Abstract

Background:

Sleep disturbances are a common and burdensome non-motor symptom in Parkinson’s disease (PD). The comparative efficacy of different exercise modalities for sleep quality in PD remains unclear. This network meta-analysis (NMA) aimed to compare and rank the effects of various exercise interventions on sleep quality in people with PD.

Methods:

This study followed the PRISMA extension statement for network meta-analyses. A systematic search was performed across Web of Science, Embase, PubMed, the Cochrane Library, Scopus, CNKI, and Wanfang databases from their inception to November 3, 2025. Eligible randomized controlled trials (RCTs) were identified, with study selection, data extraction, and bias assessment conducted independently by two reviewers. NMA was performed using Stata 19.0. Consistency was examined using design-by-treatment interaction and node-splitting approaches. Rankings were estimated using the surface under the cumulative ranking curve (SUCRA).

Results:

Of 2309 records screened, 16 RCTs involving 932 people with PD were included. Aerobic exercise (AE) significantly improved sleep quality compared with control (SMD = −0.94, 95% CI: −1.82 to −0.07). SUCRA rankings were: AE (highest) > multimodal exercise (MME) > resistance training (RT) > stretching training (ST) > mind-body exercise (MBE) > control. No significant publication bias was found (Egger’s test, P = 0.438).

Conclusion:

This NMA indicates that aerobic exercise is the most promising modality for enhancing sleep quality in people with PD and may guide non-pharmacological treatment selection in clinical practice.

Systematic review registration:

https://www.crd.york.ac.uk/prospero/, identifier CRD420261341953

1 Introduction

Parkinson’s disease (PD) is the second most common neurodegenerative disorder associated with aging, following Alzheimer’s disease (Asadpoordezaki et al., 2025). According to the Global Burden of Disease (GBD) 2021 study, approximately 1.34 million new PD cases were recorded globally in 2021, with projections indicating a rise to 1.93 million incident cases by 2030. Largely driven by population aging, the total number of individuals living with PD is forecasted to increase sharply to 25.2 million by 2050 (Su et al., 2025; Xu L. et al., 2025). The economic impact of this escalating prevalence is considerable; for instance, a study across five European countries found that patients with late-stage PD incur average social costs ranging from 12,156 to 25,649 euros every three months (Kruse et al., 2024). Among the extensive range of symptoms that add to this substantial disease burden, sleep disturbances stand out as one of the most common non-motor manifestations, occurring in more than 80% of individuals with PD. These disturbances tend to worsen progressively with the advancement of the neurodegenerative process (Iranzo et al., 2024). Despite their high prevalence, however, fewer than half of patients report these issues to physicians or receive adequate attention and management (Taximaimaiti et al., 2021). Clinically, sleep issues are typically associated with poorer outcomes, including cognitive decline, mood disorders, and a significantly reduced quality of life for patients (Anderson et al., 2025).

Treatment options for sleep disturbances in PD include both pharmacological and non-pharmacological approaches. Although several studies have confirmed that certain medications (e.g., sedative-hypnotics) offer some efficacy for sleep problems in people with PD, chronic use of these medications is frequently linked to several adverse effects, including dependence, development of tolerance, heightened risk of falls, cognitive impairment, and excessive daytime somnolence (Tang et al., 2024). In contrast, non-pharmacological interventions are increasingly favored due to their lower incidence of adverse effects, relatively sustained benefits, and cost-effectiveness. Among these, exercise-based interventions—such as multimoday exercise (MME), resistance training (RT), mind-body exercise (MBE), stretching training (ST) and others—have demonstrated potential to improve sleep quality in people with PD across multiple randomized controlled trials (Nascimento et al., 2014; Xiao and Zhuang, 2016; Silva-Batista et al., 2017; Abo-Elyazed, 2018; Cheung et al., 2018; Zhu et al., 2019; Amara et al., 2020; Li et al., 2020; Moon et al., 2020; Wu et al., 2021; Wang et al., 2022; Mei-Hua et al., 2023; Li et al., 2024; Mehta et al., 2024; Hu et al., 2025; Li et al., 2025).

