Follow Us On Facebook

Home Journal of Nursing Publish Search Advertise With Us

Bookmark the RN Journal in your Favorites File for easy reference!

Journal of Nursing

Predicting exercise adherence in cancer patients and survivors: a systematic review and meta-analysis of motivational and behavioural factors

Anne M Lunde Husebø MSc, RN, PhD Candidate, Sindre M Dyrstad PhD Associate Professor, Jon A Søreide PhD, MD Attending Surgeon, Professor, Edvin Bru PhD Professor [email protected]

Aims and objectives

To examine research findings regarding predictors of adherence to exercise programmes in cancer populations.


Cancer patients are advised to participate in daily exercise. Whether they comply with the recommendations for physical activity or not remains unclear.


A systematic review and meta-analysis.


Empirical articles published in English between 1995 and 2011 were searched in electronic databases and in reference lists, using the search terms ‘adherence’, ‘predictors’, ‘exercise’, and ‘cancer’ in varying combinations. Twelve of 541 screened abstracts met the inclusion criteria. The included studies' eligibility considering predictors of exercise adherence were reviewed. A quality assessment process evaluating the studies methodological quality was performed. Eight of the reviewed studies were considered eligible for a meta-analysis involving Pearson's r correlations.


Exercise stage of change, derived from the transtheoretical model of behaviour change (TTM) was found to be statistically significant and a strong predictor of exercise adherence. In addition, the theory of planned behaviour (TPB) construct; intention to engage in a health-changing behaviour and perceived behavioural control, demonstrated significant correlations with exercise adherence.


The review identified that both the TPB and the TTM frameworks include aspects that predicts exercise adherence in cancer patients, and thus contributes to the understanding of motivational factors of change in exercise behaviour in cancer populations. However, the strengths of predictions were relatively weak. More research is needed to identify predictors of greater importance.

Relevance to clinical practice

Surveying the patients' readiness and intention to initiate and maintain exercise levels, as well as tailoring exercise programmes to individual needs may be important for nurses in order to help patients meet exercise guidelines and stay active.

Full Article

Review procedure

The search combining ‘adherence’ and ‘exercise’ with other supplementary terms identified 541 publications. The search was refined by adding ‘cancer’, revealing twenty publications. Two reviewers examined the eligibility of the identified articles using the inclusion and exclusion criteria presented in Table 1.

Table 1. Inclusion and exclusion criteria
Inclusion criteriaExclusion criteria

Empirical articles in English

Publication years from 1995 to 2011

Primary Studies reporting findings from exercise intervention trials in cancer patients

Focus on predictors of exercise adherence in cancer patients or cancer survivors

Analysis of correlates between exercise and adherence presented in adequate statistical values

Studies scoring ≥50% on the quality assessment

Observational studies, descriptive studies, or only abstracts

Studies not reporting analysis of adherence predictors

Studies reporting postprogramme exercise adherence

Studies reporting adherence to exercise recommendation and risk of cancer

Finally, the review procedure identified 13 potential articles for inclusion. Upon further analysis, three of the identified articles failed to meet the inclusion criteria of focus on predictors of exercise adherence. Two additional articles were identified by screening reference lists of eligible articles, giving a total of 12 articles for further analysis. All of the 12 included articles were published between 2001 and 2012, and most were published between 2006 and 2012.

The purpose stated in each of the included studies was to identify key predictors of exercise adherence in cancer populations. The samples in the studies represent different cancer populations. However, Maddocks et al. (2009) found that sample characteristics like disease status and current or previous treatment in a cancer population of 7224 patients did not influence the ability to take up and complete an exercise programme. This finding suggests that it is relevant to conduct a meta-analysis across the selected studies. Eight of the 12 studies provided Pearson's correlation coefficients for the prediction of adherence and were included in a meta-analysis (Courneya et al200120022004a,b20082010, Peddle et al2009, McNeely et al2012). These eight studies comprised a total of 510 patients.

The meta-analysis procedure was conducted using the software programme Comprehensive Meta Analysis Version 2 analysis (Borenstein et al2005). A meta-analysis is beneficial, allows for a more objective assessment of the review evidence and increases understanding of inconsistent results encountered in the reviewed studies (Egger et al1997). In the meta-analysis, correlation coefficients were weighted according to the number of cases in the studies. The authors of studies giving insufficient statistical information for meta-analysis were contacted to collect the information needed; however, we were not able to retrieve the necessary information for six correlations. In cases of missing information, the correlation coefficient was set to 0·00.

Courneya et al. (2010) reported correlations for two different exercise groups, which were combined in the meta-analysis. In two studies, the predictor ‘TPB-attitude’ was reported in both ‘instrumental attitude’ and ‘affective attitude’. An estimated mean coefficient of correlations was included in the meta-analyses.

Statistically significant heterogeneity was detected for associations of PBC and pretrial exercise behaviour with adherence. Heterogeneity was tested by using the I2 estimation (Ioannidis 2008). In the meta-analysis random-effect model was implemented to compensate for between-studies heterogeneity (Deeks et al2008). Moreover, a complementary fixed-effect model was computed. According to Alderson and Green (2002), it is unlikely that there is important statistical heterogeneity if fixed-effect and random-effect meta-analysis give identical results. When comparing the results from the two models, substantial differences in coefficients and meta-correlation were not found (Table 6). There were therefore no indications of heterogeneity having a critical influence on the meta-analysis performed.

O'Connor et al. (2011) warn against excluding studies from systematic reviews just because they provide no ‘usable data’. We therefore found it relevant to also include findings concerning predictors of exercise adherence in studies not providing data suitable for the meta-analysis, but otherwise satisfied the quality criteria applied in the review. The assessment was based on the statistical significance of findings.

Quality assessment

The quality assessment was carried out by two researchers. Quality assessment criteria in accord with Cochrane Handbook for Systematic Reviews of Interventions (Higgins & Green 2011) were used in so far as they were relevant for prospective studies.

