The primary aim of this SRM was to investigate the long-term impact of PA and BCT interventions in PA maintenance of CRCS. Five randomized controlled trials with a total of 906 CRCS met the criteria for inclusion in the SRM, where a significant change with a small positive effect (SDM = 0.22, (0.09, 0.35); I2 = 15.4%) was found after completing a PA intervention combined with the use of different BCT and a follow-up period ranging from 3 to 12 months. This finding was consistent with previous meta-analyses conducted on both healthy inactive individuals [46] and cancer patients [20] or survivors [23] who completed PA and BCT interventions and a subsequent follow-up of at least 3 months with similar SMD not higher than 0.26.

Small effect sizes are common in these types of studies because improvements in PA can be observed not only between-groups, but also within the IG and, even more, within the CG. Grimmet et al. [23] conducted a meta-analysis to test the effectiveness of these interventions in different types of cancer (most of them in breast cancer), with follow-up periods starting at 3 months. The results showed significant increases in PA in both the IG (SMD = 0.49, 95% CI (0.32, 0.66)) and the CG (SMD = 0.21, 95% CI (0.08, 0.35)), as well as between-groups (SMD = 0.25, 95% CI (0.16, 0.35)). These results suggest that the effect of interventions on PA behaviour may be underestimated. On several occasions, it has been defined as “contamination” and is commonly found in randomized controlled trials of PA and dietary behaviour change [47], since participation in the study could be influenced by a highly biased behaviour change, irrespective of group allocation. Therefore, pooling these data in a meta-analysis may magnify type II errors, leading to an intervention being incorrectly considered ineffective because important changes in CG are ignored [48]. A possible solution to address this problem in future studies may be to offer the CG the possibility of performing the intervention after the study is completed. In this way, it may be possible to avoid the early predisposition of the CG to change their PA behaviour and comply with the indication not to change their usual PA level during the study. In addition, a larger sample size could also help minimize these type II errors.

This SRM also revealed very similar results to those presented by Mbous et al. [21] indicating the potential effectiveness of such interventions in CRCS. However, it did not consider the specific long-term effectiveness of these interventions, as studies without a subsequent follow-up period were also included. In contrast, our SRM suggested that PA changes in CRCS could be sustained for at least 3 months after the intervention and up to 12 months thereafter.

When the included trials were examined in more detail through descriptive analysis, although four studies also reported post-intervention results indicating higher levels of PA compared with those obtained at baseline, these increases were only statistically significant in two of them [44, 45]. By contrast, these same interventions were not effective during post-follow-up. Thus, no intervention that was effective post-intervention was effective post-follow-up. On the other side, there was one intervention that was not effective to increase PA significantly post-intervention but it was after the follow-up period [42]. This could be explained, on the one hand, by the contamination during the intervention period, and on the other hand, by the greater durability of the effect in the IG subjects. The fact that both groups increased PA levels during the intervention could have contributed to the non-statistically significant differences between groups. However, during the follow-up period, between-group differences became significant possibly because the IG maintained their increased levels while the CG decreased to almost their initial values.

Following on from the above, an indicator of maintenance could be that the level of PA did not decrease significantly from the post-intervention to the post-follow-up period in the IG. This implies that, while significant differences between-groups are to be expected, it is also important to consider what happens within each group. This fact reflects one of the limitations found in the scientific literature to identify the most effective interventions for the maintenance of PA. Therefore, it is necessary to consider those interventions that manage to maintain the increase in PA achieved during the intervention in the IG, without necessarily continuing to increase during follow-up, as successful.

Regarding measurement tools, despite the fact that self-reported PA is a valid and widely used tool to assess changes in PA with acceptable correlations to accelerometer measurement in CRCS [49, 50], the results should be interpreted with caution. In all but one of the studies included in the SRM [43], changes in PA were reported using self-report questionnaires. This could lead to an assumption of social desirability bias, which could overestimate the effectiveness of the intervention [21, 51]. Conversely, some studies have also identified an underestimation of PA measured by self-reported questionnaires compared to that indicated by an accelerometer [52]. Specifically, this was observed in one of the studies included in this SRM, where the GLTPAQ was used to assess participant eligibility and resulted in substantially lower PA levels compared to those obtained at baseline using accelerometer measurements [43]. It has been argued that the high walkability of the intervention environment could have contributed.

Among the intervention conditions of the included studies, four common features were identified: aerobic activity, use of pedometers, PA diaries and educational materials about healthy habits and cancer. In contrast, the control conditions were characterized by the use of the same educational materials, whereas, in two of the included studies, they also provided pedometers [44] and made calls [45]. These are important aspects to consider as neither of these two studies found statistically significant differences between groups, but CG significantly increased their PA levels during the intervention. This point may suggest that providing pedometers or making calls to maintain participant retention could be enough features to produce increases in PA levels of CRCS immediately after the end of the intervention period, but insufficient to maintain or increase PA in the long term.

Regarding the intervention characteristics, it is important to mention that the inclusion criteria for this SRM were not limited to aerobic activities alone. However, previous reviews have not studied strength training as a potential intervention to increase PA [1, 5, 53]. As a result, it is unclear whether strength training can be useful as a strategy to improve adherence to PA. Insufficient adherence rates to strength programmes may be a reason why it is not commonly used as an intervention to improve PA behaviour, let alone expected to do so in the long term. A recent study implemented strength training in frail CRC patients over 70 years old and found a low adherence rate to the programme [54]. In addition, as previous studies have shown, there are difficulties in monitoring strength training in a way that reflects changes in PA levels [24]. Therefore, studies are necessary to determine the effectiveness of this type of training in changing PA behaviour and maintaining the changes over time. Nevertheless, evidence suggests that combining both types of activities could be an effective intervention that improves fatigue, depression, health-related physical fitness, body composition, quality of life and survival in CRCS [5, 55, 56], as well as producing short-term positive changes in PA levels [18, 57].

