A comprehensive literature search was performed across five electronic databases: PubMed, Web of Science, Scopus, Embase, and the Cochrane Library, covering all records from database inception to May 1, 2025. The search strategy for PubMed was as follows: (“sleep” [Title/Abstract] OR “sleep quality” [Title/Abstract] OR “slumber” [Title/Abstract] OR “rest” [Title/Abstract]) AND (“exercise” [Title/Abstract] OR “physical exercise” [Title/Abstract] OR “exercise training” [Title/Abstract] OR “physical activity” [Title/Abstract] OR “workout” [Title/Abstract]) AND (“older adults” [Title/Abstract] OR “elderly” [Title/Abstract] OR “aged adults” [Title/Abstract] OR “seniors” [Title/Abstract] OR “middle-aged and older adults” [Title/Abstract]) AND (randomized controlled trial [Filter]).

The detailed search strategy for each database is presented in Supplementary Appendix A. The literature screening process was conducted independently and in duplicate by two reviewers using a blinded method. Any disagreements were resolved by consultation with a third reviewer.

Language: Only studies published in English were considered;

Review articles, conference abstracts, or case studies;

Two reviewers independently screened the literature based on pre-specified inclusion and exclusion criteria. Initially, titles and abstracts were reviewed to exclude studies irrelevant to the research topic. Full-text screening was subsequently conducted for potentially eligible studies to determine their final inclusion. Any discrepancies during the selection process were resolved through discussion with a third reviewer.

Data extraction was conducted concurrently with the full-text review. The extracted information included the first author, year of publication, study location, study design, sample size, participant age, type of exercise, exercise intensity, frequency, duration per session, intervention period, and the primary outcome measure (Pittsburgh Sleep Quality Index, PSQI). In all included studies, the PSQI was assessed both at baseline and post-intervention to evaluate changes in subjective sleep quality. For the purpose of this meta-analysis, subjective sleep quality was operationalized as the global PSQI score, unless otherwise specified. Only the global score was used for analysis; individual PSQI components (e.g., total sleep time, sleep latency) were not examined separately due to inconsistent reporting across studies.

For multiple publications originating from the same trial, only one version was included to avoid duplicate data. The publication that best matched the research objectives and provided the most complete data—typically the most recent version—was prioritized, while others were excluded. To minimize subjectivity in the classification of exercise variables, the following criteria were applied:

The risk of bias and methodological quality of included RCTs were assessed using the Cochrane Risk of Bias (RoB) tool. This tool evaluates seven domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective reporting, and other potential sources of bias.

Each domain was rated as “low risk,” “unclear risk,” or “high risk” of bias. The assessment was independently conducted by two reviewers based on the original reports. In cases of disagreement, a third reviewer participated in discussion to reach consensus, thereby ensuring objectivity and consistency in the evaluation process.

Meta-analyses were performed using R software. Subgroup analyses were conducted to examine the effects of different intervention characteristics. Heterogeneity was assessed using Cochran’s Q test, with a significance level set at α = 0.1. The degree of heterogeneity was quantified using the I² statistic: I² < 50% and p > 0.1 were interpreted as low heterogeneity, in which case a fixed-effects model was applied; I² ≥ 50% and p < 0.1 indicated substantial heterogeneity, warranting the use of a random-effects model.

The primary outcome was treated as a continuous variable. For studies employing the same sleep quality scale, weighted mean differences (WMD) with 95% confidence intervals (CI) were calculated. For studies using different scales, standardized mean differences (SMD) with 95% CI were computed and used in sensitivity analyses.

To explore the relationship between intervention dose and sleep improvement, a nonlinear dose–response meta-analysis was conducted. A random-effects one-stage mixed-effects model was employed to simultaneously estimate the dose–response relationship across all studies. Restricted cubic spline models and maximum likelihood estimation were applied at fixed percentiles (5, 50, and 95%) to model the relationship between exercise dose (expressed in MET-minutes/week) and improvements in subjective sleep quality, without assuming a predefined functional form. Exercise dose was calculated as the product of session duration, frequency, and intensity. A reference dose of zero was used for estimating relative effect sizes under varying exercise exposure conditions.

A systematic search of five databases (PubMed, Web of Science, Cochrane Library, Scopus, and Embase) yielded a total of 1,899 relevant records. After removing 642 duplicates, 1,257 studies remained for title and abstract screening. During this stage, 112 articles were excluded, including 6 non-English publications, 74 non-randomized controlled trials, and 32 with incomplete data. A total of 1,145 articles were retrieved for full-text review, of which 56 could not be obtained. Among the 1,089 reports assessed for eligibility, 1,046 were excluded for the following reasons: outcome indicators not involving the Pittsburgh Sleep Quality Index (PSQI; n = 267), study participants not meeting inclusion criteria (n = 340), inappropriate form of intervention in the control group (n = 95), disease type not in line with inclusion criteria (n = 317), meta-analysis articles (n = 23), and animal experiments (n = 4).

