Literature search
The results of the literature search and screening process are summarized in Fig. 1. The literature search totaled 1338 records; one additional record was retrieved from the Journal of National Taiwan Sports University in the Chinese language database [31]. In all, 1295 records were excluded because they did not meet all predefined inclusion criteria or were duplicated. Forty-four full-text articles were assessed for eligibility. Twenty-five were excluded because they were not randomized [31, 32], did not include relevant outcomes [33-39], did not include only female participants [40-48], included yoga as a part of a multimodal intervention (or combined with other intervention) [49-52], lacked adequate control [53], and did not include a form of yoga intervention [54, 55]. Nineteen full-text articles with 1832 participants were included in the qualitative synthesis and were included in the meta-analysis. All articles were published in English.
Study characteristics
A total of 19 studies were considered eligible for systematic reviews. Information regarding the characteristics of the sample, yoga or control group interventions, outcome measures, and results are listed in Tables 1 and 2.
Study and participant characteristics
Of the 19 RCTs that were included in Table 1, six RCTs included healthy participants [56-61], including nurses [57], teachers [58], and women in the menopausal transition period or postmenopausal period [56, 59-61]. Other 13 RCTs included breast cancer patients undergoing treatment [62-65], breast cancer patients who had completed treatment [13, 66-68], type 2 diabetes mellitus patients [69], fibromyalgia patients [70], knee osteoarthritis patients [71], restless leg syndrome patients [72], and patients experiencing dysfunctional uterine bleeding [73].
Overall, the 19 RCTs included were conducted in the United States [13, 56, 59, 60, 62-68, 71, 72], Brazil [61, 70], India [58, 73], Iran [69], and China [57]. Study participants were recruited from hospitals [57, 63-67, 69, 73], outpatient clinics [61, 62] and schools [58]. The process of recruitment also included using purchased lists and health-plan enrollment files [56, 59] and multiple other mechanisms, including flyers, newspaper advertisements, web-based announcements, brochures, public health departments, tumor registry systems, and doctor referrals [13, 60, 68, 71, 72]. One study did not reveal the source from which participants were recruited [70]. Nineteen studies included in the systematic review displayed a baseline of PSQI higher than 5 or ISI higher than 8, indicating poor sleep quality or insomnia. The only exceptions were two studies, with individual control groups in each study displaying a baseline of PSQI lower than 5 [58] or ISI lower than 8 [67]. The sample size ranged from 20 to 249 with a median of 96. Participant’s mean age ranged from 29.8 to 71.9 years, with a median of 50.1 years. All participants were women.
Intervention characteristics
Of the 19 included studies in Table 1, three reported using Iyengar Yoga [13, 60, 72]; two reported using Hatha Yoga [68, 71]; two reported using Tibetan Yoga [61, 64]; two reported using Restorative Yoga [66, 67]; one reported using Vini Yoga [56]; one reported using Yoga Relaxation with MindSound Resonance Technique [58]; one reported using yoga breathing exercise in warm water [70]; and only seven RCTs revealed yoga programs with postures, breathing, relaxation or mediation, without defining the specific style of yoga [57, 59, 62, 63, 65, 69, 73]. All RCTs included yoga postures in their yoga intervention; 16 RCTs included yoga breathing; 15 RCTs included yoga relaxation; 12 RCTs included meditation; and 7 RCTs included all contents with postures, breathing, relaxation, and meditation for the yoga intervention group [57, 59, 63, 65, 67, 69, 73]. The duration of yoga interventions ranged from 1 week to 24 weeks, with a median of 10 weeks; the frequency of yoga interventions ranged from one to five weekly sessions of 45 to 120 min. Sixteen studies compared the yoga group with waitlist control groups with no specific treatment; three studies compared the yoga group with the control group, including two studies for education groups [13, 72] and one study for social support groups [62].
Outcome measures
All studies evaluated outcomes directly at the end of interventions. All studies assessed the subjective or objective measurements of sleep quality: 16 RCTs used the PSQI; three RCTs used the ISI [59, 61, 67]; one RCT used PSG [61]; and two RCTs used actigraphy [56,64]. Safety-related events were reported in two RCTs only [59, 68].
Risk of bias
Risk of bias in individual assessments
Graphical representation of the risk-of-bias assessment is represented in Fig. 2. All studies had a high or unclear risk of bias in at least one domain. All studies claimed to be randomized; however, three studies did not reveal their content and method of random sequence [61, 66, 69]. Twelve studies did not report methods applied to perform adequate allocation [56-58, 60-66, 69, 73]. Most studies offered no data material on blinding. Three studies clearly reported that participants and personnel were blinded [13, 62, 71]. Four studies clearly reported that researchers and outcome assessments were blinded [13, 62, 68, 71]. Six studies had insufficient data on attrition rates [59-61, 69, 70, 72]. Twelve studies had a low risk of selection reporting; only two studies had a high risk of selective reporting due to several reported outcome parameters not being revealed in study protocol or duplicate publications reporting different results of the same trial [59, 61]. Six studies had a high risk of other potential sources of bias due to poor participant compliance, intervention length, sample size or baseline differences [60, 67, 70-73].
Publication bias
The meta-analysis of the effect of yoga on the sleep quality of women that involved yoga groups compared with control groups included 16 studies. The asymmetrical shape of the funnel plot indicated that subjective publication bias was detected (Fig. 3). Objective publication bias was analyzed using Egger’s Regression Test. Egger's Test consists of the regression between the accuracy of the studies and standardized effects, which are weighted by the inverse of variance. Borderline findings (P=0.05) show objective evidence on publication bias between precision and standardized effects of studies in the present study, specifically suggesting need for future studies to expound on the issue.
