Background
Restless Legs Syndrome (RLS) is characterized by unpleasant sensations of tingling, creeping, bubbling, or “tunneling bugs” [
1]. The International RLS Study Group established diagnostic criteria for RLS, including: a strong urge to move, predominantly in the legs; relief with movement, stretching, or rubbing; and symptomatic worsening at night or with inactivity [
2]. RLS is a common chronic condition, with prevalence estimates ranging from 5-10% [
3,
4], with some geographic variability [
3,
5‐
7]. Many RLS cases are secondary to other causes, such as iron deficiency, pregnancy, and neurologic, rheumatologic, or renal diseases, but more than 60% of cases are idiopathic [
8,
9]. Primary RLS prevalence in the U.S. is likely 1-3% [
9], and is likely more common in women and during the seventh and eighth decades [
5]. RLS is clinically diagnosed but can be confirmed by polysomnography or the Suggested Immobilization Test [
8], and no confirmatory laboratory tests exist [
10]. Many RLS sufferers seek care but the average time to diagnosis is two years [
11], and the RLS Epidemiology, Symptoms, and Treatment (REST) study confirmed that many are misdiagnosed and given inappropriate treatments [
3].
RLS also impacts health-related quality of life [
12‐
14]. Like those with other chronic conditions, RLS sufferers have lower scores on the Medical Outcome Study Short Form 36 (SF-36) in both clinical [
15] and population-based samples [
5,
12]. RLS sufferers may be less productive at work [
9]. RLS sufferers report negative influences on mood, energy, and daily activities [
5], and are shown to have more daytime fatigue, work difficulties, and driving impairment [
4]. Among elders, severe RLS has been associated with poorer social function, daily function, sleep quality, and emotional well-being [
16]. However, the full impact RLS has on disability has yet to be shown.
The Health and Retirement Study (HRS) is a nationally representative cohort of economic and physical health of elders [
17]. Since 1992, HRS has provided comprehensive and detailed information on many domains, including health status, employment, disability, and net worth of U.S. elders. A sub-sample was queried about RLS symptoms in 2002, providing an opportunity to report epidemiologic correlates of RLS and, more importantly, longitudinal associations between RLS and subsequent functional limitations.
Methods
The Health and Retirement Study is an ongoing cohort study consisting of a representative sample of over 20,000 Americans born prior to 1948, with interview data collected biennially on demographics, health status, employment, income and wealth, and insurance; full details are described elsewhere [
17]. In 2002, the overall response rate of the cohort was 86.9% [
17]. Sub-samples (“modules”) are asked additional questions on ancillary topics after the main survey [
18]. The module topic is revealed after respondents consent. Of 1,502 randomly selected respondents, 1,058 (70.4%) agreed to complete the 2002 RLS module. Nursing home residents and subjects outside the eligible age-range for HRS (e.g. spouses of age-eligible respondents) were excluded from the analysis (n = 50), yielding a final sample size of 1,008. HRS was reviewed by the University of Michigan Institutional Review Board [
17].
The 2002 RLS module predated validated instruments for measuring RLS outside of a clinical setting [
2]. Respondents reported presence of resting symptoms, including “crawling, tingling or achy sensations in your arms or legs”, feeling “restless, fidgety, or unable to sit still”, “feel the need to move your legs, rub your legs, or stretch your legs”, “you or bedpartner noticed twitching or kicking of your arms and legs”, or “itching sensations anywhere on your body.” Respondents reported how frequently they experienced “unpleasant feelings” in the legs – “for example, creepy-crawling or tingly feelings—that make you feel restless and keep you from getting a good night’s sleep”, with options ranging from “never” to “every night”. Subjects were also asked how frequently they “get pain or cramps in [the] legs to the point where it is uncomfortable and disturbs sleep”. Symptomatic individuals were asked if they sought medical advice, whether treatment was offered, and whether treatment helped.
