As a prevalent sleep disorder, insomnia has become a public health problem, including subjective sleep complaints (e.g., poor sleep quality, inadequate sleep time), difficulties in sleep onset/maintenance, waking up too early, or nonrefreshing sleep. Insomnia is associated with significant distress or daytime impairment [
Pharmacological treatment and cognitive behavioral therapy for insomnia (CBT-I) are widely used and have shown effectiveness. Pharmacotherapy is a traditional treatment for insomnia and has been tested and proven to improve sleep outcomes. Due to the risks of daytime residual effects and substance dependence, nonpharmacological treatments have attracted clinicians’ attention [
As an ancient practice, meditation is part of many spiritual traditions and types that emphasize training the mind, especially attention [
Tai chi, qigong, and yoga belong to meditative movements, which combined some forms of movements or body postures that focus on breathing with a clear or a calm state of mind [
In recent years, some systematic reviews have also been conducted with or without a meta-analysis of the cited issues. However, in these studies, only a small part of the evidence has been covered. They have only involved either a specific subpopulation or a certain type of therapy. Thus, it is difficult to draw broader conclusions. Furthermore, most of the existing meta-analyses have only used posttreatment scores, regardless of the existence of the baseline differences, leading to inexact results. In this study, we aim to examine the evidence that MBTs may have effects on improving the sleep health of patients with insomnia and adults who have sleep complaints and to produce an overall picture of contemporary research on this field by making a simple comparison of each intervention. We conduct this systematic review and meta-analysis of several randomized controlled trials (RCTs), which were published up to July 2018.
Literature searches were performed in PubMed, EMBASE, and the Cochrane Library, including studies published until July 2018. The following combinations of keywords were used: (
The titles and the abstracts of all publications obtained from the search strategies were screened by two reviewers. The eligibility criteria follow the PICOS framework [
Two reviewers independently screened the titles and the abstracts of the studies generated from the search to test whether these qualified for review. Next, the full texts were obtained and assessed according to prespecified eligibility criteria. If the reviewers had any disagreement, the third reviewer would resolve the issue by discussing it with them. The data were extracted by using data extraction forms, which were designed upfront. One reviewer (XW) extracted the data into the structured forms; the other reviewer (PL) verified their completeness and accuracy. The extracted data included the author(s); the publication year; the participant characteristics; the intervention types, frequency, duration and dropout rates; outcome measurements; and the main outcomes. We used Engauge Digitizer 10.4 to extract the data if they only showed figures in the study.
The Cochrane Risk of Bias tool [
Stata version MP/14.2 was used for the data analysis. Because of the various baseline values of the studies’ participants, we used the changed scores (from baseline to posttreatment) to calculate standardized mean differences (SMDs) and 95% confidence intervals (CIs). We used the global estimation of r = 0.5 as the correlation coefficient between posttreatment and pretreatment scores.
The magnitude of the SMDs indicated the following: (0-0.2) = negligible effect, (0.2-0.5) = small effect, (0.5-0.8) = moderate effect, and (0.8+) = large effect [
In total, 2646 potentially relevant records were retrieved (1,188 from PubMed, 1,185 from EMBASE, and 442 from the Cochrane Library). After eliminating duplicates, the relevant records were reduced to 2,133, and 2,044 were then excluded from the review for various reasons. Of the 89 full-text articles assessed for eligibility, 43 were retained. Additionally, six articles were included from some of the selected studies’ reference lists. Ultimately, 49 studies involving a total of 4506 participants were included in the meta-analysis. Figure
Table
Characteristics of included studies.
