Comparative efficacy of various exercise modalities on sleep quality in breast cancer survivors after primary treatment: A systematic review and dose-response network meta-analysis

Abstract

Background:

This study aims to assess the dose-response relationship between different exercise modalities and sleep quality in breast cancer survivors who have completed their primary treatment, and to propose evidence-based exercise prescriptions for clinical practice.

Methods:

Bayesian network and dose-response meta-analyses using a random-effects model were conducted to evaluate the impact of exercise on sleep quality in breast cancer survivors. Exercise dose was calculated as the product of duration, frequency, and intensity, expressed in metabolic equivalent of task minutes per week (METs-min/week), for comparison across studies.

Results:

A total of 24 studies involving 1495 patients were included. An inverted U-shaped, nonlinear dose-response relationship between exercise and sleep quality was observed, with an effective dose range of 0 to 1100 METs-min/week and maximum effect at 780 METs-min/week (standardized mean differences = 0.7, 95% CrI: 0.38, 1.06). Stratified by modality, aerobic exercise, combined aerobic-resistance training, and mind-body exercise all showed significant associations with improved sleep quality. The comprehensive ranking indicated that 500 METs-min/week of mind-body exercise (standardized mean differences = 1.06, 95% CrI: 0.3, 1.83) may be the most effective modality and dose for enhancing sleep quality.

Conclusions:

Based on the results of the dose-response network meta-analysis, we developed 9 exercise prescriptions to effectively improve sleep quality in individuals with breast cancer. Mind-body exercise may provide the most significant benefits, while aerobic and combined aerobic-resistance training are supported by more robust evidence due to larger sample sizes. However, the evidence supporting these exercise modalities is of low evidence, highlighting the need for further high-quality studies to validate the recommended doses.

1. Introduction

In 2023, breast cancer had the highest incidence of all cancers, with 2.3 million new cases worldwide.^[1] Although survival rates have improved in recent years, many survivors continue to suffer from side effects that significantly impair their quality of life.^[2–4] Poor sleep quality is one of the most frequently reported issues, and it has been associated with morbidity,^[5] mortality,^[6] fatigue,^[7] pain,^[8] and cancer-related physical distress in cancer survivors.^[9] Meta-analyses found that the global prevalence of poor sleep quality among breast cancer survivors was 62% in 2023.^[10] These sleep issues often result from the psychological and physiological effects of cancer and its long-term treatments, which can worsen inflammation and discomfort by impacting the stress response system.^[11] While pharmacological treatments remain a primary option, their uncertain efficacy and potential side effects have driven the search for safer, alternative therapies to improve survivors’ quality of life and prognosis.

Exercise offers multiple health benefits and is an effective method to improve quality of life and alleviate stress in breast cancer survivors.^[12] Recent meta-analyses have demonstrated the effectiveness of exercise in improving sleep quality in breast cancer survivors.^[13,14] Two network meta-analyses identified the most effective exercise modalities for improving sleep.^[15,16] However, these studies did not provide the effective dose range or optimal dosage for exercise. Exercise dosage (i.e., intensity, duration, and frequency) is a crucial factor in prescribing exercise. Given the vulnerability of breast cancer survivors, imprecise or inappropriate doses may not only be ineffective but could worsen sleep problems. Therefore, identifying the effective and optimal doses for different exercise types and developing tailored prescriptions is essential for clinical practice. Additionally, these studies evaluated individuals both undergoing primary treatments (e.g., chemotherapy, radiotherapy) and those who had completed such treatments, without stratifying between the 2 groups.^[15,16] Due to the severity of treatment side effects, those receiving primary treatments may experience more severe sleep disturbances. Differences in baseline sleep quality could influence the effectiveness of exercise and increase clinical heterogeneity. Furthermore, considering the physical frailty and limited exercise capacity of some patients during primary treatment, our study focuses exclusively on survivors who have completed treatment. This allows for more precise and clinically relevant recommendations for decision-making.

This study will employ a novel Bayesian-based dose-response network meta-analysis of randomized controlled trials (RCTs) to identify the most effective exercise modalities and doses for improving sleep quality in breast cancer survivors who have completed primary treatment.^[17–19] It offers 2 key advantages: the ability to determine both effective and optimal doses for various exercise modalities, and unlike traditional network meta-analysis, which typically identifies the best intervention, this approach provides a comprehensive ranking of doses across modalities, allowing for more precise clinical prioritization. The findings will inform clinical guidelines, support sleep-focused exercise prescriptions, and help shape the design of future exercise protocols in research.

