OUP user menu

Where dopamine meets opioids: a meta-analysis of the placebo effect in restless legs syndrome treatment studies

Stephany Fulda, Thomas C. Wetter
DOI: http://dx.doi.org/10.1093/brain/awm244 902-917 First published online: 11 October 2007

Summary

The restless legs syndrome (RLS) is a common sensory-motor disorder of sleep/wake motor regulation with prevalence rates between 3% and 10%. In its more severe forms, RLS is a burdening disorder with disturbed sleep and significantly impaired quality of life. Restless legs symptoms are dramatically relieved with levodopa and dopamine agonists, which are first-line treatment for this disorder. In addition, opioids have been shown to provide a marked symptomatic relief. This unique responsiveness of RLS to both dopaminergic agents and opioids places it at the crossroad of the two systems implicated in the placebo response. Indeed, in recent large-scale studies a substantial placebo response was observed. We performed a meta-analysis to provide an evidence-based estimate of the magnitude of the placebo response in RLS. Search strategies included the electronic databases PubMed and the Cochrane Clinical Trials Registry (from 1966 to March 2007), the reference lists of retrieved articles, hand-searching abstract books of sleep, neurology and movement disorder congresses and visiting clinical trial register web sites. All randomized, double-blind, placebo-controlled studies exploring a pharmacological treatment in subjects with RLS were considered. Outcome measures from five domains were extracted: RLS severity, subjective sleep parameters, sleep parameters derived from nocturnal polysomnography, periodic leg movements during sleep (PLMS) and daytime functioning. We identified 60 clinical trials and 36 of them were eligible for the meta-analysis. In 24 trials, the pooled placebo response rate was 40.09% (95% CI: 31.99–48.19). The placebo effect was large for the primary outcome measure in most studies, which is the International Restless Legs Severity Scale (1.48, CI: 1.81 to 1.14), notably smaller for other RLS severity scales, moderate for daytime functioning, small to moderate for subjective and objective sleep parameters, very small for PLMS and absent for sleep efficiency. This meta-analysis yields several implications for the planning of both clinical RLS treatment studies and basic research programs.

  • restless legs syndrome
  • placebo effect
  • meta-analysis

Introduction

The placebo response is an improvement of subjective and objective outcomes while taking an inert substance or undergoing a sham procedure. The precise mechanisms of the placebo response are not well understood (Hróbjartsson and Gøtzsche, 2001) and potential factors include regression to the mean, expectancies, non-specific effects of participating in research and physiological changes produced by placebos. The dopamine and opioid systems are thought to play a crucial role in the physiological response to a placebo (Colloca and Benedetti, 2005). The placebo effect in pain is powerful and has been known for long (Levine et al., 1978; Turner et al., 1994; Colloca and Benedetti, 2005); placebo analgesia can be blocked by naloxone, suggesting that placebos can induce the release of endogenous opioids (Levine et al., 1978; Amanzio and Benedetti, 1999; Benedetti et al., 1999). In a positron emission tomography (PET) study, analgesia induced by both a placebo and the opioid agonist remifentanil was associated with an increased activity in several pain-modulating brain regions including the rostral anterior cingulate cortex, the orbitotofrontal cortex and the anterior insula (Petrovic et al., 2002). Functional magnetic resonance imaging (fMRI) showed an increased activation in the same regions and in the dorsolateral prefrontal cortex during the anticipation phase of the placebo analgesic response, whereas placebo treatment was characterized by a decreased activation in the thalamus, anterior insula and the caudal rostral anterior cingulate cortex (Wager et al., 2004). A substantial placebo effect is also apparent in neurological disorders not directly involving pain such as Parkinson's disease (PD) (Shetty et al., 1999; Goetz et al., 2002). The dopaminergic system, particularly affected in PD, is involved in the regulation of several cognitive, behavioural and sensory-motor functions (Nieoullon, 2002), and notably in reward mechanisms (Ikemoto and Panksepp, 1999; Martin-Soelch et al., 2001). PET studies using the dopamine D2 receptor antagonist [11C]raclopride found that the placebo effect in PD is related to the release of dopamine in both the dorsal (de la Fuente-Fernández et al., 2001) and ventral striatum (de la Fuente-Fernández et al., 2002). In addition, the perception of the clinical placebo effect was related to the amount released in the dorsal striatum (de la Fuente-Fernández et al., 2001).

A disorder that is unique in the sense that it responds to both dopaminergic and opioidergic agents is the restless legs syndrome (RLS). RLS is a common sensory-motor disorder of sleep/wake motor regulation with prevalence rates estimated from population surveys between 3% and 10% (Phillips et al., 2000; Masood and Phillips, 2003). RLS is characterized by an imperative desire to move the extremities associated with paraesthesias, motor restlessness, worsening of symptoms at rest and in the evening or at night and, as a consequence, sleep disturbances (Allen et al., 2003). Additionally, most patients with RLS have periodic limb movements during sleep (PLMS) and relaxed wakefulness. In its more severe forms, RLS is a burdening disorder with significantly impaired quality of life (Hening et al., 2004). Restless legs symptoms are dramatically relieved with levodopa and dopamine agonists, which present first-line treatment (Fulda and Wetter, 2005), and this responsiveness is a supportive criterion for the diagnosis of RLS (Allen et al., 2003). In contrast to PD, RLS is not a degenerative disorder, and a dopaminergic deficit has not been proven (Paulus and Trenkwalder, 2006). So far, PET and single photon emission tomography (SPECT) studies are at best compatible with a subtle dysfunction of the dopaminergic system (Wetter et al., 2004). The sensory symptoms of RLS may be experienced as painful by a substantial number of patients (Winkelmann et al., 2000) and opioids have been shown to provide a marked symptomatic relief (Kaplan et al., 1993; Walters et al., 1993). Again, imaging results did not reveal major changes in opioid receptor binding. However, in a recent PET study using the non-selective opioid receptor ligand [11C]diprenorphine, RLS severity correlated negatively with ligand binding in the medial pain system including the thalamus, the amygdala and the anterior cingulate gyrus (von Spiczak et al., 2005).