Following full-text assessment, 26 randomized controlled trials were identified as eligible and included in the systematic review and meta-analysis (17–42). The study selection process is illustrated in Figure 1. In this study, “subjective sleep quality” was consistently defined as the global PSQI score, unless otherwise specified. Only the global score of the PSQI was used for the meta-analysis, and individual components such as total sleep time or sleep latency were not analyzed separately due to inconsistent reporting across studies.

A total of 26 randomized controlled trials (17–42) involving a total of 2,189 older adults aged between 60 and 90 years were included in the analysis.

In terms of exercise type, several studies employed multicomponent programs integrating aerobic, resistance, flexibility, and balance training (19, 22–24, 30, 37, 41), while the remaining trials adopted single-modality interventions, such as: Tai Chi: (20, 26, 29); Yoga, Pilates, or stretching: (17, 34); Aerobic walking or dance: (21, 27, 35, 36); Progressive resistance training: (31, 38).

With respect to exercise intensity, most multicomponent programs and aerobic walking interventions were conducted at a moderate intensity; Tai Chi, yoga, and stretching were generally considered low-intensity mind–body exercises; whereas Singh et al. (38) and Karimi et al. (31) implemented high-intensity progressive resistance training protocols.

Regarding intervention duration, the included studies ranged from 6 weeks to 12 months, with the majority lasting 12 to 24 weeks. Training frequency was typically set at 2–5 sessions per week across studies.

Overall, the included trials encompassed exercise modes ranging from low to high intensity and intervention durations from short to long term. Detailed information on exercise types, intensity, frequency, and duration is presented in Table 1.

All 26 included studies (17–42) were evaluated using the Cochrane Risk of Bias assessment tool.

The results indicated that most studies exhibited a low risk of bias in domains such as random sequence generation, blinding of outcome assessors, and handling of incomplete outcome data, suggesting a generally high methodological quality. However, some studies provided unclear or insufficient descriptions regarding the blinding of participants and personnel, indicating a potential risk of performance bias in these domains. Detailed results of the quality assessment are presented in Figure 2.

The funnel plot demonstrated an approximately symmetrical distribution of studies around the pooled effect estimate and showed convergence from the bottom to the top, suggesting no substantial evidence of publication bias in this meta-analysis (Figure 4).

The overall dose–response curve between exercise dose and sleep improvement in older adults exhibited an inverted U-shaped pattern. In the low-dose range (approximately 200–500 MET-min/week), the intervention effect increased progressively with the amount of exercise. The greatest effect was observed at approximately 527 MET-min/week, representing the lowest point of the curve (Hedges’ g = −0.82), with a predicted effect estimate of −0.82 (95% CI: −1.12, −0.52). As exercise volume continued to increase beyond this point, the effect began to attenuate slightly, as indicated by the upward slope of the curve (Figure 5).

A subgroup analysis based on exercise intensity was conducted across the 26 included studies (17–42).

The subgroup with low-intensity interventions demonstrated the greatest effect size [WMD = −2.79, 95% CI (−3.44, −2.14)] with moderate heterogeneity (I² = 65.3%). The moderate-intensity group yielded a similar effect size [WMD = −2.77, 95% CI (−3.30, −2.24)] but showed high heterogeneity (I² = 83.6%). The low-to-moderate intensity group reported a WMD of −2.20 (95% CI: −2.61, −1.78) with moderate heterogeneity (I² = 52.8%). The high-intensity group showed a WMD of −2.75 (95% CI: −3.55, −1.95) with no observed heterogeneity (I² = 0.0%).

By contrast, the moderate-to-high intensity group demonstrated the weakest intervention effect [WMD = −0.86, 95% CI (−1.55, −0.17)]. The differences among subgroups were statistically significant (p = 0.000), indicating that intervention intensity plays a critical role in the effectiveness of sleep improvement (Figure 7).

Subgroup analysis based on intervention frequency was performed on the same 26 studies (17–42).

The group exercising twice per week showed the greatest improvement [WMD = −2.52, 95% CI (−3.00, −2.04)]. The group exercising three times per week had a slightly smaller effect size [WMD = −2.28, 95% CI (−2.60, −1.96)]. The group exercising more than three times per week demonstrated the weakest effect [WMD = −1.69, 95% CI (−2.55, −0.83)].

Although effect sizes differed among subgroups, the differences were not statistically significant (p = 0.252), suggesting that intervention frequency may not be a key determinant of intervention efficacy (Figure 8).

