Stevinson, C., Lawlor, D. A., & Fox, K. R. (2004). Exercise interventions for cancer patients: systematic review of controlled trials. Cancer Causes and Control, 15, 1035–1056.

Databases searched were MEDLINE, EMBASE, Cochrane Controlled Trials Register, CANCERLIT, CINAHL, PsycINFO, and SPORTDiscus through December 2003.

Thirty-three studies (25 randomized trials and eight nonrandomized studies) reported in 40 articles were included in the review. Data were pooled for 10 trials that assessed physical functioning and 12 trials that assessed fatigue.

Trials were included if they tested interventions involving regular exercise of any type (e.g., aerobic, resistance, and flexibility). Exercise could be the sole intervention or could be combined with other interventions (e.g., diet counseling).  Only prospective trials with a control arm were included.  Based on an a priori decision, both nonrandomized and randomized trials were included.  There were no restrictions on the outcomes assessed in trials. Nineteen studies tested aerobic exercise interventions, of which eight used cycle ergometers and eight used walking programs.  Three trials tested resistive exercise, 10 had combined aerobic and resistive programs, and one was based on team sport activities. Most trials compared an exercise intervention with no intervention; six that did not used information training, psychological therapies, stretching, or tai chi as comparison arms.

Trials of single exercise sessions that measured acute effects were excluded, as were trials that only investigated the effects of physiotherapy.  Control arms could not comprise an intervention (e.g., usual care), an alternative intervention (e.g., counseling, relaxation), or a different type of exercise (e.g., aerobic versus flexibility exercises). Trials with healthy or historical control groups were excluded. 

Exercise interventions lasted for 10 weeks or longer in 17 trials and two weeks or less in four studies. The longest intervention period of any trial was 26 weeks. Trial quality was assessed by recording whether the following features were incorporated in the study design:  randomization, allocation concealment, blinding of the main outcome assessment, and intention-to-treat analysis.

• Thirteen trials included patients with breast cancer during or after adjuvant therapy.
• Eleven trials involved adult patients with any cancer, and one involved pediatric cancer survivors.
• The remaining trials included patients undergoing treatment for prostate, lung, colorectal, or stomach cancers; multiple myeloma; or leukemia.

Reductions in cancer-related fatigue were reported in 10 studies, although statistical significance was not reached or not tested for in three of them.  No differences between groups were reported for fatigue in six trials immediately after the intervention or several months later.  Pooling the data from the 12 trials that assessed fatigue suggested that there was no overall effect of exercise on symptoms of fatigue (standardized mean difference [standard deviation (SD)] in fatigue = -0.15 [-0.38, 0.09]).  Heterogeneity between studies was not related to randomization, allocation concealment, intention-to-treat analysis, or choice of control.  However, the effect appeared to vary by population type.  Some evidence existed that no effect was found of exercise on fatigue symptoms in trials that recruited patients with any type of cancer and that those that recruited patients with breast cancer found a modest reduction in symptoms of fatigue among those allocated to exercise (standardized mean difference [SD] = -0.52 [-0.95,-0.09]).  There was no strong evidence of small study bias in this meta-analysis by the Beggs and Egger tests.

The authors generally concluded that there were methodologic limitations associated with many of the studies and that these limitations may have contributed to the inconsistencies among the results.

Limitations included

• Lack of a randomized allocation
• Failure to conceal allocation
• Failure to specify the primary outcome measures
• Small sample sizes
• Multiple statistical testing without adjustment of significance values.

Two studies may have included participants in the control group who had declined to undertake the intervention; in five other studies, it was unclear whether eligibility (including willingness to participate) was determined prior to group allocation or whether the control group may have solely or partly consisted of those who declined to be allocated to exercise.