The unique responsiveness of RLS to both dopaminergic agents and opioids places it at the crossroad of the two systems implicated in the placebo response. In addition, substantial placebo effects have also been reported in a broad spectrum of disorders of the central nervous system such as insomnia (Perlis et al., 2005) and depression (Walsh et al., 2002; McCall et al., 2003), both conditions that are also associated with RLS (Picchietti and Winkelman, 2005). And indeed, in recent treatment trials, a large placebo effect has been observed (Allen et al., 2004; Trenkwalder et al., 2004; Walters et al., 2004). The aim of the present meta-analysis was to quantify the magnitude of the placebo effect in RLS treatment studies by combining results from studies conducted during the past 25 years. We also explored whether the placebo effect differed between the various outcome modalities assessed in RLS treatment trials such as RLS severity, periodic leg movements, subjective and objective assessment of sleep and daytime functioning.

Methods

Location and selection of studies

We searched the electronic databases PubMed and the Cochrane Clinical Trials Registry (from 1966 to March 2007) using the following key words: ‘restless legs syndrome’ and ‘placebo’. In addition, reference lists of the retrieved articles were checked and we made an extensive effort to include unpublished material and trial information published only as abstracts by hand-searching abstract books of sleep, neurology and movement disorder congresses held in the last 4 years and visiting trial register web sites of companies known or suspected to conduct trials (Supplement 1 lists all resources used for the search). All randomized, double-blind, placebo-controlled studies exploring a pharmacological treatment in subjects with RLS were considered. Exclusion criteria were the use of concomitant pharmacological treatment for RLS during the trial and a withdrawal study design. We also considered non-English publications. A detailed list of the excluded studies with the reason for exclusion is available from the authors on request.

Outcome measures

For this meta-analysis, in addition to analysing response rates, we focused on five general domains that have been addressed in RLS treatment trials: RLS severity, subjective sleep parameters, sleep parameters derived from nocturnal polysomnography, PLMS and daytime functioning. Within each domain we selected those outcomes for which at least five effect sizes could be extracted. These were available for the International Restless Legs Severity Scale (IRLS) (Walters et al. 2003) and other RLS severity scores, subjective sleep duration and sleep quality, total sleep time and sleep efficiency derived from polysomnography, the PLMS index (number of periodic leg movements per hour of sleep), daytime sleepiness and quality of life.

Data extraction and computation of effect sizes

One reviewer extracted data and another reviewer verified the data extracted. Calculation of effect sizes and variances followed the general outline given by Morris and DeShon (2002) (for computational details see Appendix I). Effect sizes were computed for all outcome measures where means and standard deviations (SD, or standard errors) were given for baseline and endpoint of the placebo trial or for the differences between baseline and endpoint. The effect size employed in this meta-analysis expressed the differences between baseline and endpoint in units of the standard deviation at baseline: Embedded Image All effect sizes were corrected for small sample bias following Hedges (1982). The sampling variance for the individual effect size was computed taking into account the correlation between baseline and endpoint (Morris and DeShon 2002) (see Appendix I). These correlations are rarely reported but can be computed by combining variances from baseline, endpoint and difference scores. The estimated correlations between baseline and placebo endpoint were 0.39 for the IRLS and other scales (n = 665, eight studies), 0.78 for subjective sleep parameters (n = 305, three studies), 0.91 for polysomnographic sleep parameters (n = 17, two studies), 0.71 for the PLMS index (n = 126, three studies) and 0.92 for daytime functioning (n = 195, two studies) (see Appendix II for specific references).

A random effects meta-analysis was conducted to yield a pooled estimate of the placebo effect and between-study heterogeneity assessed with the homogeneity index I2. In case of significant between-study heterogeneity (I2 > 25%), an attempt was made to find a homogeneous set of effect sizes by excluding studies based on study characteristics. A priori defined variables for the subgroup analysis were study design (parallel-group versus cross-over design), study duration (84 versus <84 days; ≤30 versus ≥35 days), number of treatment arms (one versus more than one), study population (idiopathic versus secondary RLS), study drug (dopaminergic versus non-dopaminergic drug) and outcome measures where applicable (e.g. periodic leg movements assessed with actigraphy versus polysomnography).

For descriptive purposes only, we also computed the corresponding effect sizes in the treatment groups for each outcome and explored whether treatment and placebo effect sizes were associated (Spearman correlation ρ). In trials employing multiple groups, treatment effects were taken from the group with the largest effect. Results are displayed as forest plots with the familiar diamond shape of the effect sizes replaced by circles to indicate that these are repeated-measures effect sizes instead of the standard independent group effect sizes. Analysis of response rates followed the general outline of Einarson (1997) but with confidence intervals of the point estimates computed according to Wilson (Agresti and Coull 1998). All analyses were performed with R (R Development Core Team, 2005) and the meta and nlme library (Pinheiro and Bates, 2000) in R.

