Digital health behaviour change interventions targeting physical activity and diet in cancer survivors: a systematic review and meta-analysis

Over 14 million people are diagnosed with cancer worldwide each year, and this is expected to rise to 22 million over the next two decades [1]. Improvements in early diagnosis and treatments mean that cancer survival is increasing. In 2012, globally there were 32 million people living beyond 5 years of diagnosis [2] and in the UK, half of people diagnosed with cancer will now survive for more than 10 years [3].

However, long-term negative consequences of cancer and treatment related side-effects are common and often debilitating. Prevalence of fatigue following a cancer diagnosis ranges from 59 to 100% depending on cancer type [4], and pain [5], sleep problems [6], physical side effects (e.g. lymphoedema) [7], weight gain [8], anxiety and depression [9, 10], fear of cancer recurrence [11] and impaired quality of life (QoL) [12] are all commonly reported. Macmillan Cancer Support, a UK cancer charity, estimates that more than 70% of cancer survivors in the UK (~1.8 million people) are also living with at least one other long-term comorbidity [13]. The most common comorbid conditions are hypertension, obesity, mental health problems and chronic heart disease [13]. The shared risk factors between cancer, obesity and cardiovascular disease (CVD) partially explain comorbidities [14]. However, there is also emerging evidence to suggest that cancer treatment can leave survivors at greater risk for developing these conditions (e.g. due to cardiovascular toxicity of cancer therapy [15]). The greater number and severity of comorbidities is linked to greater risk of death and cancer recurrence among cancer survivors [16]. There is now strong impetus to develop interventions that improve long-term outcomes for cancer survivors.

Health behaviours such as physical activity (PA), sedentary behaviour and diet are important in risk reduction and self-management of cancer, CVD and obesity. For example, a meta-analysis of 22 prospective cohort studies of 123,574 breast cancer survivors found that greater post-diagnosis PA participation reduced all-cause (hazard ratio [HR] = 0.52, 95% CI 0.43, 0.64) and breast cancer-specific mortality (HR = 0.59, 95% CI 0.45, 0.78), and breast cancer recurrence (HR = 0.79, 95% CI 0.63, 0.98) [17]. A meta-analysis of prospective studies of colorectal cancer survivors reported similar conclusions and showed that post-diagnosis PA reduced both all-cause (summary relative risk [RR] = 0.58; 95% CI 0.48, 0.70; n = 6 studies) and colorectal cancer-specific mortality (summary RR = 0.61; 95% CI 0.40, 0.92; n = 5 studies) [18]. The authors estimated that each 10 metabolic equivalent task (MET)-hour per week increase in post-diagnosis PA was associated with 24% (95% CI 11–36%) and 28% (95% CI 20–35%) decreased total mortality risk for breast and colorectal cancer survivors, respectively [18]. Mishra et al.’s meta-analysis of non-digital interventions focused on the effect of PA on health-related QoL (HRQoL) outcomes in cancer survivors (various types) and found that greater PA participation significantly improved overall HRQoL at up to 12 weeks of follow-up (11 studies, n = 826; standardised mean difference [SMD] = 0.48, 95% CI 0.16, 0.81) [19]. Individual meta-analyses of other cancer-relevant outcomes identified in this same Cochrane review also found that PA interventions improved emotional well-being/mental health and social functioning, and reduced anxiety, fatigue, pain and sleep disturbance [19]. Although limited to cross-sectional and prospective studies, there is some evidence that higher levels of sedentary time are associated with lower physical and role functioning domains of QoL, and greater reporting of comorbidities, disability and fatigue [20‐22]. As a result of the growing evidence of the benefits of PA following a cancer diagnosis, cancer survivors are encouraged to avoid inactivity as far as possible and to meet the same PA guidelines as the rest of the adult population of at least 150 min of moderate-vigorous PA (MVPA) and two instances of strength/resistance-based exercises per week [23‐25].

Diet may also influence outcomes following a cancer diagnosis. A meta-analysis of three studies (n = 9966) suggested that a low-fat diet post diagnosis can reduce breast cancer recurrence by 23% and all-cause mortality by 17% [26]. Another meta-analysis of four prospective cohort studies (n = 3675) found that high saturated fat intake increased breast cancer-specific mortality [27]. A meta-analysis of 56 observational studies in 1,784,404 cancer survivors (various types) showed greater adherence to a Mediterranean-style diet (largely based on vegetables, fruits, nuts, beans, cereal grains, olive oil and fish) was associated with lower all-cause cancer mortality for colorectal, breast, gastric, prostate, liver, head and neck, pancreatic and respiratory cancers [28]. Colorectal cancer survivors consuming a Western diet (high intake of meat, fat, refined grains and desserts) showed greater risk of recurrence and overall mortality compared to those with a prudent diet (high intake of fruits and vegetables, poultry and fish) in a prospective study of 1009 participants [29]. Similar findings have been shown in other prospective cohort studies of breast cancer survivors [30, 31]. Breast cancer survivors with better overall diet quality also reported lower levels of fatigue, independently of PA participation, at 41 months post diagnosis in a prospective cohort study [30].

