A systematic review and meta-analysis of randomized controlled trials for physical activity among colorectal cancer survivors: directions for future research

Study design

This systematic review has been registered in the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42024531424). To enhance the rigor of reporting, this review conformed to the updated guidelines for the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) in 2020 (Page et al., 2021).

Search strategy

To systematically search and integrate the existing evidence on PA interventions for CRC survivors, seven electronic databases were utilized to retrieve literature published between January 1, 2010, and March 1, 2024, including five English databases (Cochrane Library, Embase, PubMed, Scopus, and Web of Science) and two Chinese databases (China National Knowledge Infrastructure and Wan Fang Data). The date of January 1, 2010, was chosen as the beginning date for the search since it marks the publication of the initial PA guidelines targeted at cancer survivors (Schmitz et al., 2010). Each electronic database follows the principle of patient/population, intervention, comparison and outcomes (PICOS) and utilizes the subject terms combined with free terms to conduct a literature search. Furthermore, the database-specific search filters are also used in the literature retrieval to improve the quality of the final search output. The search terms included (‘physical activity intervention’ or ‘physical activity’ or ‘exercise’ or ‘exercise intervention’ or ‘aerobic exercise’ or ‘resistance exercise’ or ‘flexibility exercise’) AND (‘colorectal cancer survivors’ or ‘colorectal cancer patients’ or ‘colorectal cancer’ or ‘colon cancer’ or ‘rectal cancer’) AND (‘randomized controlled trials’). The detailed search strategies utilized in the seven databases are shown in Table S1. In addition, a manual screening was conducted in the reference lists of the included studies to guarantee thorough coverage of the eligible literature.

Eligibility criteria

The criteria for inclusion were as follows: (a) the targeted research population was the adult CRC survivors (≥18 years old); (b) PA is defined as any bodily movement that is generated by skeletal muscles and results in energy consumption (Caspersen, Powell & Christenson, 1985). Eligible interventions contained all forms of PA, and we did not impose any restrictions on the frequency, intensity, type, and time of PA interventions; (c) comparison control group received only usual care, without any measures aimed to promote PA during the trial period; (d) research outcome measure are PA levels. PA levels indicate the intensity and duration of the PA performed by participants over a period of time and typically measured by self-reported questionnaire or objective pedometers (Dowd et al., 2018); (e) the study design is RCTs (only RCTs were included in this study due to the fact that RCTs are considered to be the ideal study design for systematic reviews assessing the effectiveness of interventions (Barker et al., 2023)). Exclusion criteria included studies that (a) were conference abstracts, literature reviews, or study protocols; (b) were published in languages other than English or Chinese; (c) combined PA interventions with other interventions, e.g., dietary intervention; (d) did not set up an appropriate comparison group.

Study selection

EndNote 20 was used to eliminate duplicate citations from all retrieved studies. Two researchers (JYM and XKQ) independently conducted the literature search and subsequently assessed whether they met the selection criteria according to the titles, abstracts, and full texts. In cases of disagreement, a third independent researcher (CW) further evaluated the related disagreement, and consensus was reached through multiple discussions.

Data extraction

Data extraction was conducted using a standardized form to extract relevant information from each eligible study, including general information (authors, year of publication, and country), study characteristics (study design, sample size, intervention, and control group), patient characteristics (sex and cancer category), intervention characteristics (contents, theory, compliance, and duration), PA characteristics (frequency, intensity, type, and time), primary outcome, measurement tool, and results. In addition, we followed the previously published research to conduct subgroup analyses based on the characteristics of PA interventions (i.e., intervention characteristics and PA characteristics) in the included studies (Geng et al., 2023). The characteristics of the included studies and PA interventions are respectively presented in Tables 1 and 2.

