Quantitative evidence synthesis

Methods

Protocol and registration

We followed recommendations from the Cochrane Handbook for Systematic Reviews of Interventions.34 and pre-specified our methods in a research protocol (PROSPERO database registration number: CRD42021296922).

Search methods for identification of studies

Sensitive search strategies were designed by an information scientist using database subject headings and text word terms to identify reports of RCTs in fibromyalgia patients with sleep as an outcome. The Cochrane Highly Sensitive Search Strategy for identifying RCTs was used in MEDLINE and adapted for other electronic databases. There were no date or language restrictions in the search. The databases searched were Ovid MEDLINE, EMBASE, PyscInfo, and AMED (Allied and Complementary Medicine), EBSCO CINAHL, Clarivate Science Citation Index and the Cochrane Controlled Trials Register (CENTRAL). The searches were conducted on 1 November 2021. Reference lists of systematic reviews and included studies were checked to identify additional potentially relevant reports. Full details of the search strategies are reported in Appendix 1.

To allow studies that evaluated all potential interventions to be included in the review, a broad list of keywords was developed. This was guided by studies identified in a previous systematic review of non-pharmacological interventions for the management of fibromyalgia.31 We focused on studies that mentioned relevant sleep outcomes in the titles or abstracts of the study publication with or without numerical data. Studies were not deemed suitable for inclusion if they did not mention sleep outcomes in their titles or abstracts. Outcome data from multiple publications from the same research were linked as a single study. However, we excluded publications reporting secondary or post hoc analysis of the same outcome data of already included studies.

Inclusion and exclusion criteria

The key eligibility criteria for the quantitative evidence review are summarised using a participants, interventions, comparisons,outcomes, study design (PICOS) framework in Table 1.

Types of studies

Parallel-group, crossover and cluster RCTs assessing the effectiveness of any intervention for the management of fibromyalgia-related sleep problems were eligible for inclusion. We excluded quasi- or non-randomised studies, single-arm studies and observational studies. We included studies, published in full, regardless of the language used in the publication.

Types of setting

We included studies conducted in any relevant clinical setting, for example primary care, secondary care or community care.

Types of population (participants)

We included people of any age (both adults and children) with fibromyalgia. Regarding the age limits used to differentiate between adults and children, we accepted those reported by the authors of the identified studies. Studies in which only a subset of participants had fibromyalgia were considered for inclusion, if data were available separately for the relevant subset.

Types of interventions

We included any pharmacological or non-pharmacological intervention for the management of sleep disturbances in fibromyalgia.

Pharmacological treatments included antipsychotic (AP; e.g. quetiapine), anticonvulsant (e.g. gabapentin, pregabalin), antidepressant (e.g. amitriptyline) and hormonal (e.g. melatonin) medications. We considered eligible pharmacological treatments regardless of their dose or routes of administration. Examples of non-pharmacological interventions included exercise (e.g. aerobic, strengthening, stretching/flexibility, aquatic exercise, Tai Chi, yoga), psychological or behavioural therapy (PT/BT) (e.g. CBT), patient education, dietary and lifestyle modifications, as well as complementary and non-mainstream therapies (e.g. acupuncture).

We included interventions regardless of whether they were targeted to improve sleep or used for fibromyalgia pain management with a potential effect on sleep. Interventions that focused primarily on the management of comorbid autoimmune inflammatory conditions in people with fibromyalgia were excluded; these include, for example, disease-modifying antirheumatic drugs for rheumatoid arthritis that could treat symptoms of rheumatoid arthritis and improve the quality of sleep of coexisting fibromyalgia.

Eligible interventions included discrete interventions or multicomponent interventions, which comprise a combination of different interventions that may be delivered concurrently or sequentially. Interventions could be delivered one-to-one, in a group, in person or remotely.

Types of comparators

We investigated comparisons of active (experimental) interventions with usual care (UC), placebo or no treatment (including wait list), or with another active intervention. We excluded comparisons of two or more regimens of the same treatment (e.g. varying doses of the same drug), if placebo or another intervention group was not considered.

Types of outcomes

In consultation with clinical experts and patient partners, we decided to include the following outcomes measured at the end of the designated intervention period, or at the first assessment point after the end of the intervention.

Primary outcomes:

• sleep quality (patient’s experience of sleep and perceived sleep quality)
• adverse events (AEs).

Secondary outcomes:

• sleep efficiency (%; calculated as total sleep time / total time in bed × 100%)
• duration of sleep/total sleep time
• disease-specific quality of life.

Outcome data were considered suitable for inclusion irrespective of whether they were reported as primary or secondary end points in the published studies. For the sleep quality outcome, our primary focus was on the five PROMs on sleep quality, validated in people with fibromyalgia, that we identified as part of the update of a recently published systematic review.35 These measures included: Pittsburgh Sleep Quality Index (PSQI),36 Medical Outcomes Study Sleep Scale (MOS-SS),37 Jenkins Sleep Scale (JSS),38 Fibromyalgia Sleep Diary (FMSD)39 and Sleep Quality Numeric Rating Scale (SQ-NRS).40 Numerical rating scales (NRSs) or visual analogue scales (VASs) broadly measuring sleep quality were also included.

In the absence of an accepted quality-of-life tool specific to fibromyalgia, we used the Fibromyalgia Impact Questionnaire (FIQ)41 and the 36-Item Short Form Survey (SF-36) physical and mental health components as a proxy for disease-specific measures.42

We recorded information on AEs reported in at least 10% of participants of included studies as well as serious adverse events (SAEs); we accepted the categorisation of events reported by the authors of the included studies.

Selection of studies

Two review authors (CR and MI) independently screened a sample of 100 titles and abstracts at the beginning of the study-selection process and compared the results to ensure consistency. The remaining citations were divided into two segments and the abstracts screened by the same two review authors, where one author screened for any studies that could clearly be excluded, with the second author deciding on all studies that were labelled unsure and 10% of the studies marked as excluded. All potentially relevant articles were retrieved in full and divided between the same two review authors. One review author assessed all full-text articles for inclusion, with the second author checking all studies that were labelled unsure and 10% of the studies that were not considered suitable for inclusion. Any discrepancy or inconsistency was resolved by discussion between reviewers.

Data extraction

Data extraction forms were developed and piloted by the same two review authors with input from our patient and public involvement (PPI) collaborator (DD). Information on setting, characteristics of participants, characteristics of interventions and outcome measures were recorded. The two review authors independently extracted data from 10% of the included studies to ensure consistency and accuracy. The remaining studies were divided between the two reviewers and single data extraction was undertaken, with one review author checking the data extracted by the other review author for consistency. Any discrepancy between the review authors was resolved by discussion or consultation with a third review author (MB).

Quality assessment of included studies

The risk of bias of each included study was assessed using the revised Cochrane risk-of-bias tool (RoB2).43 To implement RoB2 assessments, we used the Excel tool available at the Risk of Bias tools website (www.riskofbias.info/welcome/rob-2-0-tool/current-version-of-rob-2). The assessment addressed the following domains:

• bias arising from the randomisation process
• bias due to deviations from intended interventions
• bias due to missing outcome data
• bias in measurement of the outcome
• bias in selection of the reported result
• overall risk of bias.

We judged each domain as ‘low risk’, ‘some concerns’ or ‘high risk’. Two review authors (CR and MI) independently assessed the first 10% of included studies to ensure consistency. During this process, we developed a document on decision-making rules with instructions on how to answer the signalling questions of the RoB2 tool. The remaining studies were then divided between the two reviewers for single assessment. Any uncertainties or disagreements arising during risk-of-bias assessment were resolved by discussion between the two review authors. While double assessment is generally recommended, the proposed approach was adopted to improve the efficiency of the systematic review process and ensure the timely completion of the project.

