This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) extension guidelines (29). It was preregistered in the International Prospective Register of Systematic Reviews (PROSPERO) (30) (registration: CRD42022342277).

A comprehensive search of the literature was performed in PubMed, Embase, Cochrane Central Register, CINAHL, ProQuest, Scopus, and Web of Science to identify articles that examined the effects of exercise intervention on sleep improvement. Combinations of Medical subject headings (MeSH) and keywords were utilized. The detailed search strategies are summarized in Supplementary Table 1. The search was performed independently by two authors (JLQ, SWS) for the period from inception until 3 June 2022. A backwards search was conducted of reference lists that included articles and relevant systematic reviews until no additional relevant articles were found.

The inclusion criteria were assessed using the PICOS approach. (1) Population: premenopausal, menopausal and postmenopausal women. (2) Intervention: potential exercise interventions might include but were not limited to the following: yoga, walking or aerobic exercise. There was no restriction on the setting, frequency, or duration of the intervention. (3) Comparison: waiting list condition, usual care or no intervention. (4) Outcome: studies that reported sleep-related outcomes, such as the Pittsburgh Sleep Quality Index (PSQI), Insomnia Severity Index (ISI) and sleep diary measures (total sleep time, wake time after sleep onset, long awakenings, sleep onset latency, wake time after sleep onset, sleep efficiency). PSQI is a validated sleep assessment tool that evaluates sleep disturbances and sleep quality over 1 month. ISI measures sleep quality over the past 2 weeks. The established cut-offs of 5 points on the PSQI (31) and 10 points on the ISI mean score (32) were used to indicate clinical levels of insomnia. (5) Study design: randomized controlled trials (RCTs). Criteria for exclusion were as follows: (1) women with musculoskeletal disorders, heart disease, arterial hypertension, diabetes, and cancer or women who received hormone replacement therapy; (2) studies carrying out health education intervention aimed at improving the physical activity of menopausal women; (3) studies integrating exercise interventions with pharmacological interventions; and (4) duplicated publications.

We imported identified references into EndNote X9 software to manage data more efficiently. First, we excluded irrelevant studies by removing duplicates and screening titles and abstracts independently by two authors (JLQ, SWS). Then, evaluation was conducted on the full texts of the remaining references based on the inclusion and exclusion criteria. Third, three authors (JLQ, SWS, MW) discussed the studies to reach a consensus on the inclusion of studies if any discrepancy appeared.

A standardized form was adopted for extracting data from eligible studies. The following information were extracted: first author, country, year of publication, participants, sample size, type of exercise, intervention frequency, session, duration, control, evaluation time points, and assessment tools. Two authors (JLQ and SWS) cross-checked the accuracy of data extracted from each study. An independent third author (MW) resolved any inconsistencies.

Review Manager (RevMan) version 5.3 was adopted to conduct all data analyses. Mean differences (MDs) and 95% confidence intervals (CIs) for all individual sleep outcomes (number of studies ≥ 2) were calculated separately if the outcomes were measured choosing the same tool. We combined studies using the standard mean difference (SMD) and 95% confidence intervals (CI) when the same outcome was measured by different methods (33, 34). According to the SMD, effect sizes of 0.2, 0.5, and 0.80 are considered small, moderate, and large, respectively (35). Between-study heterogeneity was examined by standard chi-square and quantified with I² statistics (36). If the P-value was >0.1 or I² <50%, and then a fixed effect model was conducted. Otherwise, the researchers used a random-effect model analysis (37). A subgroup analysis was also carried out based on the intervention population (with or without sleep disorders) and the intervention duration (≤3 months or >3 months). Egger's test was used to assess publication bias.

A Cochrane risk-of-bias tool was utilized by two independent authors (JLQ and SWS) to assess the risk of bias (38). Seven dimensions are outlined in the tool for estimating the risk of bias in each study. We assessed each domain individually to determine whether it posed a low risk, high risk, or unclear risk of bias. An interrater reliability score using Cronbach's alpha for risk of bias was used for the original grading done by two raters.

An overview of the study selection process are summarized in Figure 1. The initial search of seven databases identified 545 citations. One other relevant study was found after searching reference lists. Following the removal of 289 duplicate articles, we screened the titles and abstracts of the remaining 257 articles. Full text papers were ordered for 33 citations. Seventeen studies were included for the systematic review, and ten of those provided statistics for meta-analysis.

The study characteristics are summarized in Supplementary Table 2. Studies were published between 2004 and 2022. A total of 17 articles from different countries were included in the review: the USA (n = 7), Iran (n = 3), China (n = 2), Finland (n = 2), Brazil (n = 1), Spain (n = 1) and Sweden (n = 1). This review involved 2,463 participants. Exercise intervention types were grouped into the following categories: walking (20, 22, 27, 39–41) (n = 6), yoga (22, 26, 42, 43) (n = 4), aerobic exercise (19, 26, 44, 45) (n = 4), passive stretching (27, 42) (n = 2), integrated-style exercise (23, 27) (n = 2), Pilates training (21) (n = 1), resistance training (25) (n = 1), treadmill exercise (46) (n = 1) and Tai Chi Chuan (28) (n = 1). The interventions durations ranged from 3 to 12 months.

