Background

Physical activity is a well-researched intervention to alleviate health problems associated with breast cancer and reduce the risk of future comorbidities [1]. Improvements can be made in quality of life, fatigue, and physical function, counteracting many commonly experienced symptoms and also lowering the overall risk of all-cause mortality [2, 3]. In the context of physical activity trials which provide professional instruction and/or facilities, it is standard to endeavour to offset inconveniences that come with participation, for example, by assisting with travel or employing experienced study personnel. Accordingly, the key predictors of adherence to exercise interventions for people with cancer include convenience of participation and travel, the presence of professional supervision, and a high frequency of contact with research staff [4, 5]. A recent systematic review of interventions following adjuvant therapy for breast cancer showed that programs typically lasted 8 to 12 weeks, and many did not follow-up and reassess whether participants continued to be physically active after the conclusion of the intervention [6]. As adaptations due to physical activity require ongoing commitment to be maintained, the long-term effects of such trials, once the provided guidance and facilities are removed, are unclear.

Behaviour change theories have been used in physical activity trials for a wide variety of populations to promote self-management and the continuing adoption of health behaviours. These theories outline numerous interconnected processes and factors that are present when changing behaviour, potentially creating a better intervention through identification of factors that may be of particular relevance [7]. Frequently cited behaviour change theories used in health research include the social cognitive theory [8], transtheoretical model [9], and theory of planned behaviour [10]. These theories share some key behavioural constructs, including: self-efficacy i.e. one’s self-belief in their ability to achieve the behaviour change, barriers and enablers, education around the “how” and “why” of the behaviour change, and external social or environmental influences. Underpinned by these theories are behaviour change techniques, which are practical actions or strategies that facilitate change such as cognitive reframing, consideration of health outcomes, or monitoring and goal setting. Behaviour change theories are particularly relevant to programs for breast cancer survivors, due to both the physical and psychological burden of the disease presenting challenges when trying to lead a healthy lifestyle [11, 12].

Physical activity interventions that use behaviour change theories have been explored in women with breast cancer using a wide range of intervention settings and protocols. A previous systematic review, reporting on physical activity behaviour interventions up to 2013, has provided some insight into how the extent of theory use enhances intervention efficacy, while also highlighting the need for further exploration [13]. Hence, this systematic review sought to investigate the benefits of using theory-based physical activity interventions for women with breast cancer by synthesising and evaluating (i) the effects on physical activity and related health outcomes and (ii) the application of behaviour change theories.

Methods

Inclusion criteria and database search

This review was registered with the PROSPERO Register of Systematic Reviews (CRD42019121782). Trials were included if they (1) were a peer-reviewed journal article in English; (2) used a randomised or quasi-randomised study design with a control group; (3) involved women with breast cancer; (4) enrolled women in whom the majority were within three years of their last active treatment (i.e. not including hormone therapy); (5) reported an intervention targeting physical activity, with a minimum duration of 12 weeks; and (5) reported using a behaviour change theory. The criteria for types of interventions were broad and could range from unsupervised lifestyle activity programs to centre-based structured exercise. Multimodal interventions involving health behaviours in addition to physical activity were not included. For studies that reported multiple intervention arms, only arms that met the review criteria were included.

Database search

A search was conducted for published articles in MEDLINE, EMBASE, PsycINFO, the Cochrane Library, CINAHL, Web of Science, and Scopus online databases. Search strategies (Supplementary Material 1) encompassing the inclusion criteria were formulated with the assistance of an academic librarian and cross-referenced with a previous Cochrane review relating to physical activity for women with breast cancer [6]. As records may describe a behaviour change theory in its full text but not in the study title, abstract, or keywords, the search strategies did not include this criterion; instead, use of theory was screened for during the review of full texts. There were no restrictions on publication date. References of the final included papers were also searched. The initial search was undertaken in October 2018, with follow-up searches completed in November 2020 and July 2021.

Selected outcomes

The primary outcome for this review was change in physical activity, measured with questionnaires or objective activity trackers. Secondary outcomes included quality of life, fatigue, and behaviour change variables (e.g., self-efficacy or social support) measured using questionnaires, and physical function, aerobic fitness, and muscular strength.