The quality assessment tool was adopted from Jack et al. (2010). We modified this tool by implementing a criterion for sample size calculation or power calculation as described by Ingram et al. (2006), and specified the outcome measures describing adherence scores and data regarding predictors of adherence. In this review, predictors of exercise adherence in exercise groups were the exclusive focus, and the ‘sample size’ criterion was altered from ≥300 to ≥50 subjects according to the rule-of-thumb for number of cases in analyses of correlation and regression given by Green (1991). Finally, the assessment tool consisted of 12 criteria, as presented in Table 2. Criterion A (selection bias) concerns issues such as percentage of selected subjects included and likelihood of representing the target population (Ingram et al2006). Sample size information gives N in the exercise groups only (criterion D). Randomisation procedure in randomized controlled trials (RCT) is not likely to affect the prediction of exercise adherence; therefore non-randomised design was not set as a quality criterion. Only two of the included studies applied single-group designs (Courneya et al2001, Peddle et al2009).

Table 2. Quality assessment criteria (adapted from Jack et al2010 and Ingram et al2006)
Study population: selection bias (criterion A) and description of inclusion and exclusion criteria (criterion B)
Study design: prospective design (criterion C) and sample size ≥50 (criterion D)
Drop-outs: percentages of drop-outs/withdrawals (criterion E)
Outcome measures: defining adherence (criterion F), data presenting adherence score (criterion G), data presenting predictors of adherence (criterion H), and use of standardised or valid measurements (criterion I)
Analysis and data presentation: appropriate univariate crude estimates (criterion J), appropriate multivariate analysis techniques (criterion K), and sample size calculation or power calculation (criterion L)

The methodological qualities of the studies were evaluated using score points; those meeting the standard were given a score of one, while those not meeting the standard were given a score of zero. Studies that met ≥50% of the criteria (scores ≥ 6) were rated as ‘acceptable quality’, while studies that met <50% of the criteria (scores < 6) were rated as ‘low quality’ (Jack et al2010). Disagreement on scoring the studies was solved by consensus.


Characteristics of reviewed studies

Common characteristics of the reviewed trials are presented in Table 3. Five of the studies are authored by the same research group, but represent distinct studies. Sample size ranged between 19 and 160 individuals, with a median of 47 subjects calculated from number of participants randomised to the exercise groups. Five studies included both genders (Courneya et al20022004a2010, Peddle et al.2009, McNeely et al2012), one study enrolled males only (Courneya et al2004b) and six studies included females only (Courneya et al.20012008, Daley et al2007, Latka et al2009, Pinto et al2009, Swenson et al2010). Breast cancer was the most common type of cancer, addressed in seven trials. Four studies enrolled cancer patients who were offered adjuvant treatment (Courneya et al2004a,b, Swenson et al2010, McNeely et al2012) and in two studies the participants were pending treatment (Peddle et al2009, Courneya et al.2008). Three studies enrolled both cancer patients currently undergoing treatment and cancer survivors who had completed therapy (Courneya et al200120022010).

Table 3. Overview of the reviewed studies
StudyTheoretical frameworkSubjectsDesignInterventionBehavioural and motivational predictors of exercise adherenceResults
Courneyaet al. (2001) Theory of planned behaviour (TPB)

n = 24


Mean age = 51 years

Breast cancer survivorsstage I and II

Prospective convenience sample


Adherence: total number of classes attended assessed by objective attendance records

TPB constructs

Accessible beliefs

Baseline physical activity behaviour

Intention to attend was the sole determinant of exercise adherence and subjective norm, an important determinant of intention. Past exercise did not explain exercise adherence
Courneyaet al. (2002) TPB and five factor model

n = 51 (exercise group)

Both sexes

Mean age = 52 years

44% with breast cancer diagnosis

50% stage I and II

44% currently in treatment

Randomised controlled trial (RCT) Home-based Adherence: total amount of exercise performed, and percentage of the amount of prescribed exercise, self-reported

TPB constructs

Accessible beliefs

Significant correlations between overall RCT exercise and past exercise, intention, attitude, perceived behavioural control, and control beliefs. In the exercise group, 4 variables predicted exercise adherence and explained 41% of the variance: sex, extroversion, normative beliefs, and PBC
Courneyaet al.2004a) TPB and five factor model

n = 69 (exercise group)

Both sexes

Mean age = 60 years

Colon cancer stage 0–IV

Currently receiving treatment



Adherence: average percentage of sessions attended per person, self-reported.

Behavioural techniques: personalised exercise prescription and telephone calls

TPB constructs

Trans-theoretical Model exercise stage of change

Baseline physical activity behaviour

Significant correlations between exercise adherence and exercise stage and PBC
Courneyaet al.2004b) TPB

n = 82 (Exercise group)


Mean age = 68 years

Prostate cancer

Currently receiving treatment



Adherence: average percentage of sessions attended per person assessed by objective attendance records

Behavioural techniques: informal information

TPB constructs

Trans-theoretical Model exercise stage of change

Baseline physical activity behaviour

Smoking habit

Drinking habit

Independent predictors in the final equation were preprogramme overall exercise stage and intention
Daleyet al. (2007) Not reported

n = 70 (Exercise group)


Mean age unreported

Breast cancer, early stage

Completed treatment



Adherence: the level of session participation achieved by participants, recorded objectively in activity logs

Exercise stage of change

Baseline physical activity behaviour

Smoking habit

No significant associations between health behaviours and session adherence to the intervention recorded
Courneyaet al. (2008) TPB

n = 160 (Exercise groups)


Mean age = 49 years

Breast cancer stage I–IIIA, 61% had stage II Initiating treatment



Adherence: the number of sessions attended divided by the number of sessions expected, based on the length of the chemo-therapy protocol, assessed by objective attendance records

