A systematic search for studies was carried out in the PubMed, Embase, the Cochrane Library, Web of Science, and Scopus databases from their inception to 1 December 2023. The search strategy was structured according to the Peer Review of Electronic Search Strategies 2015 guidelines (17). We provide a complete list of search times, search entries, and search results in the Supplementary Material ( Supplementary Table 1 ). Two investigators (ZX and YY) independently screened the title/abstract and full text of the articles, with any disagreements resolved by discussion or adjudication by a third author (YL).

According to the principles of the “PICOS” inclusion criteria (18), the included studies comprised the following criteria: (a) Population: cancer survivors over 18 years of age; (b) Intervention: Any form of exercise; (c) Outcome: Measure the health-related quality of life in cancer survivors [ex: Functional Assessment of Cancer Therapy-General (Fact-General) (19); 36-Item Short Form Health Survey (SF-36) (20); European Organization for Research and Treatment of Cancer Quality of Life Questionnaire-Core 30 (EORTC QLQ-C30) (21); World Health Organization Quality of Life-BREF (WHOQOL-BREF) (22)]; (d) Study design: Must be randomized controlled trial (RCT); (e): Comparison: Controls could be daily care, physical health education, or other forms of exercise.

The exclusion criteria were as follows: (a) studies on children with cancer and cancer patients undergoing treatment; (b) intervention studies combining exercise with cognitive therapy, physiotherapy, massaging, diet, or medication; (c) studies that excluded trials with no control group; (d) endpoints that reported cancer-specific scales and excluded non-cancer survivors’ HRQoL scales; and (e) individuals with major depression, cognitive impairment or other diseases that may affect exercise ability and quality of life.

Two authors independently extracted data from the included studies (ZX and YY), and all authors resolved disagreements by consensus. First, the following data from each included study were extracted: first author and year of publication, sample (sample size), region (study country), age (patient age), cancer type, intervention characteristics (intervention duration/time, type of exercise), and HRQoL-related outcome indicators.

Second, data were extracted as means and standard deviations (SD) of the HRQoL-related outcome indicators before and after the intervention, based on the Cochrane Handbook for Systematic Evaluation of Intervention (23). To meet the data analysis requirements of the dose-response network meta-analysis package in the R program, we also converted the standard errors (SE). SE=SD/SQRT (Sample size/N) (24). The relevant data are provided in Supplementary File 3 . When it was not possible to retrieve the minimum required data from a published report for the dose-response meta-analysis, we contacted the authors and invited them to provide additional data.

First, we coded the interventions in a first level of categorization to classify all the interventions as equivalent to analyze the dose-response effect of overall exercise on the HRQoL of cancer survivors. Second, to analyze the optimal dose and the different forms of exercise, a second level of classification coding was performed on the interventions: Aerobic exercise (AE) (treadmill training, dance; stationary bike, etc.), mind-body exercise (MBE) (yoga; taiji; qigong);, mixed training (MT) (combined aerobic exercise and resistance exercise), resistance training (RT), and control (Placebo).

Previous studies have demonstrated the good efficacy of metabolic equivalents of task (METs) in evaluating exercise intensity (25, 26). Thus, in our study, the term “dose” referred to energy expenditure expressed in METs-min/week. We used the method validated by Ainsworth et al. (14) to calculate the different doses associated with each intervention in our study and to calculate the product of the duration of the exercise intervention, the number of sessions, and the metabolic equivalent of the task (METs-min/week). In order to better promote the connectivity of the network, we used an approximate value of 250, 500, 750, 1000, or 1500 METs-min/week for the exercise dose, which was adopted from previous studies (27).

We used the Bayesian-Model of network meta-analysis package (MBNMA) (28), and “rjags” package (29) in R version 4.0.3. (R Foundation for Statistical Computing, Vienna, Austria). We conducted the network meta-analysis to compare the overall relative effectiveness and different dose responses of the interventions under investigation. We checked the connectivity of the motion pattern grid for different motion doses based on the dose approximation (30) and we obtained agreement between the two models by comparing them with an uncorrelated mean-effects model (UME) ( Supplementary File 4.1 ) (31). Finally, we assessed transitivity using a node-splitting approach ( Supplementary File 4.2 ) (32). There were no parametric results that indicated a violation of the above key assumptions.

Regarding the dose-response models, firstly, to identify the observed effect of the overall exercise dose and different exercise doses on HRQoL, we plotted a confirmation of the non-linear relationship between overall exercise dose and cancer survivors’ HRQoL improvement ( Supplementary File 5 Figures 1, 2 ).