Study (year) | Participants | N, age | Intervention type | Control | Sleep related outcome measures | Dropouts | Study type |
---|---|---|---|---|---|---|---|
| |||||||
Gross 2011 | Chronic primary insomnia | 30, 21-65 | | PCT | PSQI, ISI, Actigraphy, sleep diary | 10% | RCT |
2.5hr/wk, 8wk | |||||||
Boettcher 2014 | Anxiety disorder | 91, 38(10.3) | | Online discussion forum | ISI | 11% | RCT |
1 module/wk, 8wk | |||||||
Garland 2014 | Insomnia comorbid with cancer | 111, 58.89(11.08) | | CBT-I | PSQI, ISI, Actigraphy, sleep diary | 50% | RCT |
90min/wk, 8wk | |||||||
Ong 2014 | Chronic insomnia | 35, 42.9(12.2) | | Self-monitoring condition | ISI, PSAS, PSG, Actigraphy, sleep diary | 31.58% | RCT |
2.5hr/wk&6hr retreat between 5-7wk, 8wk | |||||||
Black 2015 | Older adults | 49, 66.34(7.4) | | SHE program | PSQI; Athens Insomnia Scale | 12.5% | RCT |
2hr/wk, 6wk | |||||||
Alder 2017 | Adults with obesity | 194, 47(12.49) | | PMR | PSQI | 20% | RCT |
2-2.5hr ×16, 5.5 mons | |||||||
Gordon 2017 | Fibromyalgia | 148, 46.88(9.43) | | CBTG | PSQI | 27% | RCT |
2hr/wk, 8wk | |||||||
Wong 2017 | Adults with chronic primary insomnia | 216, 56.09(9.4) | | PEEC | ISI; sleep diary | 9% | RCT |
2.5hr/wk, 8wk | |||||||
Larkey 2015 | Breast cancer survivors | 101, 58.8(8.94) | | Sham Qigong | PSQI | 12.24% | RCT |
60min/wk, 12wk | |||||||
Li 2004 | Older adults with sleep complaints | 118, 75.37(7.77) | | Low-impact exercise | PSQI, ESS | 32.26% | RCT |
60min×3/wk, 24wk | |||||||
Irwin 2008 | Older adults | 52, 70.12(6.68) | | Health education | PSQI | 11.86% | RCT |
40min×3/wk, 16wk | |||||||
WANG 2010 | Elderly with cerebrovascular disorder | 34, 77.06(10.95) | | Rehabilitation program | PSQI | 5.88% | RCT |
50min/wk, 12wk | |||||||
Jones 2012 | Fibromyalgia | 101, 54.04 | | Education | PSQI | 0% | RCT |
1.5 hr×2/wk, 12wk | |||||||
Irwin 2014 | Older adults with chronic and primary insomnia | 73, 66.33(7.45) | | Sleep seminar education control | PSQI, Athens Insomnia Scale, ESS, PSG, sleep diary | 16.67% | RCT |
2hr/wk, 4months | |||||||
Bongi 2016 | Fibromyalgia | 44, 52.24(12.19) | | Educational course about FMS | PSQI | NR | RCT |
Irwin 2017 | Breast cancer survivors | 90, 59.8(8.58) | | CBT-I | PSQI, AISI, ESS, PSG, sleep diary | 15.56% | RCT |
2hr×8wk+1month skill consolidation | |||||||
Lü 2017 | Knee osteoarthritis women | 46, 64.57(3.38) | | Wellness education classes | PSQI, sleep latency, total sleep time, sleep efficiency | 8.70% | RCT |
Innes 2012 | Older Women with Restless Legs Syndrome | 20, 58.7(8.1) | | Education film intervention | PSQI | 20% | RCT |
| |||||||
| |||||||
Britton 2012 | Antidepressant users with sleep complaints | 26, 46.97(7.8) | | Wait-list control condition | PSG, sleep diary | 6.67% | RCT |
3hr/wk, 8wk | |||||||
Johns 2015 | fatigued Cancer survivors | 35, 57.29(9.3) | | Wait-list control condition | ISI | 0% | RCT |
2hr/wk, 7wk | |||||||
Bower 2015 | Younger Breast Cancer survivors | 71, <50years | | Wait-list control condition | PSQI | 10.26% | RCT |
2hr/wk, 6wk | |||||||
Lengacher 2015 | Breast cancer | 79, 57(9.7) | | Usual care | PSQI, Actigraphy, sleep diary | 0% | RCT |
2hr/wk, 6wk | |||||||
Jensen 2015 | Stressed person with poor sleep quality | 72, 42(9) | | Usual treatment | PSQI | 6% | RCT |
2.5hr/wk, 9wk | |||||||
Zhang 2015 | Older adults with chronic insomnia | 60, 78.1(2.99) | | Wait-list control condition | PSQI | 3.33% | RCT |
2hr/wk, 8wk | |||||||
