Flowchart for selection of studies
The distribution of the unweighted effects. Panels A-C represent a continuous forest plot. CI indicates confidence interval.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Herring MP, O’Connor PJ, Dishman RK. The Effect of Exercise Training on Anxiety Symptoms Among PatientsA Systematic Review. Arch Intern Med. 2010;170(4):321–331. doi:10.1001/archinternmed.2009.530
Copyright 2010 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2010
Anxiety often remains unrecognized or untreated among patients with a chronic illness. Exercise training may help improve anxiety symptoms among patients. We estimated the population effect size for exercise training effects on anxiety and determined whether selected variables of theoretical or practical importance moderate the effect.
Articles published from January 1995 to August 2007 were located using the Physical Activity Guidelines for Americans Scientific Database, supplemented by additional searches through December 2008 of the following databases: Google Scholar, MEDLINE, PsycINFO, PubMed, and Web of Science. Forty English-language articles in scholarly journals involving sedentary adults with a chronic illness were selected. They included both an anxiety outcome measured at baseline and after exercise training and random assignment to either an exercise intervention of 3 or more weeks or a comparison condition that lacked exercise. Two co-authors independently calculated the Hedges d effect sizes from studies of 2914 patients and extracted information regarding potential moderator variables. Random effects models were used to estimate sampling error and population variance for all analyses.
Compared with no treatment conditions, exercise training significantly reduced anxiety symptoms by a mean effect Δ of 0.29 (95% confidence interval, 0.23-0.36). Exercise training programs lasting no more than 12 weeks, using session durations of at least 30 minutes, and an anxiety report time frame greater than the past week resulted in the largest anxiety improvements.
Exercise training reduces anxiety symptoms among sedentary patients who have a chronic illness.
Anxiety, an unpleasant mood characterized by thoughts of worry, is an adaptive response to perceived threats that can develop into a maladaptive anxiety disorder if it becomes severe and chronic.1 Anxiety symptoms and disorders are common among individuals with a chronic illness,2-8 yet health care providers often fail to recognize or treat anxiety and may consider it to be an unimportant response to a chronic illness.9
Anxiety symptoms can have a negative impact on treatment outcomes in part because anxious patients can be less likely to adhere to prescribed medical treatments.10,11 Personal costs of anxiety among patients include reduced health-related quality of life12 and increased disability, role impairment,13 and health care visits.14
Adequate evidence is available to justify screening for anxiety problems in primary care settings and prescribing effective treatments for those likely to benefit.9,14 While pharmacological and cognitive behavioral therapies are both efficacious in reducing anxiety,15,16 there continues to be interest in alternative therapies such as relaxation and exercise.17-19
Exercise training is a healthful behavior with a minimal risk of adverse events that could be an effective and practical tool for reducing anxiety among patients.20-22 Meta-analytic reviews have summarized the association between exercise and anxiety symptoms both in samples of primarily healthy adults23-26 and exercise training studies of patients with fibromyalgia and cardiovascular disease, but these analyses did not focus on the best available evidence.27-29
We used the results from randomized controlled trials to evaluate the effects of exercise training on anxiety. One goal was to estimate the population effect size for anxiety outcomes. A second goal was to learn whether variables of theoretical or practical importance, such as features of the exercise stimulus and the method for measuring anxiety, account for variation in the estimated population effect.
This systematic review and meta-regression analysis was conducted in a manner consistent with guidelines set forth in the QUOROM statement.30
Articles published from January 1995 to August 10, 2007, were located using the Physical Activity Guidelines for Americans Scientific Database, developed and maintained by the Division of Nutrition, Physical Activity, and Obesity at the Centers for Disease Control and Prevention's National Center for Chronic Disease Prevention and Health Promotion.20 That search was supplemented by additional searches through December 2008 of the following databases: Google Scholar, MEDLINE, PsycINFO, PubMed, and Web of Science. We used the keywords “exercise,” “physical activity,” “anxiety,” “tension, “randomized trial,” and “randomized controlled trial.” Supplemental searches of the articles retrieved and those supplied by colleagues were performed manually.