Potential health correlates were derived from the parent HRS study, and full questionnaires are available at
http://hrsonline.isr.umich.edu[
17]. Subjects self-reported demographic information, including race, gender, date of birth, and educational attainment, upon study entry. Health status is evaluated with each interview wave, including medication use, chronic medical conditions (hypertension, diabetes, heart disease, stroke, lung disease, psychiatric disease, arthritis, and cancer), health behaviors, and symptoms, including general sleep concerns and pain. Bodily pain experience was categorized into three levels: severe (pain makes normal activities difficult), mild-moderate (troubled by pain but does not limit activity), or no troubling pain. Pain site was not localized. General sleep-related concerns include difficulty falling asleep, waking frequently at night, and waking early and unable to return to sleep. Depressive symptoms were assessed using a shortened version of the Center for Epidemiologic Studies Depression (CES-D) Scale [
19], with depressive symptoms noted for scores of two or more (corresponding to the 20
th most depressed percentile).
Smoking status was current, former, or never. Alcohol use was classified as current (yes/no), with additional quantification of drinking days per week and drinks per drinking day. Body mass index (BMI) was calculated using reported weight and height, and categorized as BMI < 25, 25 ≤ BMI < 30, or BMI ≥ 30 (18 missing values were classified at the median, 26.6). Self-reported health status was classified as “Excellent”, “Very Good”, “Good”, “Fair”, or “Poor”. For this analysis, the top two and the bottom two response categories were collapsed. Self-reported vigorous activity at baseline was defined as at least three vigorous activity sessions per week. In subsequent waves, frequency of physical activity was further broken into episodes of mild, moderate, or vigorous activity per week. Individuals reported paid employment outside the home and whether personal health limited their ability to work. Income was calculated from all sources, and divided into quartiles. Total medical expenses and out-of-pocket medication expenses were collected, imputed as necessary from the RAND datafiles, and divided into quartiles. Subjects self-reported falls, hospitalizations, and frequency of healthcare provider visits in the previous two years. For the analysis, falls were grouped as “None”, “One”, or “Two or more”. Frequency of provider visits were grouped as “0-3”, “4-7”, “8-11”, or “12 or more”. Self-reported medications for each of the above chronic conditions were queried, and for this analysis were categorized as “None”, “One”, “Two”, or “Three or more”.
Functional status was assessed using several standardized instruments. Variables were dichotomized to “at least some difficulty” versus no difficulty. Activities of daily living (ADL) assessed any difficulties on five tasks: bathing, eating, dressing, toileting, and transferring. Instrumental activities of daily living (IADL) included difficulties with: managing money, managing medications, preparing meals, going shopping, or using a telephone. Mobility limitations included limitations on walking and climbing stairs. Additional motor limitations were grouped based on a priori definitions. Gross motor limitations included any difficulty walking one block, climbing one flight of stairs, transferring, bathing, or requiring a walking assist device. Fine motor limitations included difficulty picking up a dime, feeding oneself, or dressing. Large muscle group limitations included sitting for two hours, rising from chair, stooping, or pushing a heavy object.
Statistical analysis
Analyses were performed using SAS, version 9.2 (SAS Institute, Inc., Cary, NC). RAND HRS analytic files, which provided processed data and derived variables [
17], were linked to raw questionnaire data. The operational definition of RLS required all three criteria: a) feeling restless, fidgety, or unable to sit still; b) the need to move your legs, rub your legs, or stretch your legs; and c) unpleasant nighttime sensations occurring at least once per week. Crude cross-sectional RLS associations with categorical demographic variables were tested using χ
2. Ordinal variables were evaluated for trend. Prevalence estimates were weighted for study sampling to reflect the U.S. non-institutionalized population born before 1947, reflecting the underlying HRS cohort.
Potential RLS correlates at baseline were explored by calculating multivariate-adjusted prevalence ratios (PR) derived from Cox proportional hazards regression with robust variance estimates [
20]. Dependent variables were modeled with a fixed time effect (uniformly set to 1). Categorical predictors were screened and considered potentially important if the likelihood ratio test was significant at p < 0.20, using forward variable selection and comparisons of the Akaike Information Criteria (AIC). Continuous independent variables were tested in various manners (linear, polynomial, ordinal, or nominal categorical), with selection of best fitting relationship based on AIC. Correlation among predictors was evaluated based on eigenvalues derived from principal component analysis of the predictor variables. Variables were selected so that that each variable contributed uniquely to total variance (i.e., condition indices associated with individual predictor variables were ≤ 30), while maximizing AIC [
21]. Additional model-building approaches, such as chunkwise variable selection [
22], did not significantly alter the final model.