2. Methods

2.1. Protocol and registration

This systematic review and network meta-analysis was prospectively registered in PROSPERO (CRD420251171697) and conducted following the PRISMA-NMA reporting standards.^[20] As this study was based solely on previously published data, ethical approval and informed consent were not required.

2.2. Search strategy

We systematically searched PubMed, Web of Science, Embase, and the Cochrane Central Register of Controlled Trials from database inception to September 1, 2025, without language restrictions. Search terms included keywords and MeSH headings covering participants (Breast Cancer), interventions (Exercise), outcomes (Sleep), and study design (Randomized Controlled Trials), combined with the operator AND. We also reviewed the reference lists of relevant reviews to identify any studies that were not included. A full description of the search strategies for each database is provided in Supplementary Table S1, Supplemental Digital Content 1.

2.3. Eligibility criteria

Selection criteria were defined according to the PICOS framework (Population, Intervention, Comparison, Outcomes, Study Design) to reduce potential selection bias.^[21]

1. Population: adult women diagnosed with breast cancer who had completed primary treatment (surgery, chemotherapy, or radiotherapy). Studies mixing breast cancer with other malignancies were excluded unless data for breast cancer survivors were separately reported.

2. Intervention: any structured or repeated physical activity designed to maintain or improve fitness, excluding relaxation-only programs. Studies combining exercise with other interventions (e.g., nutritional supplements) were excluded, unless the control group received the same non-exercise components, allowing for an accurate assessment of the exercise effects.

3. Comparison: studies were required to include a control group (e.g., usual care). For head-to-head comparisons, the comparator could be another exercise program.

4. Outcome: sleep outcomes were measured using validated instruments, such as the Pittsburgh Sleep Quality Index.

5. Study design: only randomized controlled trials were eligible. Quasi-experimental, observational, and non-randomized studies were excluded.

6. Other criteria: duplicate publications, reviews, abstracts, letters, and dissertations were excluded. When multiple papers reported the same trial, the publication with the largest sample size was retained.

2.4. Data extraction and coding

Two independent reviewers (LHL and DSW) extracted data using a standardized Excel template, capturing study details (e.g., authors, publication year, study region), participant characteristics (e.g., age, sample size), intervention specifics (e.g., exercise type, frequency, duration, length), control interventions (e.g., type, sample size), and outcomes (e.g., measurement tools, key findings). Means and standard deviations (SDs) for sleep quality were recorded for both exercise and control groups at baseline and post-intervention. To calculate SDs of change scores between pre- and post-intervention, we applied a correlation coefficient of 0.5, reflecting standard measurement reliability. Missing SDs were derived from confidence intervals (CIs), P-values, or t-values. If this information was unavailable, the corresponding authors were contacted. Given the need for standard errors in dose-response analysis, SDs were converted to SEs using the formula:

$${\displaystyle SE=\frac{SD}{\sqrt{N}}.}$$

where N represents the sample size.

Data accuracy was cross-checked by the reviewers, with a third author (CYM) validating the final dataset.

For the dose-response network meta-analysis, interventions were classified into 3 levels. Initially, they were categorized as either “Exercise” or “Control.” At the second level, exercise interventions were grouped according to type: “Aerobic Exercise (AE)” (e.g., running, cycling, walking; AE), Mind-body Exercises (e.g., tai chi, yoga, pilates), and “combined aerobic and resistance training (AE-RT)” (e.g, running combined with resistance band exercises; AE-RT). At the third level, interventions were further categorized based on the total weekly exercise dose, calculated by multiplying duration, frequency, and intensity, expressed in metabolic equivalent of task minutes per week (METs-min/week). Warm-up and cool-down periods were excluded from the total exercise time. If exercise duration increased progressively over time, the average total exercise time was used. Each intervention was then assigned to the closest predefined dose category based on its estimated METs-min/week. The interventions were grouped into 7 categories: 0 (control group), 250, 500, 750, 1000, and 1250 METs-min/week. This distribution ensured a balanced representation of exercise intensities, facilitating effective comparison while maintaining adequate network connectivity for the meta-analysis.