Zhang 2017 | Leukemnia patients in chemotherapy | 76, 39.03(9.14) | | Conventional care | PSQI | 13.16% | RCT |
30-40min/wk, 5wk | |||||||
Chen 2012 | Older people | 56, 71.75(8.13) | | No treatment | PSQI | 3.57% | RCT |
30min×3/wk, 12wk | |||||||
Lynch 2012 | Fibromyalgia | 100, 52.49(8.71) | | Wait-list control condition | PSQI | 16.98% | RCT |
40-60min/day, 8wk | |||||||
Chen 2013 | Breast cancer | 96, 45(8.1) | | Wait-list control condition | PSQI | 0% | RCT |
40min×5/wk, 5 or 6wk | |||||||
Chan 2014 | CFS patients | 150, 39(7.93) | | Wait-list control condition | PSQI | 13.33% | RCT |
1.5hr×16sessions arranged over 9wk | |||||||
McQuade 2017 | Prostate cancer patients undergoing radiotherapy | 50, 64.23(8.1) | | Wait-list control condition | PSQI | 19.2% | RCT |
40min×4/wk, during radiotherapy | |||||||
Frye 2007 | Older adults | 54, 69.2(9.26) | | Non-exercise control | PSQI | 25.8% | RCT |
60min×3/wk, 12wk | |||||||
Hosseini 2011 | Older adult residents in nursing home | 62, 69.08(5.38) | | Routine daily activity | PSQI | 12.90% | RCT |
20-25min×3/wk, 12wk | |||||||
Nguyen 2012 | Older adults | 96, 68.9(5.1) | | Routine daily activity | PSQI | 18.75% | RCT |
1hr×2/wk, 24wk | |||||||
Taylor-Piliae 2014 | Community-dwelling survivors of stroke | 101, 69.9(10) | | Usual care | PSQI | 9.43% | RCT |
60min×3/wk, 12wk | |||||||
Cohen 2004 | Lymphoma patients | 38, 51 | | Wait-list control condition | PSQI | NR | RCT |
Yoga session×1/wk, 7wk | |||||||
Manjunath 2005 | geriatric population with self-reported sleep difficulty | 46, 71.2(7.85) | | Wait-list control condition | Sleep rating questionnaire | 21.74% | RCT |
60min×6/wk, 24wk | |||||||
Chen 2009 | Older adults with sleep complaints | 139, 68.98(6.18) | | Wait-list control condition | PSQI | 7.46% | RCT |
70min×3/wk, 24wk | |||||||
Afonso 2012 | Postmenopausal women with insomnia diagnosed | 40, 50-65years | | Wait-list control condition | ISI | 37.50% | RCT |
1hr×2/wk, 4months | |||||||
Hariprasad 2013 | Elderly with sleep disturbances | 120, 75.28(6.89) | | Wait-list control condition | PSQI | 29.03% | RCT |
60min×7/wk, 24wk | |||||||
Köhn 2013 | Patients with stress-related symptoms or diagnoses | 39, 53.03(12.17) | | Standard care | ISI | 10% | RCT |
60min/wk, 12wk | |||||||
Mustian 2013 | Cancer survivors | 410, 54.1(10.33) | | Standard care | PSQI, Actigraphy | 18.45% | RCT |
75min×2/wk, 4wk | |||||||
Chandwani 2014 | Breast Cancer | 107, 52.24(9.79) | | Usual care | PSQI | 7.50% | RCT |
60min×3/wk, 6wk | |||||||
Cheung 2014 | Older women with knee osteoarthritis | 36, 71.9 | | Wait-list control condition | PSQI | 0% | RCT |
60min/wk, 8wk | |||||||
Newton 2014 | Women with menopausal vasomotor symptoms | 249, 54.24(3.67) | | Usual activity | PSQI, ISI | 1.87% | RCT |
90min/wk, 12wk | |||||||
Fang 2015 | Nurse with poor sleep in China | 120, 35.58(10.43) | | Non-yoga control group | PSQI | 11.48% | RCT |
50-60min×2/wk, 6months | |||||||
Jindani 2015 | Adults with Posttraumatic Stress | 80, 41(18-64) | | Wait-list control condition | ISI | 30% | RCT |
90min/wk, 8wk | |||||||
Cramer 2016 | Colorectal cancer patients | 54, 68.3(9.7) | | Wait-list control condition | PSQI | 22.22% | RCT |
90min/wk, 10wk | |||||||
Buchanan 2017 | Menopausal Women with Hot Flashes | 132, 54.63(3.8) | | Usual activity | Actigraphy | 40.38% | RCT |
90min/wk, 12wk | |||||||
Chaoul 2018 | Breast cancer undergoing chemotherapy | 159, 49.23(9.93) | | Usual care | PSQI | 13.5% | RCT |
75-90minutes×4/wk, 12wk |
Figure
Risk of Bias Analysis.
