A thorough review of the literature was performed using PubMed, EMBASE, Web of Science, MEDLINE, the Cochrane Library, Scopus, and key Chinese databases, including Wanfang and China National Knowledge Infrastructure, from the inception of databases to February 2025.

Search terms included:

(“cancer” OR “cancer survivors” OR “neoplasms”) AND (“lifestyle intervention” OR “physical activity” OR “exercise” OR “behavior change”) AND (“message” OR “messaging” OR “text message” OR “text messaging” OR “mobile message” OR “mobile phone messaging” OR “short message service” OR “SMS” OR “instant message”) AND (“randomized controlled trial” OR “RCT” OR “clinical trial” OR “placebo” OR “randomized” OR “randomly” OR “trial”).

In addition, the reference lists in published reviews and meta-analyses were examined to identify papers that the electronic search missed. We also performed a search in gray literature, including the preprint studies in the 8 databases, and in OpenGrey, ProQuest Dissertations & Theses, and Electronic Theses and Dissertations (EBSCO Open Dissertations). The research strategy is shown in Multimedia Appendix 1.

This review follows the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (Checklist 1) [26] to ensure a methodical approach to data collection and analysis. The review has been registered at PROSPERO (CRD42024557519), and there were no deviations from the registered protocol.

The inclusion criteria were as follows: (1) the study design should be randomized controlled trials (RCTs), including both full and pilot trials; (2) participants should be adults (≥18 y) with cancer; (3) the intervention must primarily use mobile phone messaging as the main or sole delivery channel of delivering the intervention’s content, instructions, or engagement with participants. In the case of broader interventions, mobile phone messaging must be a central and indispensable element, contributing significantly to the intervention’s intended effect (eg, accounting for ≥50% of communication touch points or engagement time, as reported in the study; (4) outcomes should include PA estimates, such as moderate-to-vigorous PA (MVPA), total PA, and step count; (5) the control or comparison condition should not involve the delivery of messaging; and (6) only studies published in English and Chinese were included. There were no restrictions on the type of cancer, treatment status, or the form of PA. Interventions where messaging is a minor or supplementary component (eg, used only for reminders or scheduling rather than delivering core content) were excluded to maintain focus on messaging-driven interventions.

All records retrieved from the databases were imported into EndNote (Clarivate), and the software’s built-in duplicate identification function was used to remove duplicates. The titles and abstracts of the retrieved papers were independently screened and cross-checked by 2 reviewers (XC and CKC). The full-text studies were obtained and assessed for eligibility against the inclusion criteria by the 2 reviewers independently. Discrepancies were resolved through discussion or, if necessary, adjudication by a third investigator (DSTC). When multiple studies describing the same RCT were identified, the study presenting the primary and most comprehensive results was selected for review.

Relevant data were extracted independently by XC and CKC. Information extracted included participant characteristics (eg, country, cancer types, mean age, gender, and cancer treatment status) and study characteristics (eg, intervention components, control conditions, intervention duration and frequency, PA estimates, theories, number of BCTs [27], and adverse events [AEs]). Regarding BCTs extraction, the standardized BCT Taxonomy v1 [27], a widely accepted framework for identifying and coding intervention components, was used for reliable identification. Data on baseline, after intervention, and follow-up PA estimates (mean and SD or SE or 95% CI), or change in PA estimates from baseline (mean change and SD) were extracted. For trials with more than 2 arms, only data from the relevant intervention group and the control group were extracted.

Risk of bias was assessed by 2 independent reviewers (XC and CKC) using the revised Cochrane Collaboration’s risk of bias tool (Risk of Bias 2) with a Microsoft Excel template [28] on the domains of (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result. Assessments were cross-checked, and any discrepancies were resolved through discussion or, if needed, adjudication by a third reviewer (DSTC).

A grade of recommendation, assessment, development, and evaluation (GRADE) approach was adopted using an online GRADEpro tool to assess the confidence of intervention effects [29]. The assessed domains included risk of bias, inconsistency, indirectness, imprecision, and publication bias.