Quality assessment

The methodological quality of the included studies was assessed using the Physiotherapy Evidence Database (PEDro) scale, and this scale is increasingly used for assessing the methodological quality of clinical trials included in systematic reviews across physiotherapy, health, and medical areas (Cashin & McAuley, 2020). A total of 11 items, each rated as zero or one, constitute the PEDro scale. For the quality of included studies, a total score of less than three indicates “poor,” four to five indicates “reasonable,” six to eight indicates “good,” and more than eight indicates “excellent” methodological quality (Cashin & McAuley, 2020). Given that the literature on PA interventions specifically for CRC survivors is still very limited, we did not exclude any studies based on their methodological quality. Two researchers (JYM and XKQ) independently assessed the methodological quality of the included studies, and any disagreement was discussed with third researcher (CW) until a consensus was reached. The detailed results of the quality assessment of the included studies are shown in Table 3.

Meta-analysis

The R and RStudio were utilized to perform statistics analysis, and meta-analysis was carried out with the meta (a kind of R package) (Balduzzi, Rücker & Schwarzer, 2019). Referring to the previously published study (Mbous, Patel & Kelly, 2020), selecting full-scale RCTs and pilot RCTs to conduct a meta-analysis to evaluate the effectiveness of PA interventions in improving the total PA levels is reasonable. The total PA levels (primary outcome) were the continuous data measured by questionnaires or accelerometers, and the measurement tools for total PA levels varied in the included studies. Due to the measurement tools for total PA levels varied in the included studies, the standardized mean deviation (SMD) was adopted in the meta-analysis. When studies reported multiple time points for outcome measures, the data of total PA levels close to the end of the intervention was extracted. The researchers extracted the sample sizes, mean of the total PA levels, and standard deviation of the total PA levels from intervention and control groups to conduct subsequent meta-analyses. Results of the meta-analysis are presented by the SMD, 95% confidence interval (CI), effect size (Z), and p-value. As for the interpretation of the value for SMD, an SMD of 0.2 represents a small effect size, an SMD of 0.5 represents a moderate effect size, and an SMD of 0.8 represents a large effect size. Heterogeneity among the included studies was examined by the χ² test. If there was no significant heterogeneity between each study (p > 0.10, I² < 50%), the fixed-effect model was used; if significant heterogeneity existed (p < 0.10, I² > 50%), the random-effects model was used for meta-analysis. In addition, we used the sensitive analysis to explore the potential source of heterogeneity, and the egger test and funnel plot were utilized to evaluate publication bias in the included studies. The p-value less than 0.05 was considered to indicate a statistically significant difference.

Evidence assessment

The Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) method was used to assess the certainty of the effect of each outcome, which was significant in the meta-analysis (Guyatt et al., 2008). The assessment was conducted using the GRADEpro GDT software. Evidence based on RCTs was considered to be high-quality evidence at the outset, and certainty of evidence may be downgraded for several reasons, including inconsistency of results, indirectness of evidence, imprecision, reporting bias, and study limitations (Guyatt et al., 2008). The certainty of evidence for the estimated effects can be categorized into four classes: high (very confident in the effect), moderate (moderately confident in the effect), low (little confidence in the effect), and very low (very little confidence in the effect) (Guyatt et al., 2008). The certainty of evidence was independently evaluated by one researcher (JYM) and confirmed by the second researcher (XKQ). The detailed results of the evidence assessment are shown in Table 4.

Study selection

The literature search initially yielded 1,986 potential studies from seven electronic databases, and two studies were identified through a manual search. After removing the duplicates, 1,230 articles remained. Then, 1,075 studies were removed because they did not meet the inclusion criteria by reading the title and abstract. Of these, 155 full-text articles were obtained for detailed eligibility assessment, and 133 studies that did not meet the eligibility criteria were excluded. Finally, 22 studies were included in this systematic review, and nine were selected for meta-analysis (see Fig. 1).

