This systematic review and meta-analysis is reported by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement and was registered at the International Prospective Register of Systematic Reviews (number CRD 42024536976).

Two independent reviewers (HZ and FL) selected relevant studies published between January 1945 and June 2024 by searching PubMed Central, Embase, Cochrane, and Web of Knowledge on 17 June 2024, respectively. In addition, ongoing clinical trials about “physical exercise for CRC patients” were searched on the international clinical trial registry¹ and conference abstracts as other sources (Supplementary Table S1). The correctness of the search string was then validated by JW. We applied no language restrictions. First, we searched PubMed for MeSH and entry terms on “colorectal cancer” and “physical exercise.” Then, concatenate their respective MeSH, entry terms, and abbreviations by the logical term “OR.” Finally, we connected the two searches using the logical term “AND,” which gave us the final search results. We used the following combined text and MeSH terms: “Colonic Neoplasms” and “Exercise.” The complete search used for PubMed was: (“Colonic Neoplasms” [MeSH Terms] OR Colonic Neoplasm [Text Word] OR Neoplasm, Colonic [Text Word] OR Neoplasms, Colonic [Text Word] OR Colon Neoplasms [Text Word] OR Colon Neoplasm [Text Word] OR Neoplasm, Colon [Text Word] OR Neoplasms, Colon [Text Word] OR Cancer of Colon [Text Word] OR Colon Cancers [Text Word] OR Colon Cancer [Text Word] OR Cancer, Colon [Text Word] OR Cancers, Colon [Text Word] OR Cancer of the Colon [Text Word] OR Colonic Cancer [Text Word] OR Cancer, Colonic [Text Word] OR Cancers, Colonic [Text Word] OR Colonic Cancers [Text Word] OR Colon Adenocarcinoma [Text Word] OR Adenocarcinoma, Colon [Text Word] OR Adenocarcinomas, Colon [Text Word] OR Colon Adenocarcinomas [Text Word] OR CRC [Text Word]) AND (“Exercise” [MeSH Terms] OR Exercises [Text Word] OR physical exercise [Text Word] OR Activities, Physical [Text Word] OR Activity, Physical [Text Word] OR Physical Activities [Text Word] OR Exercise, Physical [Text Word] OR Exercises, Physical [Text Word] OR Physical Exercise [Text Word] OR Physical Exercises [Text Word] OR Acute Exercise [Text Word] OR Acute Exercise [Text Word] OR Exercise, Acute [Text Word] OR Exercises, Acute [Text Word] OR Exercise, Isometric [Text Word] OR Exercises, Isometric [Text Word] OR Isometric Exercises [Text Word] OR Isometric Exercise [Text Word] OR Exercise, Aerobic [Text Word] OR Aerobic Exercise [Text Word] OR Aerobic Exercises [Text Word] OR Exercises, Aerobic [Text Word] OR Exercise Training [Text Word] OR Exercise Trainings [Text Word] OR Training, Exercise [Text Word] OR Trainings, Exercise [Text Word]). We considered all potentially eligible studies for review, irrespective of the primary outcome or language. To identify additional relevant studies, we also conducted a manual search of the reference lists of key articles.

Two independent reviewers (HZ and YZ) screened titles and abstracts of the retrieved articles based on the inclusion criteria. In cases where a decision could not be made based on the title and abstract, the full article was read to determine whether the study was eligible. Disagreements between them were resolved by consensus, and a final decision was reached by involving the reviewer (JW). We regarded studies as eligible for inclusion if they were RCTs conducted in adults with any stage of CRC, comparing physical exercise [comprising aerobic exercise (AE), resistance exercise (RE), or their combination] treatment to usual care treatment strategy. Considering that sudden participation in exercise for patients who have been inactive for a long period of time may cause problems such as hypoglycemia, elevated blood pressure, and muscle and joint injuries, and to avoid the influence of such abnormalities on the experimental results, we regarded studies that had at least 4 weeks’ duration of intervention and reported changes in the nutritional status and inflammatory markers of patients with CRC. Exclusion criteria were as follows: observational, retrospective studies, reviews, editorials, animal trials, comments systematic review, and meta-analysis; studies with less than 4 weeks duration of intervention and those that did not indicate exercise type.