In this meta-analysis, the specific outcome variables included the sleep quality, the insomnia severity, which were measured by subjective measures (PSQI and ISI) and sleep quantity, such as total sleep time (TST), sleep onset latency (SOL), wake time after sleep onset (WASO), and sleep efficiency (SE), which were calculated by objective measures (PSG, actigraphy) or a sleep diary. Not all the included studies reported follow-up effects, and the follow-up period also differed. Thus, our meta-analysis aimed to evaluate the immediate postintervention effects of the four types of MBTs.
Figure
Forest plots of effect estimates of MBTs versus controls on PSQI.
Forest plots of effect estimates of MBTs versus controls on ISI.
These nonsignificant outcomes needed further examination since they might be influenced by the different types of control conditions. In the included trials, the control conditions differed, including alternative active treatment control and wait-list control and other inactive control conditions. For example, the forest plots of the ISI easily showed that the SMDs obtained by Garland [
Based on the abovementioned results, we needed to conduct subgroup analyses because of the interference caused by the active control interventions. We found that when compared with the inactive control conditions, the of MBTs’ efficacy in alleviating insomnia could be fully demonstrated. The results of the subgroup analyses showed many statistically significant effects on different sleep parameters, as follows: -0.36 (95% CI, -0.56 to -0.15; p=0.001) for insomnia severity measured by the ISI, -0.58 (95% CI, -0.79 to -0.36; p<0.001) for sleep quality measured by the PSQI, and -0.44 (95% CI, -0.77 to -0.11; p=0.008) for SOL measured by a sleep diary. However, there were no statistically significant differences in the pooled results of the SMDs among SE, SOL, TST, and WASO, which were calculated by objective measures (PSG and actigraphy), as well as among SE, TST, and WASO, which were assessed by means of a sleep diary.
It is worth mentioning that the efficacy of meditation, qigong, and yoga in treating insomnia was significant when compared with inactive control conditions. Meditation, qigong, and yoga had respective SMDs of -1.06 (95% CI, -1.96 to -0.17; p=0.02), -0.61 (95% CI, -1.20 to -0.03; p=0.039), and -0.39 (95% CI, -0.59 to -0.18; p<0.001) on the PSQI ranging from small to large effects. In contrast, tai chi had a nonsignificant effect (effect size: -0.55; 95% CI: -1.23 to 0.13; p = 0.091). Regarding the heterogeneity aspects, we found that I2<50% or even I2 = 0, and p>0.1 in some subgroup analyses, such as SOL (I2=0.0%, p=0.513) and TST (I2=0.0%, p=0.419), which were assessed by objective measurements, SOL (I2=7.0%, p=0.341), which was assessed by means of a sleep diary, and ISI (I2=0.0%, p=0.838). Thus, we used the fixed effect model to conduct the abovementioned subgroup analyses and used the random effect model for the remaining subgroup analyses. All the SMDs and the heterogeneity of the subgroup analyses are shown in Tables
Comparison of outcome measures between MBTs and inactive control conditions.
Sleep parameters | Studies | SMDs (95% CI) | p-value | | p-value |
---|---|---|---|---|---|
(n) | (overall effect) | Value (%) | (heterogeneity) | ||
PSQI | 24 | -0.58 (-0.79, -0.36) | | 85.6% | <0.001 |
Meditation | 5 | -1.06 (-1.96, -0.17) | | 93.1% | <0.001 |
Tai Chi | 4 | -0.55 (-1.23, 0.13) | 0.116 | 87.7% | <0.001 |
Qigong | 4 | -0.61 (-1.20, -0.03) | | 87.1% | <0.001 |
Yoga | 10 | -0.39 (-0.59, -0.18) | | 65.6% | 0.002 |
ISI | 5 | -0.36 (-0.56, -0.15) | | 0.00% | 0.838 |
Objective-SE | 3 | 0.20 (-0.13, 0.52) | 0.232 | 51.4% | 0.041 |
Objective-SOL | 4 | -0.03 (-0.20, 0.14) | 0.728 | 0.0% | 0.513 |
Objective-TST | 3 | 0.19 (-0.07, 0.45) | 0.156 | 0.0% | 0.419 |
Objective-WASO | 4 | 0.07 (-0.50, 0.63) | 0.816 | 87.3% | <0.001 |
Self-reported-SE | 1 | 0.67 (-0.18, 1.52) | 0.123 | — | — |
Self-reported-SOL | 3 | -0.44 (-0.77, -0.11) | | 7.0% | 0.341 |
Self-reported-TST | 3 | 0.49 (-0.11, 1.09) | 0.106 | 64.8% | 0.058 |
Self-reported-WASO | 1 | -0.47 (-1.31,0.37) | 0.270 | — | — |
Exploratory of subgroup differences in SMDs in PSQI among included studies.
