Studies were considered eligible for inclusion if they were RCTs comparing supervised aerobic exercise training with a placebo, controlled comparison or standard care. For the purposes of the review, exercise was defined as a form of leisure-time physical activity that was performed on a repeated basis over an extended period of time, with the intention of improving fitness, performance or health[14]. Studies with an additional treatment arm or combined intervention (e.g., exercise with diet modification) were included only if the effects of exercise could be isolated. A priori, we excluded reports that were available only in abstract form.

Trials were included if they involved adults (17 years and older) diagnosed with cancer who were actively receiving cancer treatment or off treatment. Included studies were required to measure maximal, peak, or estimated maximal oxygen consumption (VO_2max/peak) as a study outcome.

A search was performed of the databases including OVID MEDLINE (1948 to October 2013), PubMed (1975 to October 2013), SCOPUS (1950 to October 2013), Web of Science (1950 to October 2013), EMBASE (1988 to October 2013), Cochrane Central Registry of Controlled Trials (1991 to October 2013), and LILACS (1982-October 2013). The search strategy was developed and approved by a librarian with extensive database searching knowledge and experience. We searched terms related to cancer (e.g., neoplasms, tumor), exercise (e.g., exercise, exercise therapy/or motion therapy, aerobic training), publication type (e.g., random allocation, clinical trial), and aerobic capacity (e.g., VO₂). The search strategy was modified as necessary for each database. Non-English language publications were eligible for inclusion. To locate unpublished research, we reviewed clinical trial registries and websites housing theses and dissertations. Fourteen experts in the field of cancer and exercise were contacted in order to identify any research that was not published or was pending publication. Table 1 includes an example of the MEDLINE search strategy.

The titles and abstracts were screened for eligibility by two independent evaluators (C.K. and R.B.), and coded for exclusion or potential inclusion. Potentially eligible manuscripts were obtained and the same evaluators performed a second round of screening to evaluate full eligibility criteria. Any disagreements were resolved by consensus (C.K., R.B., and M.M.). The two evaluators (C.K. and R.B.) then independently abstracted data on study participants, the intervention and control (usual care) protocols, and study outcomes, and assessed for quality. Studies were evaluated using the quality assessment framework for RCTs developed by the Cochrane Collaboration[15] to assess risk of bias in the individual studies. Sensitivity analyses were conducted to examine the effect of including studies with high risk of bias.

For the purpose of evaluating exercise prescription variables, exercise intensity was standardized to a single %VO_2max value[16-18]. For studies that used %VO_2max as the intensity prescription the average of the range was used; time spent at different intensities was factored in to create the mean value. High intensity intervals were weighted at 50% of the contributing time. Resistance exercise was not included in intensity ratings. Total exercise workload, or intensity-minutes, was calculated by multiplying the exercise intensity by the prescribed exercise volume (program duration, minutes per session and sessions per week).

Study results were pooled using random effects models. For continuous outcomes, pooled statistics were calculated using mean differences (MD) when data were on a uniform scale and using standardized MD (SMD) when data were on different scales. All results were calculated with 95%CI. The SMD was interpreted as 0.2, 0.5 and 0.8 representing small, medium and large effects on outcomes respectively[19]. Statistical heterogeneity was assessed using a χ² test that considered a P-value of less than 0.10 to indicate significant heterogeneity. I² values, ranging from 0% (homogeneity) to 100% (heterogeneity) were also calculated to quantify variability in study effect and values of 25%, 50% and 75% were used to describe low, moderate and high heterogeneity respectively[20]. Subgroup analyses and multiple regression moderator analyses were performed to explore and explain heterogeneity among studies. A priori subgroup analyses included examining the pooled effect estimate by level of supervision of exercise (group or individual), the timing of the intervention (on or off treatment), and cancer type. Meta-regression was performed to explore exercise variables of frequency, time, intensity, duration and adherence on effect estimate.

A biomedical statistician (Y.L.) provided oversight on the statistical methods, and performed the meta-regression analyses. All data were entered into Review Manager 5.2 and analyzed with SPSS v15 software utilizing meta-regression scripts created by Lipsey and Wilson and Stata/SE (version 13.0)[21]. Figures were created using Comprehensive Meta-Analysis (version 3: http://www.meta-analysis.com/index.php).

The search protocol yielded 1269 eligible studies; after removal of duplicates and screening of abstracts, 23 studies remained. Reference tracking and contacting of experts accounted for 4 additional studies. Grey literature and trial register searches yielded no further articles. Full text review of the 27 studies excluded a further 9, leaving 18 studies for qualitative and quantitative synthesis[22-39]. One study was not used for the quantitative analyses due to missing data[32] and one study was divided into two comparison groups as it involved both on and off treatment subgroups[27] (unpublished data provided by author). The remaining 17 studies, generating 18 comparisons, were included in the meta-analyses (Figure 1). Kappa statistics for the inclusion of studies was 0.9 (P < 0.001). Following discussion there was 100% agreement in scores between evaluators.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 PRISMA flow diagram of study selection process.