Quantitative synthesis and network meta-analysis

For sleep quality, there were several different outcomes identified: PSQI, MOS-SS, JSS, FMSD, SQ-NRS as well as single-item VAS/NRS measuring a similar sleep quality construct to that of the SQ-NRS. We decided to pool these outcomes together to form an overarching ‘Sleep’ outcome. However, we also performed analysis on each individual outcome as well as performing a sensitivity analysis of combining SQ-NRS and VAS. Where a study reported more than one sleep quality outcome, we specified a hierarchy based on the most reported outcome across included studies. The adopted hierarchical order was as follows: PSQI, MOS-SS, JSS, FMSD, SQ-NRS and VAS/NRS. A mixture of ‘change from baseline’ and ‘final score’ was reported; therefore, we converted the final score to ‘change from baseline’ when baseline values were available. For imputation of the change from baseline standard deviation (SD), we used a correlation coefficient as per the recommendation of the Cochrane Handbook for Systematic Reviews of Intervention.34 As we had no available data to calculate the correlation coefficient, we chose a 0.5 value and decided to perform a sensitivity analysis assuming a correlation coefficient of 0.8 to assess whether the results changed. The effect size calculated was the standardised mean difference (SMD), which divides the difference in mean between interventions by the estimated pooled between-person SD for that trial. However, due to some studies having small sample sizes we used the Hedges (adjusted) G method.44 Effect sizes reported were either SMD for the ‘Sleep’ outcome and mean differences (MDs) for the remaining outcomes along with 95% confidence intervals (CIs) or credible intervals (CrIs).

Whenever possible, we performed a pairwise and NMA of all included outcome variables. For each pairwise meta-analysis, a random-effects model was used to compare the direct evidence, with heterogeneity being assessed by I² statistic. This analysis was performed in Stata version 17.45

For each relevant outcome, a NMA was performed to combine both direct and indirect evidence using a Bayesian framework, according to guidance from the NICE Decision Support Unit in the UK and reported in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) for network meta-analyses. Random-effects models with a normal likelihood were used as all our outcomes were continuous. Convergence was assessed using history, autocorrelation and Brooks-Gelman Rubin plots. Consistency was evaluated by examining the agreement between direct and indirect evidence in all closed loops. To explore the presence of inconsistency for any treatment contrast in the network, we performed a node-splitting analysis. We also estimated the ranking probabilities of the different interventions using the surface under the cumulative ranking (SUCRA) curve, which is a numeric presentation of the likelihood that an intervention is successful, as well as presented rankograms. The network diagrams and the node-splitting analysis were performed in Stata 1745 using the network command, while all remaining analysis was done using the WinBUGS (MRC Biostatistics Unit, Cambridge, UK).46

Assessment of the certainty of the evidence

We used the Confidence In Network Meta-Analysis (CINeMA) approach,47 which is broadly based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework,48 to evaluate the certainty of evidence from the NMA on sleep quality outcome. The CINeMA framework considers six domains that may affect the level of confidence in NMA findings: (1) within-study bias, (2) reporting bias, (3) indirectness, (4) imprecision, (5) heterogeneity and (6) incoherence. For each relative treatment effect derived from the NMA, the level of concern was assessed as ‘no concerns’, ‘some concerns’ or ‘major concerns’. The final judgements across domains were summarised into four confidence ratings: ‘high’, ‘moderate’, ‘low’ or ‘very low’. As we had no available data to define the minimal clinically important difference (MCID) for the sleep quality outcome, we chose a 0.5 value for SMD, based on benchmarks for ‘medium’ effect size suggested by Cohen (1988).49

Results

Quantity of the evidence

The database search identified 4113 records of published studies. Of these, 377 records were retrieved for full-text assessment. After exclusion of 209 studies, we selected 168 studies for inclusion. Figure 1 shows the flow diagram of the selection process. A list of included studies is presented in Report Supplementary Material 1. A full list of studies excluded after evaluation of the full-text publications, alongside the main reasons for exclusion, is presented in Report Supplementary Material 2.

FIGURE 1 Preferred reporting items for systematic reviews and meta-analyses flow diagram for identification of the quantitative studies. Reproduced with permission from Page et al.50 For more information, visit: www.prisma-statement.org/. (Figure not reproduced here; see the original article.)

Of the 168 included studies, 90 assessed sleep quality using one of the PROMs validated in people with fibromyalgia (i.e. PSQI, MOS-SS, JSS, FMSD and SQ-NRS), 68 assessed sleep quality using other measures (e.g. VAS) and 10 assessed sleep efficiency and sleep duration.

The 90 studies assessing sleep quality using PROMs validated with people with fibromyalgia were eligible for the NMA, while results from the remaining 78 studies were synthesised narratively.

Description of studies

Studies eligible for the network meta-analysis

We describe the 90 studies considered for NMA below. Additional information is provided in Table 2 (baseline demographic characteristics of study participants), Report Supplementary Material 3 (list of interventions) and Appendix 2, Table 18 (inclusion and exclusion criteria, details of the intervention and baseline characteristics of study participants).

Study design

Eighty-five studies were parallel RCTs, while five were randomised crossover trials.113,115,117,128,138 Publication years ranged from 1992114 to 2021.51,72,75,76,86,100,107,112 Most studies were published in English; two in Spanish,66,92 and one each in German,118 Korean100 and French.130

Participants

All 90 studies evaluated adult populations living with fibromyalgia, including a total of 12,082 participants. The number of participants per study ranged from 1358 to 884,129 with a median of 70. Most participants across studies (94% overall) were women, with the reported mean or median age of each study group ranging from 35.182 to 57.7.70 Twenty-seven studies were conducted in Spain,52,56,58,60,63,64,66,68,75,78,79,87,88,90–95,97,99,103,108,119,130 20 in the USA,70,71,81,83,85,89,101,106,113,121,124,126–128,131,133,134,136,139,140 5 in Canada55,61,107,116,125 and 4 in Brazil.69,80,86,120 A few studies were conducted in a single European country (France,51,59,72,104 Italy,84,115,117,140 Germany,57,62,118 UK,96,102,105 Sweden,98,110 Norway112 and Switzerland114) as well as Turkey,82,109 Egypt,65,73 Iran,77 Israel,74 China,54 Japan,123 Korea,100,111 Taiwan76 and Thailand.135 One study each was conducted in multiple countries in Europe,129 Europe and the USA,132 Europe, Asia (India, Korea), Australia and the Americas (Canada, Mexico, Venezuela),137 and Asia (India, Taiwan), Canada and USA.122

Thirty-four studies (38%) reported baseline body mass index (BMI) for each study group. Reported mean or median BMI ranged from 20.579 to 33.9,89 with 9 studies having a mean BMI of 30 or higher (classified obese) in at least one of the study groups.65,80,81,83,85,89,95,136,140 Among the 30 studies (33%) which recorded information on ethnicity,52,54,56,69,74,81,83,85,89,96,99,101,103,106,116,121,122,124,131–134,136,137,139,140 the predominant majority of participants were ‘white’ or ‘Caucasian’, except for 2 studies (evaluating Tai Chi in the USA) which included people from ‘diverse’ backgrounds of whom just over half were ‘white’. The educational attainment of participants was described in 35 studies.51,54,56,57,60–63,65,67,72,73,76,81–83,85,87–90,93,96–101,103,106,109,112,113,120,138 While definition of schooling varied, making comparison across studies difficult, one study focused specifically on patients with ‘low education levels’,103 and nine studies had more than half of the recruited participants with basic or primary education.56,57,60,65,67,82,90,96,103 The other studies reported that the majority (50% or higher) of participants completed high school or higher education (e.g. college, university)51,54,61–63,72,76,83,85,87–89,93,97,98,100,101,112,138 or had an average duration of education longer than 10 years.106,109,113,120 Further information on the demographic characteristics of participants is presented in Table 2.

Twenty-four studies (27%) described the baseline clinical characteristics of participants.51–54,58,59,62,69,74,76,78,81,89,91,100,101,105,107,115,118–120,136,138 Although the type of comorbidities varied across studies, the number of participants with comorbidities was fairly balanced between treatment groups within each study. Common comorbidities (reported in > 50% of participants from these 24 studies) included psychological trauma or prolonged stress,51 depression,51 mood disorder, anxiety disorder, irritable bowel syndrome,51 back pain,52 osteoarthritis, shortness of breath, temporomandibular dysfunction and headache or migraine. Symptom severity (pain or fibromyalgia) was reported in 38 studies using various outcome measures, with the most common outcome measures used being pain VAS or NRSs.51,53,54,56,58,59,68,69,71,72,77,80,81,83–85,92,95,104,108,114,115,117–122,124,125,128,131–133,135,137,139,140 In most studies pain severity was rated to be moderate or high. Participants’ baseline characteristics are summarised in Appendix 2.

