This systematic review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)17 and MOOSE reporting guidelines.18 This study was registered in the International Prospective Register of Systematic Reviews (PROSPERO) (register number: CRD42024575962).

To define the inclusion criteria of the articles, the PICOS model was followed: (i) population: patients aged 18 years or older, male or female, with any type of cancer with or without declared metastasis undergoing surgery, chemotherapy, radiotherapy, hormone therapy, androgen deprivation, and/or immunotherapy for cancer treatment; (ii) intervention: peripheral muscle strength training during cancer therapy; (iii) comparison: usual care; (iv) outcomes: quality of life, fatigue, and/or pain; (v) study design: primary randomized controlled clinical trials (RCTs). Original, peer-reviewed articles written in English or Spanish published were included. The exclusion criteria for this study were: (i) studies that report palliative care for participants (ii) feasibility studies, pilot studies, editorials, letters, reviews and meta-analyses or (iii) preliminary analysis.

MEDLINE (via PubMed), Web of Science, CINAHL and SCOPUS were used to search for articles until January 14, 2024. The search strategy in MEDLINE was: ((((Strength training) OR (strength exercise)) OR (strengthening program)) OR (resistance training)) OR (resistance exercise)) AND (((((((neoplasms) OR (cancer)) OR (tumor)) OR (tumour)) OR (carcinoma)) OR (malignanc*)) Filters: Randomized Controlled Trial. The search strategy used in each database can be found in Supplementary Material, Table S1. In addition, a manual search of the references of the selected articles was performed to identify possible relevant studies.

Articles were selected using Rayyan web software (http://rayyan.ai)19 by two independent reviewers (P H-N and J S-M). The titles and abstracts were reviewed to determine potential articles. Then, the inclusion and exclusion criteria were applied to the full text of the potential articles to decide which ones were included. In case of discrepancies between the two reviewers, a third author (R N-C) was consulted to reach a resolution.

A standardized table was used to collect the following data: author, year of publication, sample size per group, average age per group, percentage of women per group, and cancer type and stage; concomitant cancer therapy, characteristics of the experimental intervention, control intervention, outcomes and main finding. If data were not provided in the article, attempts were made to contact the authors by email on three occasions.

Another standardized table was used to identify the reporting of resistance exercise protocols according to the Consensus on Exercise Reporting Template (CERT)20 and recommendations from previous studies,15 and to determine the characteristics of the exercise programming implemented. The data extracted were the following: frequency, duration, number of exercise, volume, intensity, inter-set rest duration, time under tension, tolerability or pain threshold, internal or external focus, exercise progression, supervised or unsupervised, exercise equipment, therapist experience, adherence assessment, and adverse events record. Exercise volume, calculated as total repetitions, was subsequently used for the dose-response analysis.

Two reviewers (J S-M and R N-C) extracted data. In case of disagreement, a third reviewer (J C) compared the extracted data and resolved the disagreement by reviewing the articles.

Risk of bias was assessed by two reviewers (J S-M and R N-C) independently using the Cochrane risk of bias (RoB-2) tool for randomized clinical trials.21 Each domain (randomization process, deviations from intended interventions, missing outcome data, outcome measurement, selection of reported outcome, and overall bias) was rated as “low,” “some concerns,” or “high” risk of bias. A third reviewer P H-N) resolved any disagreements between the two reviewers.

The certainty of the evidence was assessed by two reviewers (J S-M and R N-C) independently using the Recommendations Assessment, Development and Evaluation (GRADE) approach22 with the online tool GradePRO (https://gradepro.org). The criteria for reducing the certainty of the evidence were as follows: i) limitation of included studies: one level if 25 % or more of the included articles for each outcome were at high risk of bias as assessed by ROB-2; ii) inconsistency: one level if there was high heterogeneity (I2 ≥ 75 %); iii) indirectness: one level if there were differences between participants, interventions, outcome measures or indirect comparisons; iv) imprecision: one level if there was a wide confidence interval, crossed the line of no effect and/or a small sample size (n < 300); v) risk of publication bias: one level if there was asymmetry in the funnel plot. Additionally, between-group effect sizes (Hedges) were calculated using a Microsoft Excel template with built-in formulas based on means and standard deviations.