Methodological quality of the included studies

The 22 studies included were evaluated for methodological quality following the 11 items of the PEDro scale. Of these, 21 studies demonstrated good methodological quality with total scores ranging from six to eight, while the remaining one study had excellent quality with a score of ten (Hardcastle et al., 2023). Each study stated its eligibility criteria, ensuring compliance with randomized allocation. Hidden allocation was employed in eight studies (Hardcastle et al., 2023; Leach et al., 2023; Van Blarigan et al., 2019; Van Waart et al., 2018; Van Blarigan et al., 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Kim et al., 2019), while the remaining 14 did not report using this method (Witlox et al., 2018; Golsteijn et al., 2023; Falz et al., 2023; Cadmus-Bertram et al., 2019; Mayer et al., 2018; Lee, Kim & Jeon, 2018; Golsteijn et al., 2018; Lee et al., 2017; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Pinto et al., 2013; Ligibel et al., 2012; Courneya et al., 2016). Almost all included studies met the baseline comparison between groups, proper follow-up, comparisons between groups post-intervention, the point estimate, and variability of the PEDro scale. However, two studies did not meet the group comparisons after intervention (Leach et al., 2023; Van Waart et al., 2018). Despite this, these studies are still included in this review for subsequent analysis due to their good methodological quality and detailed statistics of interesting research outcomes. Additionally, the blindness of the participants, therapists, and assessors was the common reason that caused included studies to evaluate the “good” methodological quality. Specifically, one study reported that they blinded the participants in the research (Hardcastle et al., 2023), two studies suggested that they blinded the therapists (Hardcastle et al., 2023; Van Blarigan et al., 2022), and seven studies blinded the outcome assessors in their reports (Hardcastle et al., 2023; Leach et al., 2023; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Irwin et al., 2017; Van Vulpen et al., 2016; Pinto et al., 2013). In summary, the studies included in this review are characterized by good methodological quality.

Study characteristics of the included studies

For general information, 15 studies were full-scale RCTs (Witlox et al., 2018; Hardcastle et al., 2023, 2021; Maxwell‐Smith et al., 2019; Kim et al., 2019; Golsteijn et al., 2023; Falz et al., 2023; Mayer et al., 2018; Golsteijn et al., 2018; Lee et al., 2017; Irwin et al., 2017; Van Vulpen et al., 2016; Pinto et al., 2013; Ligibel et al., 2012; Courneya et al., 2016), and the other seven were pilot RCTs (Leach et al., 2023; Van Blarigan et al., 2019; Van Waart et al., 2018; Van Blarigan et al., 2022; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014). Of these 22 studies included, eight studies were from America (Leach et al., 2023; Van Blarigan et al., 2019, 2022; Cadmus-Bertram et al., 2019; Mayer et al., 2018; Irwin et al., 2017; Pinto et al., 2013; Ligibel et al., 2012), five studies were from The Netherlands (Witlox et al., 2018; Van Waart et al., 2018; Golsteijn et al., 2023, 2018; Van Vulpen et al., 2016), three studies were from Australia (Hardcastle et al., 2023, 2021; Maxwell‐Smith et al., 2019), three studies were from South Korea (Kim et al., 2019; Lee, Kim & Jeon, 2018; Lee et al., 2017), and the remaining three studies were from Canada (Courneya et al., 2016), Germany (Falz et al., 2023), and Sweden (Backman et al., 2014), respectively. The total sample size of the trials was 3,037, with each study consisting of 23 to 478 individuals. Each group had a balanced sample size and sex distribution in all the studies. In terms of the participant’s cancer category, 11 studies only included CRC survivors (Leach et al., 2023; Van Blarigan et al., 2019; Van Waart et al., 2018; Van Blarigan et al., 2022; Kim et al., 2019; Mayer et al., 2018; Lee, Kim & Jeon, 2018; Lee et al., 2017; Van Vulpen et al., 2016; Pinto et al., 2013; Courneya et al., 2016), and the other 11 studies included mixed cancer survivors, including CRC, breast cancer, gynecologic cancer, and prostate cancer (Witlox et al., 2018; Hardcastle et al., 2023, 2021; Maxwell‐Smith et al., 2019; Golsteijn et al., 2023; Falz et al., 2023; Cadmus-Bertram et al., 2019; Golsteijn et al., 2018; Irwin et al., 2017; Backman et al., 2014; Ligibel et al., 2012).