Meta-regression

Meta-regression employing linear mixed models with known level-1 variance (Raudenbush and Bryk, 2002) and effect sizes nested within studies was employed to explore two basic questions. First, for response rates, the role of those study characteristics used for subgroup analysis (see above) was assessed. Initially, several other trial characteristics such as percentage of drug-naïve patients, mean age, gender composition, or severity of RLS were considered, but too little information was available in the first case and not enough variation was found in the latter variables (Table 1). Second, for the RLS severity measures, we explored whether differences between the IRLS and other scores persisted after controlling for differences in study characteristics. We chose not to perform meta-regression for the other outcome measures or for all available effect sizes due to several reasons: study design and study duration were not independent and both were associated with the year of publication and sample size, thus there was considerable confounding of the potential moderator variables. This was mostly due to the fact that all newer studies (after 2002) were long-term (12 weeks), parallel-group trials with large sample sizes. In addition, the use of specific outcome measures was also associated with these trial characteristics. In particular, effect sizes for quality of life were only available for long-term, parallel-group studies and the IRLS and subjective sleep duration were mostly available from parallel-group trials.

View this table:
Table 1

Study description including study design, placebo-drug allocation, study duration, age, gender, percentage of drug-naïve patients and number of idiopathic and secondary RLS patients

DrugStudy designP : DP : D nStudy duration (days)aPlacebo group
Age (mean)Female (%)Drug naïve (%)Id : Sec nb
Dopaminergic drugs
Levodopa
    Montplaisir et al. (1996)SC CO1:1614:751501006:0
    Beneš et al.(1999)MC CO1:13228:056595028:4c
Levodopa + entacapone
    Polo et al.(2005)CO1:4281:65164
Sr-levodopa/sr-valproic acid
    Eisensehr et al. (2004)CO1:22021:0596020:0
Bromocriptine
    Walters et al. (1988)SC CO1:1630:1461336:0
Pergolide
    Wetter et al. (1999)MC CO1:12828:757573628:0
Ropinirole
    GlaxoSmithKline (2005b); Trenkwalder et al. (2004)MC PG1:1138:146845666284:0
    GlaxoSmithKline (2005c); Walters et al. (2004)MC PG1:1135:13184566257266:0
    Allen et al. (2004); GlaxoSmithKline (2005d)MC PG1:130:298453575059:0
    Kelly and Mistry (2005)MC PG1:217:3742567654:0
    Bogan et al. (2006); GlaxoSmithKline (2005a)MC PG1:1193:187845264380:0
    GlaxoSmithKline (2006b); Kushida and Tolson (2006)MC PG1:1184:175845162359:0
    GlaxoSmithKline (2006a)MC PG1:1149:154845775303:0
    Adler et al. (2004)CO1:12228:760735922:0
Pramipexole
    Montplaisir et al. (1999)SC CO1:11028:1449506010:0
    Oertel et al. (2006b)MC PG1:1114:22442566868338:0
    Partinen et al. (2006)PG1:422:8621538167107:0
    Winkelman et al. (2006)MC PG1:385:25484526481339:0
    Inoue et al. (2006)PG1:119:19d42
Cabergoline
    Kohnen et al. (2004) Stiasny-Kolster et al. (2004a)MC PG1:322:633556823685:0
    Oertel et al. (2006a)MC PG1:120:203556752040:0
Lisuride patch
    Beneš et al. (2005)MC PG1:352:156d8460b70b206:4c
Rotigotine patch
    Stiasny-Kolster et al. (2004b)MC PG1:314:49760501463:0
    Oertel et al. (2005)MC PG1:555:2854958b70b340:0
Sumanirole
    Garcia-Borreguero et al. (2007)MC PG1:452:218565358270:0
Anticonvulsant drugs
Carbamazepine
    Lundvall et al. (1983)CO1:1628:05333
    Telstad et al. (1984)MC PG1:190:84355272
Gabapentin
    Garcia-Borreguero et al. (2002)SC CO1:12242:7557322:2e
    Thorp et al. (2001)SC CO1:11642:76460:16c
XP13512
    Kushida et al. (2006)MC PG1:233:621450b62b
    XenoPort (2006)MC CO1:13614:75058
Opioids
Oxycodone
    Walters et al. (1993)MC CO1:11114:0554511:0
Other drugs
Clonazepam
    Boghen et al. (1986)SC CO1:1628:04650
    Montagna et al. (1984)SC CO1:267:35450
Clonidine
    Wagner et al. (1996)MC CO1:11014:0442010:0
Hydroquinine
    van Dijk et al. (1991)SC CO1:15914:145547
  • P: D: placebo: drug allocation, Id: Sec: idiopathic: secondary RLS, SC: single-centre, MC: multi-centre, CO: cross-over, PG: parallel-group.

  • aStudy duration: duration of wash-out phase.

  • bComplete group.

  • cUraemic RLS.

  • dThe exact number of participants in each group was not reported but estimated as being the appropriate fraction of the total numbers.

  • eIron deficiency.

Results

We identified 60 double-blind, randomized, placebo-controlled studies. Twenty-four studies were excluded for various reasons detailed in Fig. 1 and we included 36 studies published between 1983 and 2007. A detailed description of the included studies is given in Table 1. There were 17 crossover trials and 19 parallel-group trials with study durations between 1 day and 12 weeks (Table 1). The vast majority of trials included patients with idiopathic RLS, with three trials including also a low number of subjects with secondary RLS and one trial including only subjects with uraemic RLS. Summing over all trials, there were 1748 subjects in the placebo groups. The average age was 54 years (range: 44–64 years) and around 64% of the participants were female with the proportion ranging from 6% to 82%.

