The search strategy comprised seven electronic databases from their inception through March 01, 2014: Medline/PubMed, Scopus, the Cochrane Central Register of Controlled Trials, PsycINFO, CAM-Quest, CAMbase, and IndMED. The literature search was constructed around search terms for 1. “yoga” and 2. “multiple sclerosis” and adapted for each database as necessary. The complete search strategy for Medline was as follows: (“Multiple Sclerosis”[Mesh] OR “Multiple Sclerosis”[Title/Abstract] OR “MS”[Title/Abstract] OR “Disseminated Sclerosis”[Title/Abstract]) AND (“Yoga”[Mesh] OR “Yoga”[Title/Abstract] OR “Yogic”[Title/Abstract] OR “Asana”[Title/Abstract] OR “Pranayama”[Title/Abstract] OR “Dhyana”[Title/Abstract]). Additionally, reference lists of identified original articles or reviews; and the tables of contents of the International Journal of Yoga Therapy and the Journal of Yoga & Physical Therapy were searched manually.

Two reviewers independently screening and selected abstracts; potentially eligible articles were read in full by two reviewers. Disagreements were settled through a discussion with a third reviewer until consensus was reached. If necessary, additional information was obtained from the authors of the primary study.

Two reviewers independently extracted data on patients (e.g. age, gender, ethnicity), methods (e.g. randomization, allocation concealment), interventions (e.g. yoga type, frequency, and duration), control interventions (e.g. type, frequency, duration), outcomes (e.g. outcome measures, assessment time points), and results using an a priori developed data extraction form. Discrepancies were discussed with a third reviewer until consensus was reached.

Two reviewers independently assessed risk of bias using the Cochrane risk of bias tool [13]. This tool assesses risk of bias on seven domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. For each domain, risk of bias was assessed as low; unclear; or high risk of bias. Discrepancies were discussed with a third reviewer until consensus was achieved.

Statistical heterogeneity between studies was analyzed using the I² statistics, a measure of how much variance between studies can be attributed to differences between studies rather than chance. The magnitude of heterogeneity was categorized as (1) I² = 0–24%: low heterogeneity; I² = 25–49%: moderate; I² = 50–74%: substantial; and I² = 75–100%: considerable [13], [16]. The Chi² test was used to assess whether differences in results are compatible with chance alone. Given the low power of this test when only few studies or studies with low sample size are included in a meta-analysis, a P-value≤0.10 was regarded to indicate significant heterogeneity [13].

Subgroup analyses were planned for the type of yoga intervention (yoga interventions including physical postures vs. yoga interventions without physical postures). As all included studies comprised yoga postures, subgroup analyses could not be performed.

To test the robustness of significant results, sensitivity analyses were conducted for studies with high versus low risk of bias at the following domains: selection bias (random sequence generation and allocation concealment), detection bias (blinding of outcome assessment), and attrition bias (incomplete outcome data).

If present in the respective meta-analysis, subgroup and sensitivity analyses were also used to explore possible reasons for statistical heterogeneity.

If at least 10 studies were included in a meta-analysis, assessment of publication bias was originally planned by using funnel plots generated using Review Manager software [13], [17]. As less than 10 studies were included in each meta-analysis, funnel plots could not be analyzed.

The literature search retrieved 96 non-duplicate records of which 87 were excluded on the basis of title and abstract because they did not meet all pre-defined inclusion criteria (Table S1). Nine full-text articles were assessed for eligibility [18]–[26]. All nine full-text articles reporting on seven RCTs involving a total of 670 patients met the inclusion criteria and were included in the qualitative analysis and meta-analysis (Figure 1). One full-text article was published in Persian [25], the others in English.

Characteristics of the sample, interventions, outcome assessment, and results are shown in Table 1.

Risk of bias in individual studies is shown in figure 2. One study each had reported adequate random sequence generation [24] or allocation concealment [21], [22]; and three studies each reported adequate blinding of outcome assessement [21]–[24] or were free of suspected selective reported [21], [22], [26].

Since no study had low risk of selection or attrition bias, no effect remained significant in sensitivity analyses for low risk of selection or attrition bias. In studies with low risk of detection bias, the effect of yoga compared to usual care on mood did not change substantially (SMD = −0.63; 95% CI−1.25, −0.01; p = 0.05). No other effects remained significant (table 2).

In this systematic review of seven randomized trials on yoga for multiple sclerosis, evidence for positive short-term effects of yoga on fatigue and mood, but not on more objective physician-rated outcomes such as mobility or cognitive function were found. No effect was robust against potential methodological bias. Although the overall high risk of bias hinders definite conclusions, yoga seems to be equally effective as exercise interventions in improving both patient-reported and physician-rated outcomes.