Regarding the group characteristics, all selected studies were designed with appropriate research groups, including an intervention group and a control group. More than half of the included studies (12/22) for the intervention group utilized the technology-based PA interventions (e.g., online training, computer-tailored PA advice, and video conference PA intervention) (Hardcastle et al., 2023; Leach et al., 2023; Van Blarigan et al., 2019, 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Golsteijn et al., 2023; Falz et al., 2023; Cadmus-Bertram et al., 2019; Mayer et al., 2018; Golsteijn et al., 2018; Ligibel et al., 2012), five studies utilized the supervised PA intervention (Witlox et al., 2018; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Courneya et al., 2016), four studies utilized the home-based PA intervention (Kim et al., 2019; Lee, Kim & Jeon, 2018; Lee et al., 2017; Pinto et al., 2013), and the remaining one study has two intervention arms, including the Onco-Move (i.e., home-based, low-intensity, individualized, self-managed PA intervention) and OnTrack (i.e., moderate-to-high intensity, supervised PA intervention) (Van Waart et al., 2018). The intervention contents of the different types of PA interventions are detailed in Table 2. For the control group, 14 studies showed that the control group only accepted usual care during the trial period (Witlox et al., 2018; Leach et al., 2023; Van Waart et al., 2018; Kim et al., 2019; Golsteijn et al., 2023; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Golsteijn et al., 2018; Lee et al., 2017; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Pinto et al., 2013; Ligibel et al., 2012), and eight studies provided health education materials (e.g., health booklets or guidebooks) to control group participants as part of usual care (Hardcastle et al., 2023; Van Blarigan et al., 2019, 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Falz et al., 2023; Mayer et al., 2018; Courneya et al., 2016).

The measurement of PA levels is commonly divided into subjective and objective measurements (Dowd et al., 2018). Specifically, subjective measurements mainly use questionnaires (e.g., Godin Leisure-Time Exercise Questionnaire) to investigate the PA levels of participants over a period of time, and objective measurements use the motion receptors (e.g., ActiGraph GTX3+ accelerometer) to record the PA among participants and calculate their PA levels (Dowd et al., 2018). Of these 22 studies included, 11 studies used subjective questionnaires or scales (Witlox et al., 2018; Van Waart et al., 2018; Kim et al., 2019; Mayer et al., 2018; Lee, Kim & Jeon, 2018; Lee et al., 2017; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Ligibel et al., 2012; Courneya et al., 2016), seven studies utilized objective measurement tools (Hardcastle et al., 2023; Van Blarigan et al., 2019, 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Falz et al., 2023; Cadmus-Bertram et al., 2019), and the remaining four studies used subjective questionnaires combined with objective measurement tools to measure participants’ PA level (Leach et al., 2023; Golsteijn et al., 2023, 2018; Pinto et al., 2013). Among 22 studies, the results of 17 studies showed that PA interventions significantly increased PA levels in the intervention group, and the difference between each group was statistically significant (Witlox et al., 2018; Hardcastle et al., 2023; Leach et al., 2023; Van Waart et al., 2018; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Golsteijn et al., 2018; Lee et al., 2017; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Pinto et al., 2013; Ligibel et al., 2012; Courneya et al., 2016). The remaining five studies suggested that PA interventions increased PA levels in the intervention group, with no statistically significant difference between each group at the end of the intervention (Van Blarigan et al., 2019, 2022; Golsteijn et al., 2023; Falz et al., 2023; Mayer et al., 2018).