Interventions

The 90 included studies assessed a total of 97 active treatments, either alone or in combination. Most were non-pharmacological treatments (76 non-pharmacological treatments and 21 pharmacological treatments). These treatments were grouped into 45 categories according to their characteristics and mode of action (34 non-pharmacological and 11 pharmacological) (see Report Supplementary Material 3).

Exercise interventions were classified into four categories according to the American College of Sports Medicine recommendation141 and a previous systematic review of exercise training for treating osteoarthritis.142 These categories include: (1) aerobic exercise (e.g. cycling, dancing, walking, running, swimming); (2) muscle strengthening or resistance exercise (e.g. lifting hand-held weights, push-ups, squats); (3) flexibility (e.g. stretching) and neuro-motor skills training to improve balance and coordination (e.g. wobble board) and (4) Mind–body Ex that combines body movement, mental focus and controlled breathing (e.g. yoga, Tai Chi, qigong). Stretching or relaxation as part of a warm-up or cool-down component of exercise training was not classified as a distinct intervention. Exercise training was categorised as mixed exercise when it included both aerobic and anaerobic exercise. A further distinction was made for exercise training between aquatic, pool-based exercise training and land-based (exercise) (LD) training. Psychological and behavioural therapies, as well as patient education, were considered as a distinct intervention only if they were based on specific theories or delivery plans (e.g. delivered using specific instructions or programmes). A distinction was also made between interventions tailored to sleep problems (e.g. CBT for insomnia) and generic interventions with no specific focus on sleep problems (e.g. CBT for pain), and between balneotherapy and hydrotherapy, with the latter meaning water-based exercise training.

The most frequently used comparator treatment was placebo/sham treatment (39 studies) or UC (26 studies). Only 25 studies compared an active treatment with another active treatment. For the definitions of placebo and sham treatments, we accepted those provided by the authors of the included studies. We also included ‘attention control’ (typically for behavioural therapy) as placebo/sham treatment to indicate certain procedures designed to increase attention or treatment contact without providing any active therapeutic elements. Typically, pharmacological treatments were compared with placebo, while non-pharmacological treatments with sham and attention-control treatments. While in some studies appropriate sham procedures were used (e.g. sham device for neuromodulation, sham cupping therapy, sham Qigong), not all sham treatments were appropriately designed as they did not mimic closely the features of the active interventions (e.g. Manual T compared with sham ultrasound). Some of the attention-control treatments were also not clearly defined in the included studies (e.g. ‘telephone support’ as control for active neurofeedback treatment) and it could not be easily established whether the intended control programme did not comprise any active therapeutic elements (see Appendix 2). Within the definition of ‘usual care’ we included ‘standard treatment’, ‘treatment as usual’ (TAU) and ‘wait list’ control as reported by studies’ authors. It is worth noting that a wide variety of medical and non-medical interventions were categorised as ‘standard treatment’ or ‘TAU’ in the included studies. Concomitant treatments (e.g. routine medication) were permitted in most studies and were generally balanced between intervention groups. Information on adherence and compliance to the study treatment was not universally and consistently reported across studies and even when reported could not be easily incorporated into the analysis, as results were not presented separately for ‘responders’ and ‘non-responders’, for example, to allow us to assess its impact on outcomes.

The characteristics of the included interventions are summarised in Table 2 with further details provided in Appendix 2.

Outcomes

Sleep quality outcome was reported using PSQI in 53 studies,51–98,116–120 MOS-SS in 17 studies,99–106,121–123,125–130 SQ-NRS in 1 study,137 both MOS-SS and SQ-NRS in 3 studies,124,139,140 JSS in 8 studies107–109,131–135 and FMSD in 1 study.136

In addition, when sleep quality was evaluated on a NRS or VAS, using a single item, we considered it as a proxy for SQ-NRS and included it in the NMA. Seven studies met this description.110–115,138 Rating scales that did not explicitly measure ‘sleep quality’ or were based on multiple items were considered to be measuring different constructs from SQ-NRS and were not treated as its proxy measure.

In the studies focusing on non-pharmacological interventions (65 studies), the duration of interventions ranged from 2.657 to 25 weeks,78 with a median of 10 weeks. Studies focusing on pharmacological interventions (25 studies) lasted from 5 days126 to 6 months,130 with a median of 12 weeks.

Studies not eligible for the network meta-analysis

Seventy-eight studies reported sleep-related outcomes other than PROMs of sleep quality eligible for the NMA. Appendix 3, Table 19 presents a summary of characteristics of these 78 studies.

Study designs

Of the 78 studies, 71 were parallel RCTs and 7 were randomised crossover trials.143–149 Most studies were published in English, with two published in Turkish,150,151 and one study each published in Spanish,152 German153 and Italian.154

Participants

The 78 included studies involved a total of 5911 participants. One study comprised an adolescent population (n = 107), aged 12–17 years,155 while all other studies were composed of adults (total number of participants = 5804), with mean or median age of study group ranging from 28.5156 to 60.3 years.157 Study sample size ranged from 10148 to 530,158 with a median of 53.5. Across studies the vast majority of participants were women (96%).

Interventions

The 78 included studies assessed 119 different active treatments, either alone or in combination. These treatments were grouped into 45 active treatment categories (26 non-pharmacological, 19 pharmacological). Interventions and corresponding intervention categories evaluated in these studies were broadly similar to those of the studies included in the NMA. New treatment categories for non-pharmacological interventions included phototherapy and botox cervical infiltration. There were also six new pharmacological intervention categories that were not included in the NMA: serotonin receptor antagonist and dopamine receptor agonist (e.g. terguride), opioid antagonist (e.g. naltrexone), selective oestrogen receptor modulators (SERMs; e.g. raloxifene), cannabinoid (e.g. nabilone), orexin antagonists (e.g. suvorexant) and acetylcholine esterase inhibitor (e.g. pyridostigmine).

The most frequently used comparator treatment was placebo or sham treatment (PBO/Sham) (36 studies) or UC (20 studies). Twenty-two studies compared an active treatment with another active treatment.

In the studies focusing on non-pharmacological interventions (57 studies), the duration of interventions ranged from 0.7159 to 25 weeks,160 with a median of 8 weeks. Studies focusing on pharmacological interventions (21 studies) lasted from 1.3148 to 26 weeks,161 with a median of 12 weeks.

Outcome

Self-reported sleep quality was reported using a variety of measures: 29 studies used VAS or NRS,144,151,155,156,162–186 26 studies used a questionnaire,143,146,149,150,154,187–207 and 13 studies used a sleep subscale of a non-sleep-specific measure such as the FIQ.152,158,160,161,208–216 Ten studies that did not assess sleep quality outcome reported sleep efficiency or sleep duration.23,153,157,159,217–219

Risk of bias in studies eligible for the network meta-analysis

A summary of the results of the RoB2 for the studies contributing to the NMA is shown in Figure 2. The risk-of-bias judgements for each study are presented in Appendix 4, Figure 10.

FIGURE 2 Summary of risk-of-bias assessment of the included studies (RoB2). Reproduced with permission from Hudson et al.220 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Randomisation process

Around half of the included studies (46 studies) reported an appropriate method for random-sequence generation and allocation concealment and we rated them as being at low risk.51,53–58,60–62,67,73,76,80,81,85,87–91,93,95,96,98,101,102,105,107,112,114,115,117,119,122–124,125–128,131–133,139 We judged 42 studies as having ‘some concerns’, as they failed to report sufficient detail to provide assurance of an adequate method of allocation concealment (19 studies)52,59,63,68,72,74,77,79,83,86,94,97,99,100,108,109,111,120,138 or of both allocation concealment and sequence generation (23 studies).64–66,70,75,78,82,84,92,103,104,106,110,113,116,118,121,129,130,134,135,137,140 Two studies were judged to be at high risk due to a suboptimal method of allocation concealment,69 or because information on allocation concealment was insufficient and there was a baseline imbalance in age across treatment arms which could have influenced sleep quality outcome.71

Deviations from the intended interventions

As expected, for most of the non-pharmacological interventions, blinding of participants and personnel was not feasible due to the nature of the interventions and blinding of investigators and assessment staff (e.g. statisticians) was not consistently reported across studies.

Nineteen placebo-controlled pharmacological trials72,75,77,116,120–122,124,127,129,132–140 and one UC-controlled non-pharmacological trial53 were rated as being at low risk as they implemented blinding of participants and personnel and adopted an intention-to-treat analysis approach.

