Comparison of Pharmaceutical, Psychological,
and Exercise Treatments for Cancer-Related Fatigue
A Meta-analysis
Karen M. Mustian, PhD, MPH; Catherine M. Alfano, PhD; Charles Heckler, PhD, MS; Amber S. Kleckner, PhD; Ian R. Kleckner, PhD;
Corinne R. Leach, PhD; David Mohr, PhD; Oxana G. Palesh, PhD, MPH; Luke J. Peppone, PhD, MPH; Barbara F. Piper, PhD;
John Scarpato, MA; Tenbroeck Smith, MA; Lisa K. Sprod, PhD, MPH; Suzanne M. Miller, PhD
IMPORTANCE Cancer-related fatigue (CRF) remains one of the most prevalent and
troublesome adverse events experienced by patients with cancer during and after therapy.
OBJECTIVE To perform a meta-analysis to establish and compare the mean weighted effect
sizes (WESs) of the 4 most commonly recommended treatments for CRF—exercise,
psychological, combined exercise and psychological, and pharmaceutical—and to identify
independent variables associated with treatment effectiveness.
DATA SOURCES PubMed, PsycINFO, CINAHL, EMBASE, and the Cochrane Library were
searched from the inception of each database to May 31, 2016.
STUDY SELECTION Randomized clinical trials in adults with cancer were selected. Inclusion
criteria consisted of CRF severity as an outcome and testing of exercise, psychological,
exercise plus psychological, or pharmaceutical interventions.
DATA EXTRACTION AND SYNTHESIS Studies were independently reviewed by 12 raters in 3
groups using a systematic and blinded process for reconciling disagreement. Effect sizes
(Cohen d) were calculated and inversely weighted by SE.
MAIN OUTCOMES AND MEASURES Severity of CRF was the primary outcome. Study quality
was assessed using a modified 12-item version of the Physiotherapy Evidence-Based
Database scale (range, 0-12, with 12 indicating best quality).
RESULTS From 17 033 references, 113 unique studies articles (11 525 unique participants; 78%
female; mean age, 54 [range, 35-72] years) published from January 1, 1999, through May 31,
2016, had sufficient data. Studies were of good quality (mean Physiotherapy Evidence-Based
Database scale score, 8.2; range, 5-12) with no evidence of publication bias. Exercise
(WES, 0.30; 95% CI, 0.25-0.36; P < .001), psychological (WES, 0.27; 95% CI, 0.21-0.33;
P < .001), and exercise plus psychological interventions (WES, 0.26; 95% CI, 0.13-0.38;
P < .001) improved CRF during and after primary treatment, whereas pharmaceutical
interventions did not (WES, 0.09; 95% CI, 0.00-0.19; P = .05). Results also suggest that CRF
treatment effectiveness was associated with cancer stage, baseline treatment status,
experimental treatment format, experimental treatment delivery mode, psychological mode,
type of control condition, use of intention-to-treat analysis, and fatigue measures (WES
range, −0.91 to 0.99). Results suggest that the effectiveness of behavioral interventions,
specifically exercise and psychological interventions, is not attributable to time, attention,
and education, and specific intervention modes may be more effective for treating CRF at
different points in the cancer treatment trajectory (WES range, 0.09-0.22).
CONCLUSIONS AND RELEVANCE Exercise and psychological interventions are effective for
reducing CRF during and after cancer treatment, and they are significantly better than the
available pharmaceutical options. Clinicians should prescribe exercise or psychological
interventions as first-line treatments for CRF.
JAMA Oncol. 2017;3(7):961-968. doi:10.1001/jamaoncol.2016.6914
Published online March 2, 2017.
Supplemental content
Author Affiliations: Author
affiliations are listed at the end of this
article.
Corresponding Author: Karen M.
Mustian, PhD, MPH, Wilmot Cancer
Institute, Department of Surgery,
University of Rochester Medical
Center, 265 Crittend Blvd, Room
2215, Rochester, NY 14642
(karen_mustian@urmc.rochester.edu).
Research
JAMA Oncology | Original Investigation
(Reprinted) 961
© 2017 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 08/26/2022
C
ancer-related fatigue (CRF) is one of the most common
and disabling adverse effects reported by patients with
cancer during and after treatment.
