Introduction

Cancer is a leading cause of morbidity and mortality worldwide, with approximately 14 million new cases and 8.2 million cancer-related deaths in 2012 [1]. Early detection and increasingly effective treatments have led to improved survival rates for cancer. There are approximately 13 million cancer survivors in the USA today, a figure projected to increase to almost 18 million by 2022 [2]. The increase in cancer survival means that many people are coping with the side effects of cancer treatment for longer. The role of health-care professionals (HCP) has therefore expanded to include the promotion of good health behaviours to ameliorate side effects of the disease, of the treatment and of the associated lifestyle changes. One such behaviour is daily physical activity (PA).

The benefits of PA in cancer patients are well documented, including improvements in quality of life, function and a possible reduction in risk of recurrence in some cancer types [3]. Despite those known benefits, adherence to PA guidelines by cancer survivors is poor from the time of diagnosis through to survivorship [4, 5]. These low levels of PA in cancer survivors have prompted clinicians to explore novel approaches to optimise PA levels among this group.

eHealth is an emerging concept in healthcare which may present opportunities to improve PA in cancer survivors. The World Health Organisation (WHO) defines eHealth as the transfer of health resources and health care by electronic means [6]. This includes, but is not limited to, the delivery of health information through Internet and mobile technologies. Suggested benefits of using Internet technology in healthcare include convenience for users, easy storage of large amounts of information, ease of updating information and ability to provide personalised feedback [7]. A number of systematic reviews have been published which primarily focused on eHealth-based PA interventions in community-dwelling adults or in general populations from paediatric to older age groups [8,9,10,11]. Results consistently supported the effectiveness of eHealth interventions for promoting PA in those populations. One of the first systematic reviews which examined the role of technology in promoting PA in clinical populations was in type II diabetes [12]. This review reported increases in PA, highlighting the potential of eHealth among specific clinical populations.

To our knowledge, no systematic review has synthesised the literature on eHealth interventions to increase PA in people with cancer. The objective of this systematic review was to address this gap, by investigating the effectiveness of eHealth interventions to increase PA among cancer survivors.

Methods

Study design

This systematic review was conducted to identify eHealth interventions with a primary or secondary aim to increase PA in people with cancer. The systematic review follows guidelines of the “Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)” statement [13] and meets the criteria outlined in “A Measurement Tool to Assess Systematic Reviews (AMSTAR)” checklist [14]. A protocol outlining the planned search strategy and method of analysis for this review was registered online with a PROSPERO, a registry of systematic reviews (CRD42016037593).

Eligibility criteria

Experimental studies (randomised control trials, pre-post design, quasi-experimental) and observational studies, with or without controls, were eligible for inclusion if they evaluated an eHealth-based intervention (Internet and mobile technologies) delivered to cancer survivors and included PA as a primary or secondary outcome measure. Single or multi-modal interventions were included. Studies were excluded if only telephone calls, SMS or conference calls were used. Review articles were excluded.

PA is a complex multi-dimensional construct which is challenging to measure accurately [5]. PA can be measured objectively (e.g. indirect calorimetry, accelerometers, pedometers) or by using self-report methods (e.g. questionnaire, logbook). Domains of PA can be considered on a continuum from light activity (e.g. slow walking, playing most musical instruments) through to moderate level activity (e.g. brisk walking, recreational badminton) and vigorous activity (e.g. jogging, fast bicycling). Sedentary behaviour is generally referred to as low levels of activity, similar to resting levels (e.g. watching television or lying down) [15]. There are many different ways of quantifying PA. This review included PA measured by self-report, objective or direct methods and expressed PA in a number of ways, including but not limited to, MET-minutes.week−1, minutes in light, moderate and/or vigorous PA per week and meeting/not meeting PA guidelines (150 min per week of moderate/vigorous activity) [16].

Data sources and search strategy

A comprehensive search strategy was designed in collaboration with a senior medical librarian with specialist knowledge in systematic review searching (DM). The search strategy consisted of a search of six electronic databases: PubMed, CINAHL, EMBASE, PsychInfo, Web of Science and SCOPUS. Search terms included keywords and medical subject headings adapted for each database. These related to three categories: (1) the condition (e.g. ‘cancer’, ‘neoplasm’, ‘tumour’, ‘cancer survivor’), (2) technology (e.g.‘teleHealth’, ‘telerehabilitation’, ‘mobile health’, ‘Mhealth’, ‘eHealth’, ‘e-health’, ‘mobile technology’, ‘smartphone’) and (3) PA (e.g. ‘exercise’, ‘physical activity’, ‘exercise therapy’, ‘physiotherapy’). No limit on the year published was applied as it was anticipated that the search strategy would produce only articles published in approximately the last 10 years, due to the burgeoning nature of this technology. Databases were searched until March 2017. A grey literature search using Google Scholar and WorldCat search engines was performed; government reports were searched using the Google search engine and a combination of key word text. The bibliographies of all investigations selected for the review, as well as those of previous reviews, were examined to identify further studies. The search strategy is available online (Online Resource 1).

