Comparative effects of exercise modalities on sleep quality in middle-aged and older adults with sleep disorders: A systematic review with pairwise, network, and dose-responsemeta-analyses Authors: Ting Wang, Zhuang Ma, Boao Zhu, Yaqiong Tang Published: 2026 ABSTRACT AbstractBackgroundExercise may improve sleep quality in middle-aged and older adults with sleep disorders, but the comparative effects of different exercise modalities and the dose–response relationship remain unclear.MethodsWe systematically searched PubMed, Web of Science, Embase, and the Cochrane Library from inception to May 2026. Eligible studies were randomized controlled trials examining the effects of exercise interventions on sleep quality in middle-aged and older adults with sleep disorders. Two independent reviewers assessed study quality, and random-effects pairwise, network, and dose-response meta-analyses were performed to evaluate the effects of different exercise interventions and doses on sleep quality.ResultsThe analysis included 23 randomized controlled trials involving 1,478 middle-aged and older adults with sleep disorders. Pairwise meta-analysis showed that exercise reduced Pittsburgh Sleep Quality Index (PSQI) and Insomnia Severity Index (ISI) scores but had no significant effect on Epworth Sleepiness Scale (ESS) scores. Bayesian network meta-analysis showed that, compared with the control group, aerobic exercise (SMD = 1.34, 95% CrI: 0.86∼1.82), mind–body exercise (SMD = 0.64, 95% CrI: 0.30∼0.97), and combined exercise (SMD = 0.60, 95% CrI: 0.13∼1.07) may improve PSQI scores. The dose-response analysis suggested an inverted U-shaped association, with the largest predicted effect occurring near 539 MET-min/week within the observed dose range.ConclusionsLow-to moderate-certainty evidence suggests that aerobic, mind–body, and combined exercise may reduce PSQI scores, with a nonlinear dose–response relationship.Keywords: Exercise, Middle-aged and older adults, Sleep quality, Sleep disorders, Dose-response relationship Introduction Sleep disorders have become an increasingly serious public health concern among middle-aged and older adults.1 With aging, sleep architecture and circadian regulation gradually change in middle-aged and older adults, leading to shorter sleep duration, reduced sleep continuity, and weakened circadian rhythms, thereby increasing the risk of insomnia, daytime sleepiness, and sleep-related abnormalities.2 The prevalence of sleep disorders increases with age and may approach 50% among adults aged over 65 years.3 Chronic sleep disorders not only impair restorative function but are also associated with increased risks of multiple chronic conditions, including hypertension, dyslipidemia, diabetes, kidney disease, cardiovascular disease, and reproductive disorders, posing a serious threat to the health and quality of life of middle-aged and older adults 4, 5, 6, 7. Therefore, safe, effective, and scalable interventions are urgently needed to improve sleep quality and promote healthy aging. Exercise interventions, owing to their safety, low cost, and scalability, have increasingly become a key focus of non-pharmacological intervention research.8, 9, 10 Early studies were mainly individual randomized controlled trials examining whether exercise modalities such as aerobic exercise, walking, Tai Chi, and yoga could improve subjective sleep quality in middle-aged and older adults with sleep disorders.11, 12, 13 As the number of relevant RCTs increased, subsequent studies began to synthesize evidence on exercise-induced improvements in sleep quality using conventional pairwise meta-analysis. Previous meta-analyses have shown that exercise interventions have beneficial effects on subjective sleep quality.14,15 However, existing meta-analyses have mainly focused on general sleep disorder populations or older adults, leaving limited evidence specific to middle-aged and older adults with sleep disorders and uncertainty regarding the applicability of their findings. In terms of intervention modality, existing evidence has mainly focused on the overall effects of exercise on sleep quality, with insufficient comparison of the relative efficacy of different exercise modalities.15,16 Conventional pairwise meta-analyses mainly rely on direct comparisons between a single exercise modality and control, making it difficult to integrate and compare multiple exercise modalities within a unified analytical framework. Given that different exercise modalities vary in intensity, physiological stimulation, psychological regulation, and target populations 17, 18, 19, their sleep-related effects may therefore be heterogeneous. Therefore, the relative efficacy of different exercise types in middle-aged and older adults with sleep disorders requires further clarification. From an exercise prescription perspective, intervention effects may be jointly influenced by type, intensity, session duration, frequency, intervention period, and total exercise dose.10,20 However, previous studies have primarily focused on the overall effects of exercise, with limited systematic evidence on dose-response associations between exercise and sleep quality improvement. Therefore, we conducted pairwise, network, and dose-response network meta-analyses (NMAs) within a Bayesian framework to compare the effects of different exercise modalities and doses on sleep quality in middle-aged and older adults with sleep disorders. In addition, we examined the dose-response relationship between exercise dose and sleep quality improvement. Methods Protocol and registration This systematic review and network meta-analysis was prospectively registered in PROSPERO under the registration number CRD420261421238. The review was conducted and reported in accordance with the PRISMA 2020 statement and the PRISMA extension for network