PA intervention characteristics of the included studies

Among the studies included, the intervention contents of nine studies were based on a single theoretical framework, including the SCT (n = 4), TPB (n = 3), TTM (n = 1), and self-determination theory (n = 1) (Witlox et al., 2018; Leach et al., 2023; Van Blarigan et al., 2019; Van Waart et al., 2018; Van Blarigan et al., 2022; Mayer et al., 2018; Van Vulpen et al., 2016; Ligibel et al., 2012; Courneya et al., 2016). The design of the intervention contents in the four studies was based on multiple theories, such as the combination of SCT and TTM which served as the theoretical basis for designing the intervention contents (Golsteijn et al., 2023; Cadmus-Bertram et al., 2019; Golsteijn et al., 2018; Pinto et al., 2013). In addition, the remaining 11 studies did not provide the theoretical framework for the design of PA intervention (Hardcastle et al., 2023; Van Blarigan et al., 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Kim et al., 2019; Falz et al., 2023; Lee, Kim & Jeon, 2018; Lee et al., 2017; Irwin et al., 2017; Backman et al., 2014; Courneya et al., 2016). Eight studies were conducted on compliance with PA interventions, and compliance with PA interventions ranged from 61% to 100% in the included studies (Leach et al., 2023; Van Blarigan et al., 2019; Van Waart et al., 2018; Van Blarigan et al., 2022; Irwin et al., 2017; Van Vulpen et al., 2016; Backman et al., 2014; Courneya et al., 2016). The intervention duration of included studies ranged from 6 weeks to 3 years, including the 6 weeks (n = 1) (Lee, Kim & Jeon, 2018), 10 weeks (n = 1) (Backman et al., 2014), 12 weeks (n = 11) (Hardcastle et al., 2023; Leach et al., 2023; Van Blarigan et al., 2019, 2022; Hardcastle et al., 2021; Maxwell‐Smith et al., 2019; Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Irwin et al., 2017; Pinto et al., 2013), 16 weeks (n = 3) (Golsteijn et al., 2023, 2018; Ligibel et al., 2012), 18 weeks (n = 2) (Witlox et al., 2018; Van Vulpen et al., 2016), 6 months (n = 3) (Van Waart et al., 2018; Falz et al., 2023; Mayer et al., 2018), and 3 years (n = 1) (Courneya et al., 2016). The PA details (i.e., frequency, intensity, type, and time of PA) are presented in Table 2.

Overall analysis of the effect of PA interventions on total PA levels

Of the 22 studies reviewed, nine provided detailed statistics on total PA levels and were included in this meta-analysis (Leach et al., 2023; Van Waart et al., 2018; Kim et al., 2019; Falz et al., 2023; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014; Ligibel et al., 2012; Courneya et al., 2016). Specifically, a study conducted by Van Waart et al. (2018) has two intervention arms: Onco-Move and OnTrack. These two intervention arms represent different types of PA interventions, so we divided the study into two parts for meta-analysis, following the approach outlined in the previously published literature (Jung & Son, 2021). Due to the different types of measurement tools used and significant heterogeneity (p = 0.08, I² = 41%), we chose the SMD and the random-effects model to perform an overall meta-analysis of the effects of PA interventions on total PA levels (see Fig. 2). In the sensitivity analysis, the source of heterogeneity was not discovered. Results of the meta-analysis showed that PA interventions had a small-to-moderate positive effect on improving the total PA levels (SMD = 0.32, 95% CI [0.09–0.54], Z = 2.79, p = 0.005). In addition, according to the results of the egger test (p = 0.42) and funnel plot (see Fig. S1), there was no significant publication bias among the included studies.

Subgroup analysis of the effect of PA interventions on total PA levels

A total of four studies using the technology-based PA intervention (Leach et al., 2023; Falz et al., 2023; Cadmus-Bertram et al., 2019; Ligibel et al., 2012), three studies using the home-based PA intervention (Van Waart et al., 2018; Kim et al., 2019; Lee, Kim & Jeon, 2018), and three using the supervised PA intervention were included for subgroup analysis (Van Waart et al., 2018; Backman et al., 2014; Courneya et al., 2016) (see Fig. 3). Random-effects model was used due to the significant heterogeneity (p = 0.08, I² = 41%). Results of subgroup analysis showed that there were no statistically significant effects on total PA levels when the type of PA intervention was technology-based PA intervention (SMD = 0.15, 95% CI [−0.08 to 0.38], Z = 1.27, p = 0.206) and home-based PA intervention (SMD = 0.52, 95% CI [−0.06 to 1.11], Z = 1.75, p = 0.080). On the contrary, total PA levels were significantly increased when the type of PA intervention was supervised PA intervention (SMD = 0.37, 95% CI [0.11–0.63], Z = 2.82, p = 0.005).