Selection of eligible studies

Articles were retrieved and all duplicates were removed. Two researchers (CH and JM) independently screened titles and abstracts to identify studies potentially meeting the eligibility criteria. Any disagreements between researchers were resolved by consensus and/or discussion with a third researcher (JB). Full texts were retrieved and examined in detail to assess for inclusion in this review. Two researchers (CH and JM) independently screened these full texts to identify studies to be included in the final analysis. As with the screening of the titles and abstracts, any disagreements between researchers were resolved by a third researcher (JB).

Risk of bias

Two researchers (CH and JM) independently appraised the risk of bias of included studies; in cases where between-researcher disagreements could not be resolved by discussion to achieve consensus, a third reviewer (JB) arbitrated. The Downs and Black checklist was used to assess the risk of bias of observational studies [17]. This checklist contains 27 items, with a maximum possible score of 31 points. The Cochrane Collaboration’s tool [18] was used to assess risk of bias for the remaining studies,which includes the following domains: sequence generation (randomisation); allocation concealment; blinding of participants, personnel and investigator; incomplete data (e.g. losses to follow-up, intention-to treat analysis); selective outcome reporting; and other possible sources of bias.

Data extraction and analysis

Data was extracted by two researchers (CH, JM) onto standardised data abstraction forms. Disagreements between researchers were resolved by discussion to achieve consensus. Failing agreement, a third member of the research team (JB) arbitrated. Data was extracted using the following headings: methods, allocation, blinding, duration, design, setting, participants, diagnosis, age, sex, inclusion criteria, exclusion criteria, intervention, control group, primary outcomes, secondary outcomes, results in PA outcomes, results in secondary outcomes.

Aggregation of results through quantitative synthesis was planned; however, these were not completed due to the heterogeneity of studies in terms of study design, participants, interventions and outcomes. Consequently, a narrative synthesis of study interventions and results was completed. A number of sub-group analyses were also planned, including the following: self-report and objectively measured PA; intervention focus (smart phone applications vs. Web-based interventions); study design (control vs. no control group, randomised vs. non-randomised controlled trial). Ultimately, these could not be conducted due to insufficient data in the included studies.

Results

Study selection

A total of 1065 articles were identified using the searches described. Following the first screening of titles and abstracts, 43 articles remained. After review of the full-text versions of these articles, 10 studies, published between 2012 and 2017, were included in the review. The PRISMA flow chart [13] below (Fig. 1) summarises the search strategy. Randomised controlled trials (RCTs) predominated (n = 7) [19,20,21,22,23,24,25], while the remaining studies were non-controlled trials [26,27,28]. Table 1 describes methodological features of these studies.

Fig. 1
figure 1

PRISMA flow diagram

Full size image
Table 1 Study methodology
Full size table

Participant characteristics

Participant characteristics are summarised in Table 2. In total, 1994 participants were initially recruited into the included studies, 671 of these were control group participants, with an overall drop-out rate of 34.7% across studies. Just over 87% of participants were female (n = 1744); reflecting the fact that the majority of studies included patients with breast or endometrial cancer.

Table 2 Participant characteristics
Full size table

Risk of bias of included studies

The Cochrane Collaboration’s tool [18] for assessing risk of bias was used to to evaluate the seven included RCTs. Using the Cochrane tool, the overall risk of bias in the studies by Kyung Lee et al. [19], Short et al. [25], Sturgeon et al. [24] and Kanera et al. [23] was assessed as low, while the studies by O’Carroll Bantum et al. [21], Hatchett et al. [20] and Eun Uhm et al. [22] were rated as ‘unclear risk of bias’. Two non-randomised studies [26, 27] were scored for risk of bias using the Downs and Black tool [17], scoring 12, 13 respectively out of a possible score of 31. As some sections of the Downs and Black tool were not applicable to the Hooke et al. study [28], it scored 14 out of a total of 22. All 3 of these studies rated poor in quality, according to scoring categories defined by a review conducted by Hooper et al. [29]. This review stated that scores of 26–28 rated as excellent quality, 20–25 rated as good quality, 15–19 rated as fair quality and a score of ≤ 14 was considered poor quality.