RStudio software (RStudio, PBC, Boston, MA) was used for statistical analysis. The effect of RT on QoL, fatigue, and pain intensity was analyzed using a random-effects meta-analysis. The standardized mean difference (SMD) and 95 % confidence interval (CI) were considered for the analysis for each variable. For the quantitative synthesis, outcomes assessed by 3 or more studies in the same cancer type were included to avoid low-power analyses and clinical heterogeneity. To compensate for the overestimation of Cohen's d in small studies, Hedges' g was adjusted.23 The standard deviation was calculated, using standardized formulas, from the standard error (SE), the 95 % confidence interval (CI), or the p-value of a t-test when studies did not present the standard deviation.21 For consistency, comparisons were aligned in the same direction (e.g., higher score corresponds to greater fatigue). Heterogeneity was analyzed using the I2 statistic. It was considered that heterogeneity between 0–40 % is not important, while between 30–60 % corresponds to moderate heterogeneity, between 50–90 % to substantial heterogeneity, and between 75 %−100 % to considerable heterogeneity.21 The estimated SMDs were interpreted as follows: trivial effect: <0.2; small effect: 0.2–0.6; moderate effect: >0.6–1.2; large effect: >1.2–2.0; very large effect: >2.0–4.0; extremely large effect: >4.0.24 To visualize the effect sizes and CIs from the included studies, along with the calculated summary effect size, forest plots were generated. These plots were generated using the forest() function, which is also available as part of the "metafor" v. 3.0- 2 R package.

Visual inspection of the DOI plots was performed to detect publication bias.25 Additionally, a quantitative measure of the LFK index was also used.26 An LFK index within ±1 represents no asymmetry; an LFK index greater than ±1 but within ±2 represents a small asymmetry; and a LFK index greater than ±2 represents a large asymmetry.26

To assess the dose-response relationship between total exercise volume (i.e. total number of repetitions during the RT program) and changes in outcomes, a one-stage mean difference dose-response meta-analysis was performed.27 Selected effect sizes (ES) and corresponding covariances (SDs) were used to estimate study-specific dose-response curves. The SMD of the within-group changes from baseline to the final measurement for each variable and the total number of exercise repetitions were used as the dose. The dose-response relationship was defined using a restricted cubic spline model with three nodes at the 10th, 50th, and 90th percentiles. The relative efficacy of the dose studied compared to a "zero" dose is characterized by dose-response curves. The procedure has been implemented in the R package dosresmeta.28

The database searches yielded a total of 3352 potentially eligible studies and eliminated 1388 articles due to duplicates. After screening articles by title and abstract, 1902 studies were eliminated. All full-text articles were then retrieved, and after applying the inclusion criteria, 42 articles were eliminated due to incorrect study design (n = 7), incorrect population (n = 21), incorrect intervention (n = 9), incorrect outcome (n = 3), and incorrect publication type (n = 2). Ultimately, 19 studies were included in this systematic review8,29–46 (Fig. 1).

Fig. 1.

PRISMA 2020 Flow diagram.

A total of 989 participants were included, of which 56.1 % were women. Eight studies included patients with breast cancer,8,29,30,32,33,41,42,46 1 with lung cancer,30 1 with gastric cancer,30 6 with prostate cancer,31,36,38,39,44,45 1 with glioma,34 1 with nasopharyngeal cancer,35 1 with spinal bone cancer,40 and 2 studies considered several types of cancer.37,43 Most studies that report on cancer stage consider early to late stages. Regarding concomitant therapies, 2 studies considered breast surgery,29,46 12 chemotherapy,8,29,30,32–35,37,40,41,43,46 8 radiation,29,30,34,39,42,43,45,46 3 hormonal therapy,29,40,46 6 androgen deprivation,31,36,38,39,44,45 and 1 immunotherapy.40 Sixteen studies reported QoL,8,29–33,35–39,41,43–46 17 reported fatigue,8,29–35,37–39,41–46 6 reported pain intensity.36–38,40–42 The characteristics of the studies are presented in Table S2.