1-6
Cancer-related
fatigue can persist for years after treatment completion
4,6-11
and
is exacerbated by co-occurring cancer-related adverse effects
such as depression, anxiety, sleep disturbance, and pain.
3,4,12-18
Cancer-related fatigue reduces a patient’s ability to complete
medical treatments for cancer and participate in essential and
valued life activities, thus undermining quality of life and poten-
tially reducing overall survival.
6,9,19
Cancer-related fatigue has
been designated a high-priority research area by the National
Cancer Institute and is 1 of the 5 highest priority research areas
designated by the National Cancer Institute Clinical Oncology
Research Program in the United States.
20
Randomized clinical trials (RCTs)have tested exercise, psy-
chological, exercise plus psychological, and pharmaceutical in-
terventions for the amelioration of CRF.
21-35
Results of these
RCTs are promising; however, development and implemen-
tation of guidelines for clinical practice
36-38
are challenging
owing to the lack of a direct meta-analytic comparison of these
4 most commonly recommended behavioral and pharmaceu-
tical treatments for CRF. Although clinical practice guide-
lines exist for the management of CRF,
36-38
which mode of
treatment is most effective remains unclear.
To our knowledge, no prior review of CRF has applied
meta-analytic methods to compare the efficacy of all 4 major
types of treatments recommended for managing CRF, nor
has any prior review systematically explored factors that are
associated with treatment effectiveness (eg, age, type of can-
cer, during vs completed primary cancer treatment, study
quality) when managing CRF. This information can enhance
a personalized medicine approach when treating CRF and
can inform future research.
The primary purposes of this meta-analysis were to
(1) ascertain a more comprehensive and definitive estimate of
weighted effect sizes for exercise (ie, aerobic, anaerobic or
strength, or both), psychological (ie, cognitive behavioral, psy-
choeducational, or eclectic), the combination of exercise and
psychological, and pharmaceutical interventions used to treat
CRF; (2) to determine which of these 4 interventions signifi-
cantly improves CRF; and (3) to compare the magnitudes of
improvement in CRF produced by each intervention type. The
secondary purpose was to identify independent variables as-
sociated with treatment efficacy for the management of CRF.
Methods
Search Strategy
Methods and reporting for this meta-analysis adhere to the Pre-
ferred Reporting Items for Systematic Reviews and Meta-
Analyses (PRISMA) guidelines and the recommendations of 2
experts (D.M. and S.M.M.) in meta-analytic procedures on the
team.
39,40
We searched the following electronic databases:
PubMed, PsycINFO, CINAHL, EMBASE, and the Cochrane
Library. Articles published in English between the inception
of each database and May 31, 2016, were searched for con-
trolled-vocabulary terms specific to each database related to
CRF, neoplasms, questionnaires, intervention strategies, and
study design (eTable 1 in the Supplement).
Selection Strategy
Study selection strategy was rigorously defined. For inclu-
sion, studies met the following criteria: (1) use of an RCT de-
sign, (2) adult (≥18 years) participants with cancer, (3) CRF se-
verity measured as an outcome (eTable 2 in the Supplement
for fatigue measures), (4) evaluation of CRF severity not solely
as an adverse effect of cancer treatment, (5) no report on a phar-
maceutical intervention that evaluated an erythropoietin drug
because such drugs are used primarily for treating anemia and
are not recommended as a stand-alone treatment for CRF due
to adverse effects, (6) no report of a complementary and al-
ternative intervention with the exception of exercise-based
therapies (ie, yoga, tai chi), and (7) no use of reduced energy,
vitality, or vigor as the fatigue outcome measure because these
constructs are qualitatively different from CRF.