The two reviewers (HZ and YZ) used data collection forms presented in Table 1 to independently extract post-intervention outcome variables. Any discrepancies between them were resolved through iteration and discussion. The extracted data were analyzed by two investigators (HZ and FL), while the disagreements were resolved by a third investigator (JN). The extracted data included the following: (1) study characteristics, such as year of publication, sample size, and country; (2) description of the intervention, prescription of exercise program, modality, exercise time per week, frequency, and follow-up duration; (3) participant characteristics, such as age, sex, and CRC stage; (4) outcomes such as levels of BMI, body weight, body fat, waist circumference, level of proinflammatory cytokines IL-6, C-reactive protein (CRP), and TNF-α; and (5) secondary outcomes such as quality of life (QoL), fatigue, pain, insomnia, physical functioning, and emotional functioning. All the outcome indicators were presented as the mean ± standard deviation (SD), and for the data in the form of median and interquartile range (IQR), these would be counted after determining whether the data were skewed and then converted by the formula to get the form of mean ± SD (26, 27). Two independent reviewers (YW and FL) assessed the risk of bias according to the PRISMA using the Cochrane Risk of Bias tool. The two reviewers discussed their different views on the risk of bias assessment criteria and tried to reach a consensus through communication and negotiation. The quality of available evidence was assessed using the GRADEpro Guideline Development Tool (GDT).

We assessed the effect of exercise for CRC patients on two outcomes: nutrition indicators and inflammation markers. We analyzed these as continuous variables and reported the absolute differences between the arithmetic means before and after the interventions. Stata (version 17.0) and Review Manager 5.4.1, from the Cochrane Collaboration,² were utilized to conduct a quality assessment of the evidence and to calculate effect estimates for combinations of single effects from included studies, as well as to perform subgroup analyses, respectively. The mean differences in the data were calculated at 95% confidence intervals. We used the Cochran Q test to assess heterogeneity between studies. We also did I² testing to assess the magnitude of the heterogeneity between studies, with values greater than 50% regarded as being indicative of moderate-to-high heterogeneity. The results were summarized using random effects models, taking into account the potentially large heterogeneity of the population of interest as well as the details of physical exercise with respect to the clinical methodology. We only conducted descriptive analysis for studies from which the data format could not be obtained. Sensitivity analysis was performed by assessing the impact of individual studies on the overall effect size and one-by-one elimination method to explore sources of heterogeneity, followed by subgroup analyses for intervention duration, exercise types, exercise time, and patient BMIs. Datasets from inflammation markers were presented as forest plots, and the subgroup analyses were presented in statistical tables. We assessed the possibility of publication bias by constructing a funnel plot of each trial’s effect size against the standard error when the comparison included more than three studies. We assessed funnel plot asymmetry using Begg and Egger tests and defined significant publication bias as a p-value of <0.05.

The study was supported by intramural funds, with no commercial entity involved. The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