The 26 included studies (17–42) were categorized into three groups based on total intervention duration: 0–8 weeks, 8–16 weeks, and >16 weeks.

The short-term intervention group (0–8 weeks) showed the most pronounced improvement in sleep quality [WMD = −2.45, 95% CI (−2.99, −1.91)], with no heterogeneity (I² = 0.0%), suggesting highly consistent results. The moderate-term (8–16 weeks) and long-term (>16 weeks) groups yielded effect sizes of −2.32 and −2.04, respectively, but both exhibited high heterogeneity (I² = 77.4 and 82.1%, respectively). Although all three subgroups demonstrated significant improvements in sleep outcomes, the between-group difference was not statistically significant (p = 0.578), indicating that total intervention duration may not be a critical factor influencing sleep improvement (Figure 9).

A subgroup analysis based on exercise modality was performed using standardized mean differences (SMD) across the same 26 studies. Overall, exercise interventions produced a moderate improvement in sleep among older adults [SMD = −0.79, 95% CI (−0.96, −0.63), p < 0.0001], with moderate heterogeneity (I² = 60.9%).

Among intervention types, dance-based programs showed the strongest effect (SMD = −1.59), albeit with substantial heterogeneity (I² = 73.5%). Aerobic exercise also yielded a large effect (SMD = −1.03) with high heterogeneity (I² = 80.7%). In contrast, yoga and tai chi interventions demonstrated stable and consistent effects, with effect sizes of −0.54 and −0.78, respectively, and no heterogeneity (I² = 0.0%). Resistance training showed a modest effect (SMD = −0.65), but the confidence interval crossed zero, rendering the result statistically non-significant and heterogeneity relatively high (I² = 73.5%).

Although the differences among subgroups approached statistical significance (p = 0.0794), the findings suggest that dance, aerobic exercise, and tai chi are particularly effective and replicable modalities for improving sleep quality, with dance-based interventions demonstrating the greatest potential benefit (Figure 10). It should be noted that SMD values were applied solely to the global PSQI score to harmonize differences in measurement scales across studies; no other sleep outcome measures were synthesized in this analysis.

The meta-regression analysis revealed no statistically significant linear relationships between intervention effect and any of the following variables: session duration (p = 0.933), exercise intensity (p = 0.867), intervention frequency (p = 0.849), or total intervention duration (p = 0.452; Figure 11).

Grounded in the 2024 Guidelines for Elderly Health, this study investigated the dose–response relationship between physical exercise and improvements in sleep among older adults using metabolic equivalents (METs). The results provide an evidence-based foundation for designing personalized sleep intervention prescriptions. The findings demonstrated that exercise interventions significantly improved subjective sleep quality in older adults. Among various intervention types, dance-based programs, aerobic exercise, and Tai Chi were particularly effective, demonstrating large and consistent improvements in subjective sleep quality. Although this study shows that the subgroup with low-intensity intervention had the greatest effect, it also indicates that moderate-intensity exercise yields significant benefits. This result is in line with the World Health Organization (WHO) guidelines of 2020 (43). Supporting literature indicates that 150–300 min of moderate-intensity exercise per week is sufficient to substantially enhance subjective sleep quality in the elderly (13).

The nonlinear dose–response model revealed an inverted U-shaped relationship, with the maximal intervention effect occurring at approximately 527 MET-min/week. Beyond this threshold, the benefit plateaued, indicating that moderate exercise doses are sufficient to produce substantial improvements in sleep quality.

Importantly, the findings further suggested that even lower exercise doses—approximately 300 MET-min/week—could yield significant improvements in subjective sleep quality, despite falling below the WHO’s recommendation of 600 MET-min/week. This implies that near-optimal effects can be achieved within a critical dosage range, minimizing unnecessary exercise burden. Such an approach is particularly well-suited for older adults with lower physical capacity or limited exercise tolerance. Zang et al. (15) also reported that just 300 MET-min/week of low-to-moderate intensity exercise could significantly alleviate frailty symptoms in older adults, with optimal benefits observed within the 600–1,200 MET-min/week range. This underscores that moderate exercise loads are both sustainable and adaptable, offering high safety and better long-term adherence among older populations—leading to consistent health outcomes (15).

Subgroup analyses revealed that intervention frequency and duration did not significantly influence the effectiveness of sleep improvement, suggesting these parameters may be less critical. However, trends were observed regarding exercise intensity and session duration: lower-intensity exercise and sessions lasting less than 30 min tended to produce more stable and sustainable improvements. These findings align with previous studies by Reid et al. (44), and Tseng et al. (45), which emphasized that moderate-intensity activities are more manageable for older adults and are associated with better compliance and outcomes. Reid et al. (44) demonstrated that moderate aerobic exercise improved insomnia symptoms in the elderly more effectively than high-intensity activity, which, due to the associated physiological demands (e.g., elevated heart rate, breathlessness), may reduce adherence.