Fig. 1

Process of study selection of randomized controlled trials. (RLS: restless legs syndrome, IRLS: International Restless Legs Severity Scale).

Response rates

Twenty-four studies reported response rates during placebo treatment (see Apendix 2.2 for specific references). In 17 studies, response rates were given as percentage of patients rated as ‘much improved’ or ‘very much improved’ on the clinical global impression (CGI) change of condition scale by the physician. One study defined response rates as an IRLS score indicative of none or mild symptoms; two studies relied on physician-rated improvement and in a single study each response was defined on a self-made RLS symptom scale, as no RLS ‘attacks’ during 1 week, or the wish to continue on placebo medication. Finally, for one study response rates pertaining to different scales were reported and averaged within study to yield a single estimate.

The 24 studies included a total of 1527 patients in the placebo condition and 1665 patients in the treatment condition. Placebo response rates varied from 0% to 60% with substantial heterogeneity (I2 = 91.6%, Fig. 2). We considered several subgroups of studies based on study characteristics (see Methods section and Supplement 2) but were not able to find a homogeneous set of effect sizes so that a random effects model was applied. The pooled weighted response rate during placebo treatment was 40.09% [95% confidence interval (CI): 31.99–48.19]. The corresponding mean weighted response rate in the treatment groups was 68.32% (CI: 63.36–73.29), again with significant between-study heterogeneity (I2 = 77.3%). Placebo and treatment response rates did not correlate across studies (ρ 0.22, P = 0.30).

Fig. 2

Meta-analysis of response rates during placebo treatment in RLS therapy studies (CO: cross-over trial, PG: parallel-group trial, CI: confidence interval).

We conducted a linear mixed model meta-regression with fixed level-1 variances of 41 available response rates from 24 studies including 17 additional effect sizes from eight studies that reported response rates for more than one time point during the trial and with the study characteristics as covariates. Of all the study characteristics only study duration was related to the placebo response which increased with study duration both within and across studies (Supplementary Fig. 1). Given an estimated placebo rate of 22.26% (±3.02, t = 7.36, P < 0.0001) at the end of the first week, the response rate during placebo treatment is expected to increase by almost 3% per week (2.75 ± 0.19, t = 14.47, P < 0.0001).

IRLS and other RLS scores

A total of 22 studies reported placebo data regarding RLS symptom scales. The IRLS was the primary or secondary endpoint in 14 studies (see Appendix II for specific references); in four of these studies additional scales for rating restless legs symptoms have been employed as well. A further eight studies used other scores such as the RLS-6 scales, visual analogue scales of RLS severity, the CGI severity of RLS item, the number of RLS ‘attacks’ per week or various self-made RLS scores, with three studies reporting aggregated data for a 1-week diary of symptoms. Most studies that did not use the IRLS employed several different scores to assess RLS severity; these were averaged within study to yield a single effect size estimate.

Standardized repeated-measures effect sizes for the IRLS ranged from −0.04 to −2.67 with significant between-study heterogeneity (I2 = 89.0%, Fig. 3). Excluding the only crossover trial did not yield a homogeneous data set (I2 = 87.5%) nor did restricting the analysis to trials with the longest study duration (I2 = 90.5%, Supplement 2). The pooled random-effects estimator was −1.48 (CI: −1.81 to −1.14), which indicates a substantial decrease of RLS severity during placebo treatment. The corresponding pooled random-effect size for the treatment condition was −2.62 (CI: −2.97 to −2.27), again exhibiting significant between-study heterogeneity (I2 = 84.2%). There was a positive correlation between the treatment and placebo effect sizes for the IRLS (ρ = 0.53, P = 0.05).

Fig. 3

Meta-analysis of standardized repeated-measures effect sizes for the placebo effect in the IRLS scale and other RLS severity scores (IRLS: International Restless Legs Severity Scale, RLS: restless legs syndrome, CO: cross-over trial, PG: parallel-group trial, CI: confidence interval).

The standardized repeated-measures effect sizes for other RLS severity scores ranged from 0.04 to −1.21 with significant between-study heterogeneity (I2 = 75.3%, Fig. 3). Between-study heterogeneity was no longer apparent when considering only the seven crossover trials (I2 = 0%), while it was still substantial for the parallel-group trials (I2 = 87.0%). The pooled placebo effect sizes were −0.25 (CI: −0.49 to −0.01) for the crossover trials and −0.78 (CI: −1.26 to −0.30) for the parallel-group trials. The corresponding pooled treatment effect sizes were −1.57 (CI: −2.21 to −0.94; I2 = 58.8%) for the crossover trials and −1.48 (CI: −1.99 to −0.98; I2 = 86.0%) for the parallel-group trials. There was no correlation between placebo and treatment effect sizes for the crossover trials (ρ −0.04, P = 0.96), but a positive correlation for the parallel-group trials (ρ 0.70, P = 0.23).

We conducted a linear mixed model meta-regression with fixed level-1 variances of 31 available effect sizes from 22 studies including five additional effect sizes from four studies that reported RLS severity for more than one time point during the trial and with the type of scales (IRLS versus all other scores) and study characteristics as covariates. The pooled placebo effect was of greater magnitude for the IRLS than for other scales (−0.79 ± 0.29, t = −2.72, P = 0.03) when considering only marginal effects. When controlling for study duration and type of trial, the differences in effect sizes between the different scales were still evident (−0.52 ± 0.17, t = −3.01, P = 0.02) while there was only a trend for a larger placebo effects in parallel-group trials versus crossover trials (−0.55 ± 0.27, t = −2.01, P = 0.06). The placebo effect increased with study duration and each additional week was estimated to further reduce RLS severity by the standardized effect size of 0.09 (± 0.01, t = −9.62, P < 0.0001).