meta-analyses.21 Because all data were extracted from previously published studies, ethical approval and informed consent were not applicable. Search strategy and study selection To ensure a comprehensive literature search, two reviewers independently searched PubMed, Web of Science, Embase, and the Cochrane Library from database inception to May 2026. The search strategy was constructed using Medical Subject Headings (MeSH) and relevant free-text terms related to sleep disorders, exercise, middle-aged and older adults, and sleep quality. Database-specific search strategies were developed, and the full search strings are available in the supplementary materials (Tables S1–S4). Reference lists of relevant reviews and meta-analyses were additionally screened to identify eligible studies that may have been missed. Two reviewers independently screened the records, and disagreements were resolved by discussion or consultation with a third reviewer. Inclusion criteria Studies meeting all of the following criteria were included: (1)Participants: Middle-aged and older adults aged ≥45 years with either (1) a clinically diagnosed sleep disorder based on established diagnostic criteria or objective sleep testing interpreted according to accepted diagnostic thresholds, or (2) clinically significant sleep disturbances identified using validated instruments or predefined symptom-based criteria. Participants: Middle-aged and older adults aged ≥45 years with either (1) a clinically diagnosed sleep disorder based on established diagnostic criteria or objective sleep testing interpreted according to accepted diagnostic thresholds, or (2) clinically significant sleep disturbances identified using validated instruments or predefined symptom-based criteria. (2)Intervention: Any exercise intervention, such as Tai chi, walking, or High-Intensity Interval Training (HIIT). Intervention: Any exercise intervention, such as Tai chi, walking, or High-Intensity Interval Training (HIIT). (3)Control: Sedentary participants or participants receiving non-exercise interventions, such as health education or standard care, without any structured exercise program. Control: Sedentary participants or participants receiving non-exercise interventions, such as health education or standard care, without any structured exercise program. (4)Outcomes: The primary outcome was subjective sleep quality, assessed using the Pittsburgh Sleep Quality Index (PSQI). Secondary outcomes included insomnia severity, assessed using the Insomnia Severity Index (ISI), and daytime sleepiness, assessed using the Epworth Sleepiness Scale (ESS). Outcomes: The primary outcome was subjective sleep quality, assessed using the Pittsburgh Sleep Quality Index (PSQI). Secondary outcomes included insomnia severity, assessed using the Insomnia Severity Index (ISI), and daytime sleepiness, assessed using the Epworth Sleepiness Scale (ESS). (5)Study design: Randomized controlled trials (RCTs). Study design: Randomized controlled trials (RCTs). Exclusion criteria Studies meeting any of the following criteria were excluded: (1)Interventions combining exercise and medication. Interventions combining exercise and medication. (2)Studies that were conference abstracts, protocols, review articles, or preliminary reports. Studies that were conference abstracts, protocols, review articles, or preliminary reports. (3)Duplicate publications were excluded. Duplicate publications were excluded. (4)Studies that did not provide adequate data for calculating effect sizes were excluded. Studies that did not provide adequate data for calculating effect sizes were excluded. (5)Experimental studies involving non-human subjects. Experimental studies involving non-human subjects. Quality assessment Risk of bias in the included studies was assessed using the Cochrane Risk of Bias 2 tool for randomized controlled trials. This tool evaluates five domains: bias arising from the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result.22 Two reviewers independently assessed the risk of bias in the included studies, and disagreements were resolved through discussion or consultation with a third reviewer. Data extraction Data were independently extracted by two reviewers and cross-checked for accuracy, with disagreements resolved through consultation with a third reviewer. Extracted information included the first author, publication year, participant characteristics, exercise type and intervention parameters, outcome assessment instruments, and sleep-related outcome data. For studies reporting outcomes at multiple time points, data from the assessment conducted immediately after completion of the intervention were extracted. Continuous outcomes were extracted as post-intervention means and standard deviations (SDs). When SDs were not directly reported, they were calculated from confidence intervals (CIs), standard errors (SEs), or other available statistics using the formulas recommended in the Cochrane Handbook.23 Data coding and management Exercise dose was quantified as MET-min/week. Weekly dose was calculated as MET value × session duration (min) × weekly frequency, with exercise intensity assigned primarily according to the 2011 Compendium of Physical Activities.24 This compendium systematically categorizes 821 physical activities and their corresponding energy expenditure levels, providing broad coverage of commonly used exercise interventions. Weekly training frequency was determined from the reported number of sessions. Session duration was based only on the main training component; when warm-up, main exercise, and cool-down were reported separately, warm-up and cool-down periods were excluded from dose calculation. For studies without reported session duration, the mean duration from previous studies of similar exercise interventions was used.25 Based on the empirical dose