For these calculations, we used the 2024 standardized physical activity intensity tables.^[22] For example, yoga for adults is assigned 2.3 METs-min. If participants engage in 60 minutes of yoga 3 times per week, the total exercise intensity is calculated as 2.3 (intensity) × 60 (minutes) × 3 (sessions), resulting in 414 METs-min/week. This value was then rounded to the nearest predefined dose category (e.g., 500 METs-min/week).

2.5. Statistical analysis

2.5.1. Pairwise meta-analysis

We first conducted pairwise meta-analyses to evaluate the overall effect of exercise compared with non-exercise control conditions. Pairwise meta-analyses were conducted using STATA (version 16).

Given the use of various outcome measurement scales, standardized mean differences (SMD) were calculated based on the change from baseline to endpoint for both groups. Pairwise meta-analyses were performed using the inverse-variance method to pool effect sizes and corresponding 95% CIs. Following Cochrane recommendations, Hedges’ g was adopted to correct for potential small-sample bias, thereby providing more reliable estimates.^[23] Generally, an SMD of <0.2 was considered negligible, 0.2 to 0.5 small, 0.5 to 0.8 moderate, and ≥0.8 large. Heterogeneity was assessed using the I² statistic, categorized as low (25 to <50%), moderate (50 to <75%), or high (≥75%).^[24] If I² exceeded 25%, a random-effects model was applied; otherwise, a fixed-effect model was used. Statistical significance was assessed using 95% CIs, with CIs including 0 indicating non-significance for continuous outcomes. We further conducted subgroup analyses based on different exercise modalities.

2.5.2. Dose-response analysis

The dose-response analysis, examining the relationship between total exercise dose and various exercise types, was conducted using the MBNMAdose package in R (version 4.4.3, www.r-project.org). Dose-response curves were plotted using the ggplot2 package. To minimize bias and allow the data to guide results, we applied vague (non-informative) prior distributions in line with the default settings of MBNMAdose. Given the diversity of outcome measurement scales, SMDs were calculated based on changes from baseline to endpoint for both groups.

The analysis began by assessing key network meta-analysis assumptions, such as network connectivity, data consistency, and transitivity.^[25–27] We then evaluated several models to represent dose-response relationships, including emax, restricted cubic spline, log-linear, nonparametric, exponential, quadratic function, spline, and fractional polynomial models.^[28] The model’s goodness of fit was evaluated using criteria like the deviance information criterion (DIC), SD of residuals, model complexity, and residual deviation. Consistent with previous studies, we selected the model with the lowest DIC.^[29] Given the expected heterogeneity, a random-effects model was employed to account for potential study variability. The beta coefficient from this model was used to identify the effective and optimal exercise doses that significantly improve sleep quality and to predict the ranking of exercise modalities and doses based on their likelihood of producing a beneficial effect. Statistical significance was indicated by a 95% credible interval (CrI) that excludes zero in the Bayesian dose-response analysis. Lastly, we developed specific exercise prescriptions based on the identified effective dose range, referencing standardized physical activity intensity classification tables for adults.^[22]

2.5.3. Additional analyses

A sensitivity analysis was conducted by excluding studies with a high risk of bias to examine result variability. Funnel plots, along with Begg’s and Egger’s tests, were employed to assess publication bias and small-study effects.^[30,31] Symmetry in the funnel plots and P-values ≥.05 indicated no significant bias.

2.6. Risk of bias and certainty of evidence

Risk of bias was assessed using the Cochrane Risk of Bias tool (RoB 2), which covers 5 domains: randomization, deviations from interventions, missing outcome data, outcome measurement, and selective reporting. Studies were categorized as low risk, some concerns, or high risk.^[32] Two authors (LHL, DSW) independently carried out the assessment, with discrepancies resolved by a third author (CYM).

To evaluate the confidence in results, we used the CINeMA web application, which takes into account factors such as within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and inconsistency.^[33] Evidence confidence was rated as high, moderate, low, or very low.

3. Results

3.1. Study selection

A total of 4549 records were retrieved from 4 databases. After removing 936 duplicates, 3613 records remained for title and abstract screening, of which 3387 were excluded as irrelevant. The remaining 226 articles underwent full-text screening, leading to the exclusion of 205 studies based on eligibility criteria. As a result, 21 studies were included in the systematic review. Additionally, 3 studies were identified through related review articles, bringing the total to 24 studies included in the final analysis. Figure 1 illustrates the screening and selection process.