Trial screening and data extraction

All authors were involved in screening, full text review and data extraction, with each step completed independently by two authors and disagreements resolved by a third author. For publications in which the title and abstract appeared to meet the review criteria, the full text was then reviewed. Study characteristics extracted included aims, eligibility criteria, recruitment methods, intervention and control group details, adherence, and outcome measurement methods. Immediate post-intervention and follow-up (> 12 weeks post-intervention) values were extracted for the selected outcomes, along with information about intervention adherence, such as session attendance. For trials with multiple publications, we collated information across all articles and used the primary article (i.e. with the most complete, relevant outcomes for the review) as the reference article in this review.

Assessment of risk of bias

The revised Cochrane Risk of Bias 2 (RoB 2) was used to assess study risk of bias [14]. The tool covers five domains in which bias may exist: (1) randomisation process, (2) deviations from the intended intervention, (3) missing outcome data (> 5% missing for continuous outcomes), (4) the measurement of outcomes (physical activity), and (5) the selection of the reported results. The tool provides decision tree algorithms that suggest a judgement of either “low risk,” “some concerns,” or “high risk” for each of the 5 domains, and for the overall risk of bias. Risk of bias was assessed independently then discussed by two authors (ML, SK or JY).

Assessment of theory use

To evaluate the extent of behaviour change theory application, items 1–11 of Michie and Prestwich’s Theory Coding Scheme were used [15]. The remaining items were not assessed as they address methodological issues rather than the application of a behaviour change theory. Items 1–6 assess whether a specific behaviour change theory was mentioned, if the intervention was based on a single theory, and if a theory construct was used to inform inclusion criteria (e.g. recruiting low-motivation participants assessed using a theory-based questionnaire), intervention development, and/or the tailoring of the intervention to participants. Items 7–11 assess the link between intervention techniques and theory constructs or other behavioural predictors. The Theory Coding Scheme was scored by two authors (ML and JY).

In addition, the Behaviour Change Technique Taxonomy [16] was used as a means to systematically identify and categorise behaviour change-related intervention components.

Statistical analysis

All extracted continuous outcomes assessed in at least two studies were considered for meta-analyses. Analyses were performed on post-intervention and follow-up means and standard deviations (SD), with the most distal follow-up values selected to reflect the lasting effects of trials. For trials that did not report means or SDs, effect or variability values were imputed from the available data when feasible [17]. Cochrane’s guidelines [18] were followed when deciding what outcome data was appropriate for meta-analysis.

All analyses were carried out using Review Manager (v5.4). Outcomes were combined using the inverse variance random-effects method because, in addition to the possibility of random error, differences were expected from the variation in intervention protocols and population characteristics between the trials (e.g. cancer stage and treatments) [19]. Pooled effect estimates and the corresponding 95% confidence intervals (CI) for intervention versus control arms were calculated for each outcome. Effect size estimates were calculated as standardised mean differences (SMD) to allow synthesis of different measurement methods assessing one outcome (e.g. different self-reported physical activity questionnaires).

To examine the effect of different theories, subgroup analysis was conducted to compare trials according to the theory used. Sensitivity analyses were also performed to discern whether high risk of bias trials or outcomes that required data imputation significantly changed the pooled effect estimates or statistical heterogeneity [20].

Results

Search results

A total of 13,288 records were identified across the database searches (Fig. 1). After removing duplicates, 6593 titles and abstracts were screened, with 485 full texts proceeding to further review. We requested additional data via email from the corresponding authors of four studies; two authors responded, one confirming eligibility and one confirming ineligibility. From the evaluation of the full texts, 40 articles including protocols were identified as being suitable for inclusion in this review, reporting on 19 unique trials.

Fig. 1
figure 1

PRISMA flow diagram

Full size image

Participant characteristics

Details on participant characteristics across the included trials are presented in Table 1. Sixteen trials reported average age of participants with an overall mean age of 52.7 years, with one study reporting no age information [21]. Six trials (32%) recruited women with breast cancer across all stages [22,23,24,25,26,27]. The other 13 trials (68%) recruited women with early-stage or non-metastatic cancer only. Fourteen trials reported cancer stage: stage 0 = 4%, stage I = 38%, stage II = 43%, stage III = 15%, and stage IV = 1%. Eight trials (47%) were conducted while participants were undergoing active treatment [23, 26, 28,29,30,31,32,33], with the remaining trials conducted after completion of active treatment (not including hormone therapy). One trial specifically recruited women who were receiving hormone therapy [34]. Twelve trials (63%) limited inclusion into the trial based on participants’ baseline physical activity level with the allowable level varying widely, ranging from < 5 days a week of exercise to reporting no current physical activity.