RLS was then considered as an independent variable, adjusting for significant correlates from the RLS prediction model, to predict concurrent functional limitations. Disability variables were modeled as above to calculate multivariate-adjusted PR. Multi-domain measures (e.g., ADL, IADL) were positive if any component showed limitations. Next, using longitudinal disability data from three subsequent interview waves (2004, 2006, and 2008), RLS was modeled as a predictor of incident functional limitations using Cox proportional hazards regression. Hazard ratios (HR) were reported for each incident disability, limited to the sample at risk (i.e., those without the specific functional limitation in 2002). Censoring occurred at the end of follow-up, death, voluntary withdrawal from the study, or first occurrence study ineligibility (i.e., living in an institutional setting or no longer being married to a spouse in the study cohort). Adjustment for pain and medication use at baseline had the greatest effect on the point estimates. Repeated measures were considered time-varying covariates in the analysis.
Discussion
This population-based sample of U.S. elders confirms that restless legs syndrome is common, with 10.6% of our sample meeting the operational definition of weekly RLS symptoms, and potentially disabling. The prevalence estimates are consistent with other reports, although we were unable to classify patients with primary versus secondary RLS. The multinational RLS Epidemiology, Symptoms, and Treatment (REST) study found annual period prevalence of 7.2%, although 5% had weekly symptoms [
5]. The 2005 National Sleep Foundation Poll’s estimate was 9.7% for symptoms several nights per week [
4]. The MEMO study, another population-based survey, reported 9.8% prevalence in elders in Germany [
7]. Secondary RLS can be very common, such as reports of 20% in hemodialysis patients [
10], although primary RLS is less common, estimated at 2.4% in the U.S. [
9].
Among those reporting RLS symptoms, fewer than half sought medical care. Among those who discuss their symptoms, fewer than half received treatment. However, most who received treatment reported symptom improvement. This pattern confirms reports from the REST study, suggesting delays in both diagnosis of RLS and initiation of appropriate therapies [
5]. The literature suggests RLS patients have good response to therapy in 90% of patients when correctly diagnosed [
2]. Evidence-based guidelines for RLS treatment now exist [
23], although one clinic-based German study found no clinical improvement with evidence-based guideline adherence [
24]. In the REST study, 81% of RLS patients with distressing symptoms discussed their symptoms with their physicians, but only 24% of these were given any diagnosis [
3,
5]. The majority of patients with RLS were diagnosed with circulation problems, arthritis, back or spine problems, varicose veins, depression, anxiety, or trapped nerves, but only 6% were correctly diagnosed with RLS [
5].
This study identifies associations between RLS and disability using valid functional limitation measures. RLS sufferers were 46% more likely to have ADL limitations at baseline, adjusted for covariates, and several ADL domains had doubled risk (dressing oneself, transferring, and toileting). RLS sufferers were also 71% more likely to have IADL limitations, including nearly three-fold increase in difficulty making meals by themselves. RLS sufferers were more likely to have concurrent mobility limitations (25% increase) and limitations requiring large muscle groups (20% increase), but did not have significant changes in fine or gross motor functions. However, some specific limitations, such as lifting their arms, pushing heavy weights, and rising from a chair, were significantly increased. Associations with mobility measures were more prominent in more strenuous tasks (multiple blocks or multiple stair flights).