Longitudinal studies have shown that obesity increases the risk of cancer recurrence among prostate [32], colorectal [33] and breast [34] cancer patients.

Despite the wealth of evidence, cancer survivors’ engagement with health behaviours and adherence to lifestyle guidelines for cancer survivors are remarkably poor [35, 36]. The English Longitudinal Study of Ageing demonstrated that the proportion of cancer survivors who engaged in self-reported MVPA at least once per week fell from 13% before their cancer diagnosis to 9% after their cancer diagnosis (compared to a fall of 16 to 15% in the group who did not receive a cancer diagnosis between data collection waves) [37]. Wang et al. found that cancer survivors were less likely to engage in self-reported PA (adjusted odds ratio = 0.79, 95% CI 0.67, 0.93) compared to those without a cancer diagnosis [38]. Furthermore, few cancer survivors meet the minimum recommended guidelines of 150 min of MVPA per week. A study of over 9000 survivors of six types of cancer found that adherence to PA recommendations varied from 30% (uterine cancer) to 47% (skin melanoma); however, this study did use self-reported PA measures [36]. While this study reported that 35% of breast cancer survivors were meeting guidelines, another study which used accelerometers to measure PA objectively found that this can be as low as 16% and those with highest levels of comorbidities were the least active [39]. Consequently, there is a need for evidence-based interventions that are easy to access, low-cost and which therefore have the feasibility to be rolled out to reach a large number of cancer survivors.

A move towards digital health behaviour change interventions (DBCIs) has been driven by widespread and rising use of the Internet, smartphones and mobile technology [40, 41]. The most recent Ofcom Communications Market report for the UK has shown that the proportion of adults going online using a mobile phone has risen from 20% in 2009 to 66% in 2016 and 71% of adults own a smartphone [41]. DBCIs use technologies such as text messaging, email, mobile applications (apps), video-conferencing (e.g. Skype), social media, websites and online patient portals increasing access to information, connecting patients with health services and as an approach to remote delivery of behaviour change interventions. DBCIs have been used in the promotion of medication adherence [42], management of long-term conditions [43‐45], promoting smoking cessation [46] and promoting PA participation and dietary behaviours [47‐50]. A recent systematic review of 224 studies reported that Internet and mobile interventions improved diet, PA, obesity, tobacco and alcohol use up to 1 year [51]. Among cancer survivors, a recent systematic review of 27 non-face-to-face intervention studies found telephone interventions as an effective approach to delivering PA and dietary interventions [52]. However, newer digital technologies should now be evaluated in this population as only three of the studies in that systematic review included used web-based methods to deliver the intervention [53‐55]. No systematic review or meta-analysis has assessed the efficacy of DBCI interventions targeting PA, diet and/or sedentary behaviour among cancer survivors specifically.

Therefore, the primary objective of this study was to perform a systematic review and meta-analysis of health behaviour interventions using digital technologies in cancer survivors in order to assess their efficacy in promoting PA, reducing sedentary behaviour or improving dietary quality. Secondary aims were to explore any effects of DBCIs on BMI/weight, other cancer-relevant outcomes and the theoretical underpinning of included studies.

The current meta-analysis found that DBCIs resulted in an increase in MVPA participation of approximately 40 min per week. While meta-analysis was not possible for dietary outcomes, there was mixed evidence for an effect on dietary intake. No studies assessed sedentary behaviour. Meta-analyses also revealed a significant reduction in BMI, a reduction in fatigue which did not reach statistical significance, and no change was seen in cancer-specific measures of QoL. For other secondary outcomes where meta-analysis was not possible, there was mixed evidence for the effect on domains of generic QoL measures and theoretical constructs (e.g. self-efficacy). There is no evidence for an improvement in anxiety or depression, and while only two studies assessed sleep disturbance, both reported a significant improvement.