41
Review Strategy
All reviews and data extractions were performed indepen-
dently by at least 3 raters (includes all authors) considered ex-
perts in the field of cancer control and CRF. Data were ex-
tracted using online coding and Excel programs (Microsoft, Inc)
designed specifically for this project. The programs pro-
duced a list of data abstraction and coding discrepancies among
reviewers. All discrepancies were resolved by independent
third-party review and consensus; independent review was re-
quired for 6 studies, and 100% agreement was obtained for all
113 studies. Study investigators were contacted by standard-
ized email letters at least 3 times to provide information
omitted from published articles. To assess the methodologic
quality of the studies, a modified 12-item version of the Phys-
iotherapy Evidence-Based Database (PEDro) scale, devel-
oped using a Delphi expert consensus technique,
42-44
was used
because it identifies studies that are generalizable, internally
valid, and statistically interpretable. The PEDro scale (range,
0-12, with 12 indicating highest quality) accounts for unique
issues regarding blinding of the participant, assessor, or thera-
pist in behavioral trials.
42-44
Delineation of exercise interven-
tions as aerobic, anaerobic, or both and psychological inter-
ventionsas cognitivebehavioral, psychoeducational, or eclectic
was based on descriptions provided in the published articles.
Key Points
Question Which of the 4 most commonly recommended
treatments for cancer-related-fatigue—exercise, psychological,
the combination of exercise and psychological, and
pharmaceutical—is the most effective?
Findings This meta-analysis of 113 unique studies (11 525 unique
participants) found that exercise and psychological interventions
and the combination of both reduce cancer-related fatigue during
and after cancer treatment. Reduction was not due to time,
attention, or education. In contrast, pharmaceutical interventions
do not improve cancer-related fatigue to the same magnitude.
Meaning Clinicians should prescribe exercise and/or psychological
interventions as f irst-line treatments for cancer-related fatigue.
Research Original Investigation Comparison of Treatments for Cancer-Related Fatigue
962 JAMA Oncology July 2017 Volume 3, Number 7 (Reprinted) jamaoncology.com
© 2017 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 08/26/2022
Statistical Analysis
Effect sizes (Cohen d) were computed as the mean difference
in change from pretreatment to posttreatment between the ex-
perimental and control groups, divided by the pooled post-
intervention SDs. The effect sizes were combined across all in-
tervention types with weights based on a random-effects model
(Hedges random effects
45
) to facilitate generalizability of re-
sults and because we expected considerable heterogeneity.
46
Owing to the smaller number of studies, we used fixed-
effects models to combine effect sizes within each interven-
tion type and to model predictors of intervention effective-
ness. Cut points for determining small, moderate, and large
effects were defined as 0 to 0.29, 0.30 to 0.59, and 0.60 or
greater, respectively.
47
Details for the computations are given
online in the eMethods of the Supplement. All analyses were
performed using the metafor package in R (version 3.2).
48
Estimation of Intervention Effectiveness
Tests for significant differences between groups used a fixed-
effects model for categorical independent variables. Method of
moments estimation was used for analysis of continuous inde-
pendent variables in the univariate metaregression.
47
Variables
to be tested for association with intervention effectiveness were
selected a priori and included age, sex, cancer type, cancer stage,
treatmentstatus at baseline (ie, inpatient, outpatient, or mixed),
experimentaltreatment format (ie, group or individual), primary
delivery mode of experimental treatment (ie, in-person only, in-
person plus other [eg, telephone calls, mailings, or web], or no
in-person contact), exercise mode (ie, aerobic, resistance or
nonaerobic, or combined), psychological mode (ie, psychoedu-
cational, cognitive behavioral, or eclectic), type of control
comparison (ie, no intervention, standard care, or wait-list vs
placebo,time, attention, and education), allocation concealment,
intention-to-treat analysis, use of treatment fidelity protocol,
PEDro scale quality score, and fatigue scale used.
49-53
Sensitivity Analyses
Sensitivity analyses were conducted because of studies with
multiple treatment conditions that resulted in 2 or more in-
tervention comparisons (eg, treatment 1 vs control and treat-
ment 2 vs control) from the same study. To detect an artificial
reduction of heterogeneity and a bias in the overall mean ef-
fect size, we conducted analyses in which we included only 1
comparison per study at a time.
Bias Analyses
Publication bias was tested by examining funnel plots and the
trim and fill procedure of Duval and Tweedie.
54
To examine
stability of the overall effect, fail-safe number was calculated
to determine the number of studies with a null effect size that
was needed to reduce the overall effect to nonsignificance.