distribution and to improve network connectivity, active exercise doses of <375, 375–<625, 625–<875, and ≥875 MET-min/week were assigned to the 250, 500, 750, and 1000 MET-min/week nodes, respectively26; exact midpoint values were assigned to the higher node. The control condition was coded as 0 MET-min/week. These were pragmatic analytical nodes rather than clinical dose thresholds. Certainty of evidence The certainty of evidence for the meta-analysis results was assessed using the GRADE approach.27 This approach assesses the certainty of evidence across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Each domain was rated as having no serious concerns, serious concerns, or very serious concerns, corresponding to no downgrade, one-level downgrade, or two-level downgrade, respectively. The certainty of evidence for each outcome was then classified as high, moderate, low, or very low. Statistical analysis Pairwise meta-analyses In pairwise meta-analyses, the effects of exercise interventions on each sleep-related outcome were estimated separately using standardized mean differences (SMDs) with 95% confidence intervals (CIs). To ensure a consistent direction of effect, outcome scores were harmonized so that positive effect estimates indicated a beneficial effect of exercise on the corresponding sleep-related outcome. For Siu et al.’s three-arm trial, the mind–body and combined exercise arms were pooled according to Cochrane guidance for the overall pairwise analysis. In modality-specific analyses, each exercise arm was compared separately with the full no-exercise control group, and the resulting correlated estimates were not jointly included in the overall pooling. Given the anticipated clinical and methodological heterogeneity across studies, random-effects models were applied to all pairwise meta-analyses, irrespective of the observed level of statistical heterogeneity. Between-study heterogeneity was quantified using the I2 statistic and the between-study variance (τ2). For the primary outcome, PSQI, subgroup analyses were additionally conducted by sleep disorder type where sufficient data were available. Potential publication bias was assessed using Egger's28 test and Begg's test.29 Network meta-analysis Assessment of transitivity and clinical heterogeneity Before conducting the NMA, we compared participant characteristics, clinical diagnoses, baseline PSQI scores, control conditions, exercise intervention protocols, and outcome assessment methods across treatment comparisons to assess the plausibility of the transitivity assumption (Table S5). Given that obstructive sleep apnoea (OSA) differs from insomnia and other non-respiratory sleep disorders in its underlying pathophysiology and clinical characteristics,30,31 and that the limited number of relevant studies precluded the construction of a separate connected evidence network, OSA was examined as a distinct subgroup in the clinical population subgroup analysis. The NMA was conducted only for the primary outcome, PSQI. ISI and ESS were treated as secondary outcomes and, because the available studies were insufficient to form connected evidence networks, were analysed using pairwise meta-analysis only. In the network meta-analysis, the three arms of Siu et al. were assigned to the mind–body exercise, combined exercise, and no-exercise nodes, with correlations arising from the shared control group accounted for using a multi-arm trial model. Network model specification and evaluation Bayesian NMA models were then constructed using MetaInsight (v5.0.1) with the R packages “gemtc” and “BUGSnet” to integrate direct and indirect evidence and compare the relative effects of different exercise interventions on sleep quality. Intervention effects were expressed as SMDs with 95% CrIs. Network geometry was presented using network plots, where node size represented the sample size of each intervention group and edge thickness indicated the number of direct comparisons. Between-study heterogeneity was assessed using τ. Global consistency was evaluated by comparing the deviance information criterion (DIC), residual deviance, and variance parameters between the consistency model and the unrelated mean effects (UME) model. At the same time, local inconsistency was further examined using node-splitting. To assess robustness, model fit indices were compared across fixed-effect, random-effects, and random-effects UME models. Interventions were then ranked using the surface under the cumulative ranking curve (SUCRA), with higher SUCRA values indicating a greater probability of improving sleep quality.32 Dose-response meta-analysis A Bayesian random-effects dose-response NMA was conducted to evaluate the dose-response relationship between exercise dose and improvements in sleep quality.33 Before analysis, key assumptions regarding transitivity, consistency, and network connectivity were assessed, with no evidence of model assumption violations 34, 35, 36. During dose-response model construction, several candidate functions were compared, including Emax, natural cubic spline, non-parametric monotonic, and quadratic models, each fitted under common- and random-treatment-effects assumptions.37 Model selection was based on DIC, between-study standard deviation, number of parameters,38 and residual deviance to identify the best-fitting model. Based on the final model, sleep quality trends across different exercise doses were estimated, and dose-response curves were plotted. Relative effect estimates were further used to compare differences among exercise modalities in improving sleep quality.33 Effect sizes were reported as estimates with 95% CrIs. Effects were considered statistically significant when the 95% CrI did not cross zero. All analyses were performed in R software (version 4.3.1) using the MBNMAdose package, and dose-response