TPB constructs

Baseline physical activity behaviour

Smoking habit

Smoking status showed a borderline significant correlation with exercise adherence in the exercise groups. Exercise behaviour at baseline did not predict exercise programme adherence. Motivational variables did not predict exercise adherence
Peddleet al. (2009) TPB

n = 19

Both sexes

Mean age = 64 years

Non small-cell lung cancer, stage I–IIIA

Awaiting surgical resection of malignant lung lesion

Pilot study, single-group, prospective Supervised Adherence: the number of sessions attended divided by the number of sessions expected by each participant, assessed by objective attendance records

TPB constructs

Smoking habit

PBC was strongly correlated with exercise adherence, followed by subjective norm. Borderline predictors of adherence were intention, self-efficacy, and attitudes. Smoking status did not predict adherence to exercise
Latkaet al. (2009) Not reported

n = 37


Mean age = 55 years

Breast cancer stage 0–IIIA

Completed treatment


Combined supervised and home-based Adherence: Average minutes/week of moderate-intensity aerobic exercise performed from baseline to 6 months, self-reported.

Behavioural techniques: physical activity log

Trans-theoretical Model exercise stage of change

Baseline physical activity behaviour

Activity prior to baseline and exercise stage of change predicted exercise adherence
Pintoet al. (2009)

Trans-theoretical Model &

Social-Cognitive Theory (SCT)

n = 43


Mean age = 53 years

Breast cancer stage 0–II

Completed treatment



Adherence: minutes of exercise and pedometer steps and percentage of participants who met their exercise goal each week, self-reported

Trans-theoretical Model exercise stage of change

Trans-theoretical Model decisional balance

Social-Cognitive Theory

Baseline physical activity behaviour

Baseline self-efficacy for exercise was a significant positive predictor of adherence measure and Baseline PA predicted exercise steps
Courneyaet al. (2010) TPB

n = 60

Both sexes

Mean age = 53 years

100% lymphoma, 82% non-Hodgkin lymphoma

Currently receiving treatment



Adherence: number of sessions attended divided by the number of sessions expected, assessed by objective attendance, duration and intensity records.

Behavioural techniques: booked exercise sessions, telephone follow-up, and positive reinforcement, paid parking

TPB constructs

Baseline physical activity behaviour

Smoking habit

Past exercise and smoking were significant predictors of exercise adherence.

Motivational variables did not predict adherence

Swensonet al. (2010) TPB, SCT, & The Physical Activity Adherence Model

n = 36


Mean age = 47 years

Breast cancer stage I–III

Currently receiving treatment

Longitudinal and observational study of participants in the physical activity group of an RCT


Adherence: total number of steps and mean number of steps per day on days with any steps recorded, self-reported

Behavioural techniques: exercise logs, pedometers, pulse monitoring

Baseline physical activity behaviour Predictors of the adherence measure ‘mean numbers of steps per day’ were hours per week spent sleeping or reclining at baseline. The fewer steps the less exercise
McNeelyet al. (2012) TPB

n = 52

Both sexes

Mean age = 52 years

Head and neck cancer, stage I–IV

Currently receiving treatment

RCT Combined supervised and home-based Adherence: percentage of attended sessions, assessed by objective attendance records

TPB constructs

Baseline physical activity behaviour

Smoking habit

Drinking habit

Significant or borderline associations between exercise adherence and alcohol consumption. Subjects not consuming alcohol on a daily basis accomplished higher adherence. Motivational variables and baseline exercise levels did not predict adherence

The results of the quality assessment are presented in Table 4. The studies were categorised into levels with respect to their scores, with higher scores associated with higher quality. All studies were considered acceptable quality, with a median quality score of 9·0 (range, 7–11). The quality assessment process still revealed some limitations. Overall, the studies exhibited limitations regarding selection bias (criterion A), sample size (criterion D), information on drop-outs (criterion E) and power analysis (L). Accrual was the most reported problem concerning criterion A. Five studies reported accrual data; of these, four accrued <40% of eligible subjects (Courneya et al.2004a,b2008, Latka et al2009). McNeely et al. (2012) reported accrual rates between 33% and 57%, divided between two recruiting sites. One other study scored zero points on selection bias, because it used a single-group design and convenient recruiting from a special interest group, including 100% of eligible subjects (Courneya et al2001).

Table 4. Rating of methodological quality
StudyABCDEFGHIJKLQuality score

  1. A, selection bias; B, description of inclusion and exclusion criteria; C, prospective design; D, sample size ≥50; E, percentages of drop-outs/withdrawals; F, defining adherence; G, data presenting adherence score; H, data presenting predictors of adherence; I, using standardised or valid measurements; J, appropriate univariate crude estimates; K, appropriate multivariate analysis techniques; L, sample size calculation or power calculation.

Courneya et al. (2004a) 1 0 1 1 1 1 1 1 1 1 1 1 11
Courneya et al. (2004b) 1 1 1 1 1 1 1 1 1 1 1 0 11
Courneya et al. (2002) 0 1 1 1 0 1 1 1 1 1 1 1 10
Courneya et al. (2008) 1 1 1 1 0 1 1 1 1 1 1 0 10
McNeely et al. (2012) 1 1 1 1 0 1 1 1 1 1 1 0 10
Peddle et al. (2009) 1 1 1 0 1 1 1 1 1 0 1 0 9
Courneya et al. (2010) 0 0 1 1 0 1 1 1 1 1 1 1 9
Swenson et al. (2010) 0 1 1 0 1 1 1 1 1 1 1 0 9
Latka et al. (2009) 1 1 1 0 0 1 1 1 1 0 1 0 8
Pinto et al. (2009) 0 1 1 0 1 1 1 1 1 1 0 0 8
Courneya et al. (2001) 0 0 1 0 0 1 1 1 1 1 1 0 7
Daley et al. (2007) 0 0 1 0 0 1 1 1 1 1 0 1 7

Six studies contained at least 50 subjects in the exercise group (Courneya et al20022004a,b20082010, McNeely et al2012). Only three of these studies reported to have conducted power analysis (Courneya et al20022004a2010).

