Sleep characteristics of persons with chronic fatigue syndrome and non-fatigued controls: results from a population-based study

This study adhered to U.S. Department of Health and Human Services human experimentation guidelines and received Institutional Review Board approval from the CDC and collaborating institutions. All participants gave informed consent.

Between January and July 2003, we conducted a 2-day in-hospital study of adults identified with CFS from the general population of Wichita [19]. The in-hospital study enrolled people who had participated in the 1997 through 2000 Wichita Population-Based CFS Surveillance Study [4]. Participants in the in-hospital study were fatigued adults with medically/psychiatrically unexplained chronic fatigue identified during the surveillance study. Fifty-eight had been diagnosed at least once with CFS and 59 had unexplained chronic fatigue that was not CFS. Controls were randomly selected from the cohort who participated throughout surveillance, who did not have medical or psychiatric exclusions, and who had not reported fatigue of at least 1-month duration; they were matched to CFS cases on sex, age, race/ethnicity, and body mass index. Upon admission to this study, subjects were reevaluated for CFS symptoms and exclusionary medical and psychiatric conditions (discussed below). The 43 who, at the time of the in-hospital study, met 1994 criteria for CFS (discussed below) comprise the cases in this report. Controls are 43 individuals who had never reported fatigue during the surveillance study, who were not fatigued at the time of this in-hospital study and who had no exclusionary medical or psychiatric condition identified at the time of study (following section). Because current classification of CFS was not completely in accord with recruitment classification, strict matching was not maintained, though cases and controls were demographically comparable. Thirty-six (84%) of the 43 with CFS and 38 (88%) of the 43 controls were women; most (40 CFS and 42 controls) were white; their mean ages were 50.6 and 50.3 years, respectively; and body mass index was 29.4 and 29.3, respectively.

Subjects who agreed to participate were admitted to a Wichita hospital research unit for 2 days. Subjects brought all their current medications so that clinic staff could record this data and maintain medication profiles throughout the study. To identify medical conditions specified by the case definition as exclusionary for CFS [1, 11], participants provided a standardized past medical history, a review of current medications, underwent a standardized physical examination, and provided blood and urine for routine analysis [1, 11]. To identify psychiatric conditions exclusionary for CFS, licensed and specifically trained psychiatric interviewers administered the Diagnostic Interview Schedule for current Axis I disorders [20]. Exclusionary psychiatric illnesses specified by the case definition were current melancholic depression, current and lifetime bipolar disorder or psychosis, substance abuse within 2 years and eating disorders within 5 years. A panel of physicians and psychiatrists/psychologists reviewed this information and identified subjects with disorders exclusionary for the diagnosis of CFS. Subjects with no exclusionary conditions were considered to be CFS if they met empirically measured parameters [19] of the 1994 CFS case definition [1]. Non-fatigued controls met none of the parameters.

As noted, clinic staff reviewed all current (prescription and over the counter) medications that study participants were taking. Study investigators (DBR, MJD, CH, JFJ, WCR), and other Emory University Department of Psychiatry and Behavioral Sciences collaborators, reviewed all medications and classified them as affecting (inducing sleep, inhibiting sleep or with mixed effects) or not affecting sleep. Those classified as affecting sleep included analgesics (e.g., hydrocodone, Lortab, oxycodone, Propoxyphene), antidepressants (e.g., Celexa™, amitriptyline, imipramine, Lexapro™, Wellbutrin™, Effexor, Prozac™, Zoloft™, Paxil™, fluoxetine), antianxiety (Alprazolam), antihistamines (e.g., diphenhydramine, chlorpheneramine, benadryl, promethazine), decongestants (e.g., pseudoephedrine, guaifenesen), anticonvulsants (e.g., Topamax, Neurotin, clonazepam), anti-sleep phase disorder (melatonin), blood pressure controlling (e.g., Clonidine, Proamatine), antipsychotics (e.g., Seroquel, Zyprexa, Fluvoxamine), stimulants (e.g., methylphenidate, Provigil), peristaltic stimulants (Metoclopramide), and muscle relaxants (cyclobenzaprine). Medications affecting sleep were handled as a binary measure (i.e., they used or did not use one or more of those named above). Analyses took into account use of sleep affecting medications, as noted below.