Subgroups | Studies | SMDs (95% CI) | p-value | I2 | p-value | p-value |
---|---|---|---|---|---|---|
(n) | (overall effect) | Value (%) | (heterogeneity) | (group difference) | ||
Type of intervention | ||||||
Meditation | 10 | -0.57 (-1.19, 0.06) | 0.076 | 94.5% | <0.001 | 0.830 |
Tai Chi | 12 | -0.35 (-0.63, -0.07) | | 75.5% | <0.001 | |
Qigong | 4 | -0.61 (-1.20, -0.03) | | 87.1% | <0.001 | |
Yoga | 11 | -0.42 (-0.62, -0.21) | | 66.0% | 0.001 | |
Type of control | ||||||
Active control | 15 | -0.23 (-0.56, 0.10) | 0.180 | 86.3% | <0.001 | 0.080 |
Inactive control | 24 | -0.58 (-0.79, -0.36) | | 84.3% | <0.001 | |
Type of participant | ||||||
Clinical patient | 27 | -0.38 (-0.62, -0.14) | | 86.6% | <0.001 | 0.210 |
Healthy adult | 16 | -0.58 (-0.85, -0.30) | | 82.6% | <0.001 | |
Duration of intervention | ||||||
| 19 | -0.45 (-0.65, -0.25) | | 77.3% | <0.001 | 1.000 |
<12 weeks | 20 | -0.45 (-0.77, -0.13) | | 89.7% | <0.001 | |
Frequency of intervention | ||||||
| 14 | -0.35 (-0.57, -0.13) | | 71.8% | <0.001 | 0.370 |
<3 times/week | 25 | -0.51 (-0.77, -0.24) | | 89.0% | <0.001 |
Further subgroup analyses were conducted to explore the MBTs’ effects, as shown on the PSQI, among different populations. Stratified by population types, the subgroup analyses demonstrated that the studies involving clinical patients and healthy individuals both showed significant effects on sleep quality (PSQI scores), and studies involving healthy individuals had larger mean effect sizes (effect size: -0.58; 95% CI: -0.85 to -0.30; p<0.001; I2 = 82.6%) compared with studies involving clinical patients (effect size: -0.38; 95% CI; -0.62 to -0.14; p = 0.002; I2 = 86.6%). However, there was no significant difference in the pooled effect sizes between the two subgroups (
To our best knowledge, this is the largest meta-analysis with the aim of examining the effects of MBTs (meditation, tai chi, qigong, and yoga) on insomnia symptoms and sleep quality among subjects with or without diseases or pre-existing conditions. The overall effects of MBTs on improving sleep quality were significant (effect size: -0.45; 95% CI: -0.63 to -0.26; p<0.001), but the effects on reducing the severity of insomnia symptoms were not significant (effect size: -0.26; 95% CI: -0.60 to 0.09; p = 0.142). These results might be influenced by the control condition type. In some studies, researchers used some active control conditions, such as CBT-I [
We also conducted some subgroup analyses to compare the effects of MBTs based on the intervention type, the population type, and the intervention duration and frequency. For the subgroup analyses based on the population type, we compared the SMDs in the sleep quality of clinical patients and healthy people. Significant SMDs were shown in both clinical patients (effect size: -0.38; 95% CI: -0.62 to -0.14; p = 0.002) and healthy people (effect size: -0.58; 95% CI: -0.85 to -0.30; p<0.001). The effect of MBTs on the sleep quality of healthy people was obviously larger than that of clinical patients although the subgroup difference was not significant. For the clinical patients, their insomnia might be more or less related to medical disorders (e.g., knee osteoarthritis patients with chronic pain, fibromyalgia patients with non-restorative sleep, and inflammatory bowel disease [IBD] patients who must use the toilet many times/night). Thus, similar to the psychotherapies, it was difficult to solve these problems by MBTs. For the insomnia severity, MBTs had an obvious effect on reducing it among patients, but their insomnia was mostly unrelated to a medical disorder. Some examples of the treatments were MBSR or MBCT for chronic primary insomnia [
To explore the influencing factors on the effects of MBTs, we conducted subgroup analyses based on the duration and the frequency of interventions. We divided the intervention duration into
We also performed subgroup analyses among the different intervention types. The two studies [
Much evidence demonstrated that MBTs might produce benefits for different groups of people, such as insomnia patients [