The search strategy identified 3,081 articles. Among them, 15 studies (28–42), comprising 996 patients, met the inclusion criteria and were therefore included in the final analysis (Figure 1). All 15 trials were published between 2007 and 2023, and the main characteristics of the included articles are outlined in Table 1. The mean age of patients was 59.45 ± 11.48 years, the mean duration was 18.8 weeks, and the mean baseline BMI was 26.88 ± 4.77 kg/m². In summary, patients had stages 1–3 and 1–4 CRC; Six studies reported an intervention duration of >12 weeks (28, 29, 31, 32, 36, 42), while nine studies reported ≤12 weeks (30, 33–35, 37–41). In this meta-analysis, interventions lasting >12 weeks were categorized as ‘long-term exercise’, as they typically allow for physiological adaptation across systems (cardiorespiratory, metabolic, etc.) (13, 43, 44). In terms of the type of exercise training program, six articles described AE (28, 31, 32, 34, 36, 41), whereas seven were combined AE with RE (28, 33, 35, 37, 39, 40, 42). Exercise time was less than 300 min per week in nine studies (29–31, 34, 37, 38, 40–42) and more than 300 min per week in six studies (28, 32, 33, 35, 36, 39). There were four studies (28, 32, 34, 39) with exercise intensity designed at moderate-high intensity and 11 (29–31, 33, 35–38, 40–42) set at low-moderate intensity. Regarding the exercise supervision status, except for Brown, J. C. (2023) (31) and Devin, J. L. (34), the supervision status was not explicitly elucidated; the rest of the RCT studies were completed under the supervision of exercise physiologists, with a completion rate of >80%. Thus, we considered that exercise supervision had minimal effect on the heterogeneity of the experimental results (45).

Figure 2 and Supplementary Figure S1 depict the risk of bias assessment for the trials. All studies were randomized controlled trials (RCTs), and none were stopped early. Four studies reported methods using computer-generated random numbers for randomization (28, 35, 39, 42). One trial described allocation based on participant preference (“The participants in this 2-group controlled trial were allocated to either the supervised-exercise group or the usual-care group on the basis of their preference after inclusion and baseline testing”) (37), which may introduce selection bias and compromise the validity of randomization. Therefore, this approach raised concerns about potential bias in the assignment process. Four RCTs had a high risk of performance bias due to insufficient blinding of investigators or patients (29, 31, 38, 42), and two RCTs lacked blinding during data analysis (34, 39). Eight studies had a low risk of reporting bias due to systematic clinical trial registration (31, 33, 35, 38–42). One study had incomplete or inaccessible reporting of some outcome indicators (31), and another study may have had baseline imbalance in data such as age (41). All included studies exhibited low publication bias. The quality of evidence for BMI, body weight, and body fat was assessed as high, supported by adequate sample sizes and a low risk of bias. Conversely, outcomes such as waist circumference, CRP, IL-6, TNF-α, QoL (quality of life), fatigue, pain, insomnia, physical functioning, and emotional functioning were rated as moderate quality due to small sample sizes (Table 2).

In a pooled analysis of all 15 trials, physical exercise improved nutritional indicators (BMI, weight, and body fat) in CRC patients when compared to usual care. Fourteen studies (28–30, 32–42), comprising 488 exercise and 481 usual care participants, reported BMI. Figure 3a and Table 3 showed no significant between-group difference in BMI (p = 0.16), with low heterogeneity (I² = 33%). Supplementary Figure S2a demonstrated a symmetric funnel plot, and both Begg’s test (p = 0.443) and Egger’s test (p = 0.146) indicated no significant publication bias. Sensitivity analysis (Supplementary Figure S3a) showed robust results for BMI. Subgroup analyses of exercise duration, volume, intensity, baseline BMI, and type (shown in Supplementary Figure S4) revealed that trials with duration ≤12 weeks showed a BMI reduction of 0.38 kg/m² (95% CI -1.15 to −0.02, p = 0.04, I² = 33%), while those with >12 weeks showed no significant change (p = 0.45). Combined AE + RE reduced BMI by 0.85 kg/m² (95% CI -1.45 to −0.25, p = 0.005, I² = 14%), whereas aerobic-only exercise had no effect (p = 0.90). No significant differences were observed for exercise volume (< vs. ≥300 min/week, p = 0.17–0.60), baseline BMI (overweight vs. non-overweight, p = 0.07–0.97), or intensity (low-moderate vs. moderate-high, p = 0.34–0.48) (Figure 3; Table 3).