Supervision is another important component of the interventions. It is worth noting that the scientific literature reports higher adherence percentages in supervised activities [5]. However, only one of the studies in this SRM carried out the exercise in a supervised way [41], which resulted in the largest effect size for increasing levels of PA post-follow-up. Furthermore, it differed from the other interventions by facilitating access to fitness areas. This result confirms previous studies on the effectiveness of supervised exercise interventions in increasing PA in CRC patients [58]. Although other studies included in this SRM also maintained contact with participants through phone calls [42, 43, 45], text messages [44] or face-to-face meetings [43], it is unclear whether these methods alone are sufficient. A recent review suggested that supervision during the intervention could be necessary to sustain PA changes in the long term, but it could not be sufficient on its own [24]. One reason for this could be the implementation of additional BCT that supervised exercise entails. In accordance with that review and following BCT results of this SRM, some of the additional techniques that could be involved were as follows: instructions on how to perform the behaviour, demonstration of the behaviour or instantly applied feedback on behaviour. Furthermore, supervision could even include the use of other BCT that were not explicitly mentioned in the methodology of the studies but could be applied indirectly, for instance, social support, information about consequences, positive reinforcement as a form of reward or verbal persuasion about capability.

As a secondary objective of the SRM, the aim was to identify the BCT applied in the interventions of the included studies. On average, 17.2 BCT were detected, which is higher than the average found in other reviews, between 7.6 [24] and 10.3 BCT [23]. Studies with lower means obtained wider ranges (2–13 BCT [23], 2–20 BCT [24]) than those found in this SRM (15–19 BCT) with a higher mean. The reason could be that this SRM only included studies that incorporated the use of some TMBC for the design of their interventions, resulting in a higher number of implemented BCT. Our finding was consistent with the review by Avery et al. [59] who suggested that interventions supported by TMBC and using more than 10 BCT could positively affect PA behaviour.

Although the interventions were designed on the basis of 5 different TMBC, 12 common BCT were found, of which 8 coincide with techniques reported more frequently in other studies [20, 21, 23, 24]: goal setting, action planning, review of behavioural goal(s), feedback on behaviour, instructions on how to perform the behaviour, information about health consequences, practice/rehearsal of the behaviour and self-monitoring. In contrast, the same studies mentioned that other techniques such as problem-solving, social support, graded tasks and behaviour modelling were also frequently used. While the latter BCT were not common in all studies of this SRM, they are detected in at least one of the studies classified as “very promising”, with the exception of behaviour modelling, which was not applied in any of them.

It should be noted that the differential techniques demonstration of behaviour and restructuring the physical environment found between the “very promising” or “promising” and “not promising” studies did not coincide with those reported in the literature, which corresponded to action planning, social support and graded task [21, 23]. However, this result could have been highly influenced by the small number of studies included in this SRM. Furthermore, the discovery of two BCT that were not typically found in previous studies suggests a potential area for future research into the applicability of these techniques, which could be previously unknown but could have influenced our results.

Despite only BCT applied in the IG were identified, the provision of educational materials, pedometers or phone calls to maintain CG participants retention also could implicate the use of some BCT that were not being considered. Even in some studies [41], BCT were maintained during the follow-up period. Studies on inactive healthy adults have reported that an average of 5 BCT were applied in the CG leading to some change in PA behaviour in subjects who did not receive the intervention [46]. For this reason, future research should aim to identify the BCT used not only in the IG but also in the CG and during follow-up periods. This may explain why both groups experienced an improvement in PA levels, making it difficult to find significant differences between the groups. Additionally, analysing the BCT used in both groups could aid in understanding which techniques are crucial in generating increases in PA when behavioural improvement is observed in both the IG and CG.

The primary limitation of this review is the small number of studies available on the effectiveness of combined PA and BCT interventions in promoting long-term PA behaviour among CRCS. The current evidence is reduced, with concerns regarding blinding, randomization and outcome measures. Further studies with higher methodological quality are required to gain a better understanding of the efficacy of these interventions. In addition, the clustering of PA measures collected at various points throughout the follow-up period in the meta-analysis may have introduced another potential source of limitation in the study. On the other hand, we were not able to compare baseline and follow-up outcomes with post-intervention ones in the meta-analysis because there was not enough data to do so. If we had been able to analyse this, we would have been able to see whether the improvements in PA up to post-intervention were maintained, increased or decreased compared to those achieved post-follow-up, and therefore, whether people were benefiting from optimal PA adherence or whether it was lost over time. For this reason, it is necessary to conduct more randomized controlled trials that not only examine improvements in PA after interventions but also investigate the maintenance of these changes over time. Furthermore, trials comparing PA interventions, BCT interventions and the combination of both are needed to determine whether long-term adherence to PA in this population is determined by the specific PA intervention, by BCT or by the need of both.

This SRM has also important strengths. Firstly, it is the first SRM to bring together evidence on PA and BCT interventions with a follow-up of 3–12 months in CRCS. Secondly, the SRM has a moderate certainty of evidence due to low inconsistency, indirectness and imprecision. Finally, the detailed analysis of the intervention and control conditions and the BCT implemented allowed us to identify possible key characteristics for PA adherence in CRCS.