55
Results
Studies
We selected more than 17 033 titles and abstracts for initial re-
view. An article was excluded if information in the title and ab-
stract indicated it was not an RCT, it did not assess fatigue, or
it used an ineligible intervention method. We selected 351 ar-
ticles for full review. One hundred seventy-eight articles did
not meet inclusion criteria (eg, nonrandomization, assessed
vigor rather than fatigue, ineligible intervention method) and
were eliminated; 60 of the remaining 173 articles did not pro-
vide sufficient data for calculation of effect sizes, even after
querying the authors multiple times. Ultimately, we ana-
lyzed 113 unique studies (eTable 3 in the Supplement) and cal-
culated 127 effect sizes (14 articles had multiple treatment
arms). Of these 127 effect sizes, 69 evaluated exercise inter-
ventions, 34 evaluated psychological interventions, 10 evalu-
ated the combination of exercise and psychological interven-
tions, and 14 evaluated pharmaceutical interventions. Figure 1
displays the PRISMA study selection flowchart.
56,57
Participants
The 113 included studies yielded a sample of 11 525 unique par-
ticipants (78% female and 22% male). Fifty-three studies
(46.9%) were performed among women with breast cancer and
the remaining studies were performed among patients with
other cancer types. Fifty-four studies included only women
and 10 studies included only men. The mean age of partici-
pants was 54 (range, 35-72) years across all studies. Race, edu-
cational level, and partner status could not be accurately sum-
marized owing to missing data. With regard to cancer stage,
Figure 1. PRISMA Diagram
17
033 Potentially relevant articles
identified and screened for
retrieval by title and abstract
15
013 Excluded based on
title and abstract
351 Retrieved for full article review
178 Excluded based on
inclusion criteria
173 Met inclusion criteria
60 Provided insufficient data
for effect size calculations
113 Provided sufficient data
for effect size calculations
14 Have 2 intervention conditions
compared with a control condition
28 Effect sizes calculated
99 Have 1 intervention condition
compared with a control condition
99 Effect sizes calculated
127 Comparisons included in qualitative
and quantitative analysis
Flow of study screening, final inclusion, and effect size calculations are depicted.
Comparison of Treatments for Cancer-Related Fatigue Original Investigation Research
jamaoncology.com (Reprinted) JAMA Oncology July 2017 Volume 3, Number 7 963
© 2017 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 08/26/2022
50 studies (44.2%) enrolled patients with nonmetastatic can-
cer; 11 studies (9.7%), with metastatic cancer; and 33 studies
(29.2%), with metastatic and nonmetastatic cancer. The re-
maining 19 studies (16.8%) did not provide staging informa-
tion. With regard to primary treatments, 51 studies (45.1%) en-
rolled patients receiving primary treatment (defined as surgery,
chemotherapy, and/or radiotherapy) during the study inter-
vention, 45 studies (39.8%) enrolled patients who had al-
ready completed primary treatments, 15 studies (13.3%) en-
rolled patients of mixed treatment status (during and after
primary treatment), and 2 studies (1.8%) did not provide suf-
ficient information on treatment status. Recruitment for these
studies was conducted primarily in medical clinics using sys-
tematic screening and mixed recruitment strategies.
Intervention and Control Conditions
Mean (SD) sample size was 102 (95.5) at baseline with 47 (47.3)
participants in the control groups and 57 (49.0) participants
in the intervention groups at baseline. Mean duration of in-
terventions was 14 (range, 1-60) weeks, included a mean of 43
(range, 1-364) sessions, and sessions lasted a mean of 60 (range,
16-150) minutes. With regard to control interventions, 77 stud-
ies (68.1%) used standard cancer care, no intervention, or wait-
list control, whereas 36 studies (31.0%) used a placebo, time,
attention, or education control. Two pharmaceutical studies
tested paroxetine hydrochloride; 4, modafinil or ar-
modafinil; 5, methylphenidate hydrochloride or dexymeth-
ylphenidate; 1, dexamphetamine; and 1, methylpredniso-
lone. Thirty-six exercise studies tested aerobic modes of
exercise, 13 tested anaerobic modes, and 20 tested a combi-
nation of aerobic and anaerobic modes. Nineteen psychologi-
cal studies tested a cognitive behavioral method, 14 tested a
psychoeducational method, and 1 tested an eclectic method
(a unique combination of psychotherapeutic methods). Ten
studies tested a combined exercise plus psychological inter-
vention. Ninety-nine studies used a traditional 2-arm RCT de-
sign (ie, intervention vs control), whereas 14 studies used a
3-arm RCT design (ie, intervention 1 vs intervention 2 vs con-
trol). eTable 3 in the Supplement provides a detailed sum-
mary of all included studies.