To summarize the dose-response relationship between exercise and cancer survivors’ HRQoL, we compared dose-response models (Emax, Exponential, Restricted cubic spline, non-parametric, and linear models) regarding fit indices, including the deviance information criterion (DIC), between-study standard deviation, the number of parameters in the models, and the residual values. Ultimately, we opted for the random effect of restricted cubic splines to evaluate the non-linear dose-response association, as detailed in Supplementary File 5 Table 5.1 .

In order to visualize our study data and model-fitting results, deviance plots were used to confirm the robustness of our model selection ( Supplementary File 5 Figures 3, 4 ) (33). We employed a Markov Chain Monte Carlo (MCMC) model, utilizing three chains with 20,000 iterations each (the first 10,000 discarded) and a thinning factor of 10. Through this model, we leveraged Beta coefficients derived from restricted cubic spline curves to estimate both the overall and varying effects of exercise doses on HRQoL, identifying the exercise dose associated with the optimal effect. Additionally, we placed nodes at the 10th, 50th, and 90th percentiles of the exercise dose to effectively visualize the results of our model fitting process (15). Our study considered different outcome measures. Therefore, we chose the standardized mean difference (SMD) (34) as the effect size for our study. We used a 95% credible interval (95% CrI) to provide a range of values within which we are confident the true effect size lies (35). The code to reproduce the results presented in this paper can be requested by contacting the first author’s e-mail address.

It was essential to identify and address the fundamental assumptions of heterogeneity, inconsistency, and consistency. Therefore, we first used the tau-squared (τ²) test and I² statistic to evaluate the statistical heterogeneity between the studies. Moreover, we assessed global inconsistency statistically using the design-by-treatment interaction test. We separated indirect from direct evidence using the SIDE (Separating Indirect from Direct Evidence) test using the “netmeta” package. Additionally, we compared adjusted funnel plots to assess the risk of publication bias under specific circumstances. Egger’s test indicated bias when p<0.05.

Additionally, the risk of bias was assessed according to the second version of the Cochrane risk-of-bias tool for randomized trials (RoB 2) (36). Three reviewers (BQ, YB, and TW) assessed the study. The assessment items included randomized sequence generation, bias due to deviation from the intended intervention, incomplete data, bias in measurements, and selective bias in reporting results. And conducted a risk assessment of our research based on the following items:

Disagreements between reviewers were resolved by the third author (JW), who was responsible for re-examining controversial assessments and discussing with other reviewers to reach a consensus, ensuring the accuracy and consistency of the assessment results.

To assess the potential impact of high risk of bias on our findings, we performed sensitivity analyses in which we excluded studies assessed to be at high risk of bias and reanalyzed the dose-response data.

A total of 18,571 study records were found. After removing duplicates and literature that did not fit the study by title, abstract, etc., we considered 215 full-text articles that were then screened for eligibility. Figure 1 shows a flow diagram of the study selection process. Finally, 48 studies involving 3050 participants were included in our network meta-analysis. The patients included in the studies had different types of cancer and were aged between 44 to 72 years old. The basic characteristics of the included population are given in Table 1 .

Whether connectivity was met is the basis of NMA. A lack of connectivity can lead to low statistical power and misleading results when a direct comparison is not possible (32). The analysis confirmed that there was no connectivity deficit in the two networks, ensuring accurate results ( Figure 2 ).

Figure 3 shows the non-linear dose-response relationship between overall exercise dose and cancer survivors’ HRQoL. Notably, we found an inverted U-shape between the overall exercise dose and cancer survivors’ HRQoL, and an optimal peak dose for improving HRQoL existed. Overall exercise was found to be effective until 1100 METs-min/week (SMD: 0.716, 95% Crl: 0.169 to 1.280) but the optimal overall exercise dose was found to be 850 METs-min/week (SMD: 0.753, 95% Crl: 0.463 to 1.096).

Figure 4 shows the non-linear dose-response relationship between different exercise doses and cancer survivors’ HRQoL in 18 studies involving 657 cancer survivors who participated in MT. The graphs show a trend of corresponding improvement in HRQoL with increasing MT dose until 1200 METs-min/week. The optimal MT dose was found to be 970 METs-min/week (SMD: 0.883, 95% Crl: 0.455 to 1.345).