Previous meta-analyses have confirmed the overall benefits of exercise on sleep quality in people with PD, yet they either did not differentiate among specific exercise modalities or focused on a broad range of non-pharmacological interventions (including massage and music) without systematically comparing various exercise types (Tang et al., 2024; Li and Hu, 2025). Consequently, the relative efficacy of different exercise modalities—such as aerobic, resistance, stretching, multimodal, and mind−body exercise—remains unknown. However, most existing studies focus on pairwise comparisons between a single exercise modality and a control group, making it challenging to differentiate relative efficacy among the various effective exercise types. This evidence gap hinders clinicians and patients in identifying the optimal exercise regimen for improving sleep outcomes. Therefore, the present study employs a network meta-analysis (NMA) framework to comprehensively assess and compare the effects of various exercise modalities on sleep quality in people with PD. This analysis combines evidence derived from randomized trials to evaluate and rank the comparative effectiveness of various interventions. The goal is to highlight promising exercise strategies and support clinical decision-making for improving sleep quality in PD, which may have implications for patient well-being.

2 Methods

This study conducted a systematic review and meta-analysis in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) Statement (Hutton et al., 2015). In addition, the review protocol was registered in the PROSPERO registry (CRD420261341953) prior to data extraction.

2.1 Search strategy

The following electronic databases—Web of Science, Embase, PubMed, the Cochrane Library, Scopus, China National Knowledge Infrastructure (CNKI), and Wanfang—were systematically searched from their inception until 3 November 2025 to identify all relevant published articles. Search strategies consisted of a combination of Medical Subject Headings (MeSH) terms and free-text keywords related to PD, exercise interventions, and sleep disorders. These included the following: (1) Parkinson’s disease, Parkinson’s, Parkinson disease, exercise, Multimodal exercise, aerobic exercise, balance training, Baduanjin, qigong, Tai Chi, dance, sleep disorders, sleep, sleep quality, etc. The complete search strategies are fully detailed in Supplementary File 1.

2.2 Eligibility criteria

Eligibility for this review was defined based on the PICOS framework. The following criteria were required for study eligibility: (1) Population: Confirmed diagnosis of Parkinson’s disease. No restrictions were imposed regarding disease stage, age, sex, or time since diagnosis. (2) Intervention: The experimental group participated in an exercise-based intervention of any type. There were no limitations on frequency, duration, intensity, category of exercise, format, setting, or mode of delivery. (3) Comparison: The control group received either a sham activity, a delayed intervention, standard medical care, educational or supportive sessions (e.g., sleep hygiene advice), or a form of physical activity substantially different from the intervention under investigation. Studies that only compared minor variations of similar exercise approaches were excluded. (4) Outcomes: To be included, studies were required to report pre−to−post changes in sleep−related outcomes using at least one of the following instruments: Pittsburgh Sleep Quality Index (PSQI), Parkinson’s Disease Sleep Scale (PDSS), Parkinson’s Disease Sleep Scale−2 (PDSS−2), Epworth Sleepiness Scale (ESS), Mini−Sleep Questionnaire (MSQ), or Insomnia Severity Index (ISI). (5) Study Design: Randomized controlled trials (RCTs) constituted the included studies, irrespective of country of origin or publication type.

Exclusion criteria were as follows: (1) non−randomized study design; (2) Full text irretrievable or data not extractable; (3) The report was only a trial protocol or registration without enrolled participants; (4)The document type was a conference abstract, thesis, dissertation, or literature review; (5) The study duplicated previously published work or shared the same participant cohort as another report—in such cases, the most recent or most comprehensive version was retained; (6) The intervention combined exercise with any non-exercise component (e.g., cognitive training), unless the exercise effect could be isolated.

2.3 Data extraction

Two researchers (Z. D. and W. X.) independently extracted the relevant data, covering first author, year, country, sample size, gender, average age, Hoehn and Yahr stage, and disease duration. To assess inter-rater reliability, Cohen’s kappa (κ) was calculated for study selection. They also recorded intervention parameters, including session duration, training frequency, total duration, adherence, supervision, and site. Discrepancies were addressed by discussion, with consultation from a third researcher (Y. Z.) as needed. For each study, mean scores and standard deviations (SDs) were collected for outcome measures to enable effect size calculation. For studies reporting standard errors (SEs) for the experimental and control groups, we calculated standard deviations as SD = SE × . When SDs and SEs were both missing, we estimated standard deviations from other reported statistics (confidence intervals, t-values, quartiles, ranges, or p-values) using the procedures described in Section 7.7.3 of the Cochrane Handbook. If essential data were still missing after these methods, we contacted the corresponding authors up to four times across six weeks to request the information. The included studies’ characteristics are summarized in Table 1.