Figure 1.

Screening and selection process of studies for the systematic review. RCT = randomized controlled trials.

3.2. Study characteristics

A total of 1495 breast cancer survivors participated, with 750 (49.5%) in the exercise group and 755 (50.5%) in the usual care or waiting list group. The participants’ mean age ranged from 42.7 to 66.6 years (average: 54.7 years). Interventions included 6 modalities: AE (10 arms), AE-RT (9 arms), and mind body exercise (MBE) (6 arms), with durations ranging from 6 to 12 weeks. Participants’ cancer stages ranged from Stage 0 to Stage III, and 70.8% were from non-Asian regions. The Pittsburgh Sleep Quality Index was the most commonly used assessment tool (75% of studies). Detailed study characteristics are in Supplementary Table S2, Supplemental Digital Content 2.

3.3. Risk of bias

We assessed the quality of studies using the following criteria. Studies that did not specify the randomization method were rated as having “some concerns.” Since blinding participants is not feasible in exercise interventions, this domain was generally rated with caution. However, studies that reported blinding of outcome assessors or allocation personnel had a reduced risk of bias. For studies with potential missing outcome data, we rated the risk as “some concerns.” However, studies that appropriately applied statistical methods to handle missing data were rated as low risk. Most studies were prospectively registered and reported results consistent with their protocols, leading to a low risk of selection bias. Studies without preregistration were rated as “some concerns.” If a study was rated as “some concerns” or “high risk” in any domain, the overall risk was classified accordingly. In total, 7 studies were rated as having “some concerns,” 3 as “high risk,” and the remainder as “low risk.” A detailed risk of bias figure is provided in Supplementary Figure S1, Supplemental Digital Content 3.

3.4. Pairwise meta-analysis

The results of the pairwise meta-analysis are presented in Figure 2. Compared with the control group, exercise interventions were associated with significant improvements in sleep quality (SMD = 0.43, 95% CI: 0.24, 0.62; I² = 63.9%). When stratified by exercise modality, AE (SMD = 0.47, 95% CI: 0.11, 0.84; I² = 72.1%), AE-RT (SMD = 0.36, 95% CI: 0.1, 0.62; I² = 39.4%), and MBE (SMD = 0.49, 95% CI: 0.03, 0.94; I² = 76.7%) were all significantly associated with improvements in sleep quality. Overall, the results indicated small effect sizes, with small to high heterogeneity.

Figure 2.

Forest plot for the pairwise meta-analysis of exercise effects on sleep quality in breast cancer survivors. AE = aerobic exercise, AE-RT = combined aerobic and resistance training, MBE = mind-body exercise.

3.5. Dose-response meta-analysis

3.5.1. Hypothesis testing and model selection

No evidence was found to suggest that the dose-response network meta-analysis violated the assumptions of connectivity (Fig. 3), consistency (Supplementary Table S3, Supplemental Digital Content 4), or transitivity (Supplementary Table S4, Supplemental Digital Content 5). The quadratic function model, with the lowest DIC, was ultimately selected as the best-fitting approach (Supplementary Table S5, Supplemental Digital Content 6).

Figure 3.

Connectivity of the dose-level network. The first value represents the specific intervention, and the second value corresponds to the dose (METs-min/week) of that intervention. Line thickness represents the number of studies available for comparison. AE = aerobic exercise, AE-RT = combined aerobic and resistance training, MBE = mind-body exercise.

3.5.2. Dose-response relationship of overall exercise

Our analysis revealed an inverted U-shaped, nonlinear dose-response relationship between overall exercise dose and sleep quality in breast cancer survivors. No minimum exercise dose was found to significantly improve sleep quality. As the exercise dose increased, the effects improved, peaking at 780 METs-min/week (SMD = 0.7, 95% CrI: 0.38, 1.06), after which the effects gradually declined. Beyond 1100 METs-min/week, exercise ceased to be effective (Fig. 4).

Figure 4.

Dose-response relationship between total exercise (METs-min/week) and sleep quality in breast cancer survivors. The blue dashed lines represent the minimum and maximum effective doses (1100 METs-min/week), the purple shaded area indicates the significant region, and the red dots represent the optimal dose points (780 METs-min/week). METs-min/week = metabolic equivalent of task minutes per week.

