjects' classification either on phone interview or in the clinic.
Prevalence estimates and statistical analyses utilized weighted data. The survey weights maintained (through the stages of the survey) the relation between the sample and the population in each geographic stratum, and they included several adjustments that are customarily employed to reduce bias. In the process of developing weights, one step adjusted for households that did not have telephones. To estimate the proportion of households that did not have a telephone in each of the three geographic strata, we analyzed data from the 5% public-use microdata samples (PUMS) of the 2000 Census. For the metropolitan stratum, the analysis used PUMS data from De Kalb and Fulton Counties. The other two strata, however, do not correspond exactly to geographic entities for which data are available in the PUMS. Thus, the analysis for the urban stratum used data from Bibb County; the rural stratum used data from a larger combination of counties that contained the counties of that stratum. The resulting estimates of the proportion of households that did not have a telephone were 1.68% in the metropolitan stratum, 3.66% in the urban stratum, and 6.35% in the rural stratum.
Adjustments for nonresponse on the detailed telephone interview and nonresponse on the clinical evaluation kept the categories of illness separate; thus, to the extent possible, respondents accounted for nonrespondents who belonged to the same illness category (and shared other key characteristics). The adjustment factor, applied to the weight of each respondent, equaled the ratio of the sum of the weights (at that stage) of respondents and nonrespondents to the sum of the weights of respondents. Another adjustment, at the household level, used data on interruptions in telephone service to compensate for the inability of the telephone survey to reach households that did not have telephone service.
Households completing the screening interview received a base sampling weight (64.4 in the metropolitan stratum, 6.4 in the urban stratum, and 4.9 in the rural Stratum) equal to the reciprocal of the probability that the household's telephone number was selected for the sample. Base weights were reduced for multiple residential telephone numbers in the household (either by a factor of 2 or a factor of 3) and adjusted for households that did not complete screening interviews (by a factor of 1.03 in each stratum), for numbers associated with undetermined residential status (by a factor of 2.4, 2.3, and 2.1, respectively), and for non-telephone households in the population (by a factor of 1.7, 2.2, and 2.3 for households that reported interruptions in telephone service and by a factor of 1.24, 1.38, and 1.04 for households that did not report interruptions) [16, 17]. (The household weights ranged from 66 to 268 in the metropolitan stratum, from 7 to 34 in the urban stratum, and from 4 to 24 in the rural stratum.)
Subjects selected for detailed interviews received an initial interview weight equal to their household weight multiplied by the reciprocal of their probability of selection (the probability of selecting a household as a source of a subject ranged from 0.32 to 1.0 for unwell subjects and from 0.17 to 1.0 for well subjects; the probability of selecting an individual subject within the household was the reciprocal of the number of unwell, respectively well, persons in the household; subjects with prolonged fatigue were selected with certainty). Within each combination of stratum (metropolitan, urban, rural) and illness classification (fatigued, unwell, well), initial weights were adjusted for nonresponse on the detailed interview, within a total of 195 cells defined by sex, age, and race (the adjustment factors ranged from 1.05 to 2.25 and exceeded 2.0 in only 22 cells). A further adjustment in each stratum used an iterative form of post-stratification to bring the weighted totals into agreement with control totals from the 2000 Census on race and on the combination of sex and age. This process produced an interview weight for each subject who completed a detailed interview. (The interview weights ranged from 84 to 16,723 in the metropolitan stratum, from 5 to 822 in the urban stratum, and from 4 to 892 in the rural stratum.)
Each CFS-like subject who completed a clinical evaluation received a clinical-evaluation weight, which incorporated an adjustment for nonresponse on the clinical evaluation within stratum-specific cells defined by sex and age (over the 19 cells the adjustment factor ranged from 1.11 to 2.71). For chronically unwell subjects the clinical-evaluation weights incorporated a parallel adjustment for nonresponse (over the 14 cells the adjustment factor ranged from 1.52 to 2.61), preceded by an adjustment for selection of the subsample (by a factor of 3.02, 3.88, and 2.89 in the respective strata). (The clinical-evaluation weights ranged from 193 to 32,354 in the metropolitan stratum, from 8 to 1,787 in the urban stratum, and from 10 to 1,358 in the rural stratum.) Because they were selected for clinical evaluation only as a result of being matched to a CFS-like subject, well subjects did not have their own clinical-evaluation weight.
Within each stratum, prevalence was estimated using SUDAAN (SUDAAN: Research Triangle Institute, Research Triangle Park, NC)  software to calculate weighted percentages and obtain standard errors that reflected the sample design and survey weights. Prevalence estimates based on illness classifications derived from the detailed interviews used the data of all subjects who completed detailed interviews, and their respective interview weights.
In order to maintain the relation to the population, prevalence estimates based on illness classifications derived from clinical data used a combination of clinical data and interview data. For CFS-like and chronically unwell subjects who completed clinical evaluations, the data obtained from the clinical evaluations were used along with their clinical-evaluation weights. For subjects classified as CFS-like and chronically unwell who were not eligible for clinical evaluations, and also for all well subjects, the data obtained from detailed interviews were used along with their interview weights.
Because the matching process does not preserve a sampling-based connection with the population, clinical data from well subjects were not used in calculating prevalence estimates.
Weighted χ2 tests in SUDAAN were used to compare proportions of subjects diagnosed with CFS by demographic categories. P-values were calculated to evaluate the statistical significance of differences in CFS prevalence by age, sex, race, ethnicity, education and household income.