Five studies reported data on drop-outs or withdrawals (Courneya et al2004a,b, Peddle et al2009, Pinto et al2009, Swenson et al.2010), with completion rates between 79% and 91%. Adherence rates did not differ significantly between studies with a high or low drop-out rate in the exercise groups.

The quality criteria – prospective design (criterion C), defining adherence (criterion F), data presenting adherence score (criterion G), data presenting predictors of adherence (criterion H) and standardised or valid measurements (criterion I) —obtained the highest assessment scores. All of the reviewed studies provided sufficient information about these criteria. However, irregularities in defining and calculating adherence, and inconsistencies in the analysis and presentation of data regarding correlations between exercise adherence scores and predictors of adherence were observed.

Assessment of adherence

The study outcomes were specific to exercise interventions in 10 out of 12 studies. In the studies of Pinto et al. (2009) and Swenson et al. (2010), the outcomes were specific to PA interventions. In all of the studies the intervention employed strength or/and endurance training of moderate to vigorous intensity. Reported adherence to intervention programmes varied extensively, with rates from various studies ranging between 42% and 91%.

One aspect of exercise adherence examined in all trials was training volume. Training volume included the total amount of exercise performed, average minutes or steps performed per week, quantity of prescribed exercise achieved by the intervention group and total amount of exercise divided by the expected amount of exercise. Two studies (Latka et al2009, Pinto et al2009) referred to adherence meaning prescribed exercise goals. Seven studies assessed adherence by using objective attendance records to determine adherence to the exercise protocol (Courneya et al20012004b, Daley et al2007, Courneya et al2008, Peddle et al2009, Courneya et al2010, McNeely et al2012), while in five studies exercise adherence was assessed by self-reporting methods, such as exercise logs, weekly surveys and weekly phone calls to the participants (Courneya et al20022004a, Latka et al2009, Pinto et al2009, Swenson et al2010). Slightly higher adherence was observed in studies employing supervised exercise programmes than in those using home-based programmes (70·5% vs. 67·5%, respectively).

Predictors of exercise adherence

All of the 12 reviewed articles provided data concerning predictors of exercise adherence. The eight studies providing Pearson's rcorrelations are presented in Table 5. The results of the meta-analysis are presented in Table 6. Many of the reviewed studies employed motivational and behavioural theories to examine predictors of adherence, with the TPB as the most used theory. In the reviewed studies TPB-constructs and self-efficacy (SE) were all measured according to standardised guidelines (Ajzen 2001). All studies in the meta-analysis investigated TPB constructs as determinants of adherence. The results of the meta-analysis indicated that, overall, meta-correlations between the different predictors and exercise adherence ranged from very low to medium (r = −0·02 to 0·22, see Figures 1–9for Forest Plots giving information about the confidence intervals of the correlation coefficients). Intention and PBC demonstrated moderate, but statistically significant meta-correlations with exercise adherence (= 0·22 and 0·17, respectively). In the meta-analysisattitude was found to be non-significant. Subjective norm and SE were weak predictors of adherence, meta = 0·10 and 0·11, respectively, but only subjective norm was found significant. Pinto et al. (2009) investigated the SCT concept of SE and reported its significance as a predictor of home-based exercise among cancer survivors. SCT had a notable effect on both mean minutes of weekly exercise (p = 0·004) and mean pedometer steps per week (p = 0·005).

Figure 1. Forest plots of the correlations between intention and exercise adherence.


Figure 2. Forest plots of the correlations between perceived behavioural control and exercise adherence.


Figure 3. Forest plots of the correlations between attitude and exercise adherence.


Figure 4. Forest plots of the correlations between subjective norm and exercise adherence.


Figure 5. Forest plots of the correlations between self efficacy and exercise adherence.


Figure 6. Forest plots of the correlations between exercise stage of change and exercise adherence.


Figure 7. Forest Plots of the correlations between smoking habit and exercise adherence.


Figure 8. Forest Plots of the correlations between drinking habit and exercise adherence.


Figure 9. Forest Plots of the correlations between pretrial exercise behaviour and exercise adherence.


Table 5. Pearson's correlations of behavioural and motivational variables with adherence to physical exercise in studies included in the meta-analysis
StudynIntentionPerceived behavioural controlAttitudeSubjective normSelf-efficacyExercise stage of changeSmoking habitDrinking habitPretrial exercise behaviour

  1. Coefficients given are Pearson's r, affective attitude (affe), instrumental attitude (inst).

  2. n.s., reported not significant and set to 0·00; –, not included in the studies.

  3. *p < 0·05; **p < 0·01.

Courneyaet al. (2001) 24 0·59** 0·27 0·03 0·25 0·32*
Courneyaet al. (2002) 51 0·27 0·26 n.s. n.s. 0·37**
Courneyaet al. (2004a) 62 0·22 0·26* n.s. n.s. 0·43** n.s.
Courneyaet al. (2004b) 82 0·30** 0·22 −0·01 0·23* 0·31** 0·11 0·11 n.s.
Courneyaet al. (2008) 160 0·09 0·13 −0·05 0·07 −0·14 0·10
Peddle et al. (2009) 19 0·35 0·63** 0·30affe 0·51* 0·32 0·16
Courneyaet al. (2010) 60 0·21 −0·13 0·14inst −0·01 0·01 −0·22 −0·32*
McNeely et al. (2012) 52 0·11 0·03 0·05 0·15 0·15 0·01 −0·24** 0·15

Table 6. Results from meta-analyses of Pearson's correlations of behavioural and motivational variables with adherence to physical exercise
 IntentionPerceived behavioural controlAttitudeSubjective normSelf-efficacyExercise stage of changeSmoking habitDrinking habitPretrial exercise behaviour

  1. Note that the meta-analysis is based on data from eight of 12 reviewed studies providing data on Pearson's correlations.