Regarding the reporting of exercise protocols, weekly exercise frequency was reported by 18/19 studies (94.7 %), which varied within protocols from 28,29,34,35,37,41,42,46 to 543 days per week. Strength training duration was reported by 19/19 studies (100 %), which ranged from 539 to 24 weeks. Number of exercises was reported in 17/19 (89.5 %), which ranged from 129 to 1036,41,45 exercises. Number of repetitions was reported in 17/19 studies (89.5 %) from 429 to 2041 repetitions and sets in 19/19 studies (100 %) from 138,39,41 to 1030 sets. Exercise intensity was reported in 16/19 (84.2 %) of studies with 14 studies using one repetition maximum (RM)8,29–33,37,38,41–46 to prescribe intensity and two studies using Borg-based perceived exertion.34,39 Rest between sets was reported in 4/19 (21.1 %) and ranged from 134 to 330,43 minutes. No study (0 %) reported time under tension. Pain tolerance or threshold was reported in 2/1938,39 (10.5 %). No study (0 %) reported focus of attention. Regarding exercise progression, 16/19 studies (84.2 %) performed progressive strength training8,29–39,43–46 while the remainder did not provide details. Supervision modality was reported in 18/19 studies (94.7 %). Sixteen studies performed supervised exercise8,29–36,38–41,44–46 1 unsupervised,43 and 1 hybrid.37 Use of exercise equipment was reported in 13/19 studies (68.4 %). One study used elastic resistance bands,43 1 used body weight,39 5 studies used exercise machines,29,35,37,38,41 2 studies used dumbbells,30,36 3 used exercise machines plus dumbbells31–33 and 1 used body weight, resistance bands, and dumbbell barbells.39 Therapist experience in strength training was reported in 3/19 studies30,45,46 (15.8 %). On the other hand, assessment of treatment adherence was reported in 13/19 studies8,29,31–38,44–46 (68.4 %). Finally, regarding adverse events, 12/19 studies (63.2 %) reported recording adverse effects. Seven studies reported no adverse effects8,30–32,39,40,46 and 5 studies reported side effects in a few cases (n = 18, 1.8 % of a total sample), including: nausea, dizziness, weakness, diarrhea33; pain36,38,45; and fatigue, lymphedema, muscle soreness, dyspnea, and tachycardia.37 The exercise variables in the strength training protocols of each study are presented in Table S3 and Figure S1.

Among the 16 studies that reported QoL, 14 included a comparative analysis between groups. Of these, 6 (42.9 %) indicated statistically significant results in favor of the RT group,29,30,35,36,43,44 8 (57.1 %) found no differences between the RT group and the control group.8,32,33,37–39,45,46 Among the 17 studies that reported fatigue, 14 included a comparative analysis between groups. Of these, 8 (57.1 %) found statistically significant results in favor of the RT group,29,30,35,39,43–46 5 identified (35.7 %) no differences between the strength group and the control group,8,32,33,37,38 and 1 (7.1 %) reported statistically significant results in favor of the control group.34 Among the 6 studies that reported pain intensity, 4 performed a comparative analysis between groups. Of these, 2 (50 %) found statistically significant results in favor of the RT group36,40 and 2 (50 %) identified that there were no statistically significant differences between the RT group and the control group.37,38

Meta-analyses were only possible for breast and prostate cancers. In patients with breast cancer undergoing breast surgery, chemotherapy, radiotherapy, and/or hormone therapy, a statistically significant difference in favor of RT was found for fatigue reduction (605 participants, k = 6, SMD=−0.30, 95 % CI −0.46 to −0.14; p < 0.001), with a small effect size (Fig. 2). There was no statistical heterogeneity (I2 = 0 %, p = 0.494) and no significant evidence of publication bias (Egger's regression: −1.797, p = 0.072). There was a trend in favor of RT for improvement in QoL, but it was not statistically significant (611 participants, k = 6, SMD=−0.17, 95 % CI −0.03 to 0.36, p = 0.089) (Supplementary material, Figure S2). There was no statistical heterogeneity (I2 = 27.9 %, p = 0.136) and no significant evidence of publication bias (Egger regression: 1.865, p = 0.062). A statistically significant difference in favor of RT was found for pain reduction (302 participants, k = 3, SMD=−0.25, 95 % CI −0.48 to −0.02; p < 0.032), with a small effect size (Fig. 3). There was no statistical heterogeneity (I2 = 0 %, p = 0.879) and no significant evidence of publication bias (Egger's regression: 0.285, p = 0.775). The dose-response relationship between total RT volume and the effect on fatigue showed a J-shape with a maximum effect observed at 2800 repetitions of RT (304 participants, k = 6; SMD −0.028, 95 % CI: −0.057 to 0.001) (Supplementary material table S4, and Fig. 4). For QoL, a linear relationship was observed with a moderate effect at 8800 repetitions of RT (318 participants, k = 6; SMD 0.60, 95 % CI: 0.569 to 0.64) (Supplementary material table S5, and Fig. 5). A dose-response meta-analysis was not possible for pain because of the small number of trials.

Fig. 2.

Forest plot for effect of resistance training on fatigue in breast cancer. Summary of the results of the included trials comparing resistance training with usual care. The point estimate of the effect size and the sample size are shown in the small boxes with squares. The lines on either side of the box represent a 95 % confidence interval (CI). The interpretation of the SMD estimates was as follows: trivial effect 〈 0.2; small effect: 0.2–0.6; moderate effect: > 0.6–1.2; large effect: > 1.2–2.0; very large effect: > 2.0–4.0 and extremely large effect: > 4.0.

Fig. 3.