Table 1 Summary of study characteristics
Full size table

Intervention and control group characteristics

Details on the interventions are presented in Table 1. Intervention length ranged from 12 weeks to 8 months (mean = 16.5 ± 7.1 weeks). All trials included instruction on home-based physical activity, with six trials also including a centre-based exercise component [28, 30, 32,33,34,35]. Nine trials involved phone calls [23,24,25, 27, 29, 31, 36,37,38] and five trials involved group discussions [22, 30, 32, 33, 35]. Other approaches included a combination of individual counselling sessions and group discussions (n = 2) [28, 34], online/email delivery (n = 2) [21, 27], written material alone (n = 2) [39], and training nurses to engage with participants during treatment visits (n = 1) [26]. All trials that involved regular contact between participants and study personnel reported information about the personnel delivering the intervention: nine trials reported using personnel that had expertise in delivering behaviour change interventions or were trained specifically for the trial [21, 23, 24, 26, 27, 33, 36,37,38], with the remaining trials reporting non-specific information such as research staff titles or allied health profession.

Control groups either received usual care were asked not to change their physical activity level, or were provided with limited breast cancer and physical activity information. Four trials used a contact control group in which control group participants received a comparable frequency of contact but without the primary intervention components (e.g. to administer a symptom questionnaire) [25, 31, 37, 38].

Group allocation and retention

In total, 1966 participants (mean n per study = 103, range = 22–204) were allocated either to an intervention (n = 1054, mean = 55, range = 12–106) or a control group (n = 912, mean per study = 48, range = 10–102). Fifteen trials allocated participants equally between groups, while four trials allocated more participants to the intervention group due to time, financial, and/or recruitment constraints [22, 24, 25, 36]. Post-intervention, 1756 participants were retained, with similar levels of dropout from the intervention (n = 121, 12%) and control arms (n = 89, 10%). In 10 trials (53%), participants were reassessed at follow-up timepoints after the intervention ended, with the final timepoints ranging from 24 weeks to 5 years after baseline (mean = 18 months) (n = 923; intervention groups n = 471, control groups n = 452).

Adherence

Trial adherence was reported in ten trials (53%) and presented in Table 1. Adherence to scheduled sessions was reported in five of the six centre-based trials, with all ≥ 70% [28, 30, 32, 34, 35]. Adherence to home-based physical activity was reported in different formats: three trials reported adherence as the achievement of an activity level goal (ranging from 90 [31, 32] to 150 min a week [26, 29]), and three trials reported the percentage of prescribed sessions completed [24, 31, 36].

Risk of bias

Risk of bias across the included trials is presented in Fig. 2. Notably, 10 trials (53%) were classified as high risk of bias due to insufficient evidence that missing outcome data from participant dropout would not bias the results (Domain 3, > 5% missing for continuous outcomes). Eighteen trials had at least some concern for risk of bias due to having no participant blinding (Domain 4), of which 11 trials (29%) were at high risk due to the nature of the outcomes being likely influenced by knowing the group allocation. Minimal issues were identified relating to randomisation (Domain 1) and deviations from the intended intervention (Domain 2). All trials had at least some concern for risk of bias due to reporting of results, as a-priori analysis plans were not available or sufficiently detailed (Domain 5).

Fig. 2
figure 2

Risk of Bias 2.0

Full size image

The overall risk of bias judgement decisions were modified from the original Cochrane RoB 2 Tool guidance to account for unavoidable characteristics with physical activity trials, such as lack of blinding and the patient-reported nature of outcome measures. Accordingly, eight trials were graded as having overall low risk of bias and one trial was graded as having some concerns for risk of bias because of its insufficient reporting on the randomisation process. The remaining 10 trials were considered as having high risk of bias due to missing > 5% of participants’ outcome data.