Furthermore, in 16 studies involving 369 cancer survivors who participated in AE, in 7 studies involving 288 cancer survivors who participated in RT, and in 13 studies involving 311 cancer survivors who participated in MBE, we found the inverted U-shape for the AE, MBE, and RT intervention doses and cancer survivors’ HRQoL. None of these exercise doses exceeded 750 METs-min/week. AE was effective until 720 METs-min/week, and the optimal AE dose was found to be 430 METs-min/week (SMD: 0.681, 95% Crl: 0.206 to 1.099). Additionally, MBE was effective until 610 METs-min/week and the optimal MBE dose was found to be 390 METs-min/week (SMD: 0.672, 95% Crl: 0.259 to 1.087). The RT was effective until 660 METs-min/week, and the optimal RT dose was found to be 450 METs-min/week (SMD: 0.695, 95% Crl: 0.227 to 1.203).

In total, 22 studies (46%) were found to have a low risk of bias, 15 studies (31%) were found to have an unclear risk of bias, and 11 studies (23%) were found to have a high risk of bias. Figure 5 shows the result of the Cochrane Risk of Bias Tool and the study-level risk of bias assessments are presented in Supplementary File 6 .

The heterogeneity of HRQoL was moderate [I^2 = 58.9% (43.0%; 70.4%)] and the SIDE test of all outcomes showed that there was a percentage of comparisons with evidence of inconsistency ( Supplementary File 7 ). At the same time, our funnel plot results showed that there was no publication bias and the funnel plot was symmetrical ( Figure 6 ). In addition, we used Egger’s test to assess publication bias more formally. As shown in the figure, our included studies did not violate these two points.

We excluded 11 articles with a high risk of bias and reanalyzed the data. The results showed that the optimal dose of overall exercise was estimated to be 730 METs-min/week (SMD:0.727, 95%Crl: 0.461 to 1.038), 890 METs-min/week for MT (SMD:0.8418, 95%Crl: 0.226 to 1.306), 410 METs-min/week for AE (SMD:0.914, 95%Crl: 0.32 to 1.506), 410 METs-min/week for MBE (SMD:0.625, 95%Crl: 0.219 to 1.071), and 630 METs-min/week for RT (SMD:0.588, 95%Crl: 0.301 to 1.195). Additionally, these results of sensitive analysis confirmed the robustness of our study results. We provide visualization curve results in Supplementary Figures 1 , 2 in Supplementary File 9 .

Our systematic review and network meta-analysis included 48 randomized controlled trials involving 3050 cancer survivors. We determined the optimal exercise dose for overall exercise for improving cancer patients’ HRQoL, and as the dose increased, the effect took on an inverted U shape. We found that the optimal overall exercise dose to alter HRQoL in cancer survivors was 850 METs-min/week. Furthermore, in terms of different exercises, we found MT to be the best form of exercise to improve HRQoL, and we also found the optimal dose of MT (970 METs-min/week). Finally, we found that the optimal dose for improving HRQoL was 430 METs-min/week, 390 METs-min/week, and 450 METs-min/week for AE, MBE, and RT, respectively.

First, we demonstrated that there is an optimal overall exercise dose. This is consistent with previous research that showed that exercise, no matter what it is, may not only be “healthy” but may actually have therapeutic effects (37). It was also well established in a cohort study that cancer survivors who engaged in 15 MET-hours of physical activity per week had a 27% lower risk of cancer-related mortality (38).

Notably, research suggests that exercise may exert its beneficial effects on cancer through various mechanisms, such as the alteration of circulating factors such as myokines and hormones released by skeletal muscle (39). One study in particular delved into these mechanisms, shedding light on how exercise can inhibit cancer cell proliferation, induce apoptosis, regulate metabolism, and enhance the immune environment (40). In our study, we determined the effective and safe value range of the overall exercise dose, which is undoubtedly a highlight for exercise and tumor research.

