Efficacy of Exercise Interventions in Modulating Cancer-Related Fatigue among Adult Cancer Survivors: A Meta-Analysis

Currently, there are over 11 million cancer survivors in the United States (1). The 5-year survival rate for cancer survivors has steadily increased from 50% in 1974 to 66% in 2004 (1). Despite living longer after diagnosis, cancer survivors commonly report having 1 or more cancer-related symptoms that impact their quality of life and activities of daily living (e.g., 2). One of the most commonly reported symptoms by cancer survivors is cancer-related fatigue (CRF; ref. 3). CRF is a reported side effect of all types of cancer treatment (4) affecting nearly 100% of cancer survivors and persists for years after treatment cessation (5, 6). Cancer survivors often state that CRF is the most distressing symptom related to cancer or cancer treatment, more so than pain, nausea, and vomiting (2, 7, 8).

Cancer survivors often are told by medical providers to learn to live with CRF by limiting activity, conserving energy expenditure, and relying on others to complete activities of daily living (3). Yet, new evidence is accumulating that indicates cancer survivors who engage in exercise experience numerous physical and mental health benefits including increased functional capacity (4), improved quality of life (9), and diminished depression and anxiety (10). In addition, meta-analyses (11–14) and systematic reviews (15) suggest that exercise interventions may be moderately efficacious in modulating CRF.

Despite the promise of exercise in the management of CRF, an exercise prescription (Ex R_x) tailored for adult cancer survivors experiencing CRF does not exist (3, 4, 16, 17). The available Ex R_x guidelines for cancer survivors (3, 4, 16, 17) broadly focus on the general well-being of cancer survivors, encouraging 150 min/wk of aerobic exercise, 2 d/wk of strength training, and flexibility exercise on days when aerobic or resistance exercise is not performed. An American College of Sports Medicine (ACSM) panel of experts in Ex R_x for cancer survivors recently concluded exercise is safe for cancer survivors, all cancer survivors should avoid inactivity, and exercise programs should be adapted for the individual survivor on the basis of health status, cancer treatment type, targeted health outcomes, and disease trajectory (4). Yet, the panel acknowledged research in the area of Ex R_x for cancer survivors is in the developmental stage with significant research gaps in the dose of exercise required to ensure that cancer survivors receive safe and effective Ex R_x for targeted disease end points such as CRF.

We conducted a qualitative review evaluating the efficacy of exercise as an intervention to reduce CRF among adult cancer survivors. The primary purpose was to investigate which Ex R_x characteristics were associated with the greatest reductions in CRF. We also examined whether study methodologic considerations and subject characteristics combined or interacted with the dose of exercise prescribed to reduce CRF further.

Included were randomized controlled trials (RCT) that examined the effects of exercise on CRF in adult patients (≥18 years) diagnosed with any type of cancer, stage of diagnosis, and type or stage of treatment including those who have completed treatment. Exercise interventions may have occurred in any setting with or without supervision. RCTs may have compared exercise with a usual care group receiving either (a) standard, usual care (e.g., no exercise program prescribed and to maintain current activity levels) or (b) non–exercise-related information during the intervention period. (See Supplementary Appendix I for detailed systematic search information)

The outcome variable examined was patient-reported CRF (3), which studies assessed either separately or as a component of a comprehensive psychological questionnaire with a CRF subscale (see footnotes, Table 1; refs. 18–23).

Two independent raters (J.C.B, S.M.P) coded information related to the study (see Table 1). Intensity of exercise was estimated using metabolic equivalent units (MET), where 1 MET = 3.5 mL O₂·kg⁻¹·min⁻¹. Corresponding MET values for a given exercise intervention were coded from the Compendium of Physical Activity; these include low (<3 METs), moderate (3–≥6 METs), and vigorous (>6 METs) intensity levels (24). Methodologic quality was assessed via the Physiotherapy Evidence Database scale (PEDro). PEDro guidelines categorize high-quality studies from 6 to 11, fair quality from 4 to 5, and poor quality less than 4. Reliability of the raters was high across dimensions [M Cohen κ (ref. 25) = 0.78 for categorical variables, M Spearman-Brown reliability (ref. 26) = 0.90 for continuous variables]. Disagreements between coders were resolved through discussion.

Because a majority of RCTs reported continuous measures, effect sizes (d) were defined as the standardized mean difference between the exercise and control groups divided by the pooled standard deviation, correcting for sample size bias and baseline differences (27). Multiple effect sizes were calculated from individual studies when they included more than 1 exercise intervention group (e.g., aerobic and resistance training groups compared with a control group). Subsequent sensitivity analyses were conducted to confirm the dependence did not influence mean estimates (28). Consequently, the 44 included studies provided 48 exercises versus control group comparisons.