Mean age
(intervention/control)Hoehn and Yahr stageDuration of diseaseIntervention detailSession durationTraining frequencyDurationAdherenceSuperviseSiteOutcomesIntervention groupTEAMControl groupTEAM65.33 ± 8.17 vs 65.82 ± 5.192–36.0 (3.0–9.0) years vs 3.0 (1.0–7.5) yearsRT (leg press, knee
extension, chest press, overhead press, pull down) + body weight functional mobility(step-up, squat, jump squat, lunge, side
lunge, push-up, assisted pull-up, assisted dip)RTSleep hygiene suggestionCONNA3 times/week16 weeks92.2% ± 12.5%, 85% of patients > 90% vs 96.55%SupervisionCenter for Exercise Medicine-basePSQI, PSGs67.8 ± 6.8 vs 66.3 ± 8.11.7 vs 1.55.1 ± 3.9 years vs 4.6 ± 3.7 yearsMuscular resistance (bars, medicine balls, thera-bands, bobath balls, ankle weights and barbells were
progressively included) + Balance and motor coordination + Aerobic fitness (walk with obstacles)MMERoutine careCON60 min3 times/week24 weeksNASupervisionLaboratory-basedMini-Sleep Questionnaire (MSQ)63.5 ± 8.5 vs 65.8 ± 6.61–34.8 ± 2.9 yearsyogaMBEWait-list controlCON60min2 times/week12 weeksNASupervisionyoga studioPDSS68.83 ± 4.35 vs 67.95 ± 4.861.28 ± 0.45 vs 1.23 ± 0.406.63 ± 4.01 years vs 6.09 ± 3.85 yearsWu Qin XiMBESeated StretchingST90 min3 times/week24 weeksNASupervisionHome-basedPDSS68.17 ± 2.27 vs 66.52 ± 2.132.2 ± 0.21 vs 2.1 ± 0.235.45 ± 3.61 years vs 6.15 ± 2.63 yearsBaduanjin Qigong + 30 min walkingMME30 min walkingAE40 min4 times/week6 months93.75% vs 91.67%UnsupervisedHome-basedPDSS-264.6 ± 9.7 vs 64.4 ± 9.12.5 ± 0.5 vs 2.5 ± 0.410.0 ± 4.1 years vs 11.6 ± 6.0 yearsRT (leg-press, latissimus dorsi pull-down, ankle plantar flexion, chest-press, and half-
squat)RTRoutine careCON50 min2 times/week3 monthsNASupervisionCenter for Psychobiology and Exercise StudiesPSQI65.43 ± 7.27 vs 66.41 ± 7.412.51 ± 0.55 vs 2.41 ± 0.685.00 ± 5.0 vs 4.00 ± 5.5Baduanjin QigongMBERoutine careCON20–30min3 times/week12 weeks90.91% vs 90.91%SupervisionHospital-basedPSQI63.65 ± 6.02 vs 66.59 ± 8.611–24.97 ± 3.91 years vs 5.66 ± 3.81 yearsST (Entire body) + AE (Walking back and forth at home) + RT (low to medium
intensity)MMERoutine careCON10–50 min,accumulated 150 min/week3–7 times/week8 weeks55.10% vs 100%SupervisionHome-basedPSQI66.4 ± 8.1 vs 65.9 ± 5.41–34.25 ± 2.1 years vs 5.33 ± 3.36 yearsSix healing sounds’QigongMBESham QigongCON15–20 min2 times/week12 weeksNASupervisionHome-basedPDSS-266.13 ± 5.66 vs 65.27 ± 4.96NANAAE (walking on treadmill)AEUsual physical therapyCON40 min3 times/week3 monthsNANANAInsomnia Severity Index (ISI)68.53 ± 1.90 vs 67.77 ± 1.722 ± 2.2 vs 2 ± 1.24.68 ± 0.43 years vs 4.00 ± 0.39 yearsTai Chi (simplified Tai Chi training adapted
from Yang style) + Routine exerciseMMERoutine exerciseCON40–50 min3 times/week12 weeks94.74% vs 86.36%SupervisionHospital-basedPDSS72.07 ± 8.33 vs 69.80 ± 6.90 vs 67.13 ± 8.331–44.27 ± 3.31 years vs 4.80 ± 3.05 years vs 5.87 ± 2.72 yearsTai Chi (24-Form Simplified Tai Chi Chuan)MBERoutine careCON(a) 30min (b) 45min3 times/week24 weeks79.17% vs 69.44% vs 100%SupervisionHospital-basedPSQI62.7 ± 5.51 vs 61.5 ± 5.53 vs 62.8 ± 6.141–2.55.13 ± 3.11 years vs 5.44 ± 3.96 years vs 5.91 ± 4.01 years(a) Tai Chi (standardized Yi Tai Chi) (b) brisk
walking (50%–60% of the maximum heart rate)(a) MBE (b) AERoutine careCON60 min2 times/week12 month(a) 100% (b) 54.84% vs 53.13%SupervisionHospital-basedEpworth Sleepiness Scale (ESS)59.45 ± 9.37 vs 59.35 ± 8.13 vs 62.13 ± 8.211–2.539.15 ± 11.65 months vs 39.15 ± 11.65 months vs 36.80 ± 11.23 months(a) Garba Dance (b) physical
therapy (PNF)(a) AE (b) STRoutine careCON60 min5 times/week12 weeks(a) 85% (b) 85% vs 100%SupervisionHospital-basedPDSS-267.59 ± 4.94 vs 69.87 ± 3.42NANRYijinjingMBEConventional rehabilitation trainingCON30min5 times/week8 weeks92.4% vs 90.7%SupervisionHospital-basedPSQI70.47 ± 7.13 vs 71.77 ± 6.651–32–14 years vs 3–12 yearsResistance band exercise (Three items for upper limb movement and two items for lower limb movement)RTRoutine careCON20–30min3 times/week30 dayNASupervisionHospital-basedPSQI