Inclusion criteria included (1) English-language articles, (2) sedentary adult participants with a chronic illness, (3) random assignment to either an exercise intervention of at least 3 weeks or a comparison condition that lacked exercise training, and (4) an anxiety outcome measured at baseline and after exercise training.
Investigations were excluded that (1) included exercise as one part of a multicomponent intervention but did not include the additional component (eg, stress management) in a comparison condition, (2) compared exercise only with an active treatment (eg, cognitive behavioral therapy, medication, another mode of exercise), (3) focused on education promotion interventions aimed at increasing physical activity but failed to show that physical activity levels were increased, and (4) used anxiety outcome measures focused on a specific phobia. Figure 1 provides a flowchart of study selection.
Seventy-five effects were derived from 40 studies31-70: 21 from patients with cardiovascular disease, 15 from patients with fibromyalgia, 10 from patients with multiple sclerosis (MS), 9 from patients with psychological disorders, 8 from patients with cancer, 4 from patients with chronic obstructive pulmonary disease (COPD), 4 from patients with chronic pain (eg, knee osteoarthritis, back pain), and 4 from patients categorized as having “other medical illnesses” (ie, obesity, lupus, and epilepsy). The mean (SD) age was 50 (10) years. The mean percentage of women was 59% (33%). Exercise training averaged 3 (1) sessions per week, 42 (22) minutes per session, and was of 16 (10) weeks’ duration. The exercise training adherence rate averaged 78% (14%). Adherence was reported for 51 of 75 (68%) of the effects.
Two of us (M.P.H. and P.J.O.) independently assessed study quality according to randomization methods, baseline differences between treatment groups, the quality of the anxiety outcome measure, adherence, and exercise program descriptions.
Effect sizes were calculated by subtracting the mean change in the comparison condition from the mean change in the experimental condition and dividing the difference by the pooled standard deviation of baseline scores.71 Effect sizes were adjusted for small sample size bias and calculated such that a decrease in anxiety resulted in a positive effect size.71 When exact means and standard deviations were not provided (n = 3), effect sizes were estimated72 from exact P values60,62 and from a figure shown in the study.70 When a standard deviation was not reported (n = 151), it was estimated from the largest other study that used the same anxiety measure.34
Random effects models were used to aggregate mean effect size delta (Δ) and to test variation in effects according to selected moderator variables.71,73 The number of unpublished or unretrieved studies of null effect that would diminish the significance of observed effects to P > .05 was estimated as fail-safe N+.74 A 2-way (effects × raters) intraclass correlation coefficient (ICC) for absolute agreement was calculated to examine interrater reliability for the calculation of effect sizes. The initial ICC was 0.93, and discrepancies were resolved.
To provide focused research hypotheses about variation in effect size,75 6 primary moderator variables were selected (Table 1): exercise program length, session duration, and change in physical fitness20,76; type of comparison group and type of intervention used (single [exercise alone vs nonexercise comparison] vs multiple [eg, exercise + medication vs medication] interventions)77; and the time frame of anxiety report (eg, right now vs past week).23,78
Each primary moderator level was coded according to planned contrasts79 (P ≤ .05) among its levels when the number of effects (k) per level was at least 3. A two-way (effects × raters) mixed-effects model ICC with absolute agreement was calculated to assess interrater reliability for coding of moderator variables. The initial ICCs were at 0.92 or higher, and discrepancies were resolved. The 6 primary moderator variables and 2 interaction terms (program length × session duration and comparison group × intervention type) were included in mixed-effects multiple linear regression analysis with maximum-likelihood estimation,71,73 adjusting for nonindependence of multiple effects contributed by single studies.80 Tests of the regression model (QR) and its residual error (QE) are reported.
Secondary moderator variables were selected for descriptive, univariate analyses based on a logical, theoretical, or prior empirical relation with anxiety. They were organized into general categories of patient characteristics (eg, age, sex, and illness), characteristics of the exercise intervention (eg, adherence, exercise mode, frequency, and relative intensity), and the specific anxiety measure used (Table 2).