Safety of the intervention was insufficiently reported. Specifically, only one study explicitly assessed adverse events. However, exacerbations of multiple sclerosis were fewer or equal in number in the groups compared to the usual care or exercise groups. This is in line with previous cross-sectional studies [27], [28] and systematic reviews of yoga interventions in other patient populations that found no evidence for serious yoga-associated adverse events [29]–[33]. It should however be considered that yoga has occasionally been associated with serious adverse events in case studies [34].

No systematic review specifically on yoga for multiple sclerosis was available. However, a recent review on studies on mind-body medicine for multiple sclerosis published until March 2012 [35] included one RCT on yoga [24]. This review concluded that while yoga improved multiple sclerosis-related fatigue with fewer side effects than conventional treatment, more research was needed to draw definite conclusions. A recent systematic review on exercise for multiple sclerosis concluded that exercise therapy including yoga [24] may have beneficial effects on patients with multiple sclerosis without adverse events and might thus be recommended for the rehabilitation of multiple sclerosis patients [36]. A systematic review on yoga for fatigue that also conducted a meta-analysis included two trials on multiple sclerosis [24], [26] amongst trials other conditions. This meta-analysis found moderate evidence for effects of yoga on fatigue with larger effect sizes in non-cancer populations [37]. The recent guideline on complementary and alternative medicine in multiple sclerosis by the American Academy of Neurology included a systematic review on studies of yoga for multiple sclerosis [38], [39]. This review included four studies [19], [21], [24], [26] and concluded that, mainly due the low power of the available trials, the data were inadequate to assess the effect of yoga on disability, spasticity, fatigue, cognition, mood, balance, or walking speed in patients with multiple sclerosis [38], [39].

Mainly patients from the Middle East and Europe were included. Given that ethnical and geographical factors influence the incidence and severity of multiple sclerosis [1], [2], the applicability of these findings in other geographical regions is limited. Moreover, since mainly female patients were included, the results might not be fully applicable to male patients. Four studies did not report on the eligible types of multiple sclerosis [18]–[20], [24], [25]. This further limits the applicability of the results.

Overall, risk of bias of the included studies was unclear or high. Most importantly, only one study each reported adequate random sequence generation [24] or allocation concealment [21], [22]. As inadequate allocation concealment has been empirically demonstrated to be the most important source of bias in RCTs [40], this strongly limits the interpretability of results. In sensitivity analyses, no effect was robust against all potential sources of bias. The internal validity of the results is thus limited.

To the best of our knowledge, this is the first systematic review and/or meta-analysis available on yoga for patients with multiple sclerosis. Strengths of this review include the comprehensive literature search without language restrictions and the assessment of applicability of the results [41]. The primary limitation of this review is the paucity of eligible studies, rendering subgroup analyses impossible and resulting in a relatively limited overall sample size. Another major limitation is the insufficient reporting and/or low methodological quality of the included studies, limiting the interpretability of the results. Moreover, the reliability of the Cochrane risk of bias tool in systematic reviews on clinical trials of behavioral interventions has been questioned. It has been demonstrated that there is poor inter-rater reliability for several of the domains, specifically for the domain ‘blinding of participants and personnel’ [42]. Since it is generally regarded impossible to blind patients and therapists in trials of behavioral interventions such as yoga, different research groups tend to rate this domains differently if no information on blinding is provided. Accordingly, this domain was not used in sensitivity analyses. The applicability of the findings was limited. As only one study reported longer-term effects [21], [22], the results of this review are only applicable to the short-term. No unpublished studies or studies published in ‘grey literature’ were included. The usefulness of including unpublished trials is still under debate [13]. Meant to address publication bias, only few unpublished trials can normally be located for systematic reviews and the located studies may be an unrepresentative sample of all unpublished studies [13], [43]. Investigators are often unwilling to provide unfavorable results; thus publication bias may still remain an issue [13]. Moreover, unpublished studies tend to be of lower methodological quality than published studies [44] and normally lack peer-review [13].

Given that the main drawback of this review was the insufficient reporting of trial methodology, authors of prospect research should improve the reporting of yoga trials and follow commonly accepted reporting guidelines (e.g. CONSORT) [45]. Future studies should ensure rigorous methodology, mainly adequate randomization, allocation concealment, intention-to-treat analysis, and blinding of at least outcome assessors [45].

While this review found evidence for effects of yoga on fatigue and mood in patients with multiple sclerosis, due to the insufficient reporting of the included trials and the limited external validity, this evidence should be applied in clinical practice with care. Thus, no recommendation can be made regarding yoga as a routine intervention for patients with multiple sclerosis at this point. Given that yoga was not associated with severe adverse events, its practice needs not be discouraged in this patient population. Given its evident effectiveness [36], exercise is strongly encouraged by international guideline for the treatment of multiple sclerosis-associated symptoms, especially for improving fatigue [46], [47], [48]. Since fatigue seems to be equally improved by yoga and exercise, yoga might be considered a treatment option for patients who are not adherent to recommended exercise regimens.