In the sensitivity analysis, after excluding studies with a high risk of bias, the effective exercise dose range (0–1000 METs-min/week) was reduced, and the optimal dose dropped to 650 METs-min/week (SMD = 0.75, 95% CrI: 0.37, 1.16). Detailed results of the sensitivity analysis are provided in Supplementary Figure S2, Supplemental Digital Content 7.

3.5.3. Dose-response relationship of different exercise modalities

Figure 5 shows the dose-response relationship for the 3 exercise modalities. Specifically, for AE, the effective dose range was 0–1000 METs-min/week, with optimal effects at 730 METs-min/week (SMD = 0.76, 95% CrI: 0.29, 1.27). For AE-RT, the effective dose ranged from 910 to 1200 METs-min/week, with the optimal dose at 1200 METs-min/week (SMD = 0.8, 95% CrI: 0.07, 1.58). For MBE, the effective dose ranged from 360 to 500 METs-min/week, with the optimal dose at 500 METs-min/week (SMD = 1.06, 95% CrI: 0.3, 1.83). The overall predicted ranking of exercise modalities and doses indicated that MBE at 500 METs-min/week was the most effective (Supplementary Table S6, Supplemental Digital Content 8).

Figure 5.

Dose-response relationship between different exercise modalities (METs-min/week) and sleep quality in breast cancer survivors. The blue dashed lines represent the minimum and maximum effective doses, the purple shaded area indicates the significant region, and the red dots represent the optimal dose points. AE = aerobic exercise, AE-RT = combined aerobic and resistance training, MBE = mind-body exercise, METs-min/week = metabolic equivalent of task minutes per week.

3.6. Publication bias and certainty of evidence

Supplementary Figure S3, Supplemental Digital Content 9 presents a detailed analysis of publication bias. The funnel plot showed approximate symmetry, and both Egger’s and Begg’s tests did not indicate significant publication bias (P > .05), suggesting no evidence of substantial publication bias.

The CINeMA evidence rating revealed that imprecision, within-study bias, and heterogeneity across the 3 exercise modalities led to a low certainty of evidence (Supplementary Table S7, Supplemental Digital Content 10).

3.7. Evidence-based exercise prescription

Table 1 presents the effective and optimal weekly exercise accumulation required to improve sleep quality in breast cancer survivors across 9 exercise modalities. For AE-RT interventions, the total prescribed METs-min/week were evenly divided between the aerobic and resistance components. Recommended dose ranges were identified for AE (cycling, walking, and jogging), AE-RT (cycling, walking, and jogging each combined with resistance training), and MBE (yoga, Pilates, and Tai Chi).

Table 1.

Evidence-based optimized exercise prescription.

Exercise modality	Energy expenditure (METs-min)*	Recommended accumulation (min/wk)	Optimal recommended accumulation (min/wk)	Certainty of evidence (grade)
Cycling	7 (code: 01016)†	~140	100	Low
Jogging	4.8 (code: 12025)‡	~210	150	Low
Walking	3.5 (code: 17160)§	~290	210	Low
Cycling combined with resistance training	7 (code: 01016)† 3.5 (code:02054)‖	Cycling: 65~85 Resistance training: 130–170	Cycling: 85 Resistance training: 170	Low
Jogging combined with resistance training	4.8 (code: 12025)‡ 3.5 (code: 02054)‖	Jogging: 95–125 Resistance training: 130–170	Jogging: 125 Resistance training: 170	Low
Walking combined with resistance training	3.5 (code:17160)§ 3.5 (code: 02054)‖	Walking: 130~170 Resistance training: 130~170	Walking: 170 Resistance training: 170	Low
Tai Chi	3.3 (code: 15670)	110–150	150	Low
Yoga	2.3 (code: 02150)	155–220	220	Low
Pilates	2.8 (code: 02105)§	130–180	180	Low

Weekly exercise accumulation was calculated by dividing the prescribed METs-min/week by the estimated energy expenditure of each exercise modality and rounding to the nearest whole minute.

^*Exercise intensity was coded based on the 2024 standardized physical activity intensity tables.

^†Bicycling, self-selected moderate pace.

^‡Jogging, in place.

^§Walking for pleasure.

^‖Resistance (weight) training, multiple exercises, 8 to 15 repetitions at varied resistance; 3 to 6 METs represents moderate-intensity activity, and >6 METs represents high-intensity activity, serving as a reference for absolute exercise intensity. Exercise prescription recommendations are based on the currently available evidence and should be interpreted as exploratory and hypothesis-generating; they may be refined as additional evidence emerges.