Study design

A number of different study designs were employed, likely reflective of this emerging research field within cancer. Four studies investigated the effect of an eHealth intervention on the PA of cancer survivors when compared to a control group. The effectiveness of an eHealth intervention compared to current conventional programmes designed to improve PA was considered in two studies, while the 3-arm RCT by Short et al. [25] investigated their eHealth intervention over varying lengths of delivery. The remaining studies, three in total, did not use a control group, but were pilot studies investigating the feasibility of their respective eHealth PA interventions, again reflecting the novelty of such strategies.

The length of interventions ranged from 14 days to 12 months. Half of the studies (n = 5) reported short-term follow-up only, while maintenance was assessed in four studies [21, 23,24,25], with 6-month [21, 25] and 12-month follow-up reported for two studies [23, 24]. Short et al. [25] made reference to a 6-month time-point, but data was supplied for the 12-week time-point only. The study conducted by Hooke et al. [28] had the shortest intervention period of 2 weeks.

eHealth interventions

A variety of eHealth platforms designed to increase PA were described in these studies: Web-based (n = 5), Web and mobile application (n = 4) and e-mail-based (n = 1). A breakdown of each technological intervention is detailed in Table 3 below.

Table 3 Features of e-Health intervention
Full size table

In total, of the five studies utilising a Web-based intervention alone, three of these investigated the effect of an eHealth intervention on the PA of cancer survivors when compared to a control group. All five studies with a Web-based intervention reported an increase in self-reported PA or exercise. The composition of each Web-based intervention was generally similar across the studies, with all Web-based interventions including an additional educational element. This included information on PA guidelines in all studies, as well as dietary guidance in four of the studies. The study by Short et al. [25] focused on PA only, with no diet element included.

In contrast, the majority of the studies (three out of four) using mobile applications focused solely on a PA intervention, with no other element included.

Physical activity assessment

Eight studies reported significant improvements in their respective PA and exercise outcome measurements [19,20,21,22,23,24,25,26]. Diverse methods were employed to assess physical activity. One study measured PA objectively using a Fitbit to assess step count [28]. PA was assessed using self-report methods in the other nine studies (see Table 4). These included six different self-report questionnaires: 7-day PA recall instrument [20], two forms of the Godin questionnaire (Godin Exercise Questionnaire, Godin Leisure-Time Exercise Questionnaire [21, 25], Short-form International Physical Activity Questionnaire [22], Short Questionnaire to Assess Health Enhancing Physical Activity (SQUASH) [23] and Modifiable Physical Activity Questionnaire [24]. A self-log method was used in three studies: logging was through mobile application [27], completion of an exercise diary [19] or rating on a five-point scale [26].

Table 4 Physical activity outcomes
Full size table

Discussion

This systematic review comprehensively evaluated the effect of eHealth interventions on PA in cancer survivors. Overall, the review suggests that eHealth interventions may increase PA in cancer survivors, with the majority of studies (8/10) reporting improvements in PA and exercise.

eHealth in general is a rapidly emerging area of healthcare, with this review showing that the challenge now is to ascertain the optimal manner in which to integrate it into clinical practice. This is true also for eHealth interventions in the area of PA promotion in cancer survivors, with the current research evaluated by this review presenting a variety of eHealth delivery methods. The use of Web applications to deliver the intervention was the most popular delivery method, used in five of the studies as the sole delivery method, but also used in four further studies in conjunction with a mobile delivery method. Lifestyle-related mobile applications are ubiquitous in ‘non clinical’ settings, but it appears from these results, considering the low number of studies identified, that harnessing the potential of mobile-based applications in clinical practice settings with cancer survivors lags behind. Study authors did not describe whether participants accessed Web applications via mobile devices. E-mails to promote PA were only incorporated in one study [20], this study being the oldest study included in this review. This signals the rapid growth and progression of application-based multi-modal eHealth interventions, be it Web or mobile-based, which were represented strongly in this review. It is unclear if Web-only-based interventions offer more potential to increase PA compared to mobile applications or email-only interventions. Future research may investigate the optimal eHealth medium to increase PA among cancer survivors.

Further variation between studies was also seen when comparing the duration of the interventions. Short-term programmes such as the 14-day intervention of Hooke et al. [28] and the 4-week programme of Kyung Lee et al. [19] may not have been long enough to embed behavioural change. Perhaps notably, a slightly longer 6-week programme [21] reported a significant increase in vigorous level PA. However, the short-term nature of the study still means that it is unknown whether any increases in PA behaviour translated into longer term benefits. The long-term benefits and effectiveness of an eHealth programme to improve PA was, however, investigated in the two studies which had 12-month follow-up of patients [23, 24]. Both studies showed significant improvements in self-reported PA between intervention groups and controls at 12 months, providing valuable information on maintenance of behaviour in this patient group. It is particularly important in future studies to consider the importance of a follow-up period, such as that adopted by the two aforementioned studies, especially when the outcome to be measured, PA, is behavioural in nature and the aim is to affect a lifestyle change. In the context of this review, due to a wide range of intervention durations, we were not able to provide any firm conclusions on optimal intervention length.