All analyses of pooled effects were performed using the meta package in R (version 4.3.0; R Foundation for Statistical Computing). Forest plots were used to display the results of individual studies and syntheses. Overall effect sizes were calculated to pool the study results on the standardized mean difference (SMD) of the change in self-reported PA level, objective PA level, and step count between the intervention and control group at postintervention and follow-up. An SMD of 0.2, 0.5, and 0.8 corresponds to small, medium, and large effect sizes, respectively. Primarily, SMD was derived from the mean difference and SD of change. For studies that did not report mean change and SD, SMD was estimated from baseline and postintervention or follow-up values of mean and SD, whereas the SD of change was imputed based on a correlation coefficient (r) derived from the only included study that presented baseline and post-intervention means, SDs, and change (ie, r=0.68) [30]. These methods of calculation were in line with the Cochrane handbook for imputing data for SMD in systematic reviews [31]. For studies that only provided median and IQR for PA levels, mean and SD were calculated based on Luo et al [32] and Shi et al [33]. A positive SMD within the meta-analysis indicated an increased level of PA for intervention groups compared with control groups. A random effects model was used for all outcomes.

Heterogeneity was investigated in each analysis using I² values that range from 0% to 100%, with higher values indicating greater heterogeneity. Heterogeneity greater than 50% was considered substantial [34]. Potential sources of substantial heterogeneity were investigated by subgroup analysis of treatment status (posttreatment vs other status); cancer types (mixed vs single cancer); intervention period (≤3 mo vs >3 mo); message frequency (daily vs less than daily); interactive message (yes vs no); tailored messages (yes vs no), including a wearable device in the intervention (yes vs no); theory basis (yes vs no); and adoption of less or more than the median number of BCTs of studies in this review. Sensitivity analyses were performed through leave-one-out analyses, removal of outliers, and inclusion of only studies with low or some risk of bias [1]. We also assessed publication bias using funnel plots and the Egger linear regression method, with P<.05 taken as an indication of publication bias [35].

Figure 1 shows the study selection process. The search of all databases and identification through other sources resulted in a total of 1658 records. Following the removal of duplicates, the total was 1338 records. We excluded 1274 records based on titles and abstracts. Therefore, a total of 64 studies were assessed for eligibility. Fifty-one studies were excluded after applying the inclusion criteria. The primary reasons for exclusion were the absence of mobile phone messaging as a core component (n=16), a lack of reported PA outcomes (n=11), and an ineligible population (n=9). Other reasons included an ineligible intervention type (n=9), an incorrect study design (n=3), an inappropriate comparator (n=2), and a change in the study protocol (n=1). Subsequently, 13 studies met the criteria for the systematic review and meta-analysis. During the selection process, one discrepancy was resolved by a third adjudicator (DSTC), resulting in the exclusion of a study because mobile phone messaging was not a core intervention component. No eligible study was identified from the Chinese databases. All studies were published between 2018 and 2023, and they were based in 4 countries: the United States (n=9, 69%), Australia (n=2, 15%), France (n=1, 8%), and Ireland (n=1, 8%).

Table 1 summarizes participant characteristics of the included studies. A total of 777 participants from the 13 included studies were included for this review. These studies encompassed multiple cancer types, with breast and prostate cancer being the most prevalent. Specifically, 5 (38%) studies focused on mixed cancer types [30,37-40], 3 (23%) focused on breast cancer [41-43], 1 (8%) investigated endometrial cancer [44], 1 (8%) examined lung cancer [45], 2 (15%) studied prostate cancer [46,47], and 1 (8%) focused on colon or rectal cancer [48]. Four (31%) studies involved solely female patients [41-44], and 2 (15%) only involved male patients [46,47]. The mean ages of the participants ranged from 49.7 (SD 13.7) to 69.8 (SD 8.7) years. Regarding the status of cancer treatment, 69% (9/13) studies involved participants who had completed active cancer treatment, 2 (15%) recruited patients receiving concurrent cancer treatment [39,45], 1 (8%) recruited patients scheduled for treatment [42], and 1 (8%) did not specify or restrict treatment status [47]. Cancer stage was only reported in 8 (62%) studies: 3 (23%) limited to patients with stage I to III cancer [30,42,46], 2 (15%) included patients with stage 0 to III cancer [40,43], 1 (8%) aimed at patients with stage III or IV cancer only [45], 1 (8%) focused on stage II-III cancer [48], and 1 (%) recruited patients at stage I to IV [47]. Five (38%) studies did not report the cancer stages of the participants.