Body weight was another key nutritional indicator. Eleven studies comprising 370 exercise and 356 usual care participants reported body weight (29, 30, 32–36, 38, 39, 41, 42). As Table 3 and Figure 3b show, there was no significant between-group difference in body weight (p = 0.48), with low heterogeneity (I² = 45%). Supplementary Figure S2b showed a symmetric funnel plot, and both Begg’s test (p = 0.755) and Egger’s test (p = 0.910) indicated no publication bias. Sensitivity analysis (Supplementary Figure S3b) confirmed stable results for body weight. Subgroup analyses revealed that trials with duration ≤12 weeks showed a body weight reduction of −3.12 kg (95% CI -4.66 to −1.58, p < 0.0001, I² = 0%). In addition, the AE + RE group had a body weight reduction of −2.14 kg (95% CI -4.68 to −0.21, p < 0.0001, I² = 0%), while no significant changes were observed in the >12 weeks group (p = 0.35) or AE-only group (p = 0.95). In addition, no significant differences were found for exercise volume, intensity, or baseline BMI (Supplementary Figure S5; Table 3). These findings suggest that physical exercise is effective across varying baseline BMI levels, with short-duration AE + RE combinations particularly helpful for improving BMI and body weight.

Body fat, defined as the weight of body fat as a percentage of total body weight, provides a more accurate measure of obesity than BMI (46). Body fat was reported in seven trials (28–30, 32, 35, 38), comprising 386 exercise and 379 usual care participants. Table 3 and Figure 3c showed that body fat decreased by −1.36% (95% CI -2.11 to −0.61, p = 0.0004, I² = 6%) in the exercise group compared to the usual care group. Supplementary Figure S2c showed a symmetric funnel plot, with both Begg’s test (p = 0.368) and Egger’s test (p = 0.10) indicating no significant publication bias. Sensitivity analysis (Supplementary Figure S3c) confirmed the stability of body fat results. Notably, body fat was reduced by −1.75% (95% CI -2.52 to −0.99, p < 0.0001, I² = 0%) in the ≤12 weeks group, by −1.81% (95% CI -2.87 to −0.75, p = 0.0008, I² = 0%) in the <300 min/week group, and by −1.74% (95% CI -2.49 to −0.98, p < 0.0001, I² = 0%) in the low-moderate intensity group. However, no significant changes were observed in the BMI < 25 kg/m² group (p = 0.29), BMI ≥ 25 kg/m² group (p = 0.06), or moderate-high intensity group (p = 0.50). Waist circumference was measured in three studies, including 193 patients (35, 36, 38). Meta-analysis showed no significant effect of physical exercise on waist circumference compared to usual care (p = 0.94, I² = 0%) (Figure 3d), with Begg’s test (p = 1) and Egger’s test (p = 0.938) indicating no publication bias (Supplementary Figure S6; Table 3).

Evidence supporting the role of inflammation in colorectal carcinogenesis is growing (47). For example, inflammatory bowel disease, characterized by local colon inflammation, is associated with increased colorectal cancer risk (48). Chronic inflammation may initiate and promote carcinogenesis through proinflammatory cytokines (e.g., IL-6) and reactive oxygen species, which activate transcription factors driving tumor growth (49, 50). Meta-analysis results showed that physical exercise improved inflammatory profiles. Four studies (30–32, 35), including 213 exercise and 202 usual care participants, reported C-reactive protein (CRP). Figure 4a showed a significant CRP reduction of −0.33 mg/dL (95% CI -0.62 to −0.04, p = 0.03, I² = 0%) in the exercise group. Supplementary Figure S2d showed a symmetric funnel plot, with Begg’s (p = 1.0) and Egger’s (p = 0.653) tests indicating no publication bias. IL-6 was reported in three studies with marginal non-significance (p = 0.060, I² = 0%), but no significant group differences were observed in IL-6 (31, 34, 40) (3 studies, 71 participants, p = 0.32, I² = 53%) or TNF-α (31, 34, 35) (3 studies, 170 participants, p = 0.26, I² = 64%) (Figures 4b,c). High heterogeneity (I² > 50%) in IL-6 and TNF-α analyses prompted the use of the random-effects model and leave-one-out sensitivity analyses. Excluding Brown, J. C. (31), reduced IL-6 heterogeneity (I² from 53 to 34%), while excluding Devin et al. (34), eliminated TNF-α heterogeneity (I² from 64 to 0%), indicating these studies were primary heterogeneity sources. Notably, Brown, J. C. (31), did not report exercise supervision, which may have affected adherence, while Devin, J. L. (34), used high-intensity exercise, potentially influencing TNF-α dynamics. Despite high heterogeneity, publication bias was absent for both markers (IL-6: Begg’s p = 1, Egger’s p = 0.473; TNF-α: Begg’s p = 1, Egger’s p = 0.462).