Mean effect sizes (Δ) and 95% confidence intervals (CIs) were computed for continuous and categorical variables using a random effects model to account for heterogeneity of moderator effects.73
Sixty-six of 75 effects were greater than zero. The distribution of the unweighted effects shown in Figure 2 was positively skewed and leptokurtic. The mean effect size Δ was 0.29 (k = 75 [95% CI, 0.23-0.36]; z = 9.06; P < .001). The fail-safe number of effects was 1525, and a funnel plot (not shown) revealed a lack of publication bias.
The overall multiple regression model was significantly related to effect size (QR(9) = 20.89; P = .01; R2 = 0.27; QE(65) = 56.29; P = .77). Exercise program length (β = 0.33; z = 2.55; P = .01), session duration (β = 0.27; z = 2.14; P = .03), and time frame of anxiety report (β = 0.25; z = 2.01; P = .04) were independently related to effect size. In a follow-up regression model, 2-way interactions among these variables were statistically nonsignificant (exercise program length × session duration: P = .71, exercise program length × anxiety report time frame: P = .43, and session duration × anxiety report time frame: P = .95). Planned contrasts showed a larger effect in studies in which (1) exercise training duration was 3 to 12 weeks (Δ = 0.39 [95% CI, 0.28-0.49]) compared with longer durations (Δ = 0.23 [95% CI, 0.15-0.31]; z = 2.38; P = .02), (2) the exercise session duration exceeded 30 minutes (Δ = 0.36 [95% CI, 0.27-0.44]) compared with a combination of the shorter and unspecified session durations (Δ = 0.22 [95% CI, 0.13-0.31]; z = 2.09; P = .04), and (3) the time frame of anxiety report was greater than 1 week (Δ = 0.44 [95% CI, 0.29-0.59]) compared with shorter time frames (Δ = 0.26 [95% CI, 0.19-0.33]; z = 2.78; P = .005).
The number of effects (k), mean Δ effect size, 95% CI, and P value for planned contrasts of each level of primary and secondary moderators are presented in Table 3. Descriptive results for the remaining primary moderators from the overall regression model and for secondary moderators are reported as standardized regression coefficients in Table 4.
The analysis revealed that exercise training significantly decreased anxiety scores among patients with a chronic illness. The magnitude of the overall mean effect (Δ = 0.29) is similar to the effect of exercise training on fatigue symptoms among patients (Δ = 0.37)77 and on cognitive function among older adults (g = 0.30).81
Given the importance for health care providers of knowing the minimum exercise stimulus needed to improve mental health outcomes among patients,76 it is noteworthy that exercise programs of 3 to 12 weeks resulted in significantly larger decreases in anxiety (Δ = 0.39; k = 35) than programs lasting more than 12 weeks (Δ = 0.23; P = .02; k = 39). These results are generally consistent with meta-analytic reviews of the effect of exercise training on depression,82 cognitive function in older adults,81 and quality of life among patients with MS.83 These results also are comparable with the generally expected response time of pharmacological treatments of 4 to 12 weeks for individuals with anxiety.84
It is uncertain why studies with shorter program lengths had larger improvements in anxiety symptoms. One possibility is that the larger effect resulted from better adherence. A moderate inverse association was found between adherence and program length (ρ = −0.42; P = .002; k = 50). The mean (SD) adherence values for program lengths of 3 to 12 weeks (83% [11%]; k = 23) were significantly better (t(48) = 2.67; P = .01) than values for program with lengths greater than 12 weeks (73% [15%]; k = 27). A limitation, however, is that approximately one-third of the overall effects were derived from studies that did not provide information about adherence.