To our knowledge, this study is the first to synthesize the effects of mobile phone messaging–based interventions to promote PA among patients with cancer. The pooled results showed that mobile phone messaging–based intervention is effective for increasing objective PA levels at postintervention, but not for self-reported PA levels or step count. No significant effect was observed on longer-term PA improvement. Subgroup analyses suggested that targeting post-treatment patients with cancer and adopting more BCTs significantly affected intervention effects.

A statistically significant yet modest effect of mobile phone messaging–based interventions on promoting objective PA levels was identified. Notably, previous reviews of messaging interventions among adult populations revealed a relatively smaller effect in improving objective PA level (Hedge g=0.31) [25]. One of the possible explanations may be related to the use of theory. Four out of 5 (80%) RCTs measuring objective PA in this review were driven by theories [30,41,46,48], while only 1 out of 5 studies in the previous review had a theoretical basis [49]. Besides, the previous review targeted a wide range of adult populations, including those with noncommunicable diseases and those without [25]. Individuals with varying health conditions may have different needs for PA promotion, and therefore, the effect of mobile messaging may be obscured. Furthermore, the effect of mobile phone messaging on objective PA promotion as revealed by our study is also relatively larger than eHealth interventions conducted among patients with cancer (SMD=0.19) [15]. This may be explained by the special features of mobile phone messaging, such as low cost, convenience, and high accessibility to real-time reminders and support [19, 50]. Future messaging interventions and eHealth trials should explore how the interventions work, such as by conducting process evaluation and mediation analyses.

Despite the significant effect of mobile phone messaging–based interventions on objective PA, no significant effect on self-reported PA levels or step count was identified. For self-reported PA, it could be affected by information bias (ie, varied understanding of PA estimates) related to individual intelligence and educational level [51] and recall bias [52]. Therefore, self-reported PA measures have been regarded as less valid PA measurement compared to objective measures [53]. For step count, it includes estimates of PA during periods of activity, leisure activity, lower body movements, and sporadic movements in daily life and does not estimate some PA, such as biking and swimming [54]. Therefore, step count provides data on the total volume of ambulatory activity regardless of intensity and is different from the objective PA estimates, which are mostly time-based estimates of PA of particular intensities. These may account for the differences in the effect of mobile phone messaging–based interventions on objective PA and step count.

The results of subgroup analyses revealed 2 factors associated with the effect of mobile messaging on promoting step count. The first factor was targeting patients after cancer treatment. The physical and psychological demands of active cancer treatment can make it challenging for patients to maintain or increase PA levels [55], so PA promotion targeting post-treatment patients is more likely to succeed. The other factor was related to the use of more BCTs. Mobile messaging allows for enactment of a wide range of BCTs through real-time and tailored communication, which may enhance individuals’ engagement in and adherence to PA [19,56]. For BCTs, the included trials adopted a median of 10 (IQR 8.5-13.5) BCTs, with the most frequently used ones being self-monitoring of behavior, instructions on how to perform behaviors, habit formation, and nonspecific reward. With the use of more BCTs, interventions may be able to systematically address key factors that influence behavior change, such as goal-setting, self-efficacy, and self-monitoring [57,58]. Despite the importance of BCTs in behavioral change studies, only 2 of the 13 (15%) included studies explicitly reported the adoption of BCTs [37,46]. Future studies should consider improving the reporting of message development based on the adoption of BCTs and targeting patients who have finished active treatment. Also, more research is needed to determine the optimal combination of theories, BCTs, and intervention regimens (eg, timing and frequency) of messaging to influence PA behaviors.

