Dortmund Health Study

Primary aim of the Dortmund Health Study (DHS) was to determine the frequency of headache disorders, cardiovascular and other chronic diseases, and behaviour-dependent risk factors in the population of the city of Dortmund in the western part of Germany.18 From a total population of 591 000 a random sample of 3820 persons aged 25–75 years was drawn from the municipal registry. Of those sampled, 395 persons were excluded because they had moved out of the study area, died or did not have sufficient knowledge of the German language leaving 3425 persons, who were eligible and invited to participate in a personal interview at the DHS study centre. If personal participation at the centre was impossible, a questionnaire with a subset of the otherwise identical questions was mailed to the participants. The overall response at baseline was 66.9%, yielding 2291 participants (1312 with interview and 979 with questionnaire). RLS assessment at baseline was restricted to interviewed participants only, because the respective questions were not included in the questionnaire due to the reduced space available. Vital status of 13 subjects could not be identified during the follow-up. Eight had moved outside Germany and were not traceable, three of the remaining five had foreign nationalities and were likely to have moved abroad, and two were untraceable. Therefore, the analyses included the data of 1299 participants from DHS.

Study of Health in Pomerania

The Study of Health in Pomerania (SHIP) is an ongoing population-based study comprising 3 cities and 29 communities in the rural area close to the Baltic Sea (West Pomerania). It was designed to assess a broad range of health and quality-of-life indicators in the north-east region of Germany after the German reunification.19 From the total population of 212 157 residents living in the study area in 1995, a sample of 7008 men and women aged 20–79 stratified by 5-year age groups was drawn using a multistage random sampling design. The final number of subjects participating in the study was 4308 (response 68.8%). The baseline examination was conducted from 1997 to 2001, combining an interview, medical and dental examinations performed in one single visit in the study centre. RLS data at baseline were available from 4291 participants.

Women's Health Study

The Women's Health Study (WHS) was a randomised, placebo-controlled trial designed to test the risks and benefits of low-dose aspirin (100 mg every other day) and vitamin E (600 IU every other day) in the primary prevention of cardiovascular diseases (CVD) and cancer among apparently healthy women. The design and methods of the WHS have been described in detail previously.20 ,21 Briefly, a total of 39 876 US female healthcare professionals aged 45 years or older at study entry (1992–1995) without a history of CVD, cancer or other major illnesses were randomly assigned to receive active aspirin, active vitamin E, both active agents, or both placebos. Baseline information was self-reported and collected by a mailed questionnaire that asked about many cardiovascular risk factors and lifestyle variables. Twice in the first year and yearly thereafter, participants were sent follow-up questionnaires asking about study outcomes and other information during the study period. After the trial's termination in March 2004, the women who were still alive and willing to participate entered an observational follow-up. The return date of the 108-month questionnaire containing questions on RLS was defined as new baseline for this analysis. Of the 33 092 women in active follow-up at 108 months, we excluded 1722 women with missing RLS information, leaving a total of 31 370 women for this analysis.

Physicians’ Health Study

The Physicians’ Health Study I (PHS I) was a randomised, double-blind, placebo-controlled trial to test the benefits and risks of low-dose aspirin (325 mg) and β-carotene (50 mg) in the primary prevention of CVD and cancer among 22 071 apparently healthy male physicians aged 40–84 years at baseline in 1982.22 Baseline information was self-reported and collected by means of a mailed questionnaire that asked about many cardiovascular risk factors and life style variables. Every 6 months in the first year and yearly thereafter, follow-up questionnaires were sent to the participants. Since the trials’ termination in 1995, the men are continued to be followed either on an observational basis or as part of the PHS II.

Using methods successfully developed in the PHS I, the PHS II was launched in 1997.23 The PHS II is an ongoing randomised, double-blind, placebo-controlled trial to test the effects of vitamin C (500 mg), vitamin E (400 IU), β-carotene (50 mg), and a daily multivitamin (Centrum Silver) in the prevention of total and prostate cancer, CVD and age-related eye disease among 14 641 US male physicians aged 55 years and older, including a total of 7 641 PHS I participants who were willing and eligible to enter the PHS II. Baseline information was self-reported and follow-up information was collected annually by mailed questionnaires. For the purpose of this analysis, we pooled data from the PHS I and PHS II, yielding a total of 29 071 participants. The return date of the questionnaire containing the RLS questions (216-month questionnaire for PHS I participants and 12-month questionnaire for PHS II participants) was defined as new baseline for this analysis. At this time point, 24 505 men were still in active follow-up. We excluded 1579 men with missing information for all three RLS questions, leaving a total of 22 926 men at our defined baseline for our analysis.