There were four studies with blinding of participants and personnel that we rated as having ‘some concerns’ because they documented an inappropriate method of analysis (e.g. completer or per-protocol analysis).123,125,126,131

Similarly, 36 studies that did not implement blinding procedures were rated as having ‘some concerns’.51,54,56,57,59–62,65,73,76,79–81,83–86,89–91,93,94,96,98–101,103–105,107,110,112,117,119 These studies either documented an appropriate method of analysis or documented an inappropriate method of analysis but provided no specific reasons to believe that this would have a substantial impact on the results.

Nine studies with blinding71,74,78,106,108,113,114,118,128 and 21 studies without blinding52,55,58,63,64,66–70,82,87,88,92,95,97,102,109,111,115,130 were judged to be at high risk of bias because they documented an inappropriate method of analysis and failed to provide sufficient information to ensure that its impact on the results would be minimal.

Missing outcome data

There were 53 studies with high levels of missing data (10% of participants or greater) that did not provide reasons for missing data, had differential dropout, or reported reasons for missingness related to treatment efficacy.51,52,54–56,58,63,67,69,74,76,78,81,87,88,91,92,94,95,98,100–102,104–106,108,109,111–116,118,119,121–124,127–129,131–137,139,140 We rated these studies as high risk. The remaining 36 studies permitted a judgement of low risk for this risk domain.

Measurement of the outcome

Given that our primary outcome (sleep quality) is a participant-reported outcome, we judged 33 studies with blinding of participants to be at low risk of bias53,71,72,74,75,77,78,106,108,113,114,116,118,120–129,131–140 and 57 studies with no blinding of participants to be at high risk of bias.51,52,54–70,73,76,79–105,107,109–112,115,117,119,130

Selection of the reported results

For most studies the study protocol was not available and therefore we assessed whether outcome measures and analyses reported in the methods section of the published article were comparable to those reported in the Results section. Three studies were rated as having ‘some concerns’, since they provided limited information on their analysis plan.88,100,130 The remaining studies were judged to be at low risk of bias.

Overall risk of bias

Overall, the majority of the included studies were judged to be at high risk of bias in at least one domain and were given an overall judgement of high risk of bias. Seven studies were given an overall judgement of ‘some concerns’72,75,77,120,125,126,138 and one study an overall judgement of low risk.53

Effectiveness outcomes – studies eligible for the network meta-analysis

Sleep outcome

Results for the sleep outcome are presented below and in Report Supplementary Material 4. The sensitivity analysis, where we assume a higher correlation for calculating SD for those studies that did not provide a change from baseline score, showed similar results and are available on request.

Of the 90 studies that reported a sleep outcome, we were able to include 65 studies (total number of participants = 8247, total number of assessed intervention categories = 35) in the NMA (see Appendix 5, Table 20 for the number of participants according to the type of intervention). Studies were excluded from the network if they did not provide enough data (13 studies),64,66,67,77,78,91,93,101,113,132,135,139,140 if they were disconnected from the main network (4 studies),72,73,77,120 evaluated an intervention and a comparator that belong to the same category (5 studies),94–96,104,115 or did not clarify whether the outcome was an index or a subscale of a validated scale (2 studies).126,130 In addition, one study was removed from the network because of data outliers (the mean and SD were considerably different).103 The studies excluded from the NMA are summarised in Report Supplementary Material 5. The network comprises 39 studies providing PSQI outcome data, 13 providing MOS-SS data, 6 JSS data, 3 VAS data, 2 SQ-NRS and 1 study each providing FMSD and NRS data. Figure 3 shows the network plot for eligible comparisons for sleep quality. Most interventions were compared to either placebo/sham or UC. Of the 35 interventions, the majority were non-pharmacological. In the case of cross‐over trials, we only used data from the first phase prior to the cross over.

FIGURE 3 Network diagram for sleep outcome. 1, placebo/sham; 2, education + flexibility exercise LD; 3, Mind–body Ex LD; 4, aerobic exercise LD; 5, education; 6, UC; 7, aerobic exercise AQ; 8, nutrition; 9, balneotherapy; 10, generic psychological or behavioural therapy; 11, Manual T; 12, relaxation; 13, Electro T; 14, flexibility exercise LD; 15, psychological or behavioural therapy, sleep-focused; 16, Mind–body Ex AQ; 17, mixed exercise AQ; 18, weight loss; 19, neuromodulation; 20, non-mainstream practice; 21 dental splint; 22, hyperbaric oxygen therapy; 23, aerobic exercise LD + flexibility exercise LD; 24, multidisciplinary; 25, flexibility exercise LD + Manual T; 26, balneotherapy + mixed exercise AQ; 27, tricyclics; 28, AP; 29, endogenous hormones; 30, antioxidant; 31, SRI; 32, gabapentinoid; 33, analgesic; 34, central nervous system depressants; 35 strengthening exercise LD. AQ, aquatic; LD, land-based. Note: circle size represents the number of randomised participants; line width represents the number of direct comparisons. Reproduced with permission from Hudson et al.220 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Table 3 shows the results of the direct and indirect comparisons for the included interventions versus placebo/sham (see Report Supplementary Material 1 for all other comparisons; pairwise comparisons are available on request). Compared with PBO/Sham (n = 2087), there was evidence of a beneficial effect on sleep for aquatic-based aerobic exercise training (n = 59, SMD −2.63, 95% CrI −4.74 to −0.58) and land-based aerobic exercise training in combination with flexibility exercise training (n = 32; SMD −4.69, CrI −8.14 to −1.28). We also found a positive effect for hyperbaric oxygen therapy (n = 9; SMD −4.51, CrI −7.44 to −1.56) but it is worth noting that it was derived from indirect evidence and based on the assessment of only nine participants in the intervention group. Compared with PBO/Sham, there was a suggestion of a modest effect on sleep for land-based strengthening exercise training (n = 56, SMD −0.95, CrI −3.89 to 2.04), psychological or behavioural therapy with a focus on sleep (PT/BT sleep; n = 94, SMD −0.89, CrI −2.39 to 0.61), weight loss (n = 41, SMD −1.15, CrI −3.55 to 1.27), Electro T (n = 20, SMD −0.98, CrI −3.28 to 1.34), dental splint (n = 29, SMD −1.62, CrI −4.862 to 1.65), tricyclics (n = 43, SMD −1.26, CrI −4.47 to 1.93) and AP (n = 53, SMD −1.28, CrI −3.56 to 0.97); however, this could not be confirmed with certainty because of the width of the CrIs. For most of the other pharmacological and non-pharmacological interventions, there was no clear evidence of an improvement in sleep quality. In addition, there was evidence that sleep outcome did not improve in participants randomised to aquatic Mind–body Ex training (n = 75, SMD 4.26, CrI 1.76 to 6.76) compared with those who received PBO/Sham. There was evidence of inconsistency between direct and indirect evidence for UC and aquatic-based aerobic exercise compared to land-based Mind–body Ex, land-based flexibility exercise compared to UC and land-based flexibility exercise compared to aquatic-based aerobic exercise (see Appendix 6, Table 21).

Quality-of-life outcomes

Fibromyalgia Impact Questionnaire

Of the 56 studies that used the FIQ to assess quality of life, 52 (total number of participants = 7127, total number of interventions = 35) were included in the NMA while 4 studies were excluded as they did not form part of the network. Results are presented below and in Report Supplementary Material 3. Appendix 5, Table 20 shows the total number of participants for each intervention. Figure 4 shows the network plot for FIQ. Within the network many interventions were never directly compared. Placebo/sham and UC were the most common comparators.

FIGURE 4 Network diagram for fibromyalgia impact questionnaire. 1, placebo/sham; 2, education + flexibility exercise LD; 3, Mind–body Ex LD; 4, aerobic exercise LD; 5, education; 6, flexibility exercise AQ; 7, UC; 8, aerobic exercise AQ; 9, nutrition; 10, balneotherapy; 11, PT/BT generic; 12, Manual T; 13, relaxation; 14, Electro T; 15, flexibility exercise LD; 16, PT/BT sleep-focused; 17, Mind–body Ex AQ; 18, mixed exercise AQ; 19, weight loss; 20, neuromodulation; 21, non-mainstream practice; 22, hperbaric oxygen therapy; 23, aerobic exercise LD + flexibility exercise LD; 24, PT/BT generic + relaxation; 25, multidisciplinary; 26, flexibility exercise LD + Manual T; 27, balneotherapy + mixed exercise AQ; 28, tricyclics; 29, antioxidant; 30, SRI; 31, iron replacement; 32, gabapentinoid; 33, analgesic; 34, AP; 35, central nervous system depressants. AQ, aquatic; LD, land-based; PT/BT, psychological or behavioural therapy. Note: circle size represents the number of randomised participants; line width represents the number of direct comparisons. (Figure not reproduced here; see the original article.)