No adverse effects were reported in any of the studies included in this review but caution must be applied. Such interventions could potentially cause harm if inappropriate advice is provided, desired behaviour is undermined or if data is shared inappropriately [30]. All studies included were published in the last 5 years, again demonstrating the recent emergence of this research field within cancer. One of the difficulties or barriers that cancer researchers and clinicians may have in integrating eHealth in their practice may stem from the rapid progression of the area, where by the time a certain technology is researched the next new eHealth initiative is available. In this case, particularly in the promotion of PA, an eHealth intervention with a sound grounding in behavioural change theory may provide the foundation for an effective intervention, regardless of technology. As demonstrated in the results above, the majority of included studies had considered behavioural change theory and had successfully implemented components of behavioural change research into their interventions, showing the flexibility of eHealth in delivering a PA intervention. In wider eHealth literature, these results are mirrored, with a systematic review on PA eHealth interventions in cardiovascular disease reporting consistent use of behaviour change theories in 23 studies which described PA promotion [31].

Three studies in this review [26,27,28] did not employ control groups. While ideally pragmatic RCTs would be employed to evaluate new interventions, emerging literature advocates for a more fluid cycle of development and testing in such a rapidly changing context [30], as the time taken to conduct and publish RCTs means their eventual relevance is likely to be limited [32]. Studies were analysed using traditional statistical methods, but it has been suggested that approaches such as Bayaesian analyses may be best suited to this dynamic field [30]. A further sign of the infancy of this area was shown in a recent review by Harvey et al. [33] which investigated eHealth weight loss interventions in cancer survivors. The total number of eHealth studies identified was 5, with 3 of those being feasibility, pilot or single arm studies.

The underlying theme of heterogeneity between studies continued with the lack of consistency in how PA was reported in the included studies. As mentioned above, nine studies used self-report methods of assessing PA. This raises a significant likelihood of self-report bias, even if unintentional, as participants are being “cued” to think about PA. The use of direct monitoring devices (e.g. accelerometry) to measure PA could reduce this and other inherent limitations of self-report PA measurement methods [5].

A further consideration of the studies included in this review is the variability in baseline PA levels reported between each respective study. While some studies reported low baseline PA levels among their sample [20, 26, 27], there were a number whose participants reported high baseline activity levels, with some exceeding PA guidelines (see Table 4 above). It should be borne in mind that if PA levels are low at baseline, it can typically be easier to see a significant change, with the opposite being the case when the baseline PA levels are high.

The drop-out rate, as demonstrated in Table 2 above, was generally low across the studies. This is an important aspect for assessing the success of an intervention, with high drop-out rate potentially indicating an intervention that may not be suitable for the chosen population. Generally across the studies, only Short et al. [25] and McCarroll et al. [27] had high enough drop outs to warrant further concern regarding the efficacy and suitability of the programme. The study by Short et al. [25], which reported a drop-out rate of 68%, conducted its study exclusively online, with the recruitment, intervention and follow-up conducted via a website and email. The results of this study may indicate that while a study, and an intervention, conducted exclusively online may present the opportunity for a high number of potential patients for recruitment, the absence of any face to face contact with a healthcare professional may become a limitation in the retention of these patients.

This review suggests a number of ways the conduct and reporting of future eHealth studies of PA in cancer survivors could be improved. Future studies should improve measurement of PA by the use of objective measures such as pedometers and accelerometers. eHealth studies should adhere to better reporting of technological interventions to ensure that interventions can be replicated [34]. If feasible and effective, eHealth interventions may be a more scalable option to improve PA than one-to-one interventions [35] but unique challenges of this medium include pace of development, engagement with intervention and regulatory, ethical and security requirements [30].

Conclusion

The studies discussed above may constitute the first attempts to embedding technology into the cancer rehabilitation setting. Although some studies within this review showed promising results, methodological considerations pertaining to this evolving field, largely short-term follow-up, heterogeneity in interventions and varying self-report PA measures all weaken the interpretability of these studies. This means that the independent effect of individual programme components cannot be elucidated with any certainty.

This systematic review is the first, to our knowledge, to review the effectiveness of eHealth interventions in increasing PA levels among cancer survivors. Its findings provide a contemporary and reliable research base and identify gaps in this developing area to support researchers, policy makers and other stakeholders as they design and implement effective eHealth interventions to increase PA levels in cancer survivors.