  2. Coefficients given are Pearson's r.

  3. Fixed-effect and random-effect model was implemented, and coefficients were weighted according to the sample sizes of the different studies.

  4. *p < 0·05; **p < 0·01.

Fixed-effects model 0·21** 0·13** −0·02 0·10* 0·11 0·36** −0·09 −0·03 0·06
95% CI 0·12–0·29 0·04–0·21 −0·11 to 0·07 0·02–0·19 −0·07 to 0·28 0·21 to 0·50 −0·18 to 0·01 −0·20 to 0·15 −0·03 to 0·15
Random-effects model 0·22** 0·17* −0·02 0·10* 0·11 0·36** −0·06 −0·06 0·07
95% CI 0·12–0·32 0·03–0·31 −0·11 to 0·07 0·02–0·19 −0·07 to 0·28 0·21–0·50 −0·20 to 0·07 −0·38 to 0·28 −0·09 to 0·23

Exercise stage of change drawn from TTM of behaviour change was examined as an exercise adherence predictor in five of the reviewed studies, two of which were included in the meta-analysis. In both studies stage of change was measured with the same scale. Courneyaet al. (2004a,b) established statistically significant correlations between exercise adherence and exercise stage in cancer survivors, and the meta-correlation was found to be statistically significant and moderately strong. The two studies did not report which stage of change corresponded to the correlation with adherence. Latka et al. (2009) observed significant univariate associations between a higher readiness for change and adherence to the exercise prescription. The final two studies observed non-significant correlations between adherence and exercise stage of change (Daley et al2007, Pinto et al2009). The TTM construct decisional balance for exercise, associated with stage of exercise adoption, was examined as the only motivational variable in Pinto et al. (2009). It was not statistically significant in relation to a home-based exercise programme.

The association between pretrial exercise behaviour and adherence to programme exercise was investigated in eight of the studies. Pretrial exercise yielded quite heterogeneous results between studies included in the meta-analysis, resulting in a non-significant meta-correlation. Reviewed narratively this predictor achieved a contradictory result. Pinto et al. (2009) report baseline PA as a significant predictor of weekly pedometer steps (p < 0·05) among breast cancer patients. Swenson et al. (2010) observed baseline activity levels to be inversely correlated with exercise adherence, and sedentary subjects recorded significantly fewer steps per day. Daley et al. (2007) and Latka et al. (2009) report no significant correlations between pretrial exercise behaviour and adherence to programme exercise.

Smoking behaviour was included as a behavioural variable in six studies, of which five provided sufficient data for the meta-analysis (Courneya et al2004b20082010, Peddle et al2009, McNeely et al2012). Smoking status exhibited no significant associations with exercise behaviour in the meta-analysis (= −0·06). Drinking habits as a determinant of exercise adherence was examined in two studies (Courneya et al2004b, McNeely et al2012). In these two trials, alcohol consumption emerged as a weak but still statistically significant determinant of exercise adherence. McNeely et al. (2012) reported alcohol intake as the strongest predictor of adherence, establishing significant correlations between high exercise adherence and low alcohol consumption. Courneya et al. (2004b) also report alcohol consumption to be significantly and inversely associated with exercise adherence. Drinking habit, examined in the meta-analysis, did not predict exercise adherence (= −0·06).


The review and meta-analysis revealed some significant, although weak, correlations between adherence to exercise protocols and motivational and behavioural factors. A relatively strong predictor of adherence to exercise in the cancer populations studied was exercise stage of change. The transtheoretical model of behaviour change (TTM) suggests that a behaviour change intervention might be more successful if it were individualised and aimed at the individual's stage of change (Taylor 2003). Rogers et al. (2005ab) observed that breast cancer patients who exercised regularly (i.e. action or maintenance stage participants) reported exercise barriers less frequently. For cancer patients and survivors to maintain recommended exercise levels, barriers to exercise should be addressed when planning exercise interventions (Rogers et al2005b). Pinto et al. (2005) report significant increases in motivational readiness to meet exercise guidelines and adhere to the intervention programme among breast cancer patients who received tailored counselling.

Regular exercise over a long period of time is believed to increase the chance that this behaviour will become habitual (Taylor 2003). Additionally, the ability to overcome barriers to exercise may be stronger in cancer patients who exercise regularly before diagnosis. While most cancer patients decrease their physical activity levels after diagnosis, the importance of a positive attitude towards exercise including an established exercise behaviour before the cancer diagnosis has been made, increase the adherence to exercise intervention programmes (Midtgaard et al2009). In a review of 38 studies including 1088 healthy adults, Trost et al. (2002) found exercise habits to be a consistent predictor of current exercise behaviour. Findings in this review contrast somewhat with the above. Results for pretrial exercise levels varied considerably across studies included in the meta-analysis, a non-significant meta-correlation. This finding was also confirmed in the review of studies explored narratively. Because of the contrast with previous studies and also heterogeneity in findings between the reviewed studies, no firm conclusions should be made. The result may reflect that getting cancer might profoundly influence people, leading to radical change in behavioural patterns including exercise behaviour. Given this interpretation, this is a positive finding which indicates that a history of sedentary behaviour does not hinder cancer patients from becoming more physical active. However, more research is needed to test this hypothesis.

The meta-analysis identified intention as the most significant predictor among theory of planned behaviour (TPB) constructs. Godin and Kok (1996) claim that a combined effect of intention and PBC can explain one-third of variations in health behaviour, with intention being the key construct.