curves were plotted with ggplot2. Results Search result Based on the predefined search strategy and eligibility criteria, 13,058 records were identified through database searches. After removing 2375 duplicates, titles and abstracts were screened, and 60 articles underwent full-text assessment. Finally, 23 studies met the inclusion criteria and were included in data extraction, risk-of-bias assessment, and quantitative analyses. The study selection process is shown in Fig. 1. Flowchart of the selection and inclusion process, following the PRISMA Statement. Characteristics of included studies This meta-analysis included 23 studies involving 1478 participants aged 45-77 years. The included studies contributed 24 exercise intervention arms, comprising walking (n = 2), mind–body exercise (n = 9), aerobic exercise (n = 7), HIIT (n = 1), and combined exercise (n = 5). For exercise-dose classification, 14 intervention arms were assigned to the predefined MET-min/week categories: 250 (n = 3), 500 (n = 5), 750 (n = 3), and 1000 MET-min/week (n = 3). The main characteristics of the included studies and intervention arms are summarized in Table 1. Table 1Table of characteristics of included studies.Table 1StudiesCountryStudy designParticipant samples (male: female)Age (mean ± SD)Baseline sleep severitySleep quality assessment toolSleep disorder diagnosisInterventionDurationExercise dose(Hartescu et al., 2020)39United KingdomRCTEG: 5/15 CG: 6/15EG: 59.50 ± 10.59 CG: 60.10 ± 8.51ISI: Overall: 16.11 ± 4.93; EG: 15.82 ± 4.23; CG: 16.37 ± 5.58ISIInsomnia disorder; diagnosed according to RDC-I criteriaWalking30 min/session; 5 sessions/week; 26 weeksNA(Sengul et al., 2011)40TurkeyRCTEG:10/0 CG: 10/0EG:54.40 ± 6.57 CG: 48.00 ± 7.49AHI: EG: 15.19 ± 5.43 events/h; CG: 17.92 ± 6.45 events/hESSMild-to-moderate OSAS; diagnosed and classified based on PSG and AHIAerobic exercise75 min/session; 3 sessions/week; 12 weeksNA(Castro et al., 2019)41SpainRCTEG:NR CG: NREG:64.5 CG: 67AHI (median): Overall: 32 events/h; EG: 30.5 events/h; CG: 37 events/hESSModerate-to-severe OSA; diagnosed and classified based on PSG or respiratory polygraphy and AHIWalking30 min/session; 3 sessions/week; 8 weeksNA(Gokmen et al., 2019)42TurkeyRCTEG:13/12 CG: 18/7EG:50.44 ± 8.38 CG: 45.68 ± 7.64AHI: EG: 19.32 ± 7.09 events/h; CG: 18.66 ± 6.14 events/hPSQI, ESSMild-to-moderate OSA; diagnosed and classified based on overnight PSG and AHIMind-body exercise60 min/session; 3 sessions/week; 12 weeksNA(Bughin et al., 2020)43FranceRCTEG:25/9 CG: 30/4EG:53.71 ± 9.86 CG: 55.00 ± 10.27AHI inclusion range: 15–45 events/hESS, ISIModerate-to-severe OSA; diagnosed by PSG, with an AHI of 15–45 events/hCombined exercise120 min/session; 3 sessions/week; 8 weeksNA(Kline et al., 2012)44United StatesRCTEG:15/12 CG: 9/7 CG: 45.9 ± 2.2EG:47.6 ± 1.3AHI: EG: 32.2 ± 5.6 events/h; CG: 24.4 ± 5.6 events/hESSModerate-to-severe OSA; diagnosed by laboratory PSG, with an AHI ≥15 events/hCombined exercise30-45 min/session; 6 sessions/week; 12 weeksNA(Fan et al., 2020)45ChinaRCTEG:12/55 CG: 22/50EG:70.3 ± 5.7 CG: 71.8 ± 6.7PSQI: EG: 9.2 ± 3.4; CG: 9.8 ± 3.6PSQIPoor sleep quality; screened using PSQI ≥5Mind-body exercise45 min/session; 5 sessions/week; 24 weeksEG:750 CG: 0(Brandão et al., 2018)46BrazilRCTEG: NR CG: NREG:69.8 ± 7.4 CG: 69.9 ± 6.7PSQI: Overall:11.2 ± 3.2; ESS 8.6 ± 2.8;PSQIPoor sleep quality; screened using PSQI ≥5Combined exercise40 min/session; 3 sessions/week; 12 weeksEG:500 CG: 0(Lima et al., 2021)47BrazilRCTEG:2/13 CG: 4/10EG:67.2 ± 4.4 CG: 68.0 ± 6.1PSQI: EG: approximately 8.0 ± 3.9; CG: approximately 6.9 ± 4.0PSQIPoor sleep quality; screened using PSQI ≥5Aerobic exercise60 min/session; 3 sessions/week; 6 weeksEG:750 CG: 0(Durcan et al., 2014)48IrelandRCTEG:10/30 CG: 18/20EG:61 ± 8.0 CG: 59 ± 12PSQI: EG: 7.2 ± 4.4; CG: 5.6 ± 5.2PSQIRheumatoid arthritis with poor sleep quality; screened using PSQI >5Combined exercise45 min/session; 3 sessions/week; 12 weeksEG:500 CG: 0(McKenna et al., 2021)49IrelandRCTEG:0/10 CG: 0/10EG:58 ± 7.4< CG: 56 ± 7.9PSQI: EG: 13.4 ± 2.6; CG: 12.8 ± 2.9PSQIRheumatoid arthritis with poor sleep quality; screened using PSQI >5Aerobic exercise40 min/session; 3 sessions/week; 8 weeksEG:500 CG: 0(Song et al., 2023)50ChinaRCTEG:9/36 CG: 12/32EG:76.71 ± 5.96 CG: 75.20 ± 6.63PSQI: EG: approximately 11.07 ± 2.65; CG: approximately 10.66 ± 3.14PSQIMild cognitive impairment with poor sleep quality; screened using PSQI >5Aerobic exercise60 min/session; 3 sessions/week; 16 weeksEG:1000 CG: 0(Tseng et al., 2020)51ChinaRCTEG:5/15 CG: 2/18EG:61.1 ± 6.8 CG: 62.2 ± 7.4Mean PSQI: approximately 12.4PSQIPoor sleep quality; screened using CPSQI >5Aerobic exercise50 min/session; 3 sessions/week; 12 weeksEG:750 CG: 0(Filho et al., 2023)52BrazilRCTEG:10/7 CG: 9/10EG:53.2 ± 9.9 CG: 55.1 ± 9.8AHI: Overall: 42.0 ± 22.9 events/hPSQIModerate-to-severe untreated OSA; diagnosed by overnight PSG, with an AHI ≥15 events/hHIIT43 min/session; 3 sessions/week; 12 weeksNA(Gupta et al., 2023)53IndiaRCTEG:12/6 CG: 15/4EG:46.2 ± 9.4 CG: 45.7 ± 10.7AHI: EG: 51.2 ± 28.0 events/h; CG: 47.2 ± 23.0 events/hPSQI, ESSOSA; diagnosed by PSG, with an AHI >5 events/hMind-body exercise45 min/session; 5 sessions/week; 52 weeksNA(He et al., 2024)54ChinaRCTEG:2/11 CG: 0/13EG:69.85 ± 5.37 CG: 69.62 ± 4.46PSQI: EG: 11.62 ± 3.73; CG: 13.77 ± 3.00ISIPoor sleep quality; screened using PSQI >5Mind-body exercise60 min/session; 3 sessions/week; 4 weeksNA(Irwin et al., 2014)55USARCTEG:17/31 CG (Sleep education): 7/18EG:66.3 ± 7.4 CG (Sleep education): 66.4 ± 7.7PSQI: approximately 10.4–11.1PSQI, ESSChronic primary insomnia; diagnosed by structured clinical interview according to DSM-IV-TR and ICSD-2 criteriaMind-body exercise120 min/session; 1 session/week; 17 weeksEG:250 CG (Sleep education): 0(Reid et al., 2010)56USARCTEG:0/10 CG: 1/6EG:62.0 ± 4.5 CG: 63.5 ± 4.3PSQI>5PSQIChronic