Subjective sleep parameters: sleep quality and sleep duration

Twelve studies reported data on sleep quality during placebo treatment (see Appendix II for specific references). Outcomes included the Medical Outcome Study Sleep Scale item ‘sleep adequacy’ in six studies, the Schlaffragebogen A item ‘sleep quality’, the RLS-6 item ‘sleep satisfaction’, a visual analogue scale item ‘satisfaction with sleep’ and diary-derived sleep quality. Repeated-measures effect sizes ranged from 0.00 to 0.46 with significant between-study heterogeneity (I2 = 70.7%) that remained after the exclusion of the three crossover trials (I2 = 73.4%). The pooled placebo effect size estimate was 0.27 (CI: 0.18–0.36), indicative of a small increase in sleep quality during placebo treatment (Figure 4, upper panel). The corresponding pooled treatment effect size was 0.84 (CI: 0.63–1.04; I2 = 92.3%). The correlation between placebo and treatment effect sizes was negative but statistically insignificant (ρ −0.22, P = 0.50).

Fig. 4

Meta-analysis of standardized repeated-measures effect sizes for the placebo effect in sleep quality (upper panels), subjective sleep duration (upper middle panel), total sleep time derived from polysomnography (lower middle panel) and sleep efficiency derived from polysomnography (lower panel) (CO: cross-over trial, PG: parallel-group trial, CI: confidence interval).

Subjective sleep duration was reported in seven trials (see Appendix II for specific references); six of them employed the Medical Outcome Study Sleep Scale item ‘sleep quantity’ and one study reported a diary-based estimate. Repeated-measures effect sizes ranged from −0.15 to 0.32 with significant between-study heterogeneity (I2 = 78.2%) that remained after the exclusion of the only crossover study that was also the only study with a study duration less than 12 weeks (I2 = 81.4%). The pooled random-effects estimate was 0.13 (CI: 0.03–0.24) pointing to a very small, albeit statistically significant increase in subjective sleep duration during placebo treatment (Fig. 4, upper middle panel). The corresponding pooled treatment effect size was 0.35 (CI: 0.23–0.47; I2 = 76.6%) and there was a positive correlation between placebo and treatment effect sizes (ρ 0.54, P = 0.236).

Polysomnographic sleep parameters: sleep efficiency and total sleep time

Five studies reported data regarding the placebo effect on total sleep time with repeated-measures effect sizes ranging from 0.05 to 0.59 with moderate to substantial heterogeneity (I2 = 68.2%) (see Appendix II for specific references). The pooled effect size of 0.24 (CI: 0.04–0.44) was indicative of a small increase in total sleep time during placebo treatment (Fig. 4, lower middle panel). The corresponding pooled treatment effect size was 0.37 (CI: 0.25–0.49; I2 = 0%) and there was a positive but statistically insignificant correlation between placebo and treatment effect sizes (ρ 0.60, P = 0.35).

Seven studies reported data regarding sleep efficiency with effect sizes ranging homogeneously (I2 = 37.2%) from −0.21 to 0.25 around the pooled effects estimate of 0.07 (CI: −0.06–0.19) which did not significantly differ from zero (Fig. 4, lower panel) (see Appendix II for specific references). The corresponding pooled treatment effect size was 0.37 (CI: 0.26–0.49; I2 = 0%) and there was a negative association between placebo and treatment effect sizes (ρ −0.60, P = 0.24).

PLMS

Fourteen studies measured the PLMS index during placebo treatment and individual effect sizes ranged from 0.44 (increase in PLMS index) to −0.28 (see Appendix II for specific references). As shown in Fig. 5, the pooled effect size estimate was −0.11 (CI: −0.20 to −0.03) and there was no indication of significant within-study heterogeneity (I2 = 10.5%). The corresponding pooled treatment effect size was −0.88 (CI: −1.06 to −0.71; I2 = 52.4%) with no apparent correlation between placebo and treatment effect sizes (ρ 0.27, P = 0.34).

Fig. 5

Meta-analysis of standardized repeated-measures effect sizes for the placebo effect in the number of periodic leg movements per hour of sleep (PLMS index) (CO: cross-over trial, PG: parallel-group trial, CI: confidence interval).

Daytime functioning: sleepiness and quality of life

Thirteen studies contributed data concerning daytime sleepiness; six studies employed the MOS item daytime sleepiness, three studies used the Epworth Sleepiness Scale, two studies reported diary-derived measures of daytime fatigue or drowsiness and one study each used the RLS-6 item daytime tiredness or the IRLS item sleepiness (see Appendix II for specific references). A decrease in average daytime sleepiness was observed in all studies with effect sizes ranging from −0.07 to −0.96 with significant between-study heterogeneity (I2 = 92.3%). The between-study heterogeneity remained elevated even when considering only the parallel-group trials (I2 = 93.7%) or, within that group of studies, only those that used the MOS daytime sleepiness item (I2 = 95.4%, Supplement 2). The pooled random effects estimate over all studies was −0.36 (CI: −0.48 to −0.25, Supplementary Fig. 2) indicating a moderate reduction in daytime sleepiness during placebo treatment. The corresponding pooled treatment effect size was −0.63 (CI: −0.80 to −0.46) with significant between-study heterogeneity (I2 = 95.4%). Placebo and treatment effect sizes were significantly related across studies (ρ 0.87, P < 0.001).