Behaviour change theory use

Coding of the Theory Coding Scheme is shown in Table 2. Most trials specified the behaviour change theories used; the exceptions were two trials that indirectly referenced a theory [33, 39], and one trial that reported being theory-based but with no reference or further information [28]. The theories used were the social cognitive theory (n = 10), transtheoretical model (n = 9), and theory of planned behaviour (n = 2). Four trials referred to more than one theory [25, 27, 29, 38], and one referenced a wide assortment of both behavioural and psychological theories [35]. No trials screened or selected participants based on theory-related constructs. Twelve (63%) trials reported using theory to develop the intervention, and eight trials (42%) reported using theory to tailor the intervention to participants. Items 7, 8, 9, and 11 were scored as one item due to generally ambiguous reporting around theoretical constructs or predictors of physical activity; only four trials did not link intervention techniques to theoretical constructs or predictors of physical activity [24, 26, 28, 32].

Table 2 Theory Coding Scheme
Full size table

Coding of interventions using the Behaviour Change Technique Taxonomy is summarised by category in Table 3. The most frequently applied techniques included those associated with goals and planning (n = 18), shaping of knowledge (i.e. education and instruction) (n = 18), feedback and monitoring (n = 17), comparisons of outcomes (e.g. pros and cons) (n = 17), repetition and substitution (n = 14), social support (n = 11), and associations (e.g. prompts and cues) (n = 10).

Table 3 Behaviour Change Techniques Taxonomy (by category)
Full size table

Synthesis of intervention effects

Outcome measures and corresponding timepoints included in analyses are presented in Supplementary Material 2. Pooled post-intervention effect estimates showed medium improvements in self-reported physical activity (SMD = 0.57, CI = 0.33, 0.80) (Fig. 3), objectively measured physical activity (SMD = 0.52, CI = 0.14, 0.89) (Fig. 4) and physical function (SMD = 0.66, CI = 0.33, 0.98), and small improvements in quality of life (SMD = 0.18, CI = 0.08, 0.29), fatigue (SMD = 0.26, CI = 0.14, 0.37), aerobic fitness (0.30, CI = 0.10, 0.50), and muscular strength (SMD = 0.20, CI =  − 0.01, 0.41) (Supplementary Material 3). At follow-up assessments, small improvements were made for self-reported physical activity, quality of life, fatigue, and physical function. Forest plots for the meta-analysis of secondary outcomes are presented in Supplementary Material 3. Statistical heterogeneity (assessed using I2, where values > 50% may represent substantial heterogeneity) may be substantial for self-reported (I2 = 67%) and objectively measured physical activity (I2 = 75%).

Fig. 3
figure 3

Physical activity (self-reported) pooled post-intervention and follow-up effect estimates

Full size image
Fig. 4
figure 4

Physical activity (objective measure) pooled post-intervention and follow-up effect estimates

Full size image

Subgroup analyses comparing the effectiveness of trials according to each specific theory were performed but were not significant and produced wide and overlapping subgroup CIs. Sensitivity analysis for trials at high risk of bias revealed that for objectively measured physical activity and quality of life, the effect estimate for low-risk trials alone was not significant, but when high-risk trials were included, the overall effect estimate became significant. Sensitivity analysis for trials that required imputation of means or SD values showed negligible differences.

Discussion

Nineteen physical activity interventions for women with breast cancer based on behaviour change theories were identified, their results synthesised, and application of behaviour change theories appraised. Intervention groups had small to medium improvements in physical activity and related health outcomes when compared to control groups. Risk of bias was present in all trials to various extents, with missing data from participant dropout being the primary source. However, the overall risk of bias judgements were made with consideration of unavoidable factors in physical activity studies, as is the inability to blind participants allocated to the intervention. The most referenced behaviour change theories were the Social Cognitive Theory and the Transtheoretical Model. Although different theories were used with varying degrees of application, trials reported sharing the use of some common behaviour change techniques, for example goal setting and planning, and the shaping of knowledge. This corresponds with recent research that proposes shifting the focus away from established behaviour change theories, to instead analysing the effectiveness and psychosocial mechanisms of action for the behaviour change techniques themselves [40].

Included trials mostly evaluated interventions against control groups that involved minimal to no instructions to participants. It was therefore unsurprising that between-group comparisons usually favoured the intervention arm, and were consistent with findings from a broader meta-analytic review for general physical activity trials in breast cancer [6] (which included this review’s trials). Only two trial designs used control groups that resembled the intervention arm but without the theory-based components [25, 39]; although the findings favoured the theory-based arm, their results did not vary distinctively from the other included trials.