Quality of Studies
The mean PEDro scale score for all studies was 8.2 (range, 5-12),
suggesting that the studies were of good quality. In all 113 stud-
ies, inclusion and exclusion criteria for study participants were
specified; random allocation was used for group assignment,
and between-group statistical comparisons were reported for
CRF severity. Seventy studies (61.9%) used intention-to-treat
analyses; 32 studies (28.3%) concealed allocation from par-
ticipants or blinded outcome assessors; and 38 studies (32.7%)
monitored treatment quality, fidelity, and drift.
Meta-analysis Main Effects
Changes in CRF by Intervention Type
We found significant moderate improvementsin CRF (weighted
effect size [WES], 0.33; 95% CI, 0.24-0.43; P < .001) across all
113 studies, including all 4 intervention types (ie, exercise
[n = 69], psychological [n = 34], exercise plus psychological
[n = 10], and pharmaceutical [n = 14] from before to after in-
tervention). Studies that intervened with exercise demon-
strated the largest overall improvement in CRF, with signifi-
cant moderate effects (WES, 0.30; 95% CI, 0.25-0.36; P < .001).
Studies using psychological interventions exhibited similar im-
provements in CRF (WES, 0.27; 95% CI, 0.21-0.33; P < .001).
Studies that delivered the combination of exercise plus psy-
chological interventions also exhibited similar improve-
ments in CRF (WES, 0.26; 95% CI, 0.13-0.38; P < .001). Phar-
maceutical interventions yielded significant but very small
improvements in CRF (WES, 0.09; 95% CI, 0.00-0.19; P = .05).
Comparisons across all 4 intervention types revealed that ex-
ercise, psychological, and exercise plus psychological inter-
ventions produced significantly greater improvements in CRF
compared with pharmaceutical interventions, with no other
demonstrated differences between intervention types (Figure 2
and eFigure 1 in the Supplement depict forest plots).
Independent Variables Associated With
Intervention Effectiveness
We tested whether each of 15 variables listed in the Methods
section was associated with the effectiveness of all 4 inter-
vention types for improving CRF per their WES (for all data
and P values, see Table). Results suggest that intervention
effectiveness is associated with the following 8 variables:
cancer stage (nonmetastatic, metastatic, or mixed), treat-
ment status at baseline (during primary treatment, after pri-
mary treatment, and mixed), experimental treatment format
(group or individual), primary delivery mode of experimental
treatment (in-person, in-person plus other, or no in-person
contact), psychological mode (psychoeducational, cognitive
behavioral, or eclectic), type of control condition, use of
intention-to-treat analysis, and fatigue scale used. Although
improvements in CRF were reported by all patients and
Figure 2. Forest Plot of Weighted Effect Sizes (WESs)
0 0.400.30
Overall WES
0.10 0.20
Intervention
No. of
Effect Sizes WES SE (95% CI)
More favorable
All 127 0.33 0.05 (0.24-0.43)
Pharmaceutical 14 0.09 0.05 (0.00-0.19)
Exercise plus psychological 10 0.26 0.07 (0.13-0.38)
Psychological 34 0.27 0.05 (0.21-0.33)
Exercise 69 0.30 0.03 (0.25-0.36)
Overall WES across all interventions,
exercise interventions, psychological
interventions, exercise plus
psychological interventions, and
pharmaceutical interventions.
Different sizes of markers
indicate weight.