primary insomnia; included based on symptom duration, short sleep duration, and PSQI >5Aerobic exercise35 min/session; 4 sessions/week; 16 weeksEG:500 CG: 0(Tan et al., 2016)57FinlandRCTEG:24/0 CG: 21/0EG:51.2 CG: 52.6Difficulty initiating sleep, difficulty maintaining sleep, early-morning awakening, or non-restorative sleepESSChronic insomnia symptoms; screened using a modified BNSQ and quantitative symptom thresholdsAerobic exercise45 min/session; 3 sessions/week; 26 weeksNA(Lyu et al., 2026)58ChinaRCTEG:15/18 CG:16/18EG:70.4 ± 2.7 CG: 69.9 ± 2.9PSQI: EG: 7.9 ± 2.0; CG:8.2 ± 1.4PSQIMild sleep disturbance; screened using a PSQI score of 5–10Mind-body exercise60 min/session; 5 sessions/week; 12 weeksEG:1000 CG: 0(Irwin et al., 2008)12United StatesRCTEG:18/41 CG: 23/30EG:69.65 ± 6.15 CG: 70.17 ± 7.50PSQI: EG: 6.67 ± 1.54; CG: 8.18 ± 3.25PSQIModerate sleep complaints/poor sleep quality; defined as PSQI ≥5Mind-body exercise40 min/session; 3 sessions/week; 16 weeksEG:250 CG: 0(Hosseini et al., 2011)59IranRCTEG:16/15 CG: 14/17EG:68.74 ± 5.48 CG: 69.42 ± 5.34PSQI: EG: 10.33 ± 2.24; CG: 10.41 ± 2.43PSQIPoor sleep quality; screened using PSQI ≥5Mind-body exercise25 min/session; 3 sessions/week; 12 weeksEG:250 CG: 0(Siu et al., 2021)60ChinaRCTEG1:21/84 EG2:21/84 CG: 22/88EG1:67.3 ± 5.7 EG2:66.5 ± 6.4 CG: 68.0 ± 8.2PSQI: approximately 11.3–11.4; ISI: approximately 13.4–14.0PSQI, ISIChronic insomnia; diagnosed according to DSM-5 criteria using a semi-structured interviewMind-body exercise/combined exercise60 min/session; 3 sessions/week; 12 weeksEG1:500 EG2:1000 CG: 0Note: RCT: Randomized Controlled Trial; NA: not applicable; EG: Experimental Group; CG: Control Group; PSQI: Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; ESS, Epworth Sleepiness Scale. For Irwin et al. (2014), only the Tai Chi and sleep education arms were included; the active CBT-I arm was excluded from the meta-analysis. Table of characteristics of included studies. Note: RCT: Randomized Controlled Trial; NA: not applicable; EG: Experimental Group; CG: Control Group; PSQI: Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; ESS, Epworth Sleepiness Scale. For Irwin et al. (2014), only the Tai Chi and sleep education arms were included; the active CBT-I arm was excluded from the meta-analysis. Risk of bias and GRADE evidence certainty assessment results Among the 23 included studies, 5 were rated as having a high risk of bias and 18 as having some concerns (Fig. S1). Overall, bias mainly arose from missing outcome data and outcome measurement. Specifically, some studies had loss to follow-up or inadequate handling of missing data. Additionally, because sleep quality was primarily assessed using subjective scales and participant blinding was difficult in exercise interventions, outcomes may have been influenced by self-reporting and expectation effects. According to the GRADE framework, the certainty of evidence ranged from low to moderate. This was primarily due to study limitations, as several studies had some concerns or a high risk of bias, resulting in downgrading for risk of bias. Detailed evidence certainty assessments are provided in Supplementary Table S6. Publication bias test Publication bias was assessed separately for the PSQI, ESS, and ISI outcomes using Begg's test and Egger's regression test. For all three outcomes, neither test yielded statistically significant results (P > 0.05), indicating that no statistically significant evidence of publication bias was detected. However, the results for ESS and ISI should be interpreted with caution because of the limited number of included studies, which may reduce the statistical power of these tests to detect funnel plot asymmetry. Therefore, the possibility of publication bias for these outcomes cannot be excluded. The corresponding funnel plots are presented in Fig. S2–S4. Pairwise meta-analysis The results of the pairwise meta-analyses are presented in Fig. 2. Given the anticipated clinical and methodological heterogeneity across studies, all overall and subgroup analyses were conducted using random-effects models, with between-study variance estimated using the restricted maximum likelihood method. Overall, exercise significantly improved PSQI scores (SMD = 0.62, 95% CI: 0.40∼0.83; I2 = 65.8%, τ2 = 0.120) and ISI scores (SMD = 0.64, 95% CI: 0.34∼0.94; I2 = 49.2%, τ2 = 0.051), whereas the effect on ESS scores was not statistically significant (SMD = 0.70, 95% CI: −0.05∼1.46; I2 = 90.1%, τ2 = 1.057). Subgroup analyses showed significant improvements in PSQI with mind–body exercise (SMD = 0.47, 95% CI: 0.23∼0.70; I2 = 47.0%, τ2 = 0.051), aerobic exercise (SMD = 1.52, 95% CI: 0.57∼2.48; I2 = 83.0%, τ2 = 0.908), and combined exercise (SMD = 0.49, 95% CI: 0.08∼0.90; I2 = 72.8%, τ2 = 0.095). Sensitivity analyses, including the Hartung–Knapp adjustment and exclusion of the study-level estimate with |SMD| ≥3, supported the direction of the aerobic-exercise effect, although its magnitude was attenuated and substantial heterogeneity persisted; therefore, this estimate should be interpreted cautiously (Table S7). In the subgroup analysis according to sleep disorder type, exercise significantly improved PSQI scores among participants with insomnia or poor sleep quality (SMD = 0.62, 95% CI: 0.21∼1.03; I2 = 60.0%, τ2 = 0.238). However, the test for subgroup differences was not statistically significant (χ2 = 0.15, df = 2, P = 0.927), indicating no evidence that the effects of exercise differed across sleep disorder categories (Fig. S5). For ESS, a significant improvement was observed with mind–body exercise (SMD = 0.48, 95% CI: 0.01∼0.96; I2 = 44.0%, τ2 = 0.077). For ISI, significant improvements were observed with mind–body exercise (SMD = 0.78, 95% CI: 0.30∼1.25; I2 = 30.0%, τ2 = 0.052) and combined exercise (SMD = 0.57, 95% CI: 0.06∼1.08; I2 = 61.6%, τ2 = 