Exercise session durations greater than 30 minutes showed larger effects (Δ = 0.36; k = 40) than durations of 10 to 30 minutes (Δ = 0.22; k = 35). Better mental health outcomes with longer exercise session durations also have been found in studies of exercise training effects on cognitive function in older adults,81,85 and claudication pain reduction among patients with peripheral artery disease.86
As more data are generated, compelling evidence may emerge showing that the moderating effect of session duration is in part a function of its interactions with other relevant variables. We found that mean adherence for session durations of 10 to 30 minutes (74% [14%]; k = 24) was worse than for those lasting longer than 30 minutes (81% [13%]; k = 27), but the difference was not statistically significant (t(49) = −1.85; P = .07). There also is a potential interaction between session duration and the anxiety report time frame. While only 1 effect in the shorter session duration category used an anxiety measure with a report time frame of greater than 1 week, a report time frame of greater than 1 week was used in 23% of effects in the longer session duration category. The mean effect Δ for those studies was 0.62 (95% CI, 0.42-0.82).
The magnitude of the anxiolytic effects of exercise training was larger for investigations using measures with anxiety report time frames that exceeded the past week compared with investigations using measures with time frames of “past week including today” and “right now.” The present analysis may have underestimated the true effect of exercise training on anxiety because of the measurement methods used by most investigators. Although a larger mean effect was found in those studies that asked participants to report anxiety over a time frame that exceeded the prior week, approximately 80% used a shorter report time frame.
Although it is uncertain why most investigators did not use an anxiety report time frame of longer than 1 week, it may have stemmed from a misinterpretation that trait anxiety scores would be insensitive to change in response to an exercise training intervention of a few months. Trait anxiety is conceptualized as a relatively stable measure of individual differences in anxiety proneness,87 yet there is substantial evidence that trait anxiety scores are sensitive to change. Short-term interventions (up to several months) designed to reduce anxiety, including cognitive and behavioral therapies, long-term massage, and relaxation training, produce moderate-to-large reductions in trait anxiety scores.88-91 These changes are consistent with data showing that genetic factors explain only 30% to 50% of the variability in trait anxiety.92 Trait anxiety also discriminates better than state anxiety among patients and samples of people without an illness, particularly among older adults.93
State anxiety responses to an intervention theoretically depend in part on individual differences in trait anxiety.87 Only 5 studies47,55,56,59,67 reported data from both state and trait subscales of the same psychometric instrument. The mean effect Δ for trait measures was 0.56 (95% CI, 0.30-0.83) compared with 0.31 (95% CI, 0.04-0.58) for state measures in these studies. Thus, a limitation to research on the effect of exercise training on anxiety outcomes conducted with people with a chronic illness is the atheoretical nature of the anxiety measurement.
In addition, a randomized controlled trial assumes a lack of systematic error across the premeasurement to postmeasurement trials. There seems to be a greater potential for systematic error associated with the use of a “right now” time frame in exercise training studies because of the greater number of variables that could change state anxiety scores on only 1 day of testing, including psychosocial stressors, circadian timing,94,95 and physical factors such as caffeine96 or light exposure.97
Theory, evidence from our analysis, and findings from related investigations support the idea that future exercise training investigations would benefit from including anxiety outcome measures with a time frame of anxiety report greater than 1 week.
Secondary moderators were included to provide descriptive data about variables that plausibly could moderate the influence of exercise training on anxiety. There was little difference in the effect size across categories for age or sex. The effect sizes also were similar whether the exercise training did or did not meet contemporary recommendations for moderate or vigorous physical activity.98 Secondary moderator variables of special interest, type of illness and adherence, are discussed in more detail in the following 2 subsections.
Exercise training reliably reduced anxiety among subsets of patients with an illness categorized as cardiovascular, cancer, chronic pain, fibromyalgia, psychological, and pulmonary. These results are generally consistent with meta-analytic reviews of the effect of exercise training on anxiety symptoms among patients with fibromyalgia,27 coronary heart disease,28 and cardiac rehabilitation.29 It is important to note that the present analysis and conclusions assume that effects attributed to exercise training among patients were not biased by confounding from unmeasured or unreported factors such as acute exercise bouts performed within a few hours of the preintervention or postintervention testing sessions.
Exercise training results in large symptom reductions among patients with panic99 and depressive disorders (ie, mean effects of 0.85 to 1.1 for patients with depression82,100). Seven of the 9 effects in the psychological category were from studies of patients with depressive disorders.31-33,44 Because 5 of these 7 effects were derived from investigations using a short time frame of anxiety report, the aggregated mean for these 7 effects (Δ = 0.35) may have underestimated the true effect.