The results of the direct and indirect comparisons for the included interventions compared with placebo/sham (n = 2263) are shown in Table 4 (see Report Supplementary Material 6 for all other comparisons; pairwise comparisons are available on request). Improvements in quality of life assessed using the FIQ were observed for participants undertaking land-based aerobic in combination with mixed flexibility exercise training (n = 32, MD −19.91, CrI −34.89 to −4.94), multidisciplinary training (n = 81, MD −17.31, CrI −28.38 to −6.29), land-based Mind–body Ex training (n = 420, MD −16.18, CrI −22.72 to −9.73), generic psychological or behavioural therapy (PT/BT generic) with relaxation (n = 29, MD −12.07, CrI −20.75 to −3.35), psychological or behavioural therapy targeted to sleep (PT/BT sleep-focused) (n = 77, MD −11.68, CrI −20.34 to −3.11) and generic PT/BT (n = 145, MD −6.23, CrI −12.02 to −0.62). Regarding pharmacological interventions, positive effects were observed for 97 participants receiving antioxidants (n = 12, MD −17.75, CrI −34.91 to −0.61), iron replacement (n = 38, MD −15.10, CrI −30.41 to −0.06), SRIs (SRI) (n = 573) (MD −9.85, CrI −15.80 to −3.80) and central nervous system (CNS) depressants (n = 881, MD −8.83, CrI −14.77 to −2.74). In general, magnitude of effects varied across interventions. A large positive effect was also observed after hyperbaric oxygen therapy (n = 9, MD −26.29, CrI −37.56 to −15.15); however, as explained before, we are not certain of the reliability of this estimate because it was derived from the assessment of a very small sample of patients (nine patients in the intervention group). Compared with placebo/sham treatments, there was no clear evidence that interventions such as education in combination with land-based flexibility exercise training (n = 33), aquatic flexibility exercise training (n = 39), UC (n = 551), relaxation (n = 67), land-based flexibility exercise training (n = 57), aquatic Mind–body Ex training (n = 60), mixed aquatic exercise training (n = 89) and neuromodulation (n = 76) had an effect on participants’ quality of life (although the width of the CrIs indicates considerable uncertainty). It is worth noting that for some intervention comparisons (generic PT/BT therapy compared with placebo/sham, and sleep-focused PT/BT compared with education or UC), the node-splitting analysis showed significant disagreement (inconsistency) between direct and indirect estimates. See Appendix 6, Table 22 for further details.

36-Item short form survey mental component summary score

Of the 17 studies that reported SF-36 mental component summary (MCS) scores, 15 (n = 359, 13 interventions) were included in the analysis while 2 were excluded as they did not form any link in the network. Results are presented below and in Report Supplementary Material 7. Appendix 5, Table 20 shows the number of participants included in each intervention. Figure 5 shows the network for SF-36 MCS score; placebo/sham and UC were the most common comparators.

FIGURE 5 Network diagram for 36-item short form survey MCS score. 1, placebo/sham; 2, education + flexibility exercise LD; 3, Mind–body Ex LD; 4, aerobic exercise LD; 5, education; 6, UC; 7, nutrition; 8, electro T; 9, non-mainstream practice; 10, antioxidant; 11, SRI; 12, gabapentinoid; 13, CNS depressants. LD, land-based. Note: circle size represents the number of randomised participants; line width represents the number of direct comparisons. (Figure not reproduced here; see the original article.)

The direct and indirect comparisons for the included interventions compared with placebo/sham (n = 1167) are shown in Table 5 (see Report Supplementary Material 7 for all other comparisons; pairwise comparisons are available on request). Land-based Mind–body Ex (n = 281, MD 7.27, 1,11 to 13.94) and education (n = 22, MD 10.31, CrI 2.06 to 19.35) were associated with an improvement in SF-36 MCS score compared with placebo/sham. In contrast, there was evidence that SF-36 MCS scores were worse after nutrition (n = 36) than after placebo/sham (n = 1167, MD −7.96, CrI −14.83 to −1.11) but no clear evidence that SF-36 MCS scores were worse after UC and Electro T than after placebo/sham. The remaining interventions showed no clear evidence of a positive effect when compared with placebo/sham. There was no need to check the inconsistency between direct and indirect estimates, as the only two closed loops in the network were from a three-arm trial with direct comparisons between arms.

36-Item short form survey physical component summary score

Of the 17 studies that reported SF-36 physical component summary (PCS) scores, 16 (total number of participants = 401, total number of interventions = 13) were included in the analysis and one was excluded as it did form any link. Results are presented below and in Report Supplementary Material 8. Appendix 5, Table 20 shows the number of participants included in each intervention. Figure 6 shows the network plot for SF-36 PCS score; placebo/sham and UC were the most common comparators.

FIGURE 6 Network diagram for 36-item short form survey PCS score. 1, placebo/sham; 2, education + flexibility exercise LD; 3, Mind–body Ex LD; 4, aerobic exercise LD; 5, education; 6, UC; 7, nutrition; 8, electro T; 9, non-mainstream practice; 10, antioxidant; 11, SRI; 12, gabapentinoid; 13, CNS depressants. LD, land-based. Note: circle size represents the number of randomised participants; line width represents the number of direct comparisons. (Figure not reproduced here; see the original article.)

The direct and indirect comparisons for the included interventions compared with placebo/sham are shown in Table 6 (see Report Supplementary Material 8 for all other comparisons; pairwise comparisons are available on request). Compared with placebo/sham (n = 1355), a better SF-36 PCS score was recorded after land-based Mind–body Ex training (n = 281) (MD 7.61, 3.56 to 13.06), land-based aerobic exercise training (n = 75, MD 6.17, CrI 1.05 to 12.81) and use of CNS depressants (n = 874, MD 2.93, CrI 1.10 to 4.79). There was insufficient evidence that Electro T (n = 20) had a positive effect on the SF-36 PCS score compared with placebo/sham (MD −0.82, CrI −4.13 to 3.97) and there was no clear evidence that the effects of the remaining interventions were different from those of placebo/sham. There was no need to check for the presence of inconsistency between direct and indirect estimates as the only two closed loops in the network were from a single three-arm trial.

Sleep duration and efficiency

Sleep duration was reported in two studies (total number of participants = 363, total number of interventions = 3). Results are presented below and in Report Supplementary Material 6. Figure 7 shows the network. There was insufficient evidence that gabapentinoid (n = 169) increased sleep duration compared to placebo/sham (n = 179) (MD 7.40, CrI −9.84 to 24.74), while SRI (n = 15) appeared to be detrimental to sleep duration compared to placebo/sham (n = 179) (MD −24.40, CrI −59.81 to 21.96) (see Report Supplementary Material 9, for further details).

FIGURE 7 Network diagram for sleep duration. 1, placebo; 2, SRI; 3, gabapentinoid. (Figure not reproduced here; see the original article.)

Insufficient data meant that we were only able to perform a pairwise meta-analysis for sleep efficiency. Results are available on request.

Ranking of interventions

Surface under the cumulative ranking curve and rank of the interventions for sleep (see Report Supplementary Material 4), FIQ (see Report Supplementary Material 6), SF-36 MCS score (see Report Supplementary Material 7), SF-36 PCS score (see Report Supplementary Material 8) and sleep duration (see Report Supplementary Material 6) are summarised in Report Supplementary Material 10. For both sleep and FIQ, hyperbaric oxygen therapy and land-based aerobic + flexibility exercise training were ranked as the top two (these were not evaluated for SF-36). However, it is important to note that SUCRA does not consider the magnitude of differences in effects between interventions, as well as the body and quality of evidence that contributes to each treatment comparison. Moreover, between the five considered outcomes, we observed some inconsistencies. For example, antioxidant and land-based Mind–body Ex were ranked low for sleep but not for FIQ.

Non-pharmacological interventions: sensitivity analyses

We conducted sensitivity analyses by removing pharmacological interventions from the main analyses. These sensitivity analyses were only relevant for the following outcomes: sleep (see Report Supplementary Material 1), FIQ (see Report Supplementary Material 3) and SF-36 MCS and PCS score (see Report Supplementary Materials 4 and 5). Apart from only minor differences, the results of the sensitivity analyses were similar to those of the main analyses that combined pharmacological and non-pharmacological interventions. For the ranking of the non-pharmacological interventions (see Report Supplementary Material 7), there were some minor changes.

Adverse events

For the studies eligible for the inclusion of the NMA on sleep, data for common AEs reported by ≥ 10% of participants and SAEs are reported in Tables 7 and 8. Overall, the AE data were reported by two studies which assessed non-pharmacological interventions,75,77 and 18 studies which assessed pharmacological interventions.116,118–124,126,129,132,133,135–137,139,140,207 Due to the presence of heterogeneity across included studies, AEs are summarised narratively and not combined in a meta-analysis.

Common adverse events in non-pharmacological studies

Two non-pharmacological studies reported common AE data. The first of these studies, conducted by Samartin-Veiga and colleagues (2021), compared three types of active tDCS treatment [M1, dorsolateral prefrontal cortex (DLPFC) and operculo-insular cortex (OIC)] with sham treatment.75 Apart from burning, which was experienced by an equal proportion of participants in the sham and tDCS arms (28% in each treatment arm), more people experienced tickling and itching in the sham arm than in the three combined tDCS treatment arms: 56% versus 43% and 52% versus 46%, respectively. In the study by Mirzaei and colleagues (2018), more people experience dizziness in the vitamin D 50,000 IU (plus trazodone 25 mg) treatment arm than in the placebo (plus trazodone 25 mg) treatment arm: 14% versus 3% respectively.77

Common adverse events pharmacological studies

Eighteen pharmacological studies reported common AEs. A meaningful comparison across studies proved challenging as most studies assessed different doses of individual pharmacological therapies and included different control treatments. Eight studies compared either gabapentinoid versus placebo or different dose regimes.121–124,137,139,140,207 Three studies compared CNS depressants [gamma-hydroxybutyric acid 25 mg/kg body weight118 and sodium oxybate (SXB) 4.5 g and 6 g]132,133 with placebo. AP treatments were assessed in two studies: one assessed quetiapine versus placebo116 and another quetiapine versus amitriptyline.119 One study each compared the following treatments with placebo: analgesic (ASP0819):136 iron replacement (ferric carboxymaltose 15 mg/kg up to 750 mg),126 SRIs (milnacipran 200 mg/day)129 and tetracyclic antidepressants (TeCAs) (mirtazapine 15 mg/day and 30 mg/day).135 One study compared 10 mg melatonin alone or in combination with a tricyclic (amitriptyline 25 mg) with a tricyclic (amitriptyline 25 mg) alone.120

In general, a greater incidence of AEs was observed in participants receiving active treatments than PBO/Shams. Higher doses of active treatment were usually associated with higher numbers of people reporting an AE. In the original studies, the reporting of the AEs did not always specify whether they related to study treatment. Common AEs that were frequently reported across trials that evaluated gabapentinoid treatment included dizziness, drowsiness (somnolence), headache, weight gain, dry mouth and peripheral oedema. CNS depressants were often associated with headache, sweating, nausea and dizziness, and AP treatments with dry mouth, somnolence and dizziness. Frequently reported common AEs associated with the other active treatments included dry mouth, increased appetite, somnolence, dizziness, nausea, headache, hyperhidrosis, flushing and constipation.

Adverse events that were reported by a high proportion of participants included somnolence and dry mouth. The study conducted by Potvin and colleagues reported that 80% and 70% of the participants experienced somnolence and dry month, respectively, after AP treatment, while 46% of the participants receiving placebo manifested somnolence.116 Similarly, in the study conducted by Yeephu and colleagues (2013) somnolence, increased appetite and dry mouth were reported by 86%, 86% and 71% of the participants treated with 30 mg/day mirtazapine, respectively.135 In the placebo and 15 mg/day mirtazapine arms of this trial, somnolence was reported by 69% of participant in both arms, increased appetite was reported by 23% of participants receiving placebo and by 69% of participants treated with 15 mg/day mirtazapine, and dry mouth was reported by 62% of participants receiving placebo and by 92% of those treated with 15 mg/day mirtazapine.

Serious adverse events in non-pharmacological studies

Two non-pharmacological studies reported SAE data. These include three SAEs (one torn meniscus; one persistent pain after spinal operation; one flu episode) associated with the sham arm in the trial conducted by Lauche and colleagues (2016); no SAEs were reported in the other two treatment arms (UC and cupping therapy).57 In the study conducted by Maindet and colleagues (2021), slightly more SAEs were reported in the UC arm [14/108 (13%)] than in the balneotherapy treatment arm [11/110 (10%)].51 None of the SAEs were reported as being related to treatment by the study authors.

Serious adverse events in pharmacological studies

Five pharmacological studies reported SAEs.122,123,133,137,207 Five studies compared gabapentinoids with placebo and one study CNS depressants with placebo. In the study conducted by Pauer and colleagues, incidence of chest pain and subsequent withdrawal of a participant who received 450 mg/day pregabalin was considered to be related to study treatment.137 Other SAEs reported by participants treated with gabapentinoids included breast cancer,122,123 vertigo,122 viral gastroenteritis123 and musculoskeletal stiffness.123

The study conducted by Zhang and colleagues (2021) documented one episode of atrial tachycardia and one case of cerebral haemorrhage in participants receiving placebo.207 The authors report that atrial tachycardia occurred in a 71-year-old female that was not resolved by end of the study and cerebral haemorrhage occurred in a 55-year-old female that was resolved without residual symptoms. The authors did not report any further information about these SAEs.

Russell and colleagues (2011) recorded an episode of treatment-related sleep paralysis, which lasted 30 minutes and occurred 45 minutes after the participant’s first exposure to the CNS depressant SXB (2.25 g).133

In brief, the type and frequency of both common and SAEs are in line with the known safety profiles of the active treatments reported by the included pharmacological studies.

Effectiveness outcomes – studies not eligible for the network meta-analysis

Sleep outcome

Results for sleep-related outcomes from the 78 studies not included in the meta-analysis are summarised narratively and presented in Report Supplementary Material 11.

Self-reported sleep quality was reported using a variety of measures: 29 studies used VAS or NRS,144,151,155,156,162–186 26 studies used a questionnaire143,146,149,150,154,187–197,199,201–207,221 and 13 studies used a sleep subscale of a non-sleep-specific measure such as FIQ.152,158,160,161,208–216

Only nine of the VAS or NRS included a measure of ‘sleep quality’,151,155,165,172,173,176,180,182,186 while most included measures related to sleep disturbance, sleep problem, sleep disorder, unrefreshed sleep or how pain affected sleep. The 24 studies reporting a VAS or NRS for non-pharmacological interventions all used different scales. Three of these studies comparing placebo/sham with PT/BT generic,172 neuromodulation165 and non-mainstream practice (non-MSM practice),183 respectively, reported a statistically significant p-value (< 0.05), although it is not clear whether the differences were clinically important. Of the five studies reporting a VAS or NRS for pharmacological interventions, the results in one study were reported to be statistically significant, possibly favouring SERMs over placebo/sham, although the scale used in this study was not fully described.185

Among the 26 studies reporting sleep quality outcome using a questionnaire, the Insomnia Severity Index (ISI) was used by five studies149,150,195,201,202 and the Post-Sleep Inventory by three studies,189,196,197 while all other questionnaires used were reported by a single study. Of the 19 studies using a questionnaire for non-pharmacological interventions, 10 studies reported a statistically significant effect compared with PBO/Sham favouring strength LD,150 education (sleep hygiene),187 PT/BT generic (e.g. CBT),187,192,195,201 Electro T,188,191 neuromodulation,190 phototherapy + mixed exercise training193 and non-MSM practice,154 although the difference between treatment groups was relatively small. Of the seven studies with a questionnaire focusing on pharmacological interventions, three studies reported a statistically significant effect favouring gabapentinoid over placebo/sham,207 tricyclics over SSRI,204 and combined SSRI and gabapentinoid over combination of SRIs and gabapentinoid or tricyclics and gabapentinoid,206 respectively, although details of the questionnaires used were not fully reported.

Of the 13 studies that used a sleep subscale of a non-sleep-specific measure, 6 showed a statistically significant effect compared with PBO/Sham favouring multicomponent therapy,208 Electro T,209 neuromodulation,160 non-MSM practice,210 SRIs215 and acetylcholine esterase inhibitor (pyridostigmine).161

Ten studies assessed sleep efficiency or sleep duration.23,153,157,159,217–219 Two of these studies reporting sleep duration by a self-reported measure (sleep log) showed a statistically significant effect favouring mixed aquatic exercise training + aquatic Manual T over Manual T alone217 and Manual T over relaxation/meditation.218 No suitable data (presented according to treatment periods) were available from three crossover trials that measured sleep efficiency or sleep duration after pharmacological interventions.145,147,148

Adverse events

For the studies not eligible for the inclusion in the NMA on sleep, common AEs (reported by ≥ 10% of participants) reported by 11 studies and SAEs reported by one study, are presented in Report Supplementary Material 12.

Common adverse events in non-pharmacological studies

Common AEs reported by ≥ 10% of participants were documented by three non-pharmacological studies,160,195,211 comparing non-pharmacological interventions against UC, PBO/Sham. The active intervention categories included PT/BT generic (CBT),195 neuromodulation [electroencephalography neuro-biofeedback,211 and repetitive transcranial magnetic stimulation (rTMS)].160 Compared with UC, more people receiving CBT experienced increased pain [4/70 (6%) vs. 9/70 (13%), respectively]. Compared with PBO/Sham, higher proportions of people treated with neuromodulation experienced AEs. Common AEs that were reported across studies included fatigue, tiredness, pain, sleep drowsiness or change in sleep patterns and stiffness or muscle spasms.

Common adverse events in pharmacological studies

Eight pharmacological studies reported common AEs. Seven studies focused on adults147,185,186,204,214–216 and one study on adolescents aged 12 to 17 years.155 The studies that enrolled an adult population compared the following active treatments with placebo or sham (single studies): gabapentinoids (pregabalin 300–450 mg/day),147 opioid antagonist (naltrexone 4.5 mg/day),186 serotonin receptor antagonist and dopamine receptor agonist (terguride 3 mg/day maximum),216 SERMs (raloxifen, 60 mg every other day),185 SRI [duloxetine 60 mg once daily (QD) or duloxetine 60 mg twice daily]215 and SSRI (citalopram 20–40 mg daily).214 One study compared an SSRI (paroxetine 20 mg/day rising to 40 mg/day) with a tricyclic (amitriptyline 10 mg/day rising to 20 mg/day).204 The study conducted among adolescents compared gabapentinoid treatment pregabalin (75–450 mg/day) with placebo.155

More people reported experiencing AEs in the active treatment arms than in the control arms. Frequently reported common AEs included nausea, headache, dizziness, dry mouth, constipation, diarrhoea and somnolence. In the study conducted by Capaci and Hepguler (2002) comparing paroxetine with amitriptyline, more participants in the amitriptyline arm experienced dry mouth, constipation and rash than those in the paroxetine arm.204 AEs that occurred more frequently in the paroxetine arm included sleepiness, diarrhoea, nausea, dizziness and sexual dysfunction.

Serious adverse events

In general, few SAEs were reported by the authors of the individual studies. One study conducted by Arnold and colleagues (2016) comparing pregabalin (75–450 mg/day) with placebo in adolescents reported two SAEs, cholelithiasis and major depression, occurring in one participant treated with pregabalin.155 No SAEs were reported among adolescents receiving placebo in this study.

In brief, the type and frequency of the common and SAEs are in line with the known safety profiles of the active treatments reported by the pharmacological studies.

Discussion

Summary of main results

This evidence synthesis included a total of 90 RCTs assessing sleep quality using PROMs validated for use in fibromyalgia patients, and a further 78 RCTs assessing other sleep-related outcome measures. To our knowledge, this evidence synthesis and NMA is the most comprehensive approach to assess the current evidence on pharmacological and non-pharmacological interventions for fibromyalgia-related sleep problems.

The 90 RCTs that provided sleep quality PROM data involved a total of 12,082 adult participants. Most studies were parallel-arm trials with a small sample size (median 70) and short follow-up (median 12 weeks). Five studies were crossover randomised trials. The included studies evaluated 45 active treatment categories (grouped by mode of action). The most commonly evaluated treatment categories were land-based Mind–body Ex training (e.g. Tai Chi) in 13 studies,54,55,70,71,80,81,83–85,89,110,111,115 generic psychological and behavioural interventions, which were not targeted to address sleep problems (e.g. CBT, ACT, mindfulness) in 10 studies,60–63,96,100–103,112 gabapentinoids (e.g. pregabalin) in 8 studies,121–125,137,139,140 neuromodulation (e.g. tDCS) in 5 studies,72,74–76,106 nutrition (e.g. food supplement) in 5 studies,58,59,94,95,113 CNS depressants (e.g. SXB) in 4 studies,118,131–134 SRIs (e.g. milnacipran) in 4 studies127–129,134 and non-MSM practice (e.g. dry needling or electroacupuncture) in 4 in studies.57,78,93,114 Most other intervention categories were assessed only in either a few or single trials.

Considering that the scope of this evidence synthesis was to assess the effects of current interventions for treating fibromyalgia-related sleep problems, we focused specifically on studies reporting sleep outcome data. We assessed study eligibility based on the description of any sleep outcome (defined either as primary or secondary outcome) in the title and abstract of citations identified by our search strategies regardless of whether it contained numerical results. Studies that did not mention sleep outcomes in their titles and abstracts were excluded. This allowed us to identify studies that had a prominent focus on fibromyalgia-related sleep problems but may have limited inclusion of studies that assessed other fibromyalgia symptoms which could interfere with sleep, or studies that did not find significant effects on sleep outcome. We conducted a check of a 10% random selection of the excluded studies and identified no studies that reported sleep outcome data that could have been included in the NMA. Due to resource constraints, a single review author carried out study selection, data extraction, and risk-of-bias assessment, which were subsequently checked by another review author.

The findings of our NMA, which included 65 studies assessing sleep using validated PROMs, show that, compared with PBO/Sham, some forms of exercise training such as land-based aerobic exercise training combined with flexibility exercise training, and aquatic aerobic exercise training may improve sleep, although our certainty in the current evidence, assessed using CINeMA, is generally low. For all other pharmacological and non-pharmacological interventions, there was only a modest positive effect on sleep compared with PBO/Sham (CrIs indicated uncertainty and the quality of evidence is low to very low). Notably, we did not observe a significant, beneficial effect of pharmacological interventions on sleep quality.

Furthermore, compared with PBO/Sham, some interventions positively affected participants’ quality of life. Using the FIQ, an improvement in quality of life was observed among participants who undertook land-based aerobic and flexibility exercise training, multidisciplinary training, land-based Mind–body Ex training, either PT/BT generic or PT/BT tailored to sleep problems, generic PT/BT alongside relaxation, and pharmacological treatments including antioxidant, iron replacement, SRIs, and CNS depressants, although the magnitude of effect varied. An improvement in the SF-36 mental health component summary score was observed after land-based Mind–body Ex and education interventions, while an improvement in the SF-36 PCS score was observed after land-based Mind–body Ex training, land-based aerobic exercise training and use of CNS depressants.

In general, non-pharmacological treatments under investigation were reported to be reasonably well tolerated and AEs were usually reported to be of mild or moderate severity (e.g. stiffness, fatigue). For pharmacological treatments, commonly reported adverse effects included dizziness, somnolence, headache and dry mouth.