4. Discussion

4.1. Summary of overall findings

This study systematically evaluated the dose-response relationship between different exercise modalities and sleep quality in breast cancer survivors who had completed primary treatment. Furthermore, additional pairwise meta-analyses were conducted to provide complementary evidence and further strengthen the robustness of the findings. The findings revealed an inverted U-shaped dose-response relationship. Moderate exercise may promote sleep by increasing energy expenditure and the body’s need for recovery. However, excessively high exercise doses may exceed an individual’s recovery capacity,^[34] particularly in breast cancer survivors who often experience treatment-related fatigue, thereby reducing the beneficial effects of exercise on sleep quality. Previous observational studies have also reported an inverted U-shaped association between exercise and sleep, and our findings further support this dose-response pattern.^[35] We found that 1100 METs-min/week was the threshold, beyond which exercise ceased to be effective. This threshold was reduced to 1000 METs-min/week (equivalent to 140 minutes of moderate-intensity AE per week) after excluding high risk studies in the sensitivity analysis. When stratified by exercise modality, AE, AE-RT, and MBE were significantly associated with improved sleep quality in breast cancer survivors. Model-based estimates suggested that approximately 500 METs-min/week of MBE may provide the greatest improvement in sleep quality.

4.2. Comparison with previous studies

Despite numerous studies demonstrating the effectiveness of exercise in this population,^{[13–16,36,37]} the optimal exercise modality remains controversial. Hasan et al found that AE-RT was the most effective,^[16] while Song et al found walking to be the most effective.^[15] The differences may stem from variations in the included studies, as well as differences in the classification and definition of exercise modalities. Both studies employed the Surface Under the Cumulative Ranking Curve (SUCRA) to rank the effectiveness of various exercise modalities; however, this ranking method only reflects the relative efficacy of the exercises and provides no evidence of significant differences in their effectiveness.^[15,16] Our study found that MBE was the most effective for improving sleep quality in this population. MBE, which includes slow, mindful movements such as tai chi and yoga, is of moderate-intensity and well-suited for breast cancer survivors. Previous studies have demonstrated that MBE significantly improves insomnia.^[38] One study found moderate effects of yoga in breast cancer patients, including those undergoing primary treatments.^[14] Furthermore, Pilates and yoga share similar exercise characteristics, suggesting that both may be beneficial for this population. Given the limited number of available studies, these interventions were grouped into a single MBE category in the present analysis. As more evidence becomes available, future studies should adopt a more refined classification of different MBE modalities to better distinguish their individual effects. However, it should be noted that the pairwise meta-analysis showed that the lower bound of the 95% CI for MBE was close to zero, suggesting that the magnitude of the beneficial effect remains uncertain and should be interpreted with caution.

Compared to MBE, AE and AE-RT have larger sample sizes, providing stronger statistical power. Previous studies have shown that AE-RT is the most effective for improving sleep quality in breast cancer survivors ^[16], while AE has demonstrated small to moderate effects in cancer populations.^[39] Building on this, our study further defines the appropriate dose ranges. However, data for doses above 1000 METs-min/week are limited, and our dose-response analysis may not fully capture the true effects within this range. Therefore, the effects of these exercise modalities at doses exceeding 1000 METs-min/week warrant further investigation, which may yield new insights.

From a mechanistic perspective, exercise stimulates serotonin release in the brain and peripheral blood vessels, inhibiting the non-serotonergic spinal system, which promotes normal sleep and helps prevent insomnia.^[40] Some MBE may enhance parasympathetic nervous activity, contributing to improved sleep regulation.^[41] Other potential mechanisms include changes in homeostatic and immune processes, thermogenesis, energy conservation, body restoration, and mood enhancement.^[41] Moreover, when performed outdoors, exercise increases exposure to natural daylight, a powerful zeitgeber that helps resynchronize circadian rhythms, thereby promoting better nocturnal sleep.^[42] However, the exact mechanisms by which exercise influences sleep quality in breast cancer survivors remain unclear and warrant further investigation.

4.3. Clinical implications

This is the first exercise prescription aimed at improving sleep for breast cancer survivors who have completed their primary treatment, and it has 2 important clinical implications: First, clinicians can utilize our findings to create personalized exercise prescriptions based on individual patient conditions to alleviate sleep disorders. Second, future exercise intervention studies can refer to our exercise prescription framework to develop sleep-specific exercise programs and further validate the effectiveness of these dose ranges, thereby contributing to the strengthening of evidence in this field.

