Before beginning the review, we followed the checklist for systematic reviews in concurrence with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines [21] and suggestions by the Cochrane Collaboration for programming and conducting systematic reviews and meta-analyses [22].
Eligibility criteria
Types of studies
Randomized controlled trials (RCTs), randomized crossover studies, and cluster randomized trials were all eligible for this meta-analysis. No restrictions in terms of language and countries were applied.
Types of participants
Studies that included women (aged ≥18 years) with sleep problems were eligible. No restrictions on the ethnicity and comorbidity of participants were applied.
Types of interventions
No restrictions regarding yoga type, form, structure, frequency, duration or length of intervention programs were applied. Studies on cointerventions that included yoga as a part of multicomponent interventions were excluded because it would be difficult to distinguish the effects of yoga from additional modalities. Studies in control interventions that compared yoga treatments with nontreatment, usual care, wait-lists, and education without active physical exercise programs were all eligible.
Types of outcome measures
The primary outcome of this study was sleep quality. To be included in this review,
studies had to assess at least one of the sleep quality measures by using standardized
instruments and provide outcomes both at the baseline and follow-up for primary
outcomes. In particular, instruments in question include subjective measurements,
such as the Pittsburgh Sleep Quality Index (PSQI) and Insomnia Severity Index (ISI),
or objective measurements, such as polysomnography (PSG) and actigraphy. The
PSQI score have been recommended as a reliable, valid and standardized instrument
to measure and to identify quality of sleep. The widely employed
Pittsburgh Sleep Quality Index (PSQI), provides a measure of global sleep
quality, including sleep latency, sleep duration, habitual sleep efficiency, sleep
disturbances, use of sleeping medication, and daytime dysfunction [14]. The
seven components of the PSQI are standardized of areas routinely assessed
sleep complaints with possible range of 0-21 points. A global PSQI score of 5 or
higher provided a sensitive and specific measure for poor sleep quality [23]. The ISI
score is a reliable and valid instrument to quantify perceived insomnia severity. A
global ISI score of 8 or higher is indicative of some degree of insomnia, while
moderate insomnia has a score of 15-21 and severe insomnia with a score of 22-28
[24]. PSG or actigraphy reports the most complete and precise information on the
construction and distribution of sleep periods, such as total sleep time (TST), sleep
efficiency (SE), and wake time after sleep onset (WASO) [25]. Sleep quality is also
sometimes measured from PSG and actigraphy. Among these objective indices are
measures such as sleep onset latency, total sleep time, wake time after sleep onset,
sleep efficiency, and number of awakenings [26].
Secondary outcomes: The secondary outcome included in this study was the safety of the intervention, which was assessed as number of patients with adverse events (AEs), including serious adverse events or nonserious events. Serious adverse events referred to those events that caused death, life-threatening situations, hospitalization, disability or permanent damage, congenital anomaly/birth defect, or the need for medical or surgical intervention to prevent any of the aforementioned outcomes [27]. All other adverse events were regarded as nonserious.
Search methods
The search strategy comprised four electronic databases from their inception through June 01, 2019: Medline/PubMed, ClinicalKey, ScienceDirect, Embase, PsycINFO, and the Cochrane Library. The literature search was constructed around search terms for “yoga,” “women,” and “sleep” and was adapted for each database as necessary. The complete search strategy for PubMed was as follows: (“yoga" OR "asana" OR "pranayama" OR "dhyana") AND ("women" OR "female") AND ("sleep" OR "sleep quality" OR "sleep disturbance" OR "insomnia"). Additional reference lists of identified original articles or reviews, the table of the contents of the Journal of Yoga and Physical Therapy, and Journal of National Taiwan Sports University were searched manually.
Retrieved articles were scanned independently to verify their eligibility, and the entire text was assessed by two reviewers. A conflict of reviewers’ opinions on inclusion or exclusion of any article was discussed with a third reviewer to reach a consensus.
Data extraction and management
Two reviewers independently extracted data on design (e.g., article setting, author/year, country of studies, and sampling strategy), participants (e.g., age, body max index, clinical characteristics, comorbid condition, and overall sample size), interventions (e.g., yoga type, frequency of sessions per week, duration of yoga intervention, and total length of intervention time), control interventions (e.g., type, frequency, length, and duration), and outcomes (e.g., outcome measures with sleep quality and safety-related events). A conflict of reviewers’ opinions was discussed with a third reviewer until consensus was reached.
Risk of bias in individual studies
Two reviewers independently assessed the risk of bias in each study. There were seven domains of assessment for the risk of bias include in the following : (1) random sequence generation, (2) allocation concealment, (3) blinding of participants and personnel, (4) blinding of outcome assessment, (5) incomplete outcome data, (6) selective reporting, and (7) other biases using the Cochrane Systematic Review Manual risk of bias assessment tool [22]. All domains were scored as low risk, high risk, or unclear risk of bias and assessed individually. A risk of bias table was completed for each included study. A conflict of reviewers’ opinions was discussed with a third reviewer until consensus was reached.
Data assessment of overall effect size
A meta-analysis was conducted with Review Manager 5 software (Version 5.3, The Nordic Cochrane Centre, Copenhagen) and Comprehensive Meta-Analysis Software using a random effects model if at least two studies assessing this specific outcome were obtainable. For continuous outcomes, standardized mean differences (SMDs) with 95% confidence intervals (CIs) were calculated as the difference in means between groups divided by the pooled standard deviation. For studies that did not report data with standard deviations, we calculated these values from standard errors, confidence intervals, or t-values. If adequate information was available, we would plan to perform subgroup analysis. The p value of the summary effect < 0.05 were regarded as indicating statistical significance.
A negative SMD was provided a definition to display the beneficial effects of yoga intervention compared with the control intervention for sleep quality outcomes. Cohen’s categories were used to assess the significance of the overall effect size, with SMD = 0.2–0.5: small effect size; SMD = 0.5-0.8: medium effect size; and SMD > 0.8: large effect size [28].
Assessment of heterogeneity
Statistical heterogeneity between studies was analyzed using the I2 statistics and the Cochrane chi-square. The variance between studies was measured using the tau-square (Tau2). The level of heterogeneity was classified as I2 = 0%–24%: low heterogeneity; I2 = 25%–49%: moderate heterogeneity; I2 = 50%–74%: substantial heterogeneity; and I2 = 75%–100%: considerable heterogeneity. Given the low power of this test when only few studies or studies with a low sample size are included in a meta-analysis, a P value of ≤0.1 for the chi-square test was regarded as indicating significant heterogeneity [22].
Moderator analyses
Moderator and meta-regression analyses were further performed to identify possible reasons for interstudy heterogeneity. The subgroup analysis produced prespecified covariates, including outcome measurement tools, participant type, study quality, study region, participant age, intervention duration and study sample size.
Risk of publication bias
Risk of publication bias was evaluated for each meta-analysis that included at least 10 studies. Funnel plots generated using Review Manager 5 software was estimated from individual studies against each study’s standard error. Publication bias was evaluated through visual analysis, in which roughly the symmetrical funnel plot signifies no evidence of high risk of publication bias [29]. Potential publication bias was evaluated using the Egger's Intercept Test, with p values < 0.05 signifying significant bias.