The present analysis showed that exercise training reduces anxiety among patients with cancer. This observation differs from that of others who concluded that there was weak evidence for a consistent positive effect of increased physical activity on anxiety among cancer survivors.101 Our analysis differed in that it included both patients with cancer, who exercised during treatment, and survivors who exercised following treatment.
Although anxiety is a common problem among patients with MS, it is often overlooked and poorly treated.7,102 Multiple sclerosis was the only illness for which the mean effect of exercise training was not statistically significant. Although the mean effect size of 0.19 for studies of patients with MS was comparable with the mean effect of studies of patients with cancer, the associated 95% CI for patients with MS encompassed zero. Of the 10 available effects for patients with MS, 1 study55 accounted for 6 effects. The mean for those 6 effects was small (Δ = 0.07) compared with the mean for the remaining 4 effects (Δ = 0.38).
Exercise adherence is integral to the efficacy of exercise training. The ability to meaningfully assess the effects of the intervention decreases as dropout increases. Adherence to an exercise training program may be particularly difficult for patients during treatment.103 Exercise may be unacceptable to some patients (eg, patients with cancer) as indicated by poor adherence.104 The present findings indicated that adherence was also not a significant moderator (z = 1.75; P = .08) of the anxiolytic effect of exercise training. This finding may be due in part to a larger number of effects for which no corresponding adherence data were provided (k = 24).
Several research needs are suggested by the present findings. Needed are well-designed investigations into the effects of exercise training on anxiety that focus on individuals with an understudied illness, including those with an anxiety disorder, COPD, cancer, chronic pain, epilepsy, lupus, and MS. Also needed is better reporting of study features, especially clear and complete information about medication use and the exercise stimulus. Exercise training dose is a complex stimulus involving actual minutes of exercise in each session accumulated over all exercise sessions. Most investigators reported planned session duration as a proxy for actual time spent exercising, and only 1 study38 reported the degree to which patients complied with the prescribed exercise training during exercise sessions. Consequently, the true effect of exercise training on anxiety may be underestimated because of underexposure to the active feature of the intervention.
A better understanding of the role of exercise stimulus variables in maximizing positive mental health outcomes could be realized through investigations that (1) examine useful types of exercise that have been understudied, including resistance exercises; (2) compare different exercise training intensities and durations while controlling total energy expenditure to better understand the minimal and optimal dose necessary to elicit mental health benefits; and (3) select characteristics of the exercise stimulus to optimize program adherence and compliance with intensity and duration prescription. Findings also underscore the importance of including a valid measure of persistent anxiety to better quantify and understand the chronic effects of exercise training on anxiety symptoms.
Increasingly, efforts are being made to provide mental health treatments consistent with the available scientific evidence in primary care settings.105 The present results provide clinicians with solid evidence to recommend exercise training to patients as a means for reducing anxiety symptoms with minimal risk of adverse events. Exercise training may be especially useful for patients who prefer nonpharmacologic treatments106 because such preferences may influence the magnitude of the treatment outcomes.107 Perhaps most importantly, the results show that anxiety reduction is a favorable, adventitious outcome of exercise interventions that were designed as a primary treatment or adjuvant for medical conditions other than anxiety.
Correspondence: Matthew P. Herring, MS, MEd, Department of Kinesiology, Ramsey Center, The University of Georgia, 330 River Rd, Athens, GA 30602-6554 (firstname.lastname@example.org).
Accepted for Publication: October 1, 2009.
Author Contributions: Mr Herring had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Herring and O’Connor. Acquisition of data: Herring. Analysis and interpretation of data: Herring, O’Connor, and Dishman. Drafting of the manuscript: Herring. Critical revision of the manuscript for important intellectual content: Herring, O’Connor, and Dishman. Statistical analysis: Herring and Dishman. Administrative, technical, and material support: Herring. Study supervision: Herring.
Financial Disclosure: None reported.
Create a personal account or sign in to: