This SR with meta-analysis (MA) followed the guidelines for Preferred Reporting Items for Systematic reviews and Meta-Analyses and was registered in the PROSPERO database (CRD42017074982).
Data Sources and Searches
A search strategy with 3 domains (study design, disease, and intervention) was developed, and the key search terms were modified in accordance with the glossary of each database and combined using Boolean operators. A search strategy designed for MEDLINE is given in Supplement Table 1 (available at https://links.lww.com/PAIN/B513). We searched MEDLINE (Ovid), EMBASE (Ovid), AMED (Ovid), PsycINFO (Ovid), CINAHL (EBSCOhost), and Web of Science (Core collection) from their dates of inception until September 2017 and first 100 articles on Google Scholar. In addition, the bibliographies of previous SRs[32,76,82,121] were searched manually to augment the literature search. The search was updated in June 2020.
Studies were eligible for inclusion if they were RCTs comparing a nonpharmacological intervention with either usual care, waiting list, placebo, or sham treatment, in people aged 16 years or more with either physician-diagnosed fibromyalgia or meeting any of the fibromyalgia classification or diagnostic criteria.[175,200–202] A modified nonpharmacological intervention classification system was prepared based on previous reviews or guidelines[81,114] (Supplement Table 2, available at https://links.lww.com/PAIN/B513). Nonpharmacological interventions such as education, different forms of exercise, electrotherapy, balneotherapy, complementary and alternative medicine, and psychological treatments were included. If the exercise intervention included 2 or more exercise types from aerobic, strengthening, flexibility, or mind–body exercise, it was considered as mixed exercise for the purpose of this SR. Psychological treatments included cognitive behavioural treatment (CBT), mindfulness, hypnosis, acceptance and commitment therapy, and attachment-based compassion therapy. Patient education was classified as present if it was delivered either face to face or by a leaflet. In this SR, multidisciplinary treatment (MDT) referred to an intervention that included exercise intervention, patient education, and psychological treatment.
The following studies were excluded: quasirandomised studies and nonrandomised trials; studies including participants with other musculoskeletal disease (eg, rheumatoid arthritis), chronic fatigue syndrome, chronic widespread pain not meeting classification criteria for fibromyalgia or where those criteria were not applied, age < 16 years; studies evaluating combinations of pharmacological and nonpharmacological interventions or comparing pharmacological treatment with a nonpharmacological treatment; studies not assessing any prespecified outcomes; and studies only reported as conference abstract. There was no language restriction.
Disease-specific QoL assessed using Fibromyalgia Impact Questionnaire (FIQ) was the primary outcome measure. Pain, fatigue, sleep, and depression were secondary outcomes. As different outcome measures were used in different trials, we took a broad approach and data for all outcome measures for each secondary outcome were extracted. Where multiple outcome measures were used to report on an outcome, we used a prespecified hierarchy based on the OMERACT recommendations (Supplement Table 3, available at https://links.lww.com/PAIN/B513).
Citations were imported to Endnote X8. After removal of duplicates, titles and abstracts were examined against inclusion and exclusion criteria. The study selection process, including title–abstract screening and full-text screening, was conducted by one reviewer (B.K.) and subsequently another reviewer (J.K.) undertook validation on 10% of randomly selected studies. The validation exercise was completed for each step of the study selection process, and the discrepancies were discussed with the 4 senior authors as a group (A.A., M.D., W.Z., and M.H.) before proceeding with completing that particular step in the review. There was 95% agreement between 2 reviewers (more than 80% agreement as recommended by AMSTAR 2), and this was considered sufficient to continue to the next stage.
Data Extraction and Quality Assessment
A Microsoft Access database was developed, and the data were extracted and entered. The risk of bias (RoB) was evaluated using a modified version of the Cochrane RoB assessment tool consisting of a checklist of 7 items that assess RoB by evaluating the procedures of selection, detection, attrition, and reporting.
Data extraction and RoB assessment were conducted by B.K., and another author (J.K.) independently performed validation in a 10% of random sample. There was 92% agreement (more than 80% agreement as recommended by AMSTAR 2). Any disagreement between reviewers was resolved by a senior researcher (A.A.) who served as a adjudicator.
Missing data: Where required data were not published, the authors were contacted for additional information. If this was unsuccessful, the missing data were estimated using other values reported in the articles (eg, If SD was not reported, it was calculated from SE or 95% confidence interval (CI) and sample size as recommended in the Cochrane handbook). The formulae used for these calculations are included in Supplement Table 4 (available at https://links.lww.com/PAIN/B513). Where this calculation was not possible because of insufficient information, the largest SD among eligible studies for that outcome was substituted, provided the outcome scale was named in the publication. If the name of the outcome scale was not reported, the arithmetic mean of all SDs for that outcome was used as recommended by Agency for Healthcare Research and Quality (United States).
Data Synthesis and Analysis
Standardised mean difference or Cohen d was used to measure the effect size (ES) as the data were continuous. The mean change score from baseline was used to calculate the ES. All analyses were based on the random-effects model using the DerSimonian and Laird method. Heterogeneity between studies was assessed using the I2 statistic. Interpretation of an I2 value was as follows: 0% to 40%—heterogeneity might not be important, 30% to 60%—moderate heterogeneity, 50% to 90%—substantial heterogeneity, and 75% to 100%—considerable heterogeneity.
Studies were analysed separately based on their comparison types, namely, usual care, placebo or sham controlled, and A + B vs B designed studies. For the purpose of this review A + B vs B study designs were defined as studies that compared 2 interventions that were offered actively against a single intervention, for example, Relton et al. (2009), Kibar et al. (2015), Kutlu et al. (2020), etc. Studies that evaluated an intervention against continued usual care, waiting list, or no treatment were not considered as A + B designs for the purpose of this review as continued usual care would be available to participants on wait list or those not being offered any additional treatment in the trial and also to participants in the intervention arm.
Egger test and the visual inspection of funnel plot asymmetry were used to assess publication bias. The ES of each intervention category (eg, exercise, psychological therapies etc.) was examined. Additionally, the ES of different exercise types, for example, aerobic, flexibility, mind body, strengthening, and mixed (ie, exercise packages that include 2 or more types of exercise), and different psychological treatments, for example, cognitive behavioural therapy (CBT) and mindfulness, were pooled separately.
All time points at which outcomes were assessed were extracted. The most commonly reported end point (12 weeks or closest to 12 weeks) was used as the primary time point for the MA. Effect sizes for A + B vs B designs were presented separately.
Subgroup analyses were undertaken according to participant characteristics (age and body mass index [BMI]), recruitment source (hospital, community based, or mixed), source of funding (noncommercial, commercial, both, or no specific funding), and sample size to explore their impact on heterogeneity and ES. Metaregression was used to investigate the impact of various study characteristics on ES for interventions with >10 trials. A time-dependent subgroup analysis was conducted, stratifying studies according to the time point from which outcomes were meta-analysed (1–8 weeks, 9–15 weeks, and more than 16 weeks). A further subgroup analysis was performed to see the influence of time gap between end-of-treatment and outcome assessment (0-0.5 weeks, 2–6 weeks, 10–14 weeks, and 18–44 weeks).
The robustness of the results of the MA was examined by undertaking sensitivity analyses on primary outcome by excluding studies that (1) had a high RoB on allocation concealment and attrition, (2) required imputation of SD, and (3) used end point scores rather than change scores for calculating the ES. Data were analysed using StataSE 16.
Pain. 2022;163(8):1432-1445. © 2022 Lippincott Williams & Wilkins