Current literature on health psychology emphasises self-efficacy as an important predictor for exercise adherence, and claims that people with high self-efficacy are more likely to adhere to exercise programmes, and also have a stronger belief in the benefits of regular exercise (Taylor 2003). This opinion is shared by authors who claim that self-efficacy is a significant determinant of exercise behaviour in cancer patients (Rogers et al2005a). However, in contrast to this view, self-efficacy showed only weak correlations with exercise adherence in the present meta-analysis, and obtained no firm conclusion concerning significance in the complementary studies. Martin and Sinden (2001) postulate problems related to the choice of efficacy measurement as reasons for non-correlation between adherences and previous exercise behaviours.

The meta-analysis provides partial support for the opinion that TPB constructs are important motivational variables related to exercise, both in general (Hagger et al2002) and in cancer populations (Blanchard et al2002). This finding might be partly explained by the suggestion that external motivational factors are more important than internal motivation with respect to cancer patients' adherence to exercise programmes. A critique of both the TPB and social-cognitive theory is that they try to explain decision-making processes when individuals adopt a new health behaviour, without addressing how to maintain the behaviour (Martin & Sinden 2001). Varying strengths of relationships between TPB motivational variables and exercise adherence can also be explained by ceiling effects caused by a biased sample of highly motivated subjects and by little variability in the variables at baseline.

The finding that smoking status is not a significant contributor is somewhat surprising. Smoking is known to be negatively associated with adherence to health behaviour recommendations, and is significantly and inversely related to exercise adherence (Martin & Sinden 2001).

As a complex phenomenon, exercise adherence is also influenced by the exercise programme offered and by the exercise environment (WHO 2003), a claim supported by this review as it reveals large variations in adherence rates among the reviewed studies. This finding may be particularly important for assessing the quality of the studies and for calculating and interpreting correlation variables. Further research is needed to clarify the importance of external elements as determinants of exercise adherence. Supervised trials applying behavioural change techniques may counter the effects of individual internal motivation and result in increased exercise adherence (Courneya et al20082010, McNeely et al2012).

Methodological considerations

Also, this review has faced some limitations. Although a precise and well-defined literature search was conducted over an extended time period, relevant studies may have escaped the electronic search. In addition, exclusion of articles not published in peer-reviewed journals, and those not written in English may have reduced the number of available and relevant trials. Additional articles would have contributed to the validity and generalisability of the conclusions drawn from the present review.

The meta-analysis was based on relatively few trials. One explanation is the inconsistent use of statistics for estimating the prediction of adherence in the reviewed studies. This made reviewing a challenge and complicated the conclusions drawn from the review data, a problem pointed out with respect to the methodology of meta-analysis (Egger et al1997, Martin & Sinden 2001). Due to this critical issue, four out of 12 studies had to be excluded, and accordingly were 150 subjects omitted from the meta-analysis. For two of the variables correlation coefficients were available in two studies only. Combining data on predictors from only two studies in a meta-analysis can be considered a weakness in the methodology and raises the question of how many studies are required for a meta-analysis (Deeks et al.2008). Valentine et al. (2010) claims that two studies are enough, stating: ‘because all other synthesis techniques are less transparent and/or are less likely to be valid’ (p. 245). The results should nonetheless be interpreted with care.

A total of six correlations were reported as non-significant in the primary studies, and the exact coefficients were not given. In cases of missing information, the correlation coefficient was set to 0·00. If both negative and positive correlations were expected, 0·00 would be the median non-significant coefficient. This could be considered as a conservative estimate in this context, because positive correlations are more likely than negative correlations. Therefore, a replacement with 0·00 may have resulted in an underestimation of meta-correlations (Blue 1995, Armitage & Conner 2001, Hagger et al2002).

Inclusion of few studies in the meta-analysis may increase the possibility for publication bias and heterogeneity among studies. Heterogeneity may compromise conclusions regarding patients' ability to be physically active and potentially influence the reliability of the data (Egger et al1997). Significant heterogeneity between studies was detected for two of the predictions for which meta-analysis were conducted. However, random- and fixed-effect yielded only moderately different coefficients of meta-correlation for these variables. Thus, indicating that heterogeneity is a modest problem in the present study. Nonetheless, heterogeneity might indicate publication bias (Deekset al2008), and conclusions should be made with caution.

Studies reporting low accrual rates highlight the issue of selection bias and generalisation. Low accrual rates may reduce the representativeness of the sample and compromise the strength of the research results, affecting generalisation and external validity (Oldervoll et al2005).

In addition, the relatively small sample sizes and large number of predictors in the studies considered might indicate artificially strong correlations and provide less robust data regarding the relationship between exercise adherence and its determinants (Green 1991). Green (1991) recommends that researchers estimate effect size based on the characteristics of their study and employ the rules-of-thumb that include effect size or conduct power analyses.

Few studies reported data regarding dropouts or with-drawals, which made it difficult to draw conclusions about the feasibility of multivariate analysis and its results, especially when multiple variables and small samples were involved.

The studies varied in their use of adherence measurements, which complicated comparison between the characteristics of high-adherence studies to low-adherence studies. Patients who report their exercise achievements in diaries may compromise the validity of adherence data through over-reporting or recall vagueness (WHO 2003). Objective observation of the participants' attendance at exercise classes might be a more accurate measure of adherence and could affect patients' motivation and attitude towards exercise behaviour (Armitage & Conner 2001).


The present review contributes to the knowledge of factors that motivate or form barriers to exercise in cancer populations. It identifies exercise stage of change, intention and PBC as statistically significant, although moderately strong predictors of adherence to exercise intervention programmes. Findings give some support to TPB and TTM as relevant frameworks for the understanding of what motivates cancer patients to engage in exercise behaviour. Predictors of exercise adherence in cancer populations are seldom reported on. Few studies and small sample sizes increase risk of biased conclusions made in reviews and meta-analysis. To meet this challenge more research is warranted. Moreover, relatively weak predictions were identified, and, more research is needed to identify predictors that could be of greater importance.