RLS assessment

RLS is diagnosed according to the presence of specific symptoms. Participants in each study were asked to answer the following questions which were based on the minimal criteria published by the International Restless Legs Syndrome Study Group:12 ‘Do you have unpleasant leg sensations (like crawling, paraesthesias, or pain) combined with a motor restlessness and an urge to move?’, ‘Do these symptoms occur only at rest and does moving improve them?’, ‘Are these symptoms worse in the evening or at night compared with the morning?’ The three answer categories included ‘Yes’, ‘No’ or ‘Don't know’. Participants were only classified as RLS-positive if they answered all symptom questions with ‘Yes’.

In DHS and SHIP RLS assessment was conducted in face-to-face interviews by trained and certified interviewers, while in WHS and PHS RLS information was retrieved by mailed standardised questionnaires. The same questions that were used to identify RLS symptoms in each study had been previously validated24 and had already been used in prior reports.2 ,5 ,7 ,18 ,25 Comparing the questionnaire-based classification of RLS with a physician's diagnosis as a gold standard showed good agreement (unweighted κ=0.67, p<0.001) in the German MEMO (Memory and Morbidity in Augsburg Elderly) study.24 In the DHS, there was a further question related to the number of years elapsed since the onset of RLS symptoms. The frequency of RLS symptoms was also assessed in the DHS with the following possible answer categories: ‘daily’, ‘3–6 times a week’, ‘1–2 times a week’, ‘1–3 times a month’ and ‘less than once a month’. The first two categories and the remaining three categories were subsequently collapsed for the analysis.

Ascertainment of mortality

Death from all causes was defined as our main outcome for the analyses. In both German cohorts, the vital status of participants, that is, alive or deceased and the date of death, if applicable, was retrieved from the municipal registries at the follow-up. In both US cohorts, deaths of participants were identified by reports from family members or next of kin, or postal authorities, and searches of the National Death Index. Information on date and cause of death were confirmed through review of death certificates and medical records by an endpoints committee of physicians.

Socio-demographic data, lifestyle factors and comorbidities

Age, gender and lifestyle factors, that is, health-related behaviours, were assessed during the interview in the German studies and with the mailed questionnaire in the WHS and PHS. In the DHS and SHIP, comorbidities including diabetes mellitus, hypertension, cancer, myocardial infarction and stroke were assessed as self-reports with specific questions asking for a physician-made diagnosis of the respective condition. In the WHS and PHS, these comorbidities were also assessed by questionnaires, and the presence of myocardial infarction, stroke and cancer was confirmed by medical record review. Body mass index (BMI) was calculated using measured height and weight information in the German studies and self-reported height and weight information in the US cohorts. There was one question referring to the presence of leg cramps in SHIP.

Statistical analysis

Data of the four cohort studies were analysed separately using the following identical analysis approach. For each cohort, we calculated mean values for continuous and frequencies for categorical variables of baseline characteristics. Person-time was calculated from the return date of the questionnaire containing the RLS questions or date of interview, respectively, to the date of death, last documented contact or end of study, whatever occurred first. Cox proportional hazards models were used to evaluate the association between RLS status and mortality. Age-adjusted and multivariable-adjusted HRs and their 95% CIs were calculated. The multivariable models were adjusted for the following potential confounding factors: age, gender (DHS and SHIP), BMI, smoking, exercise, history of diabetes, history of hypertension, myocardial infarction, stroke and cancer. To provide comparability of results across studies, we additionally performed stratified analyses according to gender in the DHS and SHIP.

The proportional hazards assumption was tested by including an interaction term for RLS status and logarithm of follow-up time for mortality in age-adjusted models. We found no statistically significant violation.

In all multivariable models participants with missing covariate information were excluded. The models were reanalysed incorporating a missing value indicator in the outcome models for covariates if the number of participants with missing information was greater or equal to 100. We assigned participants with missing values to the covariate reference category if the number of missing information was less than 100. These analyses provided nearly identical results (data not shown).

We performed several sensitivity analyses. First, we excluded participants who reported leg cramps in SHIP since the presence of leg cramps is a potential RLS mimic. Second, we evaluated whether RLS frequency is associated with RLS mortality in the DHS. Third, we assessed whether the number of years elapsed since the onset of RLS has an effect on RLS mortality in the DHS. All analyses were performed with Stata V.11.0 (StataCorp, Texas, USA) in the DHS and SHIP. For all WHS and PHS analyses, we used SAS V.9.1.3 (SAS Institute Inc, Cary, North Carolina, USA). All p values were two-tailed and p<0.05 was considered statistically significant.

Comparisons with other studies

Some clinic-based studies found a relationship between RLS and mortality among subjects with end-stage renal disease, either in dialysed 13 ,14 or transplanted patients.15 There was no clear explanation for the findings, although the presence of RLS could be related to indicators of poorer health status, like higher level of uraemia, disease duration and comorbidity. However, those studies were conducted in a very specific patient group with very high morbidity and mortality, and thus the results can be hardly generalised to relatively healthier populations like ours.