This study is the first population-based study to longitudinally assess functional measures among adults with RLS. Despite adjustment for all baseline and time-dependent covariates, specific functional limitations were significantly increased among subjects with RLS. RLS suffers were more likely to develop specific incident disabilities, which involving the core, lower extremities, and upper extremities, such as difficulty climbing stairs (HR = 2.38), difficulty stooping (HR = 2.66), difficulty rising from a chair (HR = 2.54), difficulty sitting for 2 hours (HR = 2.17), moving large objects (HR = 1.79), raising arms (HR = 1.76), or picking up a dime (HR = 1.97). However, no aggregate measures (ADL, IADL, gross motor, fine motor, mobility, or large muscle group limitations) reached statistical significance during follow-up, after adjustment for baseline characteristics (particularly pain and chronic medication use). If pain mediates how RLS affects disability, this may reflect over-adjustment. Also, excluding disabled patients at baseline markedly decreased the population at-risk for new disability, limiting the power of the study. We did not model transitions from mild to more severe disability.
The difficulty with prolonged sitting follows with the expected perturbation of RLS symptoms during periods of inactivity. Sleep symptoms (frequent waking or waking early without being able to fall asleep) were significantly associated with RLS. We adjusted for these sleep symptoms when analyzing the potential impact of RLS on disability, although symptoms interfering with sleep quality may be an intermediate factors between RLS and disability. Multiple studies have considered role limitations and quality of life. RLS patients have been shown to have decreased scores on the Medical Outcomes Study 36-Item Short Form (SF-36) in all tested domains [
5,
15]. The National Sleep Foundation (NSF) poll noted that RLS sufferers were more likely to be unemployed and to have difficulties at work [
4]. We confirmed that employed elders outside of the home were less likely to have RLS (age- and sex- adjusted PR 0.38, 95% CI 0.21-0.68, although not significant in multivariate analysis), and those who report that there health limited their ability to work were three times more likely to have RLS. The REST study reported that RLS sufferers reported daytime sleepiness (32%), difficulty concentrating (15%), and negative influences on daily activities and personal and work life [
5]. The NSF poll also found evidence of daytime symptoms in patients with RLS [
4]. These findings were replicated in a geriatric sample, with poor sleep quality, daytime somnolence, and low social functioning due to impaired tolerance for inactivity [
16]. One Scandinavian study showed similar impact on sleep quality, but noted lower SF-12 physical domain scores but not mental domain scores [
6]. An analysis of health-related quality of life (HRQOL) in RLS patients in multiple sites in Germany suggested that sleep deficits, duration of symptoms, and household income mediated RLS associates with clinical rating scales, with RLS effects consistent with other chronic neurologic diseases [
13]. A Swedish study, limited to women, suggested statistically and clinically significant deviations in HRQOL although mental domains predominated [
14]. When HRQOL was assessed in the REST study, McKrink et al. were able to estimate factors related to RLS diagnosis, severity, and treatment using SF-36 as predictors of the domains, but no comparison to patients without RLS was possible [
12]. One recent small study of RLS patients showed substantial cognitive deficits similar to those deprived of sleep for one night, despite assessment at a time when RLS symptoms were not active [
25]. We did not find concomitant cognitive dysfunction associated with RLS (results not shown).
Our risk factor analysis yielded results consistent with other studies. Women were nearly twice as likely to have RLS based on age-adjusted estimates; however, this difference was reduced by half when adjusted for other covariates, such as pain syndromes, alcohol consumption, BMI, age, medication use, and health care utilization, suggesting that gender differences in RLS prevalence may be partially due to confounding. There was no evidence of significant effect modification by gender, although there was a relatively small number of males with RLS (n = 26). Although RLS is thought to be more common in the elderly, the prevalence of RLS trended downward with age in our sample. Several studies report increased prevalence with age, peaking around age 65 [
4,
5], consistent with our reference group, which was those aged 50–64.
RLS is known to have a strong hereditary component [
8] but may be associated with acquired factors such as increased BMI, inactivity, and low alcohol consumption [
26]. Vigorous activity was crudely associated with RLS but the association did not persist in multivariate modeling. We observed that those with some alcohol consumption tended to have decreased prevalence of RLS, similar to Phillips and colleagues [
26], despite modeling alcohol consumption in various ways. There were no independent effects for “binge drinking”, frequency of drinking, or amount per session. Cigarette smoking associations have been inconsistent [
4,
6,
7,
26,
27] and smoking status was not associated with RLS in this sample. Crude associations with arthritis, hypertension, diabetes, lung disease, and heart disease were observed, but adjustment for general health status and symptomatic variables were better predictors of RLS cross-sectionally. Although RLS may be secondary to conditions like COPD [
6], anemia, kidney disease, and diabetes, we did not observe persistent effects for medical comorbidities in this study. Psychiatric comorbidity or depressive symptoms also did not predict RLS after adjustment, despite reports that mood disorders are more common in RLS sufferers [
7,
28‐
30]. We did adjust for medication use and sleep-related symptoms, which may obscure this association. Similarity, self-reported health status had previously been noted as a predictor of RLS [
26], but did not retain significance in our study. RLS is common in patients with fibromyalgia [
31,
32], and sleep quality can be a determinant of HRQOL in these patients. Heavy medication use was associated with RLS, although we could not assess if this was causal or merely an indicator of multiple comorbidities.
Pain experience was a strong predictor of RLS, as was history of frequent falls. There may be overlap between RLS and chronic neuropathic pain pathophysiology, as both are mediated by the dopaminergic system [
33]. Neuropathic pain treatments are effective on RLS symptoms [
33], and non-opioid analgesic use predicted RLS among patients taking tricyclic and serotonin-reuptake-inhibiting antidepressants [
34]. Most RLS sufferers in that study had chronic recurrent pain, and nearly all used analgesics [
35]. RLS patients are more sensitive to pain, measured by pinprick hyperalgesia in the extremities [
36]. Furthermore, RLS patients showed worse bodily pain scores using SF-36 [
15]. Pain is not recognized in the standardized definition of RLS, but if pain is part of the clinical syndrome, increased pain may, at least partially, explain RLS associations with incident functional limitations. Our findings support the hypothesis that RLS sufferers may suffer from other chronic pain syndromes. Or, patients with neuropathic pain syndromes may some symptoms that mimic RLS [
9].
There are limitations to this work. First, the 2002 RLS module was part of the pre-existing HRS cohort and not a principle aim for the study, and data on RLS was collected at one time point and was without clinical evaluation. Risk factors and correlates, including functional limitations, were available both before and after this assessment. We did not differentiate between primary and secondary RLS. Standardized diagnostic criteria suggested by the International RLS Study Group [
2] were not strictly applied, but our case definition included the essence of the core diagnostic criteria at least weekly. Our case definition did not require the characteristics of movement providing symptom alleviation or that the symptoms worsen at night, which are considered diagnostic criteria for RLS. However, the questions did specify that the symptoms were present at rest. Our prevalence was consistent with other studies in elders, but extrapolation of our findings should be limited to those aged 54 or older. Although unmeasured confounding is always a concern, we also risk over-adjustment in multivariate modeling. Some crude estimates were markedly changed with control for pain and medication use. We do not know if pain is a mediating variable or part of the clinical syndrome rather than a true confounder. We also did not confirm chronic diseases as covariates with RLS as seen in other studies, but our general markers of health status and medication use better fit the data. We also could not assess for specific medication class effects, such as medications that cause akathisia (i.e., anti-psychotics and anti-depressants) [
37], which has clinical similarities to RLS. Some of the RLS treatments reported by subjects (e.g., “surgery or epidural”) could be potentially disabling. Although we could detect incident functional limitations, we excluded subjects with specific disabilities at baseline. For the most common disabilities, this reduced the sample size considered at risk, which also decreased the power to detect differences. Furthermore, as RLS was only assessed at one time-point, associations that may be “early” or “late” manifestations of disability would be differentially detected.
Authors’ contributions
DJC conducted the statistical analyses and drafted the manuscript. Both authors were involved in the conception of the research and interpretation of the data. RBW is a co-principle investigator for HRS, has ongoing involvement in the design and maintenance of the cohort, and was responsible for devising the survey questions use in this manuscript. Both authors read and approved of the final manuscript.