Research Original Investigation Comparison of Treatments for Cancer-Related Fatigue
964 JAMA Oncology July 2017 Volume 3, Number 7 (Reprinted) jamaoncology.com
© 2017 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 08/26/2022
Table. Factors Associated With Intervention Effectiveness on CRF
Variable
a
Overall WES (95% CI) P Value
No. of
Effect Sizes
Cancer stage at baseline, all interventions
Only nonmetastatic 0.37 (0.31 to 0.42) <.001 59
Only metastatic 0.29 (0.16 to 0.41) <.001 11
Mix of nonmetastatic and metastatic 0.10 (0.04 to 0.17) .001 35
Treatment status at baseline, all interventions
After primary treatment 0.29 (0.23 to 0.36) <.001 53
Mix during and after primary treatment 0.30 (0.19 to 0.40) <.001 15
During primary treatment 0.22 (0.17 to 0.27) <.001 57
Experimental treatment format, all interventions
Group-based
0.38 (0.31 to 0.46) <.001 35
Individual-based 0.23 (0.18 to 0.27) <.001 79
Individual-, couple-, and family-based 0.23 (−0.64 to 1.09) .61 1
Individual- and group-based 0.02 (−0.13 to 0.17) .77 5
Primary delivery mode of experimental treatment,
all interventions
Web 0.99 (0.21 to 1.78) .01 1
Telephone and print 0.46 (0.04 to 0.89) .03 1
Telephone 0.30 (0.19 to 0.41) <.001 6
In-person 0.28 (0.23 to 0.32) <.001 103
In-person and telephone 0.006 (−0.11 to 0.25) .47 7
In-person and print −0.03 (−0.22 to 0.15) .72 6
In person, telephone, and print −0.36 (−1.12 to 0.40) .35 1
Print −0.91 (−1.53 to −0.30) .004 1
Psychological mode, only psychological interventions
Eclectic 0.78 (0.29 to 1.27) .002 1
Cognitive behavioral therapy 0.37 (0.28 to 0.47) <.001 17
Behavioral 0.32 (0.13 to 0.50) .001 3
Cognitive 0.28 (−0.02 to 0.58) .07 2
Psychoeducational 0.17 (0.08 to 0.26) <.001 17
Motivational interviewing 0.10 (−0.17 to 0.37) .47 2
Cognitive behavioral stress management 0.10 (−0.18 to 0.38) .48 1
Control condition, only exercise and psychological
Standard cancer care, wait-list control 0.31 (0.26 to 0.35) <.001 88
Specific component (ie, time, attention, education) 0.24 (0.17 to 0.31) <.001 23
Use of intention-to-treat analysis, all interventions
None 0.34 (0.27 to 0.40) <.001 46
Used 0.22 (0.17 to 0.26) <.001 79
Fatigue scale, all interventions
Piper Fatigue Scale 0.64 (0.49 to 0.80) <.001 10
Brief Fatigue Inventory 0.31 (0.19 to 0.42) <.001 12
Multidimensional Fatigue Inventory 0.26 (0.13 to 0.39) <.001 9
Functional Assessment of Chronic Illness Therapy 0.22 (0.15 to 0.29) <.001 31
European Organization for Research and Treatment
of Cancer Quality of Life Questionnaire
0.12 (0.02 to 0.22) .02 13
Treatment status at baseline, separated by intervention type
During primary: exercise 0.34 (0.25 to 0.42) <.001 31
During primary: psychological 0.23 (0.15 to 0.31) <.001 18
During primary: exercise and psychological 0.01 (−0.26 to 0.28) .95 2
During primary: pharmaceutical 0.04 (−0.07 to 0.32) .51 6
After primary: exercise 0.26 (0.18 to 0.34) <.001 29
After primary: psychological 0.42 (0.29 to 0.55) <.001 13
After primary: exercise and psychological 0.32 (0.17 to 0.47) <.001 7
After primary: pharmaceutical 0.08 (−0.17 to 0.32) .55 4
(continued)
Comparison of Treatments for Cancer-Related Fatigue Original Investigation Research
jamaoncology.com (Reprinted) JAMA Oncology July 2017 Volume 3, Number 7 965
© 2017 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 08/26/2022