Seven studies assessed quality of life in RLS subjects (see Appendix II for specific references). Six of the seven studies were parallel-group, 12-week studies that employed the RLS-Quality of Life Questionnaire (Abetz et al. 2004). One study used a German disease-specific RLS quality of life questionnaire (Kohnen et al. 2002). Effect sizes ranged from 0.30 to 1.08 with significant between-study heterogeneity (I2 = 95.1%) that was not reduced after the exclusion of the only crossover study (I2 = 94.9%, Supplement 2). The pooled effect size was 0.50 (CI: 0.32–0.68, Supplementary Fig. 3) indicating a moderate improvement in quality of life during placebo treatment. The corresponding pooled treatment effect size was 0.83 (CI: 0.58–1.09; I2 = 95.7%). The rank order of treatment and placebo effect sizes across studies was in high concordance (ρ 0.96, P = 0.003).

Discussion

Our meta-analysis has revealed a substantial placebo response rate in RLS treatment studies. On average, more than one-third of RLS subjects experienced a major improvement of RLS symptoms while being treated with a placebo. This response rate was matched by a large placebo effect in overall RLS severity as measured by the IRLS. Compared to the IRLS, other scores of RLS severity had a lower—small to moderate—placebo effect even when accounting for differences in study design and study duration. There was a moderate effect for quality of life, and a smaller placebo effect was observed for daytime sleepiness, sleep quality and total sleep time. Even smaller placebo effects were apparent for subjectively estimated sleep duration and the PLMS index. A placebo effect was absent for sleep efficiency. Any discussion of differences in the magnitude of the placebo effect must, however, take the correlation between the placebo and treatment effect sizes into account. Therefore, in Fig. 6, the weighted mean placebo and treatment effect sizes are given for each outcome domain and in general the placebo effect size was proportional to the corresponding treatment effect size. Although linear mixed model analysis indicated that the placebo effect for the IRLS was larger than for other RLS scales even when taking differences in study features into account, it is apparent that this large placebo effect was also matched by a large treatment effect. Nevertheless, there were some notable exceptions to this proportionality of placebo and treatment effect sizes: the placebo effect for RLS scores other than the IRLS was considerably smaller in crossover trials than in parallel-group trials, while this difference was not observed for the treatment effects. For both sleep quality and in particular the PLMS index the treatment effects were disproportionally larger than the placebo effects.

Fig. 6

Pooled placebo effect sizes derived from the meta-analysis. Pooled treatment effect sizes derived from the same studies are depicted in grey for descriptive purposes. The signs of the effect sizes have been converted so that a positive effect size signifies an improved outcome. Confidence intervals marked with an asterisk (*) denote significant between-study heterogeneity (IRLS: International Restless Legs Severity Scale, RLS: restless legs syndrome, PLMS: periodic leg movements during sleep).

Focusing on PLMS as a primary endpoint in RLS trials could appear promising but neglects the fact that only around 80% of subjects with RLS will exhibit PLMS to some degree (Montplaisir et al., 1997), thus making generalizations regarding the complete population of RLS sufferers questionable. In addition, the true clinical significance of PLMS still remains to be determined (Mahowald, 2003). Intriguingly from a theoretical point of view, PLMS are not consciously experienced due to their occurrence during sleep and it is tempting to speculate that therefore they might not be accessible to placebo-mediated expectations of the subject. In addition, apart from the distinction between PLMS and IRLS as subjective versus objective outcome measures, the IRLS is a multidimensional assessment instrument, whereas the PLMS index is one-dimensional. This multidimensionality of the IRLS may predispose the scale to be especially sensitive to placebo but also treatment effects. Future research, analysing individual items of the IRLS, should address the question whether this sensitivity can be traced back to a specific subset of items. Furthermore, as only a few studies (four) have employed another RLS severity score alongside with the IRLS, a direct comparison of the IRLS to other measures of severity was not feasible in the present analysis. In addition, the majority of response rates included in our analysis were based on the CGI and evaluated by an experienced investigator; thus, the placebo effect was not limited to the patients’ self-report. Interestingly, for PD a recent study (Goetz et al., 2002) found a larger placebo effect for the objective part of the Unified Parkinson's Disease Rating Scale than for the subjective part, arguing against a purely subjective phenomenon.

A placebo effect has been observed in a broad spectrum of disorders of the central nervous system such as insomnia or depression and can be attributed to different mechanisms including expectation of clinical improvement and conditioning. For both insomnia and depression, comparable placebo effects have been reported (Walsh et al., 2002; McCall et al., 2003). In both disorders the placebo response increased with study duration (Walach and Maidhof, 1999; Perlis et al., 2005; Posternak and Zimmerman, 2005), a phenomenon we observed also for response rates and RLS severity. Indeed, a recent review of the potential mechanisms of the placebo response in insomnia highlighted the importance of the episodic rather than chronic pattern of insomnia symptoms with regard to the placebo response in this disorder (Perlis et al., 2005). According to these authors, because insomnia symptoms will naturally vary across time, the cooccurrence of better sleep and placebo intake represents a strong reinforcement that could in part be responsible for the observation that placebo response rates increase over time. RLS is associated with insomnia and possibly mood disturbances (Picchietti and Winkelman, 2005) and its severity exhibits day-to-day variations. Although in our meta-analysis the placebo (and treatment) effect for sleep measures was considerably smaller than the effect for RLS severity, the association with sleep and mood disturbances may predispose subjects with RLS to the occurrence of a placebo response. Research on the placebo effect in depression, however, has also shown that response to both treatment and placebo during pharmacological (Benedetti et al., 2005) and cognitive-behavioural therapy (Mayberg et al., 2000) closely matches the active treatment response that the placebo was designed to stimulate (Lidstone and Stoessl, 2007). This would rather point to the involvement of the opioid or dopamine system in the placebo response in RLS.

