Figure 1. Flowchart of study selection. CAT indicates creative arts