Prior to analysis, data were assessed for publication bias using Begg (29; z = 1.01, P = 0.31) and Egger (30; t = 0.06, P = 0.95) methods, and yielded no evidence of publication bias (Fig. 3 funnel plot, Supplementary). The trim-and-fill technique (31) identified no added or omitted studies that were necessary to normalize the effect size distribution. Analyses were conducted in Stata 10.1 with macros for meta-analysis (32). The homogeneity statistic, Q, was calculated to determine whether a weighted mean effect size (d₊) characterized a common effect size. A significant Q indicated the absence of homogeneity (i.e., more variation in effect sizes than sampling error alone would predict). To standardize Q, the I² statistic and its 95% CI were calculated (33, 34). I² ranges from 0% to 100% with low values suggesting homogeneity and large values suggesting heterogeneity. To explain variability in the effect size estimates, the relation between study-level characteristics and the magnitude of the effects, was examined in modified least-squares regression analysis with the weights equivalent to the variance for each study effect size (viz., meta-regression). Bivariate analysis was conducted using fixed-effects assumptions, and the final, multimoderator analysis was conducted using random-effects assumptions. To reduce multicollinearity in multiple moderator models, all retained continuous moderators were zero centered and categorical variables were contrast coded.

Potentially relevant reports included 7,245 articles of which 44 (n = 3,254)> satisfied the selection criteria (Figure 1). Of the studies identified, 40 provided 1 CRF effect size estimate and 4 studies provided 2 estimates, yielding 48 effect sizes among 44 studies (see Table 1 for descriptive statistics). Studies providing 2 effect sizes included 2 independent exercise intervention groups that were compared with 1 standard care group (46, 49, 55, 69). Three interventions with multiple intervention groups were randomized to aerobic exercise, resistance exercise, or control condition (49, 55, 69); whereas, the fourth study randomized participants to either supervised exercise, unsupervised exercise, or a control condition (46). The mean methodologic quality of the 44 included studies was 6.8 (1.4) of 11 (range: 3–10; Table 2).The mean age of cancer survivors was 53.8 (10.5) years and they averaged 6.7 (13.8) months postdiagnosis. The majority of cancer survivors were women (86%). Approximately, half (46%) of the cancer survivors were currently being treated with primary pharmacologic therapy during the exercise intervention. For those undergoing therapy, a majority of cancer survivors in the sample (75%) were being treated with a combination of chemotherapy and radiotherapy, whereas 13% were treated with only chemotherapy, 6% were treated with only radiation, and 6% were treated with only hormone therapy.

Twenty-five studies examined exercise interventions exclusively in breast cancer survivors (44–55, 57–68), 4 in prostate cancer survivors (69–72), 4 in lymphoma (73–76), 1 in leukemia (78), and 1 in colorectal cancer (77). The remaining 9 studies examined exercise interventions in a mixed group of cancer survivors (35–43). Twenty-four studies included only aerobic exercise (35, 38, 39, 42–44, 46, 49, 50, 52–59, 61, 65, 69, 70, 74, 77, 78), 6 studies included only resistance exercise (49, 55, 63, 68, 69, 71), 11 studies included a combination of aerobic and resistance exercise (40, 41, 48, 51, 60, 62, 64, 67, 72, 75, 76), and another 6 included neuromuscular exercise such as tai-chi or yoga (refs. 36, 37, 45, 47, 66, 73; Table 5 characteristics of included studies, Supplementary).

The average length of the exercise intervention was 11.5 (5.2) weeks. Cancer survivors exercised 3.5 (1.4) d/wk for 48.5 (22.8) minutes per session. The level of physical exertion or average intensity of the aerobic exercise interventions was 5.6 (3.0) METs, corresponding to moderate-intensity exercise (40%–60% Vo₂max), and included walking (48%), stationary cycle ergometry (30%), a combination of walking and cycling (16%), or other modalities of aerobic exercise such as the elliptical trainer or self-selected (6%). The average intensity of resistance training was 4.5 (2.0) METs, corresponding to moderate-intensity exercise (60%–80% 1-repetition maximum, 1-RM), and included weight machines, resistance bands, or free weights (75%). The remaining studies prescribed neuromuscular exercise which commonly included tai-chi or yoga (25%). Flexibility exercise was a component of the exercise in 52% of the exercise interventions. Supervision of exercise sessions was provided in 60% of the exercise interventions.

Ten studies used a theoretical basis for the exercise intervention (44, 48, 50, 54, 57–59, 61, 62, 65). Three interventions (48, 58, 62) followed the Transtheoretical model of behavior change (79, 80), 2 studies (54, 57) followed the model of self-efficacy and stages of exercise change (81), 3 studies (50, 59, 61) followed the Roy adaptation model (82), 1 study (44) followed the Payne adaptation model (83) and 1 study (65) followed the Levine conservation model (84).