It is particularly noteworthy that although high-intensity, long-duration exercise has shown potential benefits in some trials, older adults are more vulnerable to fatigue and fall-related injuries under such regimens. This has been explicitly noted by You et al. (46) and Crane et al. (47). Therefore, low-to-moderate intensity exercise with moderate session duration appears to be the most suitable approach for promoting sustained engagement and improving subjective sleep quality among older adults. Such interventions are not only safer but also better aligned with the physiological characteristics and health needs of this population.

Intervention strategies should prioritize safety while encouraging participation in enjoyable, gentle, and varied low-to-moderate intensity activities to enhance adherence and achieve long-term improvements in sleep and overall health. Flexibility in intervention design is particularly important for older adults, as it reduces fatigue, minimizes the risk of exercise-related injuries, and supports sustainable participation. Given the high prevalence of sleep disorders among elderly populations, rehabilitation specialists, exercise prescription developers, and researchers should develop personalized, moderate-intensity, and easy-to-follow programs—guided by the 2024 Guidelines for Elderly Health—to improve sleep quality and enhance overall well-being in a safe and adherent manner.

This study highlights the critical role of exercise interventions in improving subjective sleep quality among older adults, demonstrating that even low-dose exercise programs can yield significant benefits. In clinical practice, many older individuals are unable to meet the World Health Organization’s recommended levels of physical activity due to physical limitations or mobility impairments. Against this backdrop, low-dose and low-intensity exercise interventions are especially important. Such approaches not only promote gradual adaptation to physical activity but also reduce the risk of excessive fatigue or falls. A progressive, stepwise intervention model can enhance physical function while ensuring safety, thereby improving both sleep quality and overall well-being in older adults (48).

Importantly, beyond statistical significance, the magnitude of improvement observed in the present meta-analysis also demonstrated clinical relevance. The weighted mean differences (WMD) for PSQI exceeded the minimal clinically important difference (MCID), which is generally defined as a reduction of approximately 3 points. This indicates that the improvements in sleep quality observed here are not only statistically robust but also meaningful from a clinical perspective, with tangible benefits for older adults experiencing sleep difficulties.

The findings further reinforce the importance of tailoring exercise prescriptions to individual needs. Intervention programs should be designed based on the elderly individual’s health status, preferences, and living environment (13). This person-centered approach supports the evolution of sleep improvement strategies toward greater precision and inclusivity across diverse populations (49).

Although evidence-based medicine has advanced significantly in the field of exercise and aging, substantial research gaps remain in understanding the mechanisms, optimal intensity thresholds, and intervention modalities for sleep disorders. Future studies are needed to explore the adaptability and differential effects of various exercise types, intensities, and frequencies across subpopulations with varying covariates. Dose–response analysis offers valuable insights for optimizing intervention strategies and enhancing their effectiveness, and this study seeks to provide a methodological reference for clinical practitioners and rehabilitation researchers in the field of sleep health.

This study systematically evaluated the relationship between exercise dose and improvements in subjective sleep quality among older adults, and proposed a theoretical framework for personalized exercise prescriptions. The research process adhered strictly to established meta-analysis guidelines, with comprehensive literature searches conducted across multiple authoritative databases to ensure scientific rigor and representativeness. Through standardized screening procedures and stringent quality control measures, the risk of selection bias was minimized, enhancing the credibility and robustness of the conclusions.

The findings offer a practical pathway for older adults with limited physical capacity and provide evidence-based support for developing more accessible intervention strategies for sleep disorders. Moreover, the study lays a theoretical foundation for future individualized and precision-oriented rehabilitation programs, with significant potential for clinical translation and public health impact.

Nevertheless, several limitations should be acknowledged. First, the inclusion of English-language publications only may introduce language bias and limit the global generalizability of the findings. Second, although subgroup analyses were conducted, the number of studies included in some subgroups was small, potentially reducing statistical power and result stability. Third, most of the included studies lacked long-term follow-up data post-intervention, restricting our ability to assess the sustainability and long-term efficacy of exercise interventions. Additionally, many original studies failed to report adherence to the exercise protocol and intervention fidelity, which may affect the accuracy of dose–response estimations.

Future research should aim to improve study quality and reporting completeness, extend follow-up durations to evaluate long-term effects, and incorporate covariate-adjusted subgroup analyses and mechanistic investigations. Such efforts are essential for strengthening the scientific basis and practical guidance of exercise interventions tailored to older populations.