Nocturnal polysomnography and daytime multiple sleep latency testing (MSLT) were conducted in a 4-bed laboratory established at Wesley Medical Center, Wichita, KS, and consisted of polysomnography on night #1, MSLT the following day and another polysomnography on night #2. Patients were asked to arrive 3 hours before their typical bedtime on Night 1 to allow adequate time for electrode application and standard biocalibrations. "Lights out" and "lights on" time were 22:00 and 07:00, respectively. The daytime MSLT testing schedule was adjusted for other measures being collected; MSLT began at 11:00 and consisted of three additional naps at 13:00, 15:00, and 17:00.

Electrophysiological measures of wakefulness and sleep were acquired and recorded with the Flaga/Medcare N7000 digital polysomnographic system on a Windows XP platform using proprietary software (Flaga/Medcare Somnologica Studio). We employed a sampling rate of 256 Hz to allow for Fast Fourier Transform of EEG signals. Standard gold cup electrodes were employed for recording of EEG, EOG, and EMG for sleep staging and appreciation of sleep architecture. Respiration was measured with inductance plethysmography-like belts placed around the chest and abdomen. A pressure transducer, positioned in close approximation to the nares provided indices of airflow. A pulse oximeter probe was applied to either the right or left index finger, to measure arterial oxygen saturation (Sa02). Electrocardiogram (ECG) was recorded with standard snap electrodes (NeuroSupplies, Waterford, CT). The following signals were recorded: central (C₃-A₂//C₄-A₁) and occipital (O₁-A₁//O₂-A₂) EEG, left and right monopolar EOG, surface mentalis EMG, ECG (modified V3), respiratory airflow and effort and surface EMG from the right and left anterior tibialis.

The polysomnographic outcome variables used in our analyses included: total sleep time (TST) (in minutes), sleep efficiency (% of time spent in bed asleep), the percentage of TST spent in non-Rem (NREM) and REM sleep, sleep latency (in minutes) to three consecutive epochs of sleep, and REM Latency, defined as the time between the first epoch of any stage of sleep and the first epoch of REM-sleep. Brief arousals were scored following criteria set forth by the American Academy of Sleep Medicine, and the number of arousals expressed as a rate per hour of sleep adjusted for TST. Periodic leg movements both with and without accompanying arousals, were scored according to conventional criteria [22], and expressed as an index of the rate of leg movements per hour of sleep, and a separately derived index of those accompanied by an American Academy of Sleep Medicine -defined arousal [23].

Daytime sleepiness was measured with the MSLT, which has demonstrated objective sensitivity to the effects of sleep deprivation, sleep fragmentation, sleep restriction, insufficient sleep, hypersomnia, and in disease states such as sleep apnea and narcolepsy [24‐26]. Multiple sleep latency tests were performed and scored according to standard guidelines with the exception that four naps were recorded at 11:00, 13:00, 15:00, and 17:00. The mean sleep latency on the MSLT was defined as the mean time from lights out to the first 30-second epoch scored as sleep. A sleep onset REM was defined as one or more epochs of REM sleep occurring within 15 minutes of the first epoch scored as sleep. Mean MSLT values of 5 or less are considered to represent pathological sleepiness, scores between 5–10 are consistent with a degree of daytime sleepiness. Scores of 10 and above are considered normative and believed to denote a lack of daytime sleepiness. Because mean values on the MSLT may adversely be affected by a spurious sleep latency on a single nap opportunity [27] possibly due to what might be perceived as stressful inter-nap activities [28], median values were also computed for each subject.

Polysomnography data were scored by an Emory University registered polysomnology technologist (blinded as to subjects' fatigue classification). An Emory University Department of Neurology American Board of Sleep Medicine certified physician (DBR), also blinded to the subjects' fatigue classifications, interpreted results. The polysomnology technologist manually scored each recording in 30 second epochs as wake, NREM, Stages 1–4 sleep, or rapid eye movement (REM) sleep. Criteria for scoring respiratory variables were based upon those of the Sleep Heart Health Study [21]. Briefly, apnea was scored if airflow decreased to less than or equal to 25% of the immediately preceding baseline for a period of at least 10 seconds. Hypopnea was scored if either airflow or thoracic-abdominal excursion decreased by at least 30% of baseline, for at least 10 seconds, with a concomitant reduction in SaO2 of 4% or greater. The Respiratory Distress Index (apneas + hypopneas corrected for hour of sleep) was derived from these scored events. To determine the technologist's level of reproducibility, 12 randomly selected studies were scored twice, with at least a six-week interval separating the original scoring and the repeat scoring. Comparison between original and repeat scorings with Kappa analyses yielded a Kappa coefficient of 0.88.