0.089). The overall pooled effect was also statistically significant (SMD = 0.64, 95% CI: 0.34∼0.94; I2 = 49.2%, τ2 = 0.051). The substantial heterogeneity observed in several analyses, particularly for ESS and aerobic exercise on PSQI, indicates that these estimates should be interpreted with caution. Pairwise meta-analysis of the effects of exercise interventions on sleep-related outcomes in middle-aged and older adults with sleep disorders. PSQI, Pittsburgh Sleep Quality Index; ESS, Epworth Sleepiness Scale; ISI, Insomnia Severity Index; SMD, standardized mean difference; CI, confidence interval; I2, percentage of total variability attributable to between-study heterogeneity; τ2, estimated between-study variance. NA indicates that the value was not available. Network meta-analysis Fig. 3 presents the network meta-analysis results for PSQI outcomes in middle-aged and older adults with sleep disorders. Based on the SUCRA values, the interventions were ranked as follows: aerobic exercise (99.1%) > mind-body exercise (52.1%) > combined exercise (48.6%) > control (0.2%). Compared with the control group, aerobic exercise (SMD = 1.34, 95% CrI: 0.86∼1.82), mind-body exercise (SMD = 0.64, 95% CrI: 0.30∼0.97), and combined exercise (SMD = 0.60, 95% CrI: 0.13∼1.07) significantly improved PSQI outcomes. Aerobic exercise was also significantly more effective than mind-body exercise (SMD = 0.70, 95% CrI: 0.12∼1.29) and combined exercise (SMD = 0.74, 95% CrI: 0.07∼1.41), whereas no significant difference was observed between mind-body exercise and combined exercise (SMD = 0.04, 95% CrI: −0.54∼0.61). No significant inconsistency was detected in the closed loop (P > 0.05), supporting the use of the consistency model. Detailed results on inconsistency and node-splitting are provided in the supplementary materials (Tables S8–S10). Network plot of all trials reporting PSQI outcomes in middle-aged and older adults with sleep disorders, and the effects of different exercise modalities compared with the control group. Dose-response meta-analysis As shown in Fig. 4, a nonlinear inverted U-shaped dose-response relationship was observed between total exercise dose and improvements in sleep quality among middle-aged and older adults with sleep disorders. The predicted beneficial effect increased progressively with weekly exercise dose and reached its maximum at approximately 539 MET-min/week (predicted SMD = 1.09, 95% CrI, 0.63∼1.62). The 95% credible interval for the estimated optimal dose ranged from 449 to 900 MET-min/week. Beyond 539 MET-min/week, the predicted improvement gradually diminished. The dose-response analysis further indicated that approximately 225 MET-min/week represented the minimum dose associated with a significant improvement in sleep quality (SMD = 0.71, 95% CrI: 0.35∼1.12), and this significant effect was maintained up to approximately 900 MET-min/week (SMD = 0.65, 95% CrI: 0.004∼1.392). Overall, an exercise dose of approximately 225-900 MET-min/week may represent an effective dose range for improving sleep quality in this population, with the greatest predicted benefit occurring at approximately 539 MET-min/week. Sensitivity analyses using alternative models yielded broadly similar effective dose ranges, supporting the robustness of the dose–response association. Further details are provided in the Supplementary Materials (Figs. S6–S18 and Tables S11–S15). Dose–response relationship between weekly exercise dose and PSQI-assessed sleep quality in middle-aged and older adults with sleep disorders. Each point represents an observation from the original dataset. The shaded area represents the 95% CrI. The black dashed lines indicate the lower and upper bounds of the dose range in which the 95% CrI does not include zero, and the red dashed line indicates the dose level (MET-min/week) corresponding to the largest predicted effect within the observed dose range. Discussion Principal findings This study systematically evaluated the effects of exercise interventions on sleep-related outcomes in middle-aged and older adults with sleep disorders. Pairwise meta-analyses showed that exercise significantly improved PSQI and ISI scores, indicating better subjective sleep quality and reduced insomnia severity, respectively; however, no significant improvement was observed in ESS-assessed daytime sleepiness. The PSQI-based network meta-analysis indicated that aerobic exercise, mind-body exercise, and combined exercise were all superior to control conditions, with aerobic exercise producing the greatest improvement. The dose-response analysis revealed an inverted U-shaped association between total weekly exercise dose and improvements in PSQI scores. Significant benefits were observed at approximately 225-900 MET-min/week, with the maximum predicted effect occurring at approximately 539 MET-min/week. Comparison with other studies Previous reviews have evaluated the effects of exercise interventions on sleep quality,61 but most have focused on general populations or patients with other chronic conditions, limiting their applicability to middle-aged and older adults with sleep disorders. Existing studies have mainly focused on overall effects, with limited systematic comparisons of the relative efficacy of different exercise modalities. Therefore, this study integrated direct and indirect evidence to compare the effects of different exercise modalities, including mind-body exercise and aerobic exercise, on sleep quality in this population. Sleep quality is an important indicator of health status in middle-aged and older adults with sleep disorders.62,63 Sleep quality is a key health outcome in middle-aged and older adults with sleep disorders; its improvement may alleviate insomnia symptoms, daytime fatigue, and emotional distress, and reduce the risk of adverse health outcomes.64,65 Our