Relevance to clinical practice

The results of this review suggest that more attention should be paid to what will improve cancer survivors' adherence to exercise and to how motivational and behavioural predictors can play a substantial role when cancer survivors take part in health promotion behaviour, which calls for research based on motivational and behavioural theory. Our findings entail the necessity of substantial basic work for a nursing practice that focuses on what motivates survivors to take on exercise, and what will enhance their belief in managing their exercise programme. A change in lifestyle toward initiating physical health behaviour can be a challenge to cancer survivors during and after treatment. The time around diagnosis and treatment has been hypothesised as a ‘teachable moment’, which may increase the individuals' motivation to change their lifestyle. Using the findings of our review that exercise stage of change represents important predictors of exercise adherence, nurses should establish strategies for surveying patients' motivation and readiness to engage in exercise programmes. More attention should be given to the significance of external motivational factors and how these can be applied clinically in health behaviour change programmes.


Study design: AMLH, EB; data collection and analysis: AMLH, EB and manuscript preparation: AMLH, SMD, JAS, EB.

Conflict of interest

The authors have no financial, personal, political, academic or other relations that could lead to a conflict of interest.


  • Ajzen I (1991) The theory of planned behaviour. Organizational Behavior and Human Decision Processes 50, 179–211.

  • Ajzen I (2001) Nature and operation of attitudes. Annual Review of Psychology 52, 27–58.

  • Alderson P & Green S (eds) (2002) Cochrane Collaboration's Open Learning Material for Cochrane Reviewers. The Cochrane Collaboration, 2008. Available at: (accessed 1 September 2008).

  • Andrykowski MA, Beacham AO, Schmidt JE & Harper FWK (2006) Application of the theory of planned behavior to understand intentions to engage in physi-cal and psychosocial health behaviors after cancer diagnosis. Psycho-Oncology 15, 759–771.

  • Armitage CJ & Conner M (2001) Efficacy of the theory of planned behaviour: a meta-analytic review. British Journal of Social Psychology 40, 471–499.

  • Bandura A (1997) Self-Efficacy: The Exercise of Control. Freeman, New York, NY.

  • Blanchard CM, Courneya KS, Rodgers WM & Murnaghan DM (2002) Determinants of exercise intention and behavior in survivors of breast and prostate cancer: an application of the theory of planned behavior. Cancer Nursing 25, 88–95.

  • Blue C (1995) The predictive capacity of the theory of reasoned action and the theory of planned behavior in exercise research: an integrated literature review. Research in Nursing & Health 18, 105–121.

  • Bogg T (2008) Conscientiousness, the transtheoretical model of change, and exercise: a neo-socioanalytic integration of trait and social-cognitive frameworks in the prediction of behavior. Journal of Personality 76, 775–801.

  • Borenstein M, Hedges L, Higgins J & Rothstein H (2005) Comprehensive Meta-Analysis, version 2. Biostat, Englewood, NJ.

  • Courneya KS, Blanchard CM & Laing DM (2001) Exercise adherence in breast cancer survivors training for a dragon boat race competition: a preliminary investigation. Psycho-Oncology 10, 444–452.

  • Courneya KS, Friedenreich CM, Sela R, Quinney HA & Rhodes R (2002) Correlates of adherence and contamination in a randomized controlled trial of exercise in cancer survivors: an application of the theory of planned behavior and the five factor model of personality. Annals of Behavioral Medicine 24, 257–268.

  • Courneya KS, Friedenreich CM, Quinney HA, Fields ALA, Jones LW & Fairey AS (2004a) Predictors of adherence and contamination in a randomized trial of exercise in colorectal cancer survivors. Psycho-Oncology 13, 857–866.

  • Courneya KS, Segal RJ, Reid RD, Jones LW, Malone SC, Venner PM, Parliament MB, Scott CG, Quinney HA & Wells GA (2004b)Three independent factors predicted adherence in a randomized controlled trial of resistance exercise training among prostate cancer survivors. Journal of Clinical Epidemiology 57, 571–579.

  • Courneya KS, Segal RJ, Gelmon K, Reid RD, Mackey JR, Friedenreich CM, Proulx C, Lane K, Ladha AB, Vallance JK & McKenzie DC(2008) Predictors of supervised exercise adherence during breast cancer chemotherapy. Medicine and Science in Sports and Exercise 40, 1180–1187.

  • Courneya KS, Stevinson C, McNeely ML, Sellar CM, Peddle CJ, Friedenreich CM, Mazurek A, Chua N, Tankel K, Basi S & Reiman T(2010) Predictors of adherence to supervised exercise in lymphoma patients participating in a randomized controlled trial. Annals of Behavioral Medicine 40, 30–39.

  • Daley AJ, Crank H, Mutrie N & Saxton JM (2007) Determinants of adherence to exercise in women treated for breast cancer.European Journal of Oncology Nursing 11, 392–399.

  • Deeks JJ, Higgins JPT & Altman DG (2008) Analysing data and undertaking meta-analyses. In Cochrane Handbook for Systematic Reviews of Interventions, version 5.0.1 (Higgins JPT & Green S eds). The Cochrane Collaboration. Available at: (accessed 1 September 2008).

  • Egger M, Smith GD & Phillips AN (1997) Meta-analysis: potentials and promise. British Medical Journal 315, 1371–1377.

  • Godin G & Kok G (1996) The theory of planned behavior: a review of its applications to health-related behaviors. American Journal of Health Promotion 11, 87–98.

  • Green SB (1991) How many subjects does it take to do a regression analysis? Multivariate Behavioral Research 26, 499–510.

  • Hagger MS, Chatzisarantis NLD & Biddle SJH (2002) A meta-analytic review of the theories of reasoned action and planned behavior in physical activity: predictive validity and the contribution of additional variables. Journal of Sport and Exercise Psychology 24,3–32.