According to our additional subgroup analyses, the effect of MBTs on the sleep quality of healthy adults was larger compared with clinical patients. This result might be influenced by the patients’ characteristics. For those patients whose insomnia is caused by other medical disorders, MBTs may not achieve the desired effect. Treating their related medical disorder is the fundamental way to reduce their insomnia. Therefore, for these patients, MBTs might only be used as adjuvant therapies. In sum, MBTs could be treated as effective preventive interventions for insomnia in both healthy and clinical populations. MBTs could also be used as adjuvant or alternative therapies in treating insomnia with or without comorbidity, respectively. However, because secondary insomnia is always associated with physical or mental disorders, which is not the case of primary insomnia, this difference might interfere with the outcomes. Further studies should separate primary insomnia from secondary insomnia to explore the MBTs’ effect on insomnia in clinical populations. Our other subgroup analyses showed that the effects of MBTs might be influenced by the intervention duration but not the frequency, and these results should be confirmed in the future research.
Mild to moderate dropout rates were also founded in these studies. According to the included studies, the dropout rates greatly varied; 6 studies [
Our study had several strengths. First, we included 49 studies in this meta-analysis, which produced more comprehensive and broader conclusions. This review included both healthy and clinical populations, ranging from young and middle-aged to older people. Second, both subjective and objective outcomes were analyzed. We extracted outcomes from a sleep questionnaire, a sleep diary, PSG, and actigraphy to conduct an overall meta-analysis, which covered both sleep quality and sleep quantity. Third, we analyzed the effects of tai chi and qigong separately, leading to more explicit results, and we further clarified the effects of each intervention on sleep quality and insomnia.
Although the findings of this meta-analysis suggested some promising clinical benefits of MBTs for alleviating insomnia, there were also several limitations. First, we only included studies published in English, which might have influenced our results to some extent and limited the generalizability of our findings. For example, the studies on the intervention of qigong were mostly included in Chinese databases; thus, the evidence on the effect of qigong on insomnia was inadequate. Second, our subgroup analysis might not have been sufficiently robust to obtain the actual effect because of the limited studies and the relatively small sample size. Third, the studies included in this meta-analysis had significant heterogeneity. The study quality, various population types, the intervention duration and frequency, and even the severity of insomnia or sleep complaints might influence heterogeneity. Finally, we only used the immediate posttreatment outcomes to examine the effects of the four types of MBTs on insomnia, but some studies showed improvements in sleep quality in the follow-up period.
In conclusion, this systematic review and meta-analysis provided evidence that MBTs could be effective in treating insomnia and improving the sleep quality of healthy subjects and clinical patients. As two different types of MBTs, tai chi and qigong were analyzed separately and produced a minor difference in outcomes. These results might indicate that tai chi and qigong, as two different types of MBTs, should not be equated. Our findings on the larger effect of MBTs on the sleep quality of healthy adults compared with clinical patients should also be further explored. However, we only included studies published in English, which also had varying levels of quality. Further research should include high-quality and well-controlled RCTs, published in English and other languages. Future studies should conduct more detailed subgroup analyses to confirm the accuracy of the effect sizes of MBTs; the changes observed in the follow-up period should also be considered.
This study has been presented as conference abstract in the 24th Annual Meeting of Chinese Society of Psychosomatic Medicine & International Psychosomatic Medicine Forum, At Shijiazhuang, China.
The authors declared no potential conflicts of interest.
Xiang Wang and Chen Pan conceived the study. The literature search and screening data were done by Xiang Wang. Data extraction and quality assessment were carried out independently by Xiang Wang and Peihuan Li. Xiang Wang and Yan Wu analyzed and interpreted data and Xiang Wang wrote the manuscript. Yunlong Deng and Lisha Dai revised the manuscript. All authors read and approved the final manuscript.
This research was supported by the New Xiangya Talent Project of the Third Xiangya Hospital of Central South University (grant no 20150302).