In general there was high or unclear risk of bias for selection (allocation concealment) and detection bias (lack of blinding of outcome assessors) and low risk of bias for attrition (handling of incomplete data) and reporting bias (outcome reporting) among the included studies (Figure 2). Sensitivity analyses were performed after excluding studies with a high or unclear risk of bias for allocation concealment (n = 10)[24,31-39] and for use of blinded outcome assessment (n = 11)[24,26,27,31-36,38,39]. The results showed minimal differences in the pooled effect estimates for aerobic capacity based on risk of bias. For allocation concealment, the pooled effect estimate increased by 0.6 (SMD: 0.80; 95%CI: 0.51, 1.25) whilst for blinding of outcome assessment the estimate decreased by 0.4 (SMD: 0.7; 95%CI: 0.35, 1.05). After excluding studies with a high or unclear risk of bias for any factor (n = 13), the pooled effect estimate decreased by 0.8 (SMD: 0.66, 95%CI: 0.22, 1.11).

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Risk of bias summary.

The 18 included studies involved 1149 participants of which 576 were randomized to receive an aerobic exercise intervention and the remaining 573 received usual care or no exercise. Participants were on average 53 years of age and 76% were female. Survivors of breast cancer were most commonly studied in both breast cancer specific trials and mixed cancer type trials (14 studies)[22-26,28,29,33-39] accounting for 686 participants (60%) of the total participants in the review. Further details on the included studies are provided in Table 2.

Ten studies consisted exclusively of aerobic exercise training[23-27,29,30,33,34,38], six studies included a resistance exercise component with or without flexibility training[22,28,31,36,37,39], one included physiotherapy exercises and relaxation[35], and one included flexibility training plus a dietary intervention[32]. Exercise interventions consisted primarily of cycling[23-31,34,39] or walking/jogging[23,24,32,35,37-39]. Five studies[22,23,28,35,38] offered exercise programs in a class setting (group exercise format) and the remaining 13 studies[24-27,29-34,36,37,39] were individualized exercise programs, although further detail on the level of supervision was not often provided. Eight studies were carried out during active cancer treatment[22,26,29-31,33,34,38], nine in the post treatment phase[23-25,28,32,35-37,39] and one included participants both on and off treatment[27]. The duration of exercise programs ranged from 4-6 wk to 26 wk with individual exercise sessions ranging from 20-90 min including warm up and cool down. Seventeen studies prescribed aerobic exercise that was of moderate intensity with 4 of these studies[22,27,29,34] including high intensity intervals. One study combined both low and moderate intensity intervention groups into a single intervention group for their analysis due to the small sample size of the study[24]. Further information on the exercise prescription variables is provided in Table 3.

All eighteen studies reported VO_2peak, with 13 studies (14 comparisons) indexing this outcome to body weight (mL/kg per minute)[23-30,35-39], 4 studies measuring absolute (L/min)[22,31,33,34], and 1 study measuring percent change in VO_2peak (mL/kg per minute)[32]. The study measuring percent change in VO_2peak was excluded from analysis due to insufficient data on measures of variability.

Pooling of all 18 comparisons showed a moderate-to-large effect estimate (SMD: 0.74; 95%CI: 0.52, 0.96; P < 0.001) in favour of supervised aerobic exercise training; however, moderate heterogeneity was found among the included studies (I² = 63%; P < 0.001) (Figure 3). Pooling of the 13 studies (14 comparisons) reporting VO_2peak (mL/kg per minute) showed a statistically significant mean difference in VO_2peak of 3.13 mL/kg per minute (95%CI: 2.21, 4.05; P < 0.001) in favour of supervised aerobic exercise training; however, again moderate heterogeneity was found among the included studies (I² = 58%; P < 0.001).

Open in New Tab   Full Size Figure   Download Figure

Figure 3 VO₂ effect size.

Subgroup analyses were performed for level of supervision, treatment timing and cancer type (Table 4). A significantly smaller effect estimate (P = 0.003) was found for group/ class-led exercise studies[22,23,35,38] (SMD: 0.36; 95%CI: 0.17, 0.56) when compared to studies involving individualized exercise programs[24-31,33,34,36,37,39] (SMD: 0.87; 95%CI: 0.60, 1.15). Non-significant effects (P = 0.11) were observed between on and off treatment studies. Statistically significant differences in pooled effect estimates were observed between cancer types with a significantly larger beneficial effect found among studies including survivors with hematological cancers (P < 0.001)[27,31] when compared to other cancer tumor groups (breast cancer, lung cancer and mixed cancer).