Table 3 summarizes weighted mean effect sizes, d₊, for all cancer types collectively, as well as cancer type individually. This analysis indicated that exercise reduced CRF (Table 3 and Fig. 2), yet its impact did not attain significance for survivors of lymphoma, colorectal, or leukemia cancer, which may have lacked sufficient statistical power to detect a difference. Pooled, the effect sizes for the 48 interventions lacked homogeneity, as did the collection of studies addressing breast cancer survivors.

Bivariate regression analyses examined potential sample, methodologic, and exercise intervention characteristics. Significant bivariate models were then integrated into a combined moderator model to explain unique study variance (Table 4). When integrated, the following moderators no longer remained significant: session length (minutes), number of exercise sessions, and treatment with radiation therapy. Four moderators impacting CRF modulation in adult cancer survivors remained significant. Reductions in CRF were greater to the extent interventions: (i) adhered to a theoretical model (compared with those that did not do so; β = 0.48, P = <0.001); (ii) sampled older cancer survivors (β = 0.24, P = 0.04); and (iii), the greatest reductions in CRF occurred with moderate-intensity (3–6 METs, 60%–80%, 1-RM) resistance exercise (β = 0.60, P = 0.01), particularly for higher quality interventions (interaction β = 0.23, P < 0.05). In contrast, lower quality interventions were efficacious in reducing CRF at low (<3 METs) and moderate-intensity (3–6 METs, 60%–80% 1-RM) resistance exercise. Intensity of resistance exercise, use of theory, age, and methodologic quality together explained 52% of the variance among exercise interventions for adult cancer survivors. The estimates in Table 4 reveal that exercise interventions of moderate-intensity (3–6 METs, 60%–80% 1-RM) resistance exercise were successful in reducing CRF, regardless of the use of theory in the exercise intervention, age of the cancer survivor, and methodologic intervention quality. In contrast, interventions of low-intensity resistance (<3 METs, <60% 1-RM) exercise showed no significant reduction of CRF when theory was absent or in high methodologic quality interventions. Time since diagnosis, aerobic exercise, flexibility exercise, or supervision of exercise sessions did not moderate CRF modulation.

Overall, we found that exercise moderately reduced CRF among cancer survivors with an effect size of 0.31 (95% CI: 0.22–0.40), consistent with prior reviews (12, 15). Of note is our new finding that resistance exercise has a positive, quadratic, and exercise intensity dose response effect on CRF. For cancer survivors engaging in moderate-intensity, resistance exercise (3–6 METs, 60%–80% 1-RM) reduced CRF more so than those engaging in lower intensity resistance or aerobic exercise of any level of physical exertion. Another interesting finding was that exercise interventions based on a theoretical model of behavior change or adaptation were more successful in reducing CRF than those interventions not based on such models. Age was also related to CRF reduction, with older cancer survivors reducing CRF to greater levels than younger cancer survivors. Lastly, RCTs of stronger methodologic quality (i.e., higher PEDro score) reduced CRF less than those of weaker methodologic quality. Our findings about exercise interventions based on theoretical models and of higher methodologic quality support previous meta-analytic work examining the influence of exercise on CRF (11). They also update the literature with a larger, more diverse sample of cancer survivors, and types of exercise interventions (11).

Subgroup analysis relating to cancer type revealed that exercise moderately reduced CRF, 0.39 (95% CI: 0.27–0.51) and 0.42 (95% CI: 0.27–0.57), among breast and prostate cancer survivors, respectively. These findings update and support previous meta-analytic reviews advocating the use of exercise as a nonpharmacologic intervention to reduce CRF among breast and prostate cancer survivors (11, 12). Subgroup analysis among leukemia, lymphoma, and colorectal cancer survivors yielded nonsignificant reductions in CRF.

Four meta-analyses have been conducted examining the effect of exercise on CRF (11–14). Two of these meta-analysis have examined the mean reduction of exercise on CRF (13, 14) without accounting for exercise characteristics that may moderate the efficacy of exercise on CRF. The remaining 2 meta-analyses (11, 12) have examined moderators relating to the efficacy of exercise in reducing CRF; however, these meta-analyses were composed of a smaller number of studies [i.e., 17 (11) and 18 studies (12)] and did not examine specific Ex R_x characteristics included in our analysis that may impact CRF modulation. In our meta-analysis of 48 interventions, we found that exercise intensity was a significant moderator of CRF among adult cancer survivors participating in resistance training programs. A positive, quadratic pattern emerged suggesting that moderate-intensity resistance exercise interventions were more efficacious in diminishing CRF than those of lower intensity or aerobic exercise of any level of intensity. Our finding of the efficacy of resistance exercise reducing CRF was somewhat unexpected. Current exercise guidelines for cancer survivors emphasize the importance of participating in aerobic exercise, complimented with resistance and flexibility exercises (ACSM Roundtable; ref. 4) and often make no (National Comprehensive Cancer Network; ref. 3) or minimal mention (American Cancer Society; ref. 17) of resistance exercise.