Forest plot for effect of resistance training on pain intensity in breast cancer. Summary of the results of the included trials comparing resistance training with usual care. The point estimate of the effect size and the sample size are shown in the small boxes with squares. The lines on either side of the box represent a 95 % confidence interval (CI). The interpretation of the SMD estimates was as follows: trivial effect < 0.2; small effect: 0.2–0.6; moderate effect: > 0.6–1.2; large effect: > 1.2–2.0; very large effect: > 2.0–4.0 and extremely large effect: > 4.0.

Fig. 4.

Dose-response association between RT volume and the effect on fatigue. Each study included corresponds to a point estimate. The difference in means corresponds to the change within the group between the baseline and the last measurement. Studies with larger circles contributed more to the overall effect size than other studies. The solid line represents the estimate for the mean difference between baseline and last follow-up, while the dashed lines represent the 95 % CI.

Fig. 5.

Dose-response association between RT volume and the effect on QoL. Each study included corresponds to a point estimate. The difference in means corresponds to the change within the group between the baseline and the last measurement. Studies with larger circles contributed more to the overall effect size than other studies. The solid line represents the estimate for the mean difference between baseline and last follow-up, while the dashed lines represent the 95 % CI.

In patients with prostate cancer undergoing radiotherapy and/or androgen deprivation therapy, there was a trend in favor of RT for fatigue reduction, but it was not statistically significant (340 participants, k = 4, SMD=−0.20, 95 % CI −0.02 to 0.41, p = 0.072) (Supplementary material, Figure S3). There was no statistical heterogeneity (I2 = 0 %, p = 0.540) and no significant evidence of publication bias (Egger's regression: 0.528, p = 0.598). There was no statistically significant difference in favor of RT for improvement in QoL (341 participants, k = 4, SMD = 0.15, 95 % CI −0.06 to 0.37, p = 0.154) (Supplementary material, Figure S4). There was no statistical heterogeneity (I2 = 0 %, p = 0.969) and no significant evidence of publication bias (Egger's regression: −0.164, p = 0.870). There was insufficient data to perform a meta-analysis for pain outcome. A dose-response meta-analysis was not possible for prostate cancer because of the small number of trials.

For the overall risk of bias, no study was found to have a low risk of bias, 10.52 % of the studies were categorized as having some concerns, and 89.47 % had a high risk of bias. For the domain of the randomization process, 63.15 % had a low risk of bias and 36.84 % had some concerns. For deviations from the intended interventions, 68.42 % had some concerns and 31.57 % had a high risk of bias.

For the domain of missing outcome data, 84.21 % had a low risk of bias and 15.78 % had a high risk of bias. In the domain of measurement of the outcome, 15.78 % had a low risk of bias and 84.21 % had a high risk of bias. Finally, in the domain of selection of the reported result, 63.15 % had a low risk of bias and 36.84 % had some concerns. The evaluation of each study for each domain using RoB-2 is provided in the Supplementary Material (Supplementary material, Figure S5).

The results of the analyses in breast cancer patients indicate that there is low certainty of evidence (downgraded for risk of bias) with small effect size for fatigue and pain intensity, and very low certainty of evidence (downgraded for risk of bias and imprecision) with trivial effect size for QoL. The results of the analyses in prostate cancer patients indicate that there is very low certainty of evidence (downgraded for risk of bias and imprecision) with trivial effect size for fatigue and QoL (Supplementary material, Figure S6).

Our study is in line with previous meta-analyses that have reported benefits of RT on patients undergoing cancer therapy. McGovern et al.11 found RT to be an effective adjunctive therapy for improving muscle strength and lean mass in cancer patients undergoing chemotherapy and/or radiation therapy. However, this study analyzed the effects of multiple cancer types (e.g., breast cancer, colorectal cancer, head and neck cancer, prostate leukemia, and gastrointestinal cancer) without stratifying by cancer type. Tian et al.12 concluded that RT appears to be a promising approach, compared with stretching exercises and usual care, to improve body composition and physical function in prostate cancer patients to offset their treatment-related side effects. Metcalfe et al.14 reported that RT can improve lower-body strength in patients undergoing chemotherapy (for germ cell cancer, breast cancer, colon cancer, pancreatic cancer, and lymphoma) treatment compared to control, however, they found no improvements in fatigue or QoL, physical function, or upper-body strength. In particular, the fact that the Metcalfe et al. study did not report improvements in fatigue can be explained by the fact that they combined different types of cancer in the meta-analysis. In contrast, our stratified meta-analysis found improvements in fatigue in breast cancer, but not in prostate cancer. In this regard, studies have indicated that improvements in fatigue potentially occur due to improvements in cardiac output, metabolic adaptations, and the recruitment of skeletal muscle motor units.51 However, the results of the fatigue meta-analysis differed depending on the type of cancer patients had, which may be because the level of fatigue may depend on the type of cancer,52 which could influence the effects of RT depending on the baseline level of fatigue. Importantly, dose-response analysis showed a J-shaped association between total exercise volume and the effect on fatigue, reinforcing the importance of considering exercise dosage when designing RT interventions.