The International Multidisciplinary Roundtable’s Consensus Statement recommends that cancer survivors can safely engage in aerobic, resistance, and AE-RT, which can significantly improve cancer-related health outcomes, such as anxiety, depression, fatigue, physical functioning, and health-related quality of life.^[43] For sleep, the guidelines suggest moderate-intensity AE for 30 to 40 minutes, 3 to 4 times a week.^[43] Our meta-analysis further supports these recommendations. Additionally, we provided more detailed exercise prescriptions for AE-RT, AE, and MBE, offering broader guidance for clinical practice.

In summary, MBE seems to be the most effective, while AE and AE-RT are supported by more substantial evidence due to larger sample sizes. Clinicians should customize exercise prescriptions based on individual patient conditions, fitness levels, and preferences, adjusting as needed based on feedback.

4.4. Limtations

Although our study offers several strengths, there are a few limitations that should be acknowledged. The number of included studies was limited, particularly for MBE. The insufficient sample sizes may reduce statistical power, thereby diminishing confidence in the overall evidence. In network meta-analysis, the scarcity of direct comparison studies necessitates reliance on statistical model-derived estimates for many comparisons. Future RCTs should prioritize direct comparisons of the effects of different exercise modalities on sleep improvement to strengthen the evidence base. Although no significant publication bias was observed, the fact that all eligible studies were published in English may have introduced language bias. Additionally, the impact of small-sample size studies cannot be ruled out. Future research could mitigate this by searching multiple language databases and systematically including gray literature. The use of approximate dose categories may compromise the precision of dose estimation; therefore, the dose-response relationship should be considered exploratory. While using METs to assess exercise dose has certain advantages, such as standardizing various exercise types into a fixed intensity scale and assisting in the development of clinical exercise prescriptions, it also has notable limitations. MET calculations rely on the average resting metabolic rate, which overlooks individual factors like age, gender, weight, and physical condition that influence energy expenditure. This approach is also difficult to apply to complex exercises with fluctuating intensities and is challenging to measure accurately without specialized equipment. Moreover, standard METs values do not account for health conditions that may alter an individual’s response to exercise. Therefore, while METs offer a useful estimate of exercise intensity, more personalized methods, such as heart rate monitoring, are recommended for more precise assessments. Due to statistical heterogeneity, within-study bias, and imprecision, the certainty of evidence for 3 recommended exercise modalities was rated as low, which means that our dose-response relationship is based on these lower-certainty levels of evidence. This necessitates a cautious interpretation of our conclusions. Potential sources of heterogeneity include baseline sleep levels, age, cancer stage and severity, individual responses to exercise, and the quality of exercise protocol implementation. Future research should further explore heterogeneity, and including only populations with sleep disorders may help reduce imprecision. Additionally, adherence to a predesigned reporting plan, along with improvements in research design quality, is crucial for enhancing the overall quality of the evidence.

5. Conclusions

Exercise is linked to improved sleep quality in breast cancer survivors who have completed their primary treatment, following an inverted U-shaped pattern. The effective dose ranges from 0 to 1100 METs-min/week (equivalent to 0–150 minutes of moderate-intensity AE per week), with maximum efficacy achieved at 780 METs-min/week (110 minutes of AE per week). When stratified by exercise type, AE, AE-RT, and MBE all show significant improvements in sleep quality. Model-based estimates suggested that approximately 500 METs-min/week of MBE may provide the greatest improvement in sleep quality.

However, given that the evidence supporting all 3 exercise modalities was rated as low certainty, these conclusions should be interpreted with caution. This emphasizes the urgent need for high-quality RCTs to further substantiate these findings and verify the efficacy of the recommended dose ranges, which may lead to new insights.

Acknowledgments

The authors would like to thank all those who contributed to this manuscript.

Author contributions

Conceptualization: Hailin Li.

Data curation: Hailin Li, Shouwei Dong, Yimin Cheng.

Formal analysis: Hailin Li.

Supervision: Hailin Li.

Writing – original draft: Hailin Li, Shouwei Dong.

Writing – review & editing: Hailin Li, Shouwei Dong, Yimin Cheng.