A possible mechanism for the effectiveness of resistance exercise in reducing CRF among breast and prostate cancer survivors is the attenuation of the progressive muscle wasting and disruptions in muscle metabolism that occur with cancer and associated treatments (85). Several hypotheses related to muscle protein synthesis, ATP dysregulation, cytokine dysregulation, and progressive muscle wasting have all been postulated as mechanistic underpinnings of CRF (85, 86). Moderate-intensity resistance training increases muscle protein synthesis (87), improves cytokine response (88), and diminishes the rate of sarcopenia (89) among healthy human populations as well as those with compromised muscle function such as those with cerebral palsy and other musculoskeletal disorders (90). Further, recent evidence suggests that resistance exercise may provide health benefits such as improved total body muscular strength, self-esteem, and vitality in breast and prostate cancer survivors (49, 72, 91).

Another interesting finding was that older cancer survivors reduced CRF to greater levels than younger cancer survivors engaging in any form of exercise. This finding is of particular importance as most cancer survivors are older ≥65 years (1), yet most exercise interventions have focused on younger cancer survivors (4). Older cancer survivors are frequently challenged with age-related declines in health (i.e., sarcopenia, decreased functional capacity) as well as cancer-related declines in health (e.g., cachexia, body composition changes, decreased bone mineral density; ref. 92). Exercise has been shown to elicit favorable health outcomes among older prostate cancer survivors including, increased lean body mass and muscle strength, and increase distance walked in 6 minutes (72). Improving the status of these health parameters (e.g., body composition, muscular strength, and cardiorespiratory fitness) may influence the mediation of CRF among other populations of cancer survivors.

Exercise interventions that adhered to a theoretical model of behavior change (86, 88) or adaptation model (82) achieved larger reductions in CRF than those that did not adhere to such models. Theoretical models provide empirically supported frameworks that inform behavior change and may offer useful information about determinants of exercise behavior (93, 94). An understanding of exercise behavior and behavioral determinants among cancer survivors may help clinicians identify specific intervention strategies to facilitation adoption and maintenance of an existing exercise program in this population. Theoretical models of adaptation for cancer survivors may be efficacious in improving psychological components of mental health (e.g., distress of cancer diagnosis) potentially influencing CRF modulation. Despite the promise of such interventions, relatively few of the studies implementing a theoretical framework elaborated on the specific role of theory in the exercise intervention. Therefore, the current meta-analysis is limited in its ability to determine the specific underpinnings of theory mediating the reduction in CRF.

This study is subject to several limitations. Despite our comprehensive review of the literature examining CRF in all types of cancer, our search yielded 28 of the 48 exercise interventions that targeted breast (58%) and prostate cancer (10%) survivors exclusively. The large number of interventions examining the impact of exercise on CRF modulation among breast cancer survivors limits the generalizability of our findings to other types of cancer survivors. Moreover, we acknowledge that theories of behavior change and adaptation models are hypothesized to influence fatigue through different mechanisms. As noted, we combined them into a single category because there were relatively few instantiations of theory-led interventions. Despite this limitation, the efficacy of the application of either behavior changes or adaptation models is promising when compared with those not adhering to a prespecified theory or model.

Another limitation relates to the major finding of this meta-analysis, that moderate-intensity resistance exercise may be beneficial in reducing CRF. In particular, no study examined resistance exercise interventions greater than 6 METs (>80% 1-RM). It remains unknown if more vigorous-intensity resistance training would provide greater or lesser reductions in CRF. We did not evaluate adherence to the exercise interventions in this meta-analysis because most studies did not report this information. This variable should have important moderating effects on CRF modulation.

In summary, we confirm with the largest meta-analysis of RCTs conducted to date that moderate resistance exercise reduces CRF among adult cancer survivors, particularly breast and prostate cancer survivors and those of older age. Cancer survivors engaging in moderate-intensity resistance exercise modulated CRF levels more than those engaging in low-intensity resistance exercise or low to moderate intensity, aerobic exercise. Further, the most efficacious exercise interventions were based on behavior change and adaptation theory. Our findings reinforce the notion that exercise interventions for adult cancer survivors should be individualized based on the targeted health outcome and possibly cancer type. In addition, exercise interventions should be multi-dimensional, combining sound exercise as well as behavioral science.

No potential conflicts of interest were disclosed.

We thank Robert D. Siegel, M.D., Gray Cancer Center, Hartford Hospital for reviewing this manuscript and providing valuable feedback.

This research was supported by University of Connecticut Research Foundation Grant 433527 (PIs: B.T. Johnson and L.S. Pescatello)

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.