Apart from the general limitations inherent to meta-analyses (Lyman and Kuderer, 2005) there are specific methodological issues to mention regarding our present study. First, our data basis is limited insofar as we identified 60 controlled trials but could only include 36. Even for these included trials, not all potentially available effect sizes could be extracted due to the data not being reported in an appropriate form. Given that a more detailed reporting would be more likely in the case of significant treatment effects, i.e. larger treatment-placebo differences, our present results could even be an underestimation of the magnitude of the placebo effect. A second issue pertains to the choice of the effect size measure. For the present analysis, we standardized the change from baseline to endpoint by the standard deviation at baseline. Thus, all other parameters being equal, the effect size will be larger if the standard deviation is smaller. If an outcome measure is also used as an inclusion criterion, a range and thus variance restriction is to be expected, effect sizes will be larger and regression to the mean is promoted. This pertains to the IRLS scale for which a minimum value of 10 or 15 was generally required. Finally, we included trials conducted within the last 25 years. While the aim was to guarantee the completeness of the data basis, there was also a notable change in methodology across this large period of time. Standards for treatment trials have changed in terms of design, duration and sample size. In addition, significant research progresses such as the development of the IRLS scale are reflected by the fact that this scale was the primary endpoint in almost all newer studies. For this reason, the use of meta-regression to explore the influence of potential moderators of the placebo effect was restricted to only two outcome domains and even their results must be treated with appropriate caution. Thus, the present analysis cannot claim to provide any insights into the role of possible moderators of the placebo effect.

This meta-analysis has several implications for the planning of both clinical RLS treatment studies and basic research programmes. RLS treatment studies have to reckon with a substantial placebo effect that increases with study duration. Thus, any long-term controlled trial should include a larger number of subjects into the trial. The present meta-analysis can only serve as a starting point for research into the nature of the placebo effect in RLS. To explore potential moderators and predictors of the placebo effect, more detailed data and preferably an individual patient data meta-analysis is needed, which, however, requires the cooperation of the respective drug companies to share their data. Potential moderator variables that could influence the placebo response are gender and age, the fact of whether subjects are drug-naïve or pre-treated or the presence of associated symptoms or comorbidities.

A further promising area of research would be the imaging of the opioid and dopamine systems in RLS in relation to the placebo response. So far, these research projects have been restricted to subjects with PD (de la Fuente-Fernández et al., 2001, 2002) and to subjects with pain disorders or experimentally induced pain (Colloca and Benedetti, 2005). Here, RLS offers the unique possibility to study both systems in the same group of subjects and may help to disentangle their respective contributions to the placebo effect. Indeed, the multitude of subjective and objective outcome measures used in assessing treatment efficacy in RLS, the association with sleep and mood disorders, the occurrence of symptoms both consciously experienced and not accessible for the subject, together with the responsiveness to dopaminergic as well as opioidergic agents, make RLS a model disease to study the placebo effect systematically. Finally, one necessary research project must be the inclusion of a no-treatment control group in a future trial. Without such data there is no way to differentiate between the natural course of the disease as opposed to the placebo effect. In summary, placebo treatment has a strong impact on outcome measures in RLS treatment studies, which may not only point to promising research avenues but may also challenge us to incorporate the therapeutic placebo effect into clinical practice.

Supplementary material

Supplementary material is available at Brain online.

Acknowledgement

We thank Dorothee Skottke for help in preparing the manuscript.

Appendix

Calculation of effect sizes and variances followed the general outline given by Morris and DeShon (2002). Effect sizes (ES) were computed from means and standard deviations as ES = (MbMe)/(SDb) with Mb the mean at baseline, Me the mean at endpoint and SDb the standard deviation at baseline. In a minority of cases only the standard deviation of the difference between baseline and endpoint was available. In that case, the effect size in the change score metric ESdiff = (MbMe)/(SDdiff) was transformed into the raw score metric ES = ESdiff [2 (1 − r)]1/2 with SDdiff the standard deviation of difference scores and r the correlation between baseline and endpoint. All effect sizes were corrected for small sample bias following Hedges (1982). The sampling variance of the effect size was computed as Ves = [2(1 − r)/n][(n − 1)/(n − 3)][1 + (n/2 (1 − r)) ESg 2] − ESg 2/c(df)2 with n the number of paired observations, ESg the population effect sizes, and c(df) the bias function c(df) = 1 − [3/(4 df − 1)] with df = n − 1. The correlation between baseline and endpoint was computed from standard deviations according to r = (SDb 2 + SDe 2 + SDdiff 2)/(2 × SDb SDe) with SDe the standard deviation at the end of the trial.

Studies contributing to the meta-analysis

Studies used to compute correlations

RLS severity: Montagna et al. (1984), Boghen et al. (1986), Walters et al. (1993), Trenkwalder et al. (2004), Stiasny-Kolster et al. (2004b), Bogan et al. (2006), Oertel et al. (2006b), GlaxoSmithKline (2006b).

Subjective sleep parameters: Montagna et al. (1984), Bogan et al. (2006), Oertel et al. (2006b).

Polysomnographic sleep parameters: Walters et al. (1988), Walters et al. (1993).

PLMS index: Walters et al. (1988), Walters et al. (1993), Bogan et al. (2006).

Daytime functioning: Walters et al. (1988), Bogan et al. (2006).