To our knowledge, this is the first meta-analysis to assess the current evidence with regards to DBCIs targeting PA and/or diet among cancer survivors. An increase in approximately 40 min of MVPA per week is important given that this represents a substantial proportion (27%) of cancer survivors’ recommended weekly MVPA participation [23‐25]. While there is limited evidence on the dose-response relationship between MVPA and mortality and recurrence outcomes, Schmid et al. estimated that each 10 MET-hour per week increase in post-diagnosis PA (approximately equivalent to the 150-min MVPA/week guideline) was associated with 24% (95% CI 11–36%) and 28% (95% CI 20–35%) decreased total mortality risk for breast and colorectal cancer survivors, respectively [18]. Schmid et al. also reported that breast or colorectal cancer survivors who increased their PA by any amount between pre and post diagnosis showed a decreased total mortality risk (RR = 0.61, 95% CI 0.46–0.80) compared to cancer survivors who did not change their PA level or who were insufficiently active pre diagnosis [18]. Therefore, even small increases in MVPA post diagnosis are likely to be beneficial for cancer survivors.

It is noteworthy that a third of the included studies were published in 2016, illustrating the rise in research interest of the effectiveness of DBCIs. It is interesting to compare the findings of this study with other meta-analyses using non-digital PA interventions among cancer survivors. While we did not find a significant reduction in fatigue (SMD = −0.23), Mishra et al. did find a significant improvement, both for survivors who had completed treatment (SMD = −0.82) [19] and who were still undergoing active treatment (SMD = −0.73) [74]. We also did not find any improvement in cancer-specific measures of QoL, while Mishra et al. reported positive improvements in both of their meta-analyses. Mishra et al. also found significantly improved anxiety and depression, sleep quality and improvements in some domains of generic measures of QoL (e.g. social, physical and role functioning) [19, 74]. The small number of studies assessing these outcomes meant that meta-analyses for these outcomes were not possible in the current study. However, in the studies that did assess these outcomes, we did not find any evidence for an improvement in anxiety and depression, but both studies assessing sleep reported significant improvements. We also found mixed evidence for individual domains of generic QoL measures. It is possible that the non-digital interventions included in Mishra et al.’s studies result in larger effect sizes as many of the interventions are supervised by trained staff or involve some level of human interaction, which may foster higher levels of engagement and adherence to the intervention. However, due to the small number of low-quality studies included in this review, there is a need for more high-quality RCTs, with objective measures of PA, long-term follow-up and larger sample sizes before reliable comparisons between non-digital and DBCIs can be made.

Sustained engagement with DBCIs was a significant problem for a number of the studies included in this review. For instance, in the study conducted by Short et al., retention at the 3-month follow-up was only 32% (156/492) and 11% (53/492) at the 6-month follow-up [72]. Furthermore, while 75% of the sample completed at least one action plan, the average number of action plans completed was only 2.2. Similarly, 50% of participants completed the week 1 module compared to 10% for the week 9 module in Forbes et al.’s study [63]. A systematic review has shown that there is a positive relationship between participants’ adherence to/engagement with digital interventions and positive physical health outcomes across a range of populations and behaviours, suggesting that efforts to improve effective engagement with DBCIs could improve behaviour change outcomes [75]. It is possible that suboptimal engagement with the DBCIs in the studies included in this review may explain the reduced effects on outcomes compared to those observed in Mishra et al.’s review of non-digital PA interventions [19, 74]. Future DBCI studies should integrate techniques or components that maintain effective engagement with the intervention for its duration. There is some evidence that technology-based strategies (e.g. reminders, prompts) can encourage user engagement [76]. Other aspects which have been identified as important for engagement include ease of use, design aesthetic, feedback, function, ability to change design to suit own preferences, tailored information and unique mobile phone features [77]. Similarly, less time consumption, user-friendly design, real-time feedback, individualised elements, detailed information and health professional involvement may also improve effectiveness of DBCIs, in particular mobile apps [78]. Furthermore, future studies should aim to better understand the link between engagement and effectiveness of DBCIs targeting PA and diet in cancer survivors and define, evaluate and report engagement more consistently so as to better understand techniques that foster effective engagement and mechanisms of action [79].

The majority (10/15) of the studies used an online portal or website to deliver the intervention, and, while one of these websites was mobile-enabled [65], only two studies used mobile apps [68, 69]. This is interesting given the findings of a recent review of 23 interventions using mobile apps that found that 17 of the included studies reported a significant effect on behaviour change in the general population [78]. It is possible that interventions using mobile apps may be more effective than other types of DBCIs due to the widespread usage of and constant access to smartphones and the Internet. The most recent Ofcom Communications Market report conducted in the UK reports that 71% of UK adults own a smartphone and 66% use their smartphone to access the Internet most frequently [41]. Therefore, it would be interesting for future studies to use mobile apps as a mode of intervention delivery and compare the effectiveness of mobile app interventions compared to other DBCIs as they may foster higher levels of engagement.