The sensitivity analyses demonstrated that the overall results were relatively robust. The significant effect of mobile phone messaging–based interventions on objective PA and the nonsignificant effect on self-reported PA persisted after the exclusion of obvious outliers and studies with a high risk of bias. However, the effect on step count, which was nonsignificant in the primary analysis, became significant following the removal of the studies by Kenfield et al [46] and Villaron et al [39]. This shift may be attributed to the high risk of bias in these studies, which potentially introduced distortion into the pooled estimate. Furthermore, leave-one-out sensitivity analysis revealed that the significant effect on objective PA was contingent upon the inclusion of two influential studies: Gell et al [30] and Kenfield et al [46]. The instability observed upon their removal can be explained by their distinct characteristics. The study by Kenfield et al [46] was judged to be at high risk of bias, which may have biased the overall result. In contrast, the study by Gell et al [30], while methodologically sounder, presented an exceptionally large effect size and carried substantial weight in the meta-analysis due to its sample size. More methodologically rigorous trials should be conducted to test the effects of mobile phone messaging-based interventions on PA promotion.

This review has some limitations. First, the geographical scope of the studies in this review was confined to Western and high-income countries, so the generalizability of the findings could be restricted. Second, despite comprehensive efforts, the literature search may have missed potentially relevant papers. Third, the nature of mobile phone messaging–based interventions made it hard to blind participants, which may have biased the effects of the interventions. Fourth, the subgroup analyses were exploratory, with no formal tests for interaction due to the limited number of studies; this approach is prone to ecological fallacy and overinterpretation. In addition, substantial heterogeneity in some subgroups and the small number of studies further hindered the robust interpretation of those findings. Fifth, only 2 studies estimated PA levels at follow-up, leading to unclear long-term effects of mobile phone messaging–based interventions. Finally, although our search included major English and Chinese databases to capture a broad evidence base, the exclusion of studies in other languages (eg, Spanish and Portuguese) introduced a potential for language bias.

Mobile phone messaging–based interventions could be recommended to improve PA among patients with cancer, given their significant effect on objective PA, an outcome that is generally considered to be more valid than self-reported PA and step count [53]. Health care providers should improve their capacity for designing and implementing mobile phone messaging–based interventions on PA promotion. Mobile phone messaging–based interventions that are designed based on theories, adopt various BCTs, and target patients who have finished active treatment are more likely to elicit significant benefits on diverse PA estimates. Importantly, mobile phone messaging–based interventions should be implemented among patients who have no contraindications to unsupervised exercise. Education on safety precautions for home-based exercise should be provided to avoid potential exercise-related AEs.

In terms of research implications, future studies should prioritize more methodologically rigorous RCTs to strengthen the evidence base for mobile phone messaging–based interventions, given that only 30.8% of the included trials were at low risk of bias. Development processes for messages in the intervention should also be rigorous and clearly reported by specifying the theoretical basis and BCTs adopted. Furthermore, studies focusing on exploring differential intervention regimens (eg, message frequency and duration, and wearable devices) should be conducted to promote understanding of the optimal messaging intervention design. Besides, longer follow-up should be considered for future research to estimate the long-term effects of mobile phone messaging–based interventions. Finally, instant messaging is increasingly used in other messaging interventions on behavior change, such as smoking cessation [59]; however, it is not adopted by any of the included studies in this review. Future messaging interventions for promoting PA in patients with cancer can adopt instant messaging as the message delivery platform because it is free of charge, routinely used in daily life, and has a broad range of functionalities (eg, sharing large files in various media formats, such as images, videos, documents, and voice messages).

Our findings indicate that the effect of mobile phone messaging–based interventions varies among different PA outcomes. While messaging significantly improved objective PA with a small effect size, its effect on self-reported PA and step count was insignificant. Mobile phone messaging–based interventions adopting more BCTs and targeting patients who have completed active cancer treatment were more likely to improve step count. More methodologically rigorous trials are needed to test the long-term effect of mobile phone messaging–based intervention on PA and to explore the effects of different intervention regimens.