The development and application of theoretical practices for physical activity behaviour are intended to promote both long-term and autonomous change, through in-depth deliberation of individuals’ intrinsic and extrinsic context, such as personal goals, preferences, and underlying attitudes towards the behaviour [7]. As outlined in the Theory Coding Scheme, trials using behaviour change theories should aim to incorporate them in their entirety, and also effectively assess any changes in the behaviour. The results from this review indicate that theory application, although varied, was not incorporated in its entirety. Several trials mentioned a theoretical construct as a predictor of the behaviour, used theory to select or develop intervention techniques, and used theory to tailor the intervention to participants, but descriptions were typically concise and the use of theory appeared to be secondary to other intervention components, namely the instructed exercise parameters. In some cases, the prescribed intervention will serve participants’ preferences and needs, but not all personal and unique contexts will be accommodated if the support they require is not offered within the program [41]. Although no included trials screened participants based on theory-relevant outcomes, 59% of trials selectively recruited participants who were not regularly physically active. Such inclusion criteria should be of priority, akin to how behaviour change interventions for other lifestyle health behaviours would naturally concentrate on relevant participants, for example smoking cessation interventions. In many cases, the reporting around theory application was clear in the introduction but diminished in the methods and results, as demonstrated by how few trials linked theory constructs to the intervention, such that links between the theory constructs, intervention components and interpretation of the results were ambiguous or lacking. This is aligned with the continuing discourse on the delineation between interventions that are merely “theory-informed” versus truly “theory-based” [42]; a more comprehensive implementation of theory has been suggested to enhance the effectiveness of an intervention in producing behaviour change [13]. The Theory Coding Scheme ratings from this review were similar to behaviour change trials in other populations [43], indicating that health behaviour studies generally do not fulfil all the items and hence are not maximising their potential.

Behaviour change techniques known to be pertinent to physical activity behaviour, namely the monitoring of the behaviour, goal setting and goal revision, and receiving feedback [44], were used in the majority of included trials. Despite being an important component of behaviour change for people with cancer [45], the use of social support, both practically (e.g. accompanying during physical activity) or emotionally (e.g. encouragement), was less commonly reported. In particular, none of the trials that cited the Transtheoretical Model reported the involvement of social support, despite it being a discrete concept within the model. Variability of behaviour change techniques across trials may be due to trials placing differing levels of priority on behaviour change within the intervention, ranging from using the theory as a core component for the study aims [22] to a supplementary means to enhance participant adherence to the overall intervention [26].

More generally, this review also reaffirms known challenges when delivering interventions in physical activity and behavioural research. The potential for discrepancies between a trial protocol and how it is delivered in practice is somewhat unavoidable, which has particular implications for any attempts to identify causal or correlational associations between use of theory, the intervention content, and its effects [46]. Use of evaluative tools like the Theory Coding Scheme relies on the information reported in manuscripts, and it is possible that interventions were extensively theory-based but details were not reported in the published article [47]. In addition, where intervention staff are trained behavioural coaches, behaviour change techniques may also be inadvertently implemented, when not part of the intended protocol. Similarly, contamination of the control, where control group participants independently increase physical activity despite being requested at baseline to not make changes, is acknowledged to potentially be a common occurrence and difficult to account for when making between-group comparisons [48]. Fidelity between the reporting of protocols and practice thus should be considered when evaluating intervention delivery and drawing conclusions on a trial’s findings.

The findings of this review are constrained by some limitations. First, the year of publication for the included trials extended to as early as 2005, and as a body of research, behaviour change for physical activity has been continually evolving and recommendations around the use and reporting of behaviour change theories are still being established. Therefore, the appraisal of earlier trials in comparison to more recent ones may not be equitable. Second, the most used measures of physical activity level in the included trials were self-reported, for which validity is debated [49]. However, this is less important when evaluating between-group differences as opposed to pre/post-intervention values alone. Third, there was marked heterogeneity between the behaviour change variables which were meta-analysed, along with lacking information about the development, validity, or reliability of the outcome measures, which together reduces the validity of the analysis. However, a meta-analysis was conducted to provide an approximate effect estimate.

Conclusions

The current research around the use of behaviour change theories in physical activity interventions for women with breast cancer was appraised. Although effective in increasing physical activity, the application and reporting of theories in the included trials was mixed, with many studies lacking in detail. To better understand the value of adopting behaviour change theories to increase physical activity, evaluation of the effectiveness of defined behavioural constructs and techniques is required.