Meta-analyses revealed differences between exercise intervention and control groups when it came to sleep outcomes, including sleep quality, insomnia and sleep problems, in menopausal women, as displayed in Figure 2. There were five trials with PSQI post-assessment. A random-effects model was chosen due to the significangt heterogeneity (I² = 90%). The pooled MD was −0.93 (95% CI = −2.73 to 0.87, Z = 1.01, P = 0.31), indicating that the exercise intervention had no significant effects on improving sleep quality. Exercise intervention led to significant reductions in insomnia symptoms in five RCTs among menopausal women (SMD = −0.91, 95% CI = −1.45 to −0.36, Z = 3.27, P = 0.001) compared with the control group. The heterogeneity was high (I² = 89%), and therefore a random-effects model was applied. Regarding sleep problems measured by WHQ, two trials showed that exercise had favorable effects on the WHQ scores (MD = −0.09, 95% CI = −0.17 to −0.01, Z = 2.20, P = 0.03) compared to controls. As there was no significant heterogeneity, we used a fixed-effects model (I² = 0%, P = 0.73).

Subgroup analyses were performed to determine whether the effect sizes for sleep quality and severity of insomnia changed based on the intervention population (with or without sleep disorders) and the intervention duration (≤3 months or >3 months). For sleep quality, different intervention populations showed significant differences (see Supplementary Figure 1). Positive effects of exercise intervention on sleep quality improvement were found among menopausal women with sleep disorders (MD = −3.70, 95% CI = −4.60 to −2.80, Z = 8.10, P < 0.00001) in one study. However, exercise intervention appeared to be ineffective in menopausal women without sleep disorders (MD = −0.06, 95% CI = −0.73 to 0.61, Z = 0.17, P = 0.86) in five studies. For the severity of insomnia, two studies applying exercise intervention among women with sleep disorders (SMD = −2.47, 95% CI = −3.89 to −1.05, Z = 3.41, P = 0.0007) had a larger effect size than those without sleep disorders in four studies (SMD = −0.38, 95% CI = −0.72 to −0.04, Z = 2.22, P = 0.03).

Supplementary Figure 2 displays the forest plots of the subgroup analyses based on the intervention duration (≤3 months or >3 months). For sleep quality, different intervention durations result in significant differences. Four studies with ≤3 months of exercise intervention duration (MD = −0.20, 95% CI = −3.51 to −0.49, Z = 2.59, P = 0.005) appeared to be much more effective on PSQI score reduction than two studies with >3 months of intervention duration (MD = 0.95, 95% CI = 0.05 to 1.86, Z = 2.06, P = 0.31). For the severity of insomnia, three studies conducting >3 months of intervention (SMD = −1.95, 95% CI = −3.10 to −0.80, Z = 3.33, P = 0.0009) had a higher effect size than three studies with ≤3 months of intervention duration (SMD = −0.19, 95% CI = −0.36 to −0.02, Z = 2.23, P = 0.03).

Figures 3, 4 illustrate the results of the risk of bias assessment using the Cochrane tool. The majority of studies had detailed descriptions of the generation of randomization sequence, concealment of allocations, incomplete outcome data and selective outcome reporting processes, and the risk of bias was found to be low for most trials in terms of these aspects. The nature of the intervention made it impossible to blind study participants or researchers. Hence, performance bias is present in all of the studies. A total of seven trials reported blinding of outcome assessment (21, 22, 25, 26, 28, 45, 46). The Cronbach's alpha for the seven assessment dimensions ranged from 0.8 to 1.0.

Using funnel plots for <10 articles is inappropriate. Hence, Egger's tests were conducted for the outcomes sleep quality and insomnia. Egger's showed no evidence of publication bias for sleep quality (t = −0.3, P = 0.777). However, significant publication bias was observed with regard to insomnia (t = −7.76, P = 0.01).

There is a growing concern over the health management of menopausal women (1, 3, 59). This study provides the latest evidence of the efficacy of exercise intervention on improving sleep in menopausal women and indicates some sleep management strategies for this population. The inclusion of RCTs in this study provided high standards for evidence-based research and was a strength of this study. Our research was conducted by two independent researchers and followed the PROSPERO statements as well as a prospective registered protocol, which ensured the rigor of this study.

There are some limitations in this study. First, significant publication bias was observed in regard to insomnia (P = 0.01), specifically suggesting that negative results may not have been published in articles. Future research is needed to explore this issue. Second, we did not perform subgroup analysis based on exercise types due to the limited number of studies. This may have led to significant heterogeneity in our study. Research shows that different types of exercise may lead to different intervention efficacies (24). Future studies should carry out further analysis based on exercise types to select the most effective exercise type to diminish sleep problems in menopausal women. Third, in all included studies, participants or personnel were not blinded, which increased the risk of bias. Consequently, caution should be taken when interpreting the findings of our study due to potential biases. Fourth, we did not examine the long-term effects of exercise intervention due to the limited number of studies and different assessment tools. Clinical trials with long-term follow-up intervals are needed to explore the long-term effectiveness of exercise intervention on sleep. Finally, most of the included studies only assessed subjective sleep using questionnaires such as the PSQI and ISI. The lack of objective sleep measures may influence the reliability of the results. Polysomnography (PSG) has been considered a standard tool for objectively quantifying sleep and evaluating most sleep disorders (60). Actigraphy is valid and reliable for assessing sleep patterns (61). These tools could be used to evaluate objective sleep in future studies.