Meta regression was performed analyzing the effect estimate with exercise parameters of exercise workload and participant adherence as potential moderators. These two variables, workload and adherence, explained 65.8% (P = 0.04) of the between-study variance in effect estimate among the included studies (Figure 4).

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Meta-regression analysis: Workload, adherence.

Nine studies reported data for health-related quality of life as measured by the Functional Assessment of Cancer Therapy-General (FACT-G) scale[23,25,29,31], the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire: EORTC-QLQ-C30[22,28,30,39] and Medical Outcomes Survey: Short Form: SF36[38]. Pooling of all nine studies demonstrated a non-significant effect on quality of life (SMD: 0.3; 95%CI: -0.11, 0.71; P = 0.16), with high heterogeneity found among studies (I² = 80%; P < 0.001). Further details are provided in Table 5.

The major limitations of this meta-analysis were the assumptions revolving around exercise prescription factors. All intensity values represented average values obtained and were standardized to an estimated %VO_2max value. Conversions are imperfect as are average values created from studies using intervals and step protocols. Therefore we acknowledge that there is some associated error in our intensity estimates. As well, no data were provided on actual adherence to intensity among participants in the individual studies to allow more precise estimation of intensity. Thus our crude estimates of targeted intensity functioned merely as a means to determine relative ranking for between study comparisons. Assumptions were also made that resistance exercise provided minimal contributions to VO_2peak. A further limitation of our meta-analysis was the small number of included studies, which permitted the analysis of only two moderator variables. Thus, further research is needed particularly in survivors of cancers other than breast cancer.

Studies included in this review were generally of good methodological quality with low risk of bias. However, further attention to study quality is needed, as many studies did not adequately report methods for allocation concealment and use of blinded assessment, limiting our ability to evaluate the impact of risk of bias across studies. Of note, the estimated effect size was lower when excluding studies at high risk of bias; thus, our findings may represent an overestimate of the effect of supervised exercise on aerobic capacity.

A final limitation is that the mechanism(s) responsible for the improvement in VO_2peak along the oxygen cascade were not studied in any of the studies included in our review; thus, the favourable finding in VO_2peak may be due to improved convective and/or diffusive oxygen transport coupled with improved oxygen utilization by the active muscles[48].

Supervised aerobic exercise training was found to have a moderate-to-large beneficial effect on VO_2peak. Aerobic capacity increased in a dose response fashion with overall workload, with larger effects found in studies prescribing a higher overall workload of aerobic exercise. Larger benefits were also seen in studies with better participant attendance and among survivors of haematological cancers. There is a need for further randomized controlled trials examining supervised aerobic exercise interventions in understudied but common cancers such as prostate, lung and colorectal cancer.

Evidence is accumulating to support the benefit of exercise to improve the physical functioning and quality of life of survivors. Currently, the optimal exercise prescription is unknown and the effect of variations in exercise training parameters on cancer-specific outcomes are poorly understood. Therefore, questions remain over how to best tailor exercise prescriptions to optimize the health outcomes of survivors at different times through the cancer continuum.

A previous meta-analysis included data from six randomized controlled trials and reported a significant benefit from supervised aerobic exercise training, compared with usual care, on VO_2peak (2.90 mL/kg per minute; 95%CI: 1.16, 4.64; P = 0.01). However, statistical and clinical heterogeneity was found among the exercise trials included in their review, and therefore further research was indicated to build on and extend the current knowledge in the field.

Pooling of the 13 studies (14 comparisons) reporting VO_2peak (mL/kg per minute) showed a statistically significant mean difference in VO_2peak of 3.13 mL/kg per minute (95%CI: 2.21, 4.05; P < 0.001) in favour of supervised aerobic exercise training; however, again moderate heterogeneity was found among the included studies (I² = 58%; P < 0.001). Meta-regression was performed analyzing the effect estimate with exercise parameters of exercise workload and participant adherence as potential moderators. These two variables, workload and adherence, explained 65.8% (P = 0.04) of the between-study variance in effect estimate among the included studies.

Supervised aerobic exercise training is an effective intervention to improve aerobic capacity in survivors of cancer. Aerobic capacity increased in a dose response fashion with overall workload, with larger effects found in studies prescribing a higher overall workload of aerobic exercise. Larger benefits were also seen in studies with better participant attendance and among survivors of haematological cancers.

Aerobic capacity (VO_2max) is the maximum volume of oxygen that the body can consume during maximal exercise, using at least 60% of the musculature, and while breathing air at sea level. Aerobic capacity is best increased by habitual aerobic exercise training that is of a moderate-to-vigorous intensity.

An excellent systematic review.