PSQI-based network meta-analysis showed that aerobic exercise significantly improved sleep quality in middle-aged and older adults with sleep disorders (SMD = 1.34, 95% CrI: 0.86 to 1.82). Aerobic exercise also ranked highest among the evaluated exercise modalities, with a SUCRA value of 99.1%, suggesting that it may be one of the most effective exercise interventions for improving sleep quality in this population. An RCT by Reid et al. found that 16 weeks of moderate-intensity aerobic exercise significantly improved total PSQI score, sleep latency, sleep duration, and sleep efficiency in older adults with chronic insomnia.56 This effect may be related to increased energy expenditure, enhanced sleep homeostatic pressure, post-exercise reductions in body temperature, and improved negative mood, thereby facilitating sleep initiation and maintenance.66 Therefore, aerobic exercise may serve as an important non-pharmacological strategy for sleep management in middle-aged and older adults with sleep disorders. Moreover, mind-body exercise may also improve subjective sleep quality by reducing psychological stress and autonomic arousal through breathing regulation, meditative relaxation, and low-intensity physical activity.67,68 Previous RCTs have shown that multicomponent exercise not only improves muscular endurance, cardiorespiratory fitness, balance, and overall functional capacity in middle-aged and older adults, but also reduces PSQI total scores and sleep disturbance scores while improving sleep efficiency.69 Meanwhile, exercise-induced reductions in psychological distress may also contribute to its beneficial effects on sleep.70 Therefore, the improvements in physiological function and psychological well-being induced by combined exercise may act synergistically to reduce nocturnal sleep disturbances and thereby enhance subjective sleep quality. Notably, this study did not find a statistically significant improvement in PSQI scores following exercise interventions among middle-aged and older adults with OSA. This finding may partly reflect the limited number of included studies and participants in this subgroup, resulting in insufficient statistical power. In addition, exercise may have only modest effects on the core respiratory disturbances characteristic of OSA. Previous meta-analytic evidence has shown that exercise training reduces the apnea–hypopnea index (AHI) by an average of approximately 6.27 events/h.71 In the RCT conducted by Kline et al., the AHI decreased from 32.2 to 24.6 events/h; however, participants still exhibited moderate residual OSA after the intervention.72 Moreover, the PSQI primarily assesses subjective sleep quality, and its correlations with objective indices of OSA severity, such as the AHI and oxygen desaturation index, have been reported to be weak or statistically non-significant.73 Therefore, the relatively modest physiological improvements induced by exercise may not necessarily translate into statistically significant changes in PSQI scores. The PSQI-based dose-response analysis revealed an inverted U-shaped nonlinear association between total exercise dose and improvements in sleep quality. As the weekly exercise dose increased, the predicted improvement in PSQI-assessed sleep quality became progressively greater, reaching a peak at approximately 539 MET-min/week. Beyond this point, the predicted benefit gradually diminished with further increases in exercise dose. Overall, weekly exercise doses of approximately 225-900 MET-min were associated with statistically significant improvements in PSQI scores, with the greatest predicted benefit occurring at approximately 539 MET-min/week. This finding is broadly consistent with previous dose-response meta-analyses. Xue et al. identified an optimal exercise dose of approximately 440 MET-min/week for improving sleep quality in older adults,74 whereas another nonlinear meta-analysis estimated the optimal dose to be approximately 527 MET-min/week.75 Collectively, these findings suggest that the sleep-related benefits of exercise do not increase indefinitely with escalating dose but may instead occur within an appropriate therapeutic window. One plausible explanation is that moderate exercise enhances homeostatic sleep pressure, promotes circadian regulation, alleviates psychological stress, and improves sleep continuity, thereby shortening sleep-onset latency, increasing sleep efficiency, and enhancing overall sleep quality.76 At higher exercise loads, however, the accompanying increases in physiological stress and recovery demands may partially offset these benefits.77 Previous research has shown that excessive endurance-training loads can reduce sleep efficiency and sleep duration, while functional overreaching may also impair cardiac parasympathetic modulation during slow-wave sleep.78 Accordingly, approximately 539 MET-min/week may represent a model-derived dose level associated with the greatest predicted improvement within the range of evidence examined in the present study, although this estimate requires further validation in middle-aged and older adults with sleep disorders. For the secondary outcomes, pairwise meta-analysis showed that exercise interventions significantly reduced ISI scores, whereas the improvement in ESS scores did not reach statistical significance, suggesting that the effects of exercise may differ between insomnia severity and daytime sleepiness. The beneficial effects of exercise on insomnia symptoms may be mediated through multiple mechanisms. On the one hand, regular exercise may increase energy expenditure and strengthen homeostatic sleep drive, thereby shortening sleep-onset latency and reducing wakefulness after sleep onset.76 On the other hand, exercise may alleviate anxiety, depression, and psychological distress, thereby reducing sleep-related cognitive and