Furthermore, we found that the optimal form was MT (combined aerobic exercise and resistance training) for improving HRQoL in cancer survivors. The optimal MT dose was 970 METs-min/week. This finding not only aligns with previous analyses that unanimously recognize MT as the most effective form of exercise for enhancing HRQoL but also corresponds with international statements regarding exercise guidelines for cancer survivors (13, 41). In the included studies, MT consists of aerobic exercise and resistance training. In our study, we determined the optimal dose for resistance training to be 450 METs-min/week. Previous studies have indicated that resistance training correlates with a decreased risk of all-cause mortality, cardiovascular disease, and cancer-specific mortality (42). Furthermore, recent reviews of study outcomes suggest that non-traditional resistance training can yield equivalent or greater improvements in muscle strength, hypertrophy, body function, and composition compared to traditional methods. Studies on eccentric, cluster, and blood flow restriction training in older adults and other clinical populations provide robust evidence supporting its potential application in cancer research (43). This underscores the significance of integrating resistance exercise into the treatment regimen for cancer patients, as our study unequivocally confirms its efficacy in enhancing HRQoL within this demographic. Notably, our research identified the optimal intensity for resistance exercise, marking a substantial advancement over prior studies. Moreover, the optimal aerobic exercise dose to improve HRQoL was found to be 430 METs-min/week. We explored the main potential mechanism: an animal study has demonstrated that exercise is linked to the mobilization and redistribution of natural killer (NK) cells through adrenaline and interleukin-6 (IL-6), ultimately regulating tumorigenesis. This research result suggests that aerobic exercise can reduce the risk of cancer and disease recurrence by over 60% (44). Our research categorizes tai chi, qigong, and yoga as MBE (45). We found that the optimal dose of MBE was 390 METs-min/week. A previous study confirmed an improvement in the HRQoL of cancer patients through tai chi and qigong (46). On this basis, we determined the dose of MBE. However, the underlying mechanisms by which mind-body exercise improves symptoms in cancer patients are uncertain.

In addition, we conducted heterogeneity, inconsistency, and publication bias tests through network meta-analysis. Our results did not find any statistical differences in our data. We found that the heterogeneity was moderate and used a random effects model to merge the data. These key steps also indirectly boosted the credibility of our research from statistics to practice.

While our study provides valuable insights, it is important to acknowledge its limitations. First, we did not conduct meticulous statistical heterogeneity analysis to explore possible heterogeneity among the studies.

Furthermore, in the preliminary analysis, we found that MT was the most effective in improving the HRQoL of cancer survivors. However, after sensitivity analysis, AE showed a more significant effect after excluding the studies with a high risk of bias. This change in results may reflect the impact of the risk of bias in different studies on the preliminary results. Specifically, some studies with high bias may have overestimated the effect of MT, causing the preliminary analysis results to favor MT.

This suggests that in high-quality, low-bias studies, AE may have an advantage in improving health-related quality of life. This finding reminds us that study quality and risk of bias should be considered when selecting interventions, and suggests that the relative effectiveness of different types of exercise may need to be re-evaluated in different research scenarios.

Nevertheless, this does not negate the potential value of MT, especially in specific patient groups. However, based on the results of the sensitivity analysis, AE may be a more robust and widely applicable option. Future studies should further explore the effects of these interventions in different cancer survivor groups to determine the optimal treatment plan.

Additionally, we did not report basic information regarding patient medication use, and it is plausible that patients received treatment not solely due to exercise-related improvements in outcomes. Moreover, our findings focused exclusively on the HRQoL of cancer patients and may not extend to all symptoms associated with cancer. We also note that the generalizability of our results may be limited as we observed an insufficient number of randomized controlled studies conducted in Asian populations. Although most exercise doses were reported following extensive deliberations between two reviewers, there remains the possibility of some misclassification at the dose level. Therefore, approximations were employed for dose characterization. Another potential limitation is the inclusion of cancer survivors with comorbidities (e.g., major depression or cognitive impairment) that may have underlying factors that may affect quality of life. These comorbidities may have had an impact on the intervention effect and may have confounded the relationship between exercise dose and HRQoL improvement to some extent. Although excluding these individuals would have resulted in a more homogeneous sample and a clearer understanding of the optimal exercise dose, we chose to include them in the study to increase the external validity of the study. Many cancer survivors face multiple health challenges simultaneously, and including these individuals made our findings more generalizable. Future studies may consider stratifying participants according to specific comorbidities or controlling for these variables in the study design to more precisely assess the effects of the exercise intervention. Finally, there may be possible differences in diagnosis stage and cancer type due to a higher proportion of female participants in the sample as certain cancer types (such as breast cancer) are more common in women, resulting in a relatively large number of female cancer survivors. This gender imbalance may have had an impact on the research results as a result of differences in physiology, psychology, and response to exercise intervention. Therefore, the inconsistent diagnosis of cancer patients in our study and the mixture of various cancer types may lead to doubts about the reliability of our results. In future RCT experiments, we should actively verify the true and effective value of our core research results, and provide recommended guidelines for exercise prescription for the next RCT experiment.