Although there was a trend towards improved QoL in both breast and prostate cancer patients, the difference was not statistically significant compared to treatment as usual. This can probably be explained by the multidimensional nature of QoL. However, it is important to note that just over a third of the included studies in total reported improvements in QoL. In addition, the dose-response analysis showed a linear trend, i.e., a greater effect on quality of life with higher exercise volume (total number of repetitions), emphasizing the necessity of optimizing training parameters to achieve meaningful improvements. Thus, improvements in physical function and symptoms due to physical exercise,53–58 could also influence the improvement in the perception of quality of cancer patients.59

On the other hand, our study identified that resistance exercise improved pain intensity in patients with breast cancer. These changes, although small, may be due to multiple factors (involving biological, physical and psychosocial factors).60 For example, although biomarkers were not analyzed in the present study, previous studies in other population have identified that exercise generates a modulation of the endogenous pain inhibition system61 through the release of beta-endorphins and endocannabinoids,62–64 and the inhibition of excitatory neurotransmitters and pro-inflammatory cytokines in pain pathways.65 In addition, a change in behavior and coping with pain has been identified due to the maintenance of physical exercise, which contributes to the decrease in pain intensity.66

Our review highlights a critical issue in the field, namely the inconsistent and often incomplete reporting of exercise prescription variables in resistance training protocols for cancer patients. This lack of detailed reporting limits the comparability between studies limits and the translation of research findings into clinical practice. Future studies should prioritize detailed reporting of these variables to enhance the quality and applicability of research in this field.

For example, only 10.5 % of the included studies report a pain threshold or tolerability during exercise. For people with chronic musculoskeletal pain, painful exercises have been reported to have a small but significant short-term benefit over pain-free exercises.67 Nonetheless, this merits further investigation in patients undergoing cancer therapy.

On the other hand, time under tension, understood as the duration of muscular force production, which was not reported by any study, may play an important role in improving strength, as 6 s per repetition has been shown to produce the greatest benefits among older adults.68 Also, there were no studies reporting on the cognitive modality (internal or external focus) used. Internal focus (emphasis on muscle contraction rather than the movement of weight) can increase muscle activity,69 as well as possibly lead to a greater increase in muscle thickness.70 In contrast, external focus (e.g., dual tasks) has been observed to improve muscle endurance.71 Finally, rest time between sets was also poorly reported among the included studies. Although recent research suggests that rest time does not affect variables such as muscle hypertrophy in healthy individuals,72 this particular variable may be of interest due to the fatigue that cancer patients may experience as a result of their treatment. Addressing the identified gaps in exercise prescription reporting is essential for advancing the quality of resistance training interventions. The standardization and systematic reporting of these variables would not only enhance the reproducibility and comparability of clinical trials but also facilitate the design of more precise, patient-centered protocols, thereby strengthening the implementation of evidence-based practice in oncology exercise prescription.

The strengths of this systematic review were: 1) All assessments were self-reported using validated questionnaires, so the subjective components of the responses may reflect participants' self-perception and sense of self-knowledge in the final results. 2) All included studies were randomized clinical trials. 3) Effects were analyzed specifically for cancer types, providing more specific results. This systematic review and meta-analysis has some limitations: 1) There is heterogeneity in the instruments used, which may affect the results due to their psychometric properties.76 2) RT programs present great heterogeneity between studies, so it is not possible to specifically determine which are the best exercises. We considered the information on exercise protocols in the trials to be reliable, and our aim was to analyze adherence to exercise variables as published. 3) The type of cancer also varied widely and there were only enough studies to perform a meta-analysis on breast and prostate cancer. Furthermore, it was not possible to specify the exact stages and type of treatment the cancer patients underwent, which may influence the results; 4) The number of studies included in the analyses was small, which may affect the robustness of the results; 5) Most studies had a high risk of bias, so the individual results of each study affect the pooled analysis; 6) Due to the very low and low certainty of the evidence, future randomized controlled trials need to be of higher quality to provide more robust results; 7) Outcomes included in the analyses are assessed with self-reported questionnaires, creating a subjective component of assessment; and 8) After searching the databases, only English-language studies were included, which could influence the results due to the possible non-inclusion of relevant studies in other languages.