emotional hyperarousal and subsequently improving difficulties with sleep initiation and maintenance 30,79,80. Hartescu et al. found that engaging in at least 150 min of moderate-to-vigorous physical activity per week significantly reduced ISI scores in individuals with insomnia, providing direct evidence that exercise can alleviate insomnia symptoms.81 In contrast, the present study found no significant change in ESS scores. This may be attributable to the limited number of available studies and small sample sizes, as well as the fact that daytime sleepiness is influenced by multiple factors, including OSA severity, nocturnal sleep duration, medication use, and comorbid conditions. Thus, although the current evidence supports the beneficial effects of exercise on insomnia severity, its effects on daytime sleepiness remain uncertain and should be further examined in large-scale, high-quality randomized controlled trials. Clinical implications This study provides preliminary clinical insights. The network meta-analysis suggested that aerobic exercise was associated with the largest estimated improvement in sleep quality among middle-aged and older adults with sleep disorders; however, the limited number of comparisons and substantial heterogeneity preclude definitive conclusions regarding its superiority over other exercise modalities. The dose–response analysis suggested an inverted U-shaped association between total weekly exercise dose and improvement in PSQI scores. Under the selected model, improvements were predicted across approximately 225-900 MET-min/week, with the maximum predicted effect occurring at approximately 539 MET-min/week. However, this peak dose represents an exploratory, model-dependent estimate rather than a clinically validated prescription target, given its wide 95% CrI, the absence of studies with a low overall risk of bias, between-study heterogeneity, imprecision, and possible small-study effects. This estimate may help inform future dose-ranging trials, while clinical exercise prescriptions should continue to be individualized according to patients’ health status, exercise capacity, preferences, and adherence. Strengths and limitations This study integrated pairwise meta-analysis, network meta-analysis, and dose-response meta-analysis to systematically compare the effects of different exercise modalities on sleep quality in middle-aged and older adults with sleep disorders and to examine the association between total exercise dose and intervention effects. Although these findings may inform the development of exercise prescriptions, they should be interpreted with caution. First, owing to incomplete reporting in the original studies, this review was unable to adequately control for several potential effect modifiers, including the type and severity of sleep disorders, comorbidities, psychological status, physical function, medication use, and lifestyle factors. These factors may have influenced the effect estimates and reduced comparability across studies. Second, some included studies had small sample sizes, and considerable heterogeneity was observed in exercise protocols, control conditions, and outcome assessments, which may have limited the robustness and external generalizability of the findings. In addition, the number of studies within each exercise-modality subgroup was limited, and the available dose gradients were insufficient; therefore, modality-specific dose-response analyses were not conducted, and the effective dose ranges and dose levels associated with the greatest predicted effects for individual exercise modalities remain unclear. For the ISI, ESS, and subgroups defined by sleep-disorder type, only pairwise meta-analyses were performed because of the limited numbers of studies and participants. Exercise modality remained partially associated with population and geographic characteristics, and residual confounding could not be excluded. Finally, language restrictions and the lack of long-term follow-up may have affected the completeness of the evidence. Conclusions Low-to moderate-certainty evidence suggests that exercise interventions may improve PSQI-assessed sleep quality and ISI-assessed insomnia severity in middle-aged and older adults with sleep disorders, whereas no clear benefit was observed for ESS-assessed daytime sleepiness. In the PSQI network, aerobic exercise yielded the largest estimated effect, although treatment rankings remain uncertain because of sparse and heterogeneous evidence. The dose-response analysis suggested a nonlinear association, with the quadratic model estimating the largest predicted effect near 539 MET-min/week; however, this exploratory estimate was model-dependent and should not be interpreted as a definitive optimal dose or modality-specific prescription. Further high-quality RCTs are needed to confirm efficacy, safety, dose ranges, and clinical applicability and to support individualized exercise prescriptions. Funding This study received no funding. Author contributions Ting Wang: Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Writing – original draft, Writing – review & editing, Project administration. Zhuang Ma: Investigation, Data curation, Validation, Writing – review & editing. Boao Zhu: Investigation, Data curation, Validation, Writing – review & editing. Yaqiong Tang: Supervision, Validation, Writing – review & editing. All authors read and approved the final manuscript. Competing interests The authors declare that they have no competing interests. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.jesf.2026.200526. Supplementary data The following are the Supplementary data to this article: Multimedia component 1 Multimedia component 2 Multimedia component 3 References Associated Data Supplementary Materials Multimedia component 1 Multimedia component 2 Multimedia component 3