In prior reports of community-dwelling elderly subjects, RLS symptoms indicated higher mortality only among women.16 Mallon et al found that RLS combined with daytime sleepiness was associated with increased mortality risk among women in a middle-aged population, although RLS without sleepiness was not significantly related to mortality.17 There was no clear explanation for these findings, though it has been speculated that disturbed sleep, which frequently accompanies severe RLS, might contribute to the decreased lifespan of subjects with RLS. We had no data about daytime sleepiness or other sleep variables available to determine whether these factors modify the relationship between RLS and mortality.

Differences in sampling and population characteristics might have contributed to the discrepant findings between the previous and the present studies. More importantly, at the time of the baseline assessment of the two earlier studies, standard criteria for RLS did not exist. The urge to move the legs and relief of symptoms due to leg movement, two core symptoms of RLS, were not assessed. It should be also noted that the follow-up time of the study of Mallon et al17 was 20 years, much longer than any of our studies.

In contrast to these previous results, recently published analyses from the WHS and PHS cohorts do not suggest that RLS is associated with incident cardiovascular disease events including CVD mortality, either in women or men.25 However, another recent report suggests that RLS with a duration at least three years is an independent predictor of coronary heart disease among women.26

Subjective and objective sleep disturbances are very common among subjects with RLS.27 ,28 Both short sleep duration and frequent insomnia symptoms predict mortality according to population-based and community-based studies.29 ,30 Thus we cannot rule out that patients experiencing RLS combined with chronic severe sleep deprivation and/or insomnia may have a higher mortality risk over time. Future prospective studies with assessment of RLS severity as well as various sleep-related factors should further investigate this hypothesis.

Strengths and limitations

One of the strengths of the present work is the analysis of four independently conducted cohort studies. The lack of association observed across the four different studies indicates the robustness of our finding. Furthermore, RLS cases were classified according to the minimal criteria for RLS. There were differences in the sampling procedures, sample characteristics and baseline data collection across the studies. However, the fact that the results of a null association were the same despite these differences underlines the robustness of the lack of association between RLS and mortality. Finally, we were able to take several important potential confounders into account, including behavioural risk factors and comorbid conditions.

Several limitations have to be considered when interpreting our results. An important limitation is that, as in all large epidemiological studies, case classification was based on a set of self-administered diagnostic questions instead of a complete clinical interview and examination. However, the set of RLS questions used in our cohorts has been successfully used and validated in previous studies and the observed prevalences in our cohorts are similar to those reported in other population-based studies.31 We also note that conducting clinical interviews in large studies like these would have been extremely difficult. Still, the use of screening questions could have led to misclassification, and RLS mimics could have been falsely identified as RLS cases. We adjusted for the presence of diabetes, therefore diabetic neuropathy, a common potential RLS mimic, was not likely to influence the findings. In SHIP, excluding participants with leg cramps, another frequent RLS mimic, did not substantially alter the results. Nevertheless, there might be other potential mimics of RLS symptoms that could have an effect on the association between RLS and mortality. Due to the low number of outcome events in the DHS, the estimates for the association of RLS with mortality were less precise than in the other three cohorts. It is possible that a longer follow-up time might have revealed an association between RLS and mortality. Information on frequency and duration of RLS symptoms was available in only one of the cohorts. Despite adjustments for a large number of potential confounders, residual and unmeasureable confounding remains possible as our study is observational. However, we are not aware of any confounding factor that, if included in our final models, would result in increased risk of all-cause mortality among patients with RLS. Participants in all four cohorts were predominately white, which may limit the generalisability to other populations.

Clinical implications

Results of these four independent large cohort studies do not suggest that RLS is a risk factor for mortality. However, previous studies indicate that RLS is associated with a variety of comorbidities, such as diabetes, higher body mass index and cardiovascular diseases, which are established risk factors for mortality. Patients with RLS should be screened for RLS associated comorbidities and treated if necessary.

Unanswered questions and future research

RLS is a highly prevalent disease among the elderly, but the mechanisms causing the disease and its relationship with a diverse set of comorbidities remain poorly understood. In addition, sleep deprivation and insomnia accompanying RLS may also influence mortality and comorbidity. Further studies are warranted to establish risk factors for incident RLS to be able to develop sufficient prevention strategies and to understand the complex relationship between RLS and its comorbid conditions. In the DHS, the frequency and duration of RLS symptoms were not related to mortality, even though the effect size increased with symptom frequency. Since this was the smallest study with the youngest population, the potential role of RLS severity needs further clarification. Finally, RLS as a potential mortality hazard should be further investigated with longer follow-up time and in high-risk populations, for example among elderly subjects.

In summary, results of four independent large, prospective cohort studies indicate that RLS is not associated with increased risk of all-cause mortality. Given the high prevalence of RLS in western populations, this is a positive result for all living with RLS.