Self-monitoring, goal-setting, credible source and feedback on behaviour were the most frequently described BCTs used in the included studies. Due to the heterogeneity in intervention type, mode of delivery, behavioural outcomes and measurement approaches, it is difficult to interpret which BCTs were most effective at changing PA/dietary behaviour. Kanera et al.’s study used the most BCTs (n = 16) but also used a tailored if-then algorithm within the intervention to automatically tailor content to participants; therefore, it is difficult to ascertain what intervention components or approaches to delivery lead to increased effectiveness. A recent meta-analysis revealed that theory-based interventions are significantly more effective at improving PA [80]. SCT was the most commonly reported theoretical basis of the interventions; however, several other behaviour change theories were used across the studies. The level of reporting of the extent to which theory was incorporated into the development of the interventions varied across studies, but was generally poor. Only three studies did not report any theoretical underpinning. Future studies should aim to explicitly report how theory is used to develop intervention techniques and tailor the intervention to participants. Measurement and exploration of changes in targeted theoretical constructs (e.g. via mediation analyses) can aid understanding of why interventions may or may not be effective [59] and can be used to refine theoretical models of behaviour change. The lack of clarity about the intervention content and theoretical underpinning and the lack of measurement on theoretical constructs mean it is difficult to unpick how or why the interventions which improved behaviour in this review were effective. Future DBCI studies should clearly report any theoretical underpinning and behaviour change techniques used, for instance by using Michie et al.’s Theory Coding Scheme [59] and Behaviour Change Technique Taxonomy [56].

There are several limitations to this review. Primarily, the data extracted for the meta-analyses reflect unadjusted models. While the forest plot for MVPA minutes/week (Fig. 2) illustrates that Bantum et al. [61], Rabin et al. [54] and Kanera et al. [71] found a significant effect, the original publications show that this no longer remained significant when adjusting for baseline values and/or other covariates (e.g. demographics, disease characteristics) [54, 61], or when adjusted models are controlled for multiple testing [71]. Bantum et al. [61] did find a significant increase in vigorous PA participation in the adjusted model, but not when combined with moderate PA. Similar issues arise for the meta-analyses for fatigue and BMI/weight. The studies that show significant findings for fatigue and BMI/weight in the current meta-analyses do not report significant findings when adjusted for baseline and/or covariates in the original publications. Therefore, pooling the adjusted results would reduce the overall effect size substantially, and it is likely that this would no longer remain significant. Furthermore, a combined MVPA variable was chosen to assess PA outcomes to maximise the number of studies that could be included in the meta-analysis and to coincide with the American College of Sports Medicine’s recommendation that cancer survivors should follow the PA guidelines for the general population (at least 150 min of at least moderate intensity PA per week) [25]. However, this means the current meta-analysis does not differentiate between different intensities of PA: it may be beneficial to evaluate the effect on outcomes at different intensities of PA. Visual inspection of funnel plots suggested possible publication bias for smaller studies assessing MVPA outcomes; it is possible that our choice to only include published studies may have increased publication bias. There was no suggestion of publication bias for BMI/weight, fatigue or cancer-specific QoL measures.

The risk of bias in included studies was high. The inclusion of one-arm, pre-post studies substantially increases the risk of bias; however, we felt that the novelty of this area of research warranted the inclusion of these studies and that this added valuable insight into the current state of the literature. Few studies assessed outcomes other than PA, where only half of the studies could be included in a meta-analysis. Heterogeneity across studies was very high, likely due to the variability of types of DBCIs, intervention content, cancer type and populations, outcome measurement tools, etc. With the addition of future studies, more specific inclusion criteria could be used to assess effectiveness of more similar studies. All of the PA findings used self-report data, which while easy to use, inexpensive and validated, often hugely underestimate or overestimate PA participation [81]. Therefore, we advise that these results are interpreted with caution until the number of published studies assessing DBCIs increases and inclusion criteria for future systematic reviews can be more stringent for low-quality trials.

To the best of our knowledge, this is the first meta-analysis of DBCIs designed to improve PA and diet among cancer survivors. While the review shows some evidence of an improvement in MVPA, a reduction in BMI and a trend towards significance for fatigue, large, high-quality RCTs, with objective measures of PA and long-term follow-up, are lacking. Future studies should aim to address these limitations, but the approach of using digital technology in this context appears promising.