Response rates

CGI: Beneš and TULIR study group (2005), Bogan et al. (2006), Garcia-Borreguero et al. (2007), GlaxoSmithKline (2005d), GlaxoSmithKline (2006a), Inoue et al. (2006), Kushida et al. (2006), Kushida and Tolson (2006), Oertel et al. (2005), Oertel et al. (2006a), Oertel et al. (2006b), Partinen et al. 2006), Stiasny-Kolster et al. 2004b), Trenkwalder et al. (2004), Walters et al. (2004), Winkelman et al. (2006), XenoPort (2006).

Other: IRLS: Garcia-Borreguero et al. (2002), Physician-rating: Boghen et al. (1986), Lundvall et al. (1983), Self-made RLS symptom scale: Thorp et al. (2001), No RLS ‘attacks’: Telstad et al. (1984), Wish to continue: van Dijk et al. 1991), Different scales: Kohnen et al. (2004).

Multiple time points: Telstad et al. (1984), Walters et al. (2004), GlaxoSmithKline (2005a), Winkelman et al. 2006), GlaxoSmithKline (2006a), GlaxoSmithKline (2006b), Oertel et al. (2006b), Bogan et al. (2006).

IRLS and other RLS scores

IRLS: Adler et al. (2004), Trenkwalder et al. (2004), Walters et al. (2004), Stiasny-Kolster et al. (2004a), Stiasny-Kolster et al. (2004b), Kelly and Mistry (2005), Oertel et al. (2005), GlaxoSmithKline (2006b), GlaxoSmithKline (2005d), Bogan et al. (2006), Partinen et al. (2006), Winkelman et al. (2006), Oertel et al. (2006a), Oertel et al. (2006b).

IRLS and other scales: Stiasny-Kolster et al. (2004a), Stiasny-Kolster et al. (2004b), Oertel et al. (2006a), Oertel et al. (2006b).

Other scales: RLS-6 scales: Stiasny-Kolster et al. (2004a), Stiasny-Kolster et al. (2004b), Oertel et al. (2006a), Visual analogue scales: Oertel et al. (2006b), CGI severity item: Wetter et al. (1999), Number of RLS ‘attacks’: Telstad et al. (1984), Diaries: Montagna et al. (1984), Walters et al. (1993), Montplaisir et al. (1999), Self-made RLS scores: Boghen et al. (1986), Wagner et al. (1996), Beneš et al. (1999), Wetter et al. (1999).

Multiple time points: Trenkwalder et al. (2004), Walters et al. (2004), Bogan et al. (2006), GlaxoSmithKline (2006b).

Subjective sleep parameters

Sleep quality: Medical Outcome Study Sleep Scale: Allen et al. (2004; GlaxoSmithKline (2005b), GlaxoSmithKline (2005c), Bogan et al. (2006), GlaxoSmithKline (2006a), GlaxoSmithKline (2006b), Schlaffragebogen A: Beneš et al. (1999), Wetter et al. (1999), Oertel et al. (2006a; RLS-6 item: Stiasny-Kolster et al. (2004a), Visual analogue scale: Oertel et al. (2006b), Diary: Boghen et al. (1986).

Subjective sleep duration: Medical Outcome Study Sleep Scale: Allen et al. (2004), GlaxoSmithKline (2005b), GlaxoSmithKline (2005c), Bogan et al. (2006), GlaxoSmithKline (2006a), GlaxoSmithKline (2006b), Diary: Beneš et al. (1999).

Polysomnographic sleep parameters

Total sleep time: Walters et al. (s1988), Wagner et al. (1996), Allen et al. (2004), Eisensehr et al. (2004), Oertel et al. (2006a).

Sleep efficiency: Walters et al. (1988), Walters et al. (1993), Wagner et al. (1996), Allen et al. (2004), Eisensehr et al. (2004), Oertel et al. (2006a).

PLMS

Walters et al. (1988), Montplaisir et al. (1996), Walters et al. (1993), Wagner et al. (1996), Beneš et al. (1999), Montplaisir et al. (1999), Allen et al. (2004), Eisensehr et al. (2004), Polo et al. (2005), Bogan et al. (2006), GlaxoSmithKline (2006a), Oertel et al. (2006a), GlaxoSmithKline (2006b), Garcia-Borreguero et al. (2007).

Daytime functioning

Sleepiness: Medical Outcome Study Sleep Scale: Allen et al. (2004), GlaxoSmithKline (2005b), GlaxoSmithKline (2005c), Bogan et al. (2006), GlaxoSmithKline (2006a), GlaxoSmithKline (2006b), Epworth Sleepiness Scale: Adler et al. (2004), Stiasny-Kolster et al. (2004b), Winkelman et al. (2006), Diary: Walters et al. (1993), Wagner et al. (1996), RLS-6: Oertel et al. (2006a), IRLS: Oertel et al. (2006b).

Quality of life: RLS-Quality of Life Questionnaire: Walters et al. (2004), GlaxoSmithKline (2005b), GlaxoSmithKline (2005d), Bogan et al. (2006), Winkelman et al. (2006), GlaxoSmithKline (2006a), German RLS quality of life: Oertel et al. (2006a).

Footnotes

  • Abbreviations:
    Abbreviations:
    CGI
    Clinical Global Impression
    CI
    confidence interval
    CO
    cross-over trial
    ES
    effect size
    IRLS
    International Restless Legs Severity Scale
    PD
    Parkinson's disease
    PG
    parallel-group trial
    PLMS
    periodic leg movements during sleep
    RLS
    restless legs syndrome

References

View Abstract