A total of five studies with theory-based interventions (Leach et al., 2023; Van Waart et al., 2018; Cadmus-Bertram et al., 2019; Ligibel et al., 2012; Courneya et al., 2016) and five studies with non-theory-based interventions (Van Waart et al., 2018; Kim et al., 2019; Falz et al., 2023; Lee, Kim & Jeon, 2018; Backman et al., 2014) were included for subgroup analysis (see Fig. 4). Random-effects model was used in both two subgroups due to the significant heterogeneity (p = 0.08, I² = 41%). Results of subgroup analysis showed that theory-based PA interventions were not significantly effective in improving total PA levels (SMD = 0.23, 95% CI [−0.03 to 0.49], Z = 1.75, p = 0.08). In contrast, non-theory-based PA interventions had a significant effect in improving total PA levels (SMD = 0.46, 95% CI [0.04–0.87], Z = 2.17, p = 0.03).

A total of five studies with more than or equal to three intervention components (Van Waart et al., 2018; Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Courneya et al., 2016) and five studies with less than three intervention components (Leach et al., 2023; Van Waart et al., 2018; Falz et al., 2023; Backman et al., 2014; Ligibel et al., 2012) were included for subgroup analysis (see Fig. 5). Random-effects model was used due to the significant heterogeneity (p = 0.08, I² = 41%). Results of subgroup analysis showed that there was a statistically significant effect for total PA levels when the number of PA intervention components was more than or equal to three (SMD = 0.47, 95% CI [0.17–0.77], Z = 3.06, p = 0.002). In contrast, PA intervention with less than three intervention components was not significantly effective in improving total PA levels (SMD = 0.13, 95% CI [−0.11 to 0.37], Z = 1.03, p = 0.302).

The different frequencies of PA were employed in these nine included studies, including the daily (n = 5) (Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014; Ligibel et al., 2012), twice a week (n = 2) (Leach et al., 2023; Falz et al., 2023), five times per week (n = 1) (Van Waart et al., 2018), and determined by participants (n = 1) (Courneya et al., 2016). These studies were included for subgroup analysis to determine the appropriate frequency of PA in improving total PA levels (see Fig. 6). Random-effects model was used in these four subgroups due to the significant heterogeneity (p = 0.08, I² = 41%). Results of subgroup analysis revealed that there were no statistically significant effects for total PA levels when the frequency of PA was Twice a week (SMD = 0.01, 95% CI [−0.32 to 0.33], Z = 0.04, p = 0.967) and five times per week (SMD = −0.33, 95% CI [−1.07 to 0.41], Z = −0.87, p = 0.383). Because there was only one study in the subgroup of “determined by participants”, it was impossible to calculate the effect size of participant-determined PA frequency for improving total PA levels. In addition, total PA levels were significantly increased when the frequency of PA was daily (SMD = 0.53, 95% CI [0.29–0.77], Z = 4.28, p < 0.001).