Characteristics of the included studies.

RCT, Randomized controlled trial; NA, not applicable; AE, aerobic exercise; MME, Multimodal exercise; MBE, mind -body Exercise; RT, resistance training; ST, Stretching training; NR, not reported; vs: versus.

2.4 Quality assessment

Risk of bias for the 16 randomized controlled trials was independently assessed by two investigators (Z.D. and W.X.) using the Cochrane RoB 2 tool. In cases of inconsistency, a third reviewer (H.C.) was consulted to reach consensus. The assessment covered domains related to the randomization procedure, adherence to assigned interventions, completeness of outcome data, outcome measurement, and selective reporting. An overall judgment for each study was subsequently assigned as low risk, some concerns, or high risk. Additionally, the certainty of evidence for the primary outcome was evaluated using the GRADE (Grading of Recommendations Assessment, Development and valuation) framework, which classifies evidence as high, moderate, low, or very low.

2.5 Statistical analysis

Using Stata (version 19.0), we performed network meta-analysis. A network plot was created to visualize direct comparisons among exercise modalities, where nodes (scaled by sample size) represent interventions and connecting lines (weighted by study count) indicate direct comparisons.

A p-value > 0.05 was considered to indicate no inconsistency between direct and indirect evidence; therefore, a consistency model was applied. Otherwise, an inconsistency model was used. A random-effects model was employed to account for between-study heterogeneity.

Effect sizes were expressed as standardized mean differences (SMD) with 95% confidence intervals for continuous outcomes (pre-post changes in sleep quality scales). Pairwise comparison results were presented in a league table. We obtained intervention rankings via the surface under the cumulative ranking (SUCRA) curve, where SUCRA scores span 0% to 100%; higher scores reflect a greater chance of being among the most effective interventions. Such rankings, however, require cautious interpretation when clinically meaningful differences between interventions are absent. Heterogeneity was assessed using Cochran’s Q test (p < 0.10 indicating significant heterogeneity) and the I² statistic (interpreted as: 0–40% low, 40–60% moderate, 60–75% substantial, 75–100% considerable). We explicitly explored consistency versus inconsistency models (p < 0.05 indicating inconsistency) using the design-by-treatment interaction model and node-splitting analysis. To evaluate the transitivity assumption, a network meta-regression was performed using mean age as a potential effect modifier.

For publication bias assessment, we generated comparison-adjusted funnel plots and visually checked them for symmetry. Egger’s test was additionally used to quantify asymmetry, with a p−value above 0.05 indicating no significant publication bias.

3 Results3.1 Study selection

The investigators (Z.D. and W.X.) each used a comparable method to evaluate the titles and abstracts of the references obtained through the implementation of the aforementioned search technique. Our initial database search yielded 2309 records. Following stepwise screening, 16 trials met the eligibility criteria for the meta-analysis. Figure 1 presents the PRISMA flowchart, which details the study selection process. The basic characteristics of the 16 included RCTs are summarized in Table 1. Inter-rater reliability for study selection was high, with a Cohen’s kappa (κ) of 0.94 (95% CI: 0.89–0.99). A total of 932 participants were enrolled across all trials. The studies were conducted in the United States, China, Brazil, Egypt, and India. All trials were randomized controlled trials, most of which employed a single-blind design, while three used an assessor-blind design. Overall, 58.9% of participants were male and 41.1% were female. The mean age of participants ranged from 59.4 to 72.1 years across studies. Disease duration ranged from 4.0 to 11.6 years, and the Hoehn and Yahr stage was predominantly between 1 and 3. Training frequency ranged from 2 to 5 times per week (mean, 3.1 times/week), and intervention duration ranged from 8 weeks to 12 months. Session duration varied from 15 to 90 min.