Analysis of overall effects
Primary outcomes
The random effects model was applied to analyze the 19 RCTs outcomes by different sleep outcome measurement tools. The meta-analysis of combined data conducted with Comprehensive Meta-Analysis, showed a significant improvement in sleep problems (SMD =-0.327, 95% CI= -0.506 to -0.148, P<0.001). However, significant heterogeneity existed among all the studies (Q=43.152, I² = 58.287%, P=0.001). Therefore, moderator and meta-regression analyses were conducted to further explore the determinants of the heterogeneity.
The meta-analysis revealed the effects of yoga compared with the control group on the sleep quality and insomnia of women using the PSQI or ISI, as displayed in Fig. 4. Sixteen RCTs revealed evidence for effects of yoga compared with the control group in improving sleep quality in women using the PSQI (SMD = −0.54; 95% CI = −0.89 to −0.19; P = 0.003). However, three RCTs revealed no effects of yoga compared with the control group in reducing the severity of insomnia in women using ISI (SMD = −0.13; 95% CI = −0.74 to 0.48; P = 0.69). Two RCTs revealed no effects of yoga compared with control group in improving sleep efficiency (SMD = 0.85; 95% CI = −0.56 to 0.26; P = 2.26) or total sleep time (SMD = -0.06; 95% CI = −0.26 to 0.13; P = -0.59) in women using actigraphy.
Secondary outcomes (safety)
Only two studies reported safety-related events. Two events revealed in one study could potentially be attributed to yoga intervention: two women reported the recurrence of chronic back and/or shoulder problems [68]. In another study, adverse events reported did not differ between the yoga intervention group and the control group (P = 0.41). These adverse events included muscle aches and strains (6.7%, yoga group; 10.3%, control group), low back pain (4.2%, yoga group; 3.1%, control group), or changes in strength or sensation in arms or legs (5.5% yoga group; 8.9% control group). Dropouts were not regarded as being adverse events because it was not explicitly telling as the reason for dropout in the original study. No serious adverse effects were reported in the included studies.
Subgroup analyses
Participants were divided into two separate subgroups. Meta-analyses revealed the effects of yoga compared with the control group for women with breast cancer in Fig. 5. Seven RCTs revealed no evidence for the effect of yoga compared with the control group in improving sleep quality for women with breast cancer using the PSQI (SMD = −0.15; 95% CI = −0.31 to 0.01; P = 0.5). Four RCTs revealed no evidence for effects of yoga compared with the control group in improving sleep quality for women undergoing treatment for breast cancer (SMD = −0.08; 95% CI = −0.29 to 0.13; P = 0.45). Three RCTs revealed no evidence for positive effects of yoga in terms of improving sleep quality for women with breast cancer who had completed treatment compared with the control group (SMD = −0.25; 95% CI = −0.50 to 0.00; P = 0.31).
The meta-analysis showed evidence of the positive effects of yoga on sleep quality compared with control groups for peri/postmenopausal women as displayed in Fig. 6. Four RCTs revealed no evidence for effects of yoga compared with control groups in improving sleep quality in peri/postmenopausal women using the PSQI (SMD = −0.31; 95% CI = −0.95 to 0.33; P = 0.34). Two RCTs revealed no evidence for effects of yoga compared with the control group in reducing severity of insomnia in peri/postmenopausal using ISI (SMD = −0.29; 95% CI = −1.23 to 0.65; P = 0.55).
Moderator analyses, Meta-regression
Moderator analyses and meta-regression are presented in Table3. Significant factors in observed heterogeneity were identified for yoga on sleep quality and insomnia in women with sleep problems. Studies that used PSQI as Outcome measurement tool showed a greater reduction in sleep problems than other studies that used other instruments as outcome measurement tools (SMD= -0.374 vs. 0.035, P=0.002). Participants without breast cancer showed more improvement in sleep problems than participants with breast cancer (SMD= -0.420 vs. -0.234, P<0.001). Studies without peri/postmenopausal women showed more improvement in sleep problems than studies with peri/postmenopausal women (SMD= -0.423 vs. -0.05, P=0.004). Regression analyses revealed a positive correlation with total length of class hours (p = 0.003), indicating that the longer of total length class hours, the more likely the chance to have significant results. Regression analyses revealed a negative correlation with mean age (p = 0.003) and sample size (p = 0.032) of study, indicating that the younger, and smaller sample sizes were more likely to have significant results.
Sensitivity analyses
In the included studies with low risk of selection bias, reporting bias, and other bias, the effect of yoga group compared to control group on women sleep PSQI did not change substantially, including random sequence generation bias (SMD = −0.45; 95% CI = −0.84 to −0.11; P = 0.01; heterogeneity: I² = 88%; χ2 = 107.43, P < 0.00001), allocation concealment bias (SMD = −0.77; 95% CI = −1.37 to −0.16 ; P = 0.01; heterogeneity: I² = 88%; χ2 = 40.95, P < 0.00001), selective reporting bias (standard mean difference = −0.59; 95% CI = −1.10 to −0.08; P = 0.02; heterogeneity: I² = 88%; χ2 = 93.11, P < 0.00001) and other bias (standard mean difference = −0.53; 95% CI = −1.03 to −0.04; P = 0.03; heterogeneity: I² = 86%; χ2 = 44.03, P < 0.00001). The effect compared with the control group remained significant in terms of sensitivity analyses of performance bias, detection bias, or attrition bias after eliminating high risk bias or uncertain risk bias of the studies.