The different intensities of PA were employed in these nine included studies, and one study did not report the intensity of PA in its research (Backman et al., 2014). Moderate-to-vigorous physical activity (MVPA) (Kim et al., 2019; Cadmus-Bertram et al., 2019), moderate physical activity (MPA) (Lee, Kim & Jeon, 2018; Ligibel et al., 2012), the intensity of PA determined by the perceived exertion (Van Waart et al., 2018; Falz et al., 2023), and the intensity of PA determined by participants (Leach et al., 2023; Courneya et al., 2016) were included for subgroup analysis to the determine the appropriate intensity of PA in improving total PA levels (see Fig. 7). Random-effects model was used in these four subgroups due to the significant heterogeneity (p = 0.06, I² = 47%). Results of subgroup analysis revealed that there were no statistically significant effects for total PA levels when the intensity of PA was determined by the perceived exertion (SMD = 0, 95% CI [−0.32 to 0.32], Z = 0.01, p = 0.99) or participants (SMD = 0.12, 95% CI [−0.58, 0.83], Z = 0.34, p = 0.735). On the contrary, total PA levels were significantly increased when the intensity of PA was MVPA (SMD = 0.52, 95% CI [0.08–0.97], Z = 2.29, p = 0.022) or MPA (SMD = 0.55, 95% CI [0.03, 1.07], Z = 2.08, p = 0.038). In summary, the pooled results showed that the intensity of PA was MVPA had more significant effects on increasing total PA levels than MPA (Z = 2.29 > 2.08, p < 0.05).

The different types of PA were employed in the nine included studies. Aerobic exercise (Cadmus-Bertram et al., 2019; Backman et al., 2014; Courneya et al., 2016), resistance exercise (Falz et al., 2023), aerobic combined with resistance exercise (Van Waart et al., 2018; Kim et al., 2019; Lee, Kim & Jeon, 2018), and type of PA determined by participants (Leach et al., 2023; Ligibel et al., 2012) were included for subgroup analysis to determine the appropriate type of PA in improving total PA levels (see Fig. 8). We utilized the fixed-effects model to perform the subgroup analysis as a result of the low heterogeneity (p = 0.12, I² = 37%). Results of subgroup analysis revealed no statistically significant effects for total PA levels when the type of PA was determined by the participant (SMD = 0.17, 95% CI [−0.18 to 0.53], Z = 0.96, p = 0.336). Only one study was included in the subgroup of “resistance exercise” (Falz et al., 2023), so it was not possible to calculate the effect size. In contrast, aerobic exercise alone (SMD = 0.38, 95% CI [0.14–0.62], Z = 3.08, p = 0.002) and particularly when combined with resistance exercise (SMD = 0.69, 95% CI [0.37, 1.02], Z = 4.19, p < 0.001) showed more substantial benefits for increasing total PA levels.

The different time of PA was employed in these nine included studies, and two studies did not report the time of PA in their research (Cadmus-Bertram et al., 2019; Courneya et al., 2016). The time of PA was 30 min (Van Waart et al., 2018; Kim et al., 2019; Falz et al., 2023; Ligibel et al., 2012) and 60 min (Leach et al., 2023; Lee, Kim & Jeon, 2018; Backman et al., 2014) were included for subgroup analysis to determine the appropriate time of PA in improving total PA levels (see Fig. 9). Random-effects model was used in the subgroup analysis due to the significant heterogeneity (p = 0.04, I² =53%). Results of subgroup analysis revealed that there were no statistically significant effects for total PA levels when the time of PA was 30 min (SMD = 0.23, 95% CI [−0.10 to 0.56], Z = 1.36, p = 0.174) or 60 min (SMD = 0.38, 95% CI [−0.47 to 1.23], Z = 0.87, p = 0.382). In summary, the appropriate time of PA was not determined in this subgroup analysis due to the effect of each subgroup on improving the total PA levels is not significant.

Certainty of evidence

The GRADE test evaluated the certainty of the effect for each significant outcome in the meta-analysis. Evidence on the effect of PA interventions on increasing total PA levels was rated as moderate quality because most of the included studies were at significant risk of bias due to the lack of concealed allocation (Falz et al., 2023; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014; Ligibel et al., 2012; Courneya et al., 2016) and the use of unblinded assessors (Van Waart et al., 2018; Kim et al., 2019; Falz et al., 2023; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014; Ligibel et al., 2012; Courneya et al., 2016). For the characteristics of PA interventions, evidence on the effect of supervised PA intervention on increasing total PA levels was rated as low quality because most of the included studies presented serious risk of bias (Van Waart et al., 2018; Backman et al., 2014; Courneya et al., 2016) and imprecision. Evidence on the effect of PA intervention with more than three intervention components on increasing total PA levels was rated as moderate quality, and the serious risk of bias in the included studies was the main reason for downgrading this evidence (Van Waart et al., 2018; Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Courneya et al., 2016). For the type of PA, evidence on the effect of PA interventions (frequency = daily), PA intervention (intensity = MVPA), and PA intervention (type = aerobic combined with resistance exercise) on increasing total PA levels was rated as low quality due to the serious risk of bias (Kim et al., 2019; Cadmus-Bertram et al., 2019; Lee, Kim & Jeon, 2018; Backman et al., 2014; Ligibel et al., 2012) and serious imprecision in the included studies. Therefore, a cautious interpretation and application of the above evidence is required due to the small sample size and high heterogeneity of the included studies.