Exercise-induced improvements in QoL were multifaceted. It has been shown that physical exercise can alleviate symptoms such as fatigue, pain, and insomnia, which cancer patients commonly experience. It can also enhance psychological wellbeing by reducing anxiety and depression, fostering a sense of control and empowerment, and improving social interactions and support. These psychosocial benefits are crucial, as mental health significantly influences overall health outcomes and QoL in cancer patients (51, 52). Therefore, this study examined the secondary outcomes (e.g., QoL, fatigue, pain, etc.) of exercise in CRC patients. A total of three studies (36, 39, 42), comprising 41 participants who received physical exercise and 40 usual care subjects, reported QoL. Supplementary Figure S7 and Table 3 showed no significant benefit (p = 0.91, I² = 0%) for QoL in the exercise group. In addition, some signature indicators in the EORTC QLQ-C30 form were statistically analyzed (53). Physical functioning and emotional functioning were reported in three trials (37, 39, 42), comprising 53 participants who received physical exercise and 56 usual care subjects. The meta-analysis showed that physical functioning increased by 6.51 (95% CI 0.41 to 12.61, p = 0.04, I² = 0) in the physical exercise group compared with the usual care group, while emotional functioning increased by 10.1 (95% CI 2.02 to 18.18, p = 0.01, I² = 41%), as Table 3 shows. Moreover, there was no significant benefit for fatigue (p = 0.36, I² = 7%), pain (p = 0.60, I² = 34%), and insomnia (p = 0.41, I² = 0) in the exercise group. Furthermore, subgroup analyses showed no significant differences in fatigue and pain indicators across different exercise durations, types, and intensities. All secondary outcomes showed I² values <50%, indicating low heterogeneity. Begg’s test and Egger’s test values were >0.05, which indicate no significant publication bias.

A limitation of this analysis is that the long-term durability of this treatment is unknown; included trials ranged in duration from 4 weeks to 12 months (mean 18.8 weeks). Second, although most of the included studies were published in high-impact journals, there were study features that carry a potential risk of bias, such as the fact that only eight studies published clinical trial registration. In addition, most of the trials included in this meta-analysis were based on small sample sizes, except for three outcome indicators, namely BMI, body weight, and body fat. Meta-analyses with small sample sizes are prone to bias in tests of heterogeneity based on chi-square tests (e.g., the I² statistic); for instance, in sensitivity analyses of IL-6/TNF-α, the stability of study-specific effects and meta-regression results was compromised due to insufficient data, potentially leading to false conclusions about heterogeneity (e.g., underestimating true variability). In addition, the small sample size limited the feasibility of subgroup analyses of data on pain and fatigue in this study. Furthermore, the lack of standalone RE trials (only two studies) prevented subgroup analysis of RE-specific effects, possibly due to unpublished negative results. Last but not least, many of the outcome metrics reported in fewer studies could not be analyzed, such as skeletal muscle mass, glycated hemoglobin A1c (HbA1c), IGF-1, and other inflammation markers such as serum amyloid A (SAA) and procalcitonin (PCT), which hinders understanding of exercise’s biological pathways in CRC. Therefore, further large-scale (≥200 participants per arm), multi-center studies with longer exercise interventions and more sensitive indicators are needed to validate the effects of physical exercise on CRC patients.