Strengths and limitations

To determine the impact of our findings on current clinical practice it is important to take into consideration the limitations of our NMAs. Overall, our analyses were hampered by the limitations of the current evidence base; specifically, the lack of head-to-head comparisons for active treatments. Most interventions were compared to either a placebo, sham or UC; however, the control treatments were often inadequate to test the study hypothesis (e.g. the choice of ‘usual care/waiting list’ as comparator treatment for hyperbaric oxygen therapy). We were also unable to include all studies identified by our search strategies in the NMA because some interventions were outliers or disconnected from other interventions in the network. Most of the studies that contributed to the network were small, with short-term follow-ups (around 3 months), and assessed a diverse range of interventions. The 90 studies that reported a PROM of sleep quality and were eligible for inclusion in the NMA assessed a total of 97 different active interventions. To make the NMA feasible, we grouped these active interventions into 45 categories. Specifically, among the 65 studies that contributed to the sleep quality NMA, 35 different treatment categories were considered but only a limited number of studies, usually of small sample size, were available for each treatment comparison. Most studies included fewer than 100 participants and some even fewer than 20 participants in the active treatment group. It is important to acknowledge that while interventions included in the same category were similar as we grouped them according to their characteristics and mode of action, inevitably they were not the same. It is also worth noting that many of the studies included in the NMA were conducted in the USA and Spain, with only three studies conducted in the UK.

While the diverse range of existing interventions may suggest that there is no one-size-fits-all approach for managing sleep problems in fibromyalgia, it hampers the possibility to reach any definite conclusion. The characteristics of exercise-based interventions varied considerably across studies in terms of intensity, frequency and duration, as well as description of the specific elements of the physical training/programme, making treatment comparisons very challenging and impossible to establish which exercise protocols or ‘active ingredients’ are more likely to be effective for the treatment of sleep problems in people with fibromyalgia. Moreover, there was variability in terms of mode of delivery, with some interventions delivered as ‘group activity’ and others at an individual level. We were also only able to analyse average treatment effects and not relevant clinical and demographic modifiers at the patient level (e.g. severity of disease, duration of illness, lifestyle, extent and nature of sleep disturbances, and level of physical activity before and during treatment). It is also worth noting that while pharmacological interventions were usually assessed against placebo, non-pharmacological interventions were compared with control interventions in very different ways in the included studies. Some of the included non-pharmacological studies failed to compare their active interventions to sham procedures that involved appropriate control strategies in terms of exposure time (frequency and duration) and ‘attention’ received by the therapist/instructor. Appropriate sham controls have been used in similar clinical areas and are considered particularly useful for studies with subjective or self-reported end points (e.g. improvement of symptoms) and when the risk of the sham procedures is low (e.g. less intensive, or generic physical activity/procedure).222,223 Furthermore, the recent CoPPS Statement on the development, implementation and reporting of control interventions in efficacy trials of physical, psychological and self-management therapies recommends designing control interventions that are as similar as possible to the interventions under investigation, apart from the components whose effect the trials aim to study.224 The use of inadequate controls (e.g. no intervention, UC) may result in an exaggeration of effect estimates or produce misleading results and often is methodologically inadequate to test the study hypothesis. Therefore, the diverse characteristics of both active and control interventions across included studies as well as the intrinsic limitations of the design of most studies included in the NMA (no adequate control treatments, small sample size, short duration) may explain differences with respect to the effects of interventions and the incoherence we observed between some direct and indirect estimates of effects.

Moreover, our primary outcome, sleep quality, was not measured consistently across included studies; several different PROMs were used (i.e. PSQI, MOS-SS, JSS, FMSD, SQ-NRS as well as single-item VAS/NRS measuring a similar sleep quality construct to that of the SQ-NRS). As there is no consensus on which is the best outcome measure to use in the field of fibromyalgia, we decided to combine studies irrespective of the way sleep quality was measured provided that a validated instrument was used. This again might have contributed to increasing heterogeneity and inconsistency in the network and limited the reliability of our findings. Furthermore, for some outcome measures that assessed sleep quality, the interpretation of results was complicated by the fact that we did not find reliable information on what can be considered a minimally important clinical difference.

According to CINeMA, for many comparisons included in our NMA, our certainty of the evidence was rated as low to very low. The level of certainty was downgraded primarily due to within-study bias, imprecision, heterogeneity and incoherence. For within-study bias, most studies were judged to be at an overall high risk of bias mostly due to an inadequate reporting of randomisation and allocation concealment methods as well as issues related to missing outcome data. For studies evaluating non-pharmacological treatments, blinding of participants and personnel is particularly challenging due to the nature of interventions. However, unblinded patients who undergo a specific procedure/treatment may have higher expectations about improvement while, in contrast, those who are aware of receiving no intervention, a different control intervention or UC may have much lower expectations or a nocebo response. These higher and lower expectations can influence outcome measurement, especially when symptoms are subjective, self-reported and susceptible to changes. The certainty level was downgraded for imprecision because of the low number of studies available for each comparison and their small sample size. Moreover, heterogeneity and incoherence across comparisons also yielded a downgrading in the level of certainty. Given our CINeMA findings, we were unable to conduct sensitivity analyses restricted to high-quality studies.

The SUCRA approach is the most common method for ranking NMA results and identifies the ‘best’ and ‘worst’ treatment in terms of the probability of success for each analysed outcome. However, given the low and very low quality of the evidence included in our NMA and the small amount of available evidence for most comparisons, the results of our SUCRA ranking may generate misleading and unreliable inferences and, therefore, we have little confidence in them. Moreover, it is worth stressing that one limitation of the SUCRA approach is that it does not consider the magnitude and precision of effect estimates.

The 78 trials which evaluated sleep outcomes using non-PROM instruments or tools, and therefore were not included in the NMA, involved a total of 5911 randomised participants (5804 adults and 107 adolescents). The reporting of sleep outcomes was not consistent across studies and various scales and questionnaires were used. Apart from the Insomnia Severity Index (ISI) reported in five studies149,150,195,201,202 and the Post Sleep Inventory reported in three studies,189,196,197 no other common outcome measures were assessed in the remaining studies. These studies presented the same limitations as those included in the NMA and even though we summarised their results, we could not draw any firm inferences about treatment effects.

It is also worth noting that in the included studies most participants were middle-aged women from high-income countries. Information on ethnicity and level of education was often not reported (NR). Therefore, our findings cannot be generalised to all people with fibromyalgia and different settings.

In brief, regarding the effects of interventions for the management of sleep problems in people with fibromyalgia, the current evidence base is very fragmented, with many diverse interventions often assessed by small, individual trials generally of poor quality. The considerable heterogeneity in terms of study design, characteristics of interventions and exercise/practice protocols hampers the possibility to draw firm conclusions. There is an indication that some interventions may produce positive effects on sleep or on quality of life in people with fibromyalgia. The beneficial effects of physical activity on health and well-being are well known and it makes sense to encourage fibromyalgia patients to follow healthy lifestyle recommendations. However, the limitations of the current evidence base preclude any specific recommendation in terms of optimal interventions for managing sleep problems. It is difficult to fathom why land-based aerobic training in combination with flexibility training may be effective but land-based aerobic training alone or flexibility training alone is not; similarly, it is difficult to understand why aquatic-based aerobic exercise training may be effective, but aquatic Mind–body Ex training and aquatic mixed exercise are not. It is also difficult to figure out why some pharmacological interventions may improve quality of life but not sleep. Our original plan was to conduct a component NMA to disentangle the effect of each component of the interventions assessed by the included studies, but this proved impossible due to the lack of suitable data.

There are several published systematic reviews assessing different forms of exercise training and other non-pharmacological interventions for the management of fibromyalgia symptoms.225–228 Although their primary focus is not on sleep problems, they all point out similar limitations to those we have observed here, including the heterogeneity among included studies in terms of protocols, the insufficient evidence to establish the effectiveness of one intervention compared with another, the lack of proper comparator treatments, the lack of appropriate statistical power in most studies and the low-to-moderate quality of the evidence base. A systematic review conducted by Kundakci et al. in 2022, which aimed to evaluate the effectiveness of non-pharmacological treatments for fibromyalgia, revealed that all types of exercise, except for flexibility exercises, helped ease pain. Both mind–body and strengthening exercises were found to be helpful in reducing fatigue, while aerobic and strengthening exercises were found to be useful in improving sleep.31 However, the authors identified only a limited number of studies usually of small sample sizes for each form of exercise (10 for aerobic exercise, 9 for strengthening and 2 for flexibility) and found considerable heterogeneity and high risk of bias among included studies. They also acknowledged that outcome measures, intervention programmes and control interventions varied considerably across studies, which is in line with our results. Findings for other types of interventions were found to be less convincing.