The CFS case definition specifies sleep apnea and narcolepsy as conditions that could explain fatigue and symptoms of CFS and therefore exclude the diagnosis of CFS [1, 11].

Data were analyzed by Systat (Systat Software Inc, Richmond, CA) and SAS (SAS Institute Inc, Cary, NC). we used A 2-factor analysis of variance using PROC GLM was to measure the associations between case status and medication use (yes/no) with polysomnographic variables. Log transformed values of polysomnographic variables were used when necessary to satisfy the assumption of normally distributed outcomes. Mean values for each polysomnographic variable were adjusted for medication use using the least square method (LSMEANS). All mean values presented in this paper represent arithmetic means. We also used standard and exact logistic regression models to compute odds ratios as estimates of relative risks and 95% confidence intervals for CFS associated with dichotomous polysomnographic variables (cut-offs based on 25^th or 75^th percentiles). Measurement of clinical sleep variables included a high number of zero values. Zero-inflated Poisson Regression was used to regress case status and medication use (yes/no) on continuous values of clinical sleep variables. For this final analysis, we used SAS version 9.0 (PROC NLMIXED) and an inflation probability determined by the regressors. Analyses were also performed excluding participants taking medications that affect sleep. Estimates were unchanged when analyses excluded participants taking sleep-affecting medications. For this reason, the results presented in this report do not exclude participants taking such medications, but rather adjust for them in the analysis. We used the χ² statistic or Fisher's exact test to evaluate associations between CFS and dichotomous variables.

Eleven study participants had sleep disorders exclusionary for CFS (obstructive apnea n = 8, narcolepsy n = 3). Persons with CFS had a higher frequency of exclusionary sleep disorders (8 of 43, 18.6%) than non-fatigued controls (3 of 43, 7%), but the difference was not statistically significant (p = .16). The remainder of this presentation considered the remaining 35 individuals with CFS and 40 controls.

There were no statistically significant differences in standard polysomnographic measurements between those with CFS and controls on either night 1 or night 2. As expected, total sleep time increased in both groups between nights 1 and 2; latency to sleep onset and to REM onset decreased in both groups; and, waking after sleep onset was less common in both groups on night 2 (data not shown). As night 1 was considered to be an adaptation night, data is shown for night 2 (Table 1). Mean values, adjusted for medication use, did not differ between participants with CFS and non-fatigued controls. Interestingly, regardless of case status, medication use was independently associated with both REM latency and Stage 1 % total sleep time. Both REM latency and Stage-1 percent of total sleep time were significantly longer in study participants who reported using any sleep medications at the time of the study (P = .02, P = .01, respectively). In addition to comparing polysomnographic measurements as continuous variables between people with CFS and non-fatigued controls, we used regular and exact logistic regression to examine possible associations. We dichotomized measurements based on 25^th and 75^th percentiles among non-fatigued controls. As with the previous analyses there were no significant differences (data not shown). Virtually identical results were obtained when analysis was restricted to subjects who did not use sleep-altering medications. Finally, evaluation of multiple sleep latency testing studies yielded similar distributions of classifications; (39% normal, 35% borderline and 26% abnormal) between CFS and controls.

Subjects with CFS had significantly more episodes of obstructive apnea and a higher Respiratory Distress Index than did the non-fatigued controls (Table 2). Nonetheless, the difference between the groups in mean obstructive episodes per hour of total sleep was not of a magnitude recognized to have a clinical impact. All other measures of disordered breathing and periodic leg movements were not different between the two groups. Use of medications affecting sleep was independently associated with a higher rate of hypopnea and leg movements episodes per hour, after adjusting for case status (P = .03, P = .05, respectively). However, use of sleep altering medication was associated with a lower rate of obstructive apneic episodes per hour (P <.0001).