  • Higgins JPT & Green S (eds) (2011) Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0. The Cochrane Collaboration. Available at: (accessed 1 March 2011).

  • Ingram C, Courneya KS & Kingston D (2006) The effects of exercise on body weight and composition in breast cancer survivors: an integrative systematic review. Oncology Nursing Forum 33, 937–950.

  • Ioannidis JPA (2008) Interpretation of tests of heterogeneity and bias in meta-analysis. Journal of Evaluation in Clinical Practice 14,951–957.

  • Jack K, McLean SM, Moffett JK & Gardiner E (2010) Barriers to treatment adherence in physiotherapy outpatient clinics: a systematic review. Manuel Therapy 15, 220–228.

  • Knols R, Aaronson NK, Uebelhart D, Fransen J & Aufdemkampe G (2005) Physical exercise in cancer patients during and after medical treatment: a systematic review of randomized and controlled clinical trials. Journal Clinical Oncology 23, 3830–3842.

  • Latka RN, Alvarez-Reeves M, Cadmus L & Irwin ML (2009) Adherence to a randomized controlled trial of aerobic exercise in breast cancer survivors: the Yale exercise and survivorship study. Journal of Cancer Survivorship 3, 148–157.

  • Maddocks M, Mockett S & Wilcock A (2009) Is exercise an acceptable and practical therapy for people with or cured of cancer? A systematic review. Cancer Treatment Review 35, 383–390.

  • Maddux JE, DuCharme KA & Gochman DS (1997) Handbook of Health Behavior Research 1: Personal and Social Determinants. Plenum Press, New York, NY.

  • Martin KA & Sinden AR (2001) Who will stay and who will go? A review of older adults' adherence to randomized controlled trials of exercise. Journal of Aging and Physical Activity 9, 91–114.

  • McNeely ML, Parliament MB, Seikaly H, Jha N, Magee DJ, Haykowsky MJ & Courneya KS (2012) Predictors of adherence to an exercise program for shoulder pain and dysfunction in head and neck cancer survivors. Supportive Care in Cancer 20, 515–522.

  • Midtgaard J, Baadsgaard M, Møller T, Rasmussen B, Quist M, Andersen C & Adamsen L (2009) Self-reported physical activity behavior: exercise motivation and information among Danish adult cancer patients undergoing chemotherapy. European Journal of Oncology Nursing 13, 116–121.

  • O'Connor D, Green S & Higgins JPT (2011) Defining the review question and developing criteria for including studies. In Cochrane Handbook of Systematic Reviews of Intervention, version 5.1.0 (Higgins JPT & Green S eds). The Cochrane Collaboration. Available at: (accessed 1 March 2011).

  • Oldervoll LM, Loge JH, Paltiel H, Asp MB, Vidvei U, Hjemstad MJ & Kaasa S (2005) Are palliative cancer patients willing and able to participate in a physical exercise program? Palliative and Supportive Care 3, 281–287.

  • Peddle CJ, Jones LW, Eves ND, Reiman T, Sellar CM, Winton T & Courneya KS (2009) Correlates of adherence to supervised exercise in patients awaiting surgical removal of malignant lung lesions: results and a pilot study. Oncology Nursing Forum 36,287–295.

  • Pinto BM, Frierson GM, Rabin C, Trunzo JJ & Marcus BH (2005) Home-based physical activity intervention for breast cancer patients.Journal Clinical Oncology 23, 3577–3587.

  • Pinto BM, Rabin C & Dunsiger S (2009) Home-based exercise among cancer survivors: adherence and its predictors. Psycho-Oncology 18, 369–376.

  • Rogers LQ, Shah P, Dunnington G, Greive A, Shanmugham A, Dawson D & Courneya KS (2005a) Social cognitive theory and physical activity during breast cancer treatment. Oncology Nursing Forum 32, 807–815.

  • Rogers LQ, Courneya KS, Shah P, Dunnington G & Hopkins-Price P (2005b) Exercise stage of change, barriers, expectations, values and preferences among breast cancer patients during treatment: a pilot study. European Journal of Cancer Care 16, 55–66.

  • Schünemann HJ, Oxman AD, Higgins JPT, Vist GE, Glasziou P & Guyatt GH (2011) Presenting results and ‘Summary of findings’ tables. In Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0 (Higgins JPT & Green S eds). The Cochrane Collaboration. Available at: (accessed 1 March 2011).

  • Swenson KK, Nissen MJ & Henly SJ (2010) Physical Activity in Women Receiving Chemotherapy for Breast Cancer: adherence to a Walking Intervention. Oncology Nursing Forum 37, 321–330.

  • Taylor S (2003) Health Psychology, 5th edn. McGraw-Hill, New York, NY.

  • Trost SG, Owen N, Bauman AE, Sallis JF & Brown W (2002) Correlates of adults' participation in physical activity: review and update.Medical and Science in Sports and Exercise 34, 1996–2001.

  • Valentine JC, Pigott TD & Rothstein HR (2010) How many studies do you need? A primer on statistical power for meta-analysisJournal of Educational and Behavioral Statistics 35, 215–247.

  • World Health Organization (WHO) (2003) Adherence to Long-Term Therapies. Evidence for Action. WHO Library Cataloguing-in-Publication Data, Geneva.

« Back to the Journal of Nursing

© 2000-2018 // Times Publishing, LLC
Use of this website constitutes acceptance of our 
Terms of Service Information, articles, Press Releases, videos, and links are published as a convenience to our visitors. Articles are to be used only as a reference aid. They are not intended to be used as a substitute for the exercise of professional judgment. Any questions or comments about the article should be directed to the author. We take no responsibility and give no guarantees, warranties or representations, implied or otherwise, for the content or accuracy.
Palm Tree Creative