Suggestions for future PA intervention studies

Based on the characteristics of the included PA intervention studies and the results of the meta-analysis, the following recommendations are provided to guide the development of future PA interventions:

• (1)

  Conducted country: the vast majority of PA intervention studies in our study were conducted in the western countries, such as the United States and The Netherlands (n = 13, 60%), highlighting the need for future interventions to focus on the eastern countries, particularly China.

• (2)

  Target population: this review determined the effectiveness of PA intervention in increasing total PA levels for CRC survivors, suggesting that future PA interventions could be applied to CRC survivors with low PA levels.

• (3)

  Research design: most of the included studies have methodological quality problems since these included studies did not use hidden allocation (n = 14) and impose blinding on participants or assessors (n = 21). In order to ensure the methodological quality of research, it is important that future studies utilize more robust study designs (e.g., RCTs) and adherence to the Consolidated Standards of Reporting Trials (CONSORT) checklist.

• (4)

  Theory application: results of the meta-analysis showed that non-theory-based PA interventions have a better effect on increasing total PA levels in CRC survivors than theory-based PA interventions (Z = 2.17 > 1.75, p < 0.05). One possible reason is that theory-based PA intervention studies included in this review only indicated their chosen intervention theory (Van Waart et al., 2018; Ligibel et al., 2012; Courneya et al., 2016) but did not integrate the theoretical framework with the design of the intervention program. Thus, future research should carefully select behavior change theories and describe how the theoretical frameworks were integrated into the study design of PA interventions, providing a valuable reference for similar intervention studies.

• (5)

  Intervention components: results of the meta-analysis showed that PA interventions with more than or equal to three intervention components have a better effect on increasing total PA levels in CRC survivors than PA interventions with less than three intervention components (Z = 3.06 > 1.03, p < 0.05). Compared to conventional PA interventions, PA interventions with multiple intervention components (n ≥ 3) focus on the psychological condition of the participants and provide enough social support, thus contributing to improved intervention adherence. Therefore, it is recommended that future studies employ PA interventions with multiple intervention components (n ≥ 3).

• (6)

  Outcome measure: PA levels in half of the included studies (n = 11, 50%) were measured using self-report questionnaires or scales, which may have led to recall bias. Although activity loggers or accelerometers are not the gold standard for the measurement of PA levels, but they are an objective and relatively reliable instrument that can be applied to future studies.

Implications for clinical practice

It demonstrates that PA interventions can significantly improve PA levels for CRC survivors, and evidence on the effect of PA interventions on increasing total PA levels was rated as moderate quality. Recent research also confirmed that PA has a solid theoretical basis and strong evidence for reducing symptom burden and improving physical function in cancer patients, leading to improved health-related QoL (Maddocks, 2020). Moreover, implementing exercise programs during adjuvant treatment for CRC survivors can yield cost savings of 4,321 euros, highlighting the cost-effectiveness of such interventions (May et al., 2017). In clinical practice, CRC survivors typically face a wide range of disease-related and treatment-related challenges that often lead to a lack of PA during cancer survivorship (Fisher et al., 2016). Therefore, PA interventions should be recognized by clinical practitioners as an important and cost-effective strategy for improving PA levels and other health-related outcomes and should be implemented in the care of CRC survivors.