PRISMA flow diagram of the search process for studies.

3.2 Risk of bias assessment

Sixteen RCTs were assessed for risk of bias, one study was considered low risk across all five domains, reflecting high methodological quality (Moon et al., 2020). Inter-rater agreement for risk of bias assessment reached a Cohen’s kappa of 1.00. The remaining 15 studies were judged to have some concerns overall, primarily due to the inherent limitation of exercise interventions precluding blinding of participants and personnel. Four studies were rated low risk for measurement bias, as they used assessor-blind designs or objective measurements such as polysomnography (Amara et al., 2020; Li et al., 2024; Mehta et al., 2024; Hu et al., 2025). One study had some concerns regarding randomization due to insufficient description of the randomization method (Abo-Elyazed, 2018). All studies demonstrated adequate management of missing outcome data and selective reporting. Importantly, none of the included RCTs were considered high risk for bias. The detailed risk of bias assessment for each RCT is presented in Figure 2.

ROB2 risk of bias plot.

3.3 Network meta-analysis

Figure 3 displays the network plot summarizing the available evidence on various exercise interventions for sleep quality in people with PD. In the network plot, the size of each node reflects the total number of participants assigned to the corresponding intervention, while the width of the connecting edges indicates the number of direct comparisons between interventions. As shown in the network, The most commonly evaluated interventions included MBE, RT, MME, and aerobic exercise (AE), based on the available direct comparison. Compared with CON, AE yielded an SMD of -0.94 (95% CI: -1.82 to -0.07) for sleep improvement. Detailed pairwise results are in Table 2. The global inconsistency test showed no significant difference between the consistency and inconsistency models (χ² = 4.70, p = 0.0956). Meta-regression analysis indicated that mean age did not significantly influence the treatment effect (p = 0.67), supporting the transitivity assumption (Table 3). However, node-splitting analysis revealed local inconsistencies in three comparisons: CON vs AE, CON vs MME, and AE vs MME (p < 0.05) (Table 4).

Network diagram.

AEMMERTSTMBECONAE—MME-0.15 (-1.19, 0.88)—RT-0.32 (-1.62, 0.98)-0.17 (-1.45, 1.12)—ST-0.42 (-1.91, 1.07)-0.26 (-1.73, 1.21)-0.10 (-1.65, 1.46)—MBE-0.44 (-1.50, 0.63)-0.29 (-1.32, 0.75)-0.12 (-1.27, 1.03)-0.02 (-1.23, 1.18)—CON-0.94 (-1.82, -0.07)*-0.79 (-1.63, 0.05)-0.62 (-1.60, 0.36)-0.53 (-1.73, 0.68)-0.51 (-1.11, 0.10)—

AE, Aerobic exercise; MME, multimodal exercise; RT, resistance training; ST, stretching training; MBE, mind-body exercise.

*Data are mean difference (95% CI). For CON vs. AE: -0.94 (-1.82, -0.07). Negative values favor AE; CI not crossing zero implies p < 0.05.

CovariateCoefficient95% CIP-valueτ²I²Mean age (per 1 year)-0.021(-0.118, 0.076)0.670.2356.2%Constant-0.94(-1.58, -0.30)0.004––

Network meta-regression results with mean age as a covariate.

SUCRA rankings after adjusting for age.

AE: 90.2%, MME: 72.0%, RT: 62.5%, MBE: 41.8%, ST: 37.6%, CON: 5.9%.

Meta-regression analysis indicated that mean age did not significantly influence the treatment effect (p = 0.67), supporting the transitivity assumption.

SideDirect coef.Std. err.Indirect coef.Std. err.Difference coef.Std. err.P>ztauA B1.36521.4523134-.6219843.81649341.987194.93312240.033*.667239A D-1.012347.7045362.9800926.616995-1.99244.93651170.033*.6673912B C-.4561601.330916-1.179151.225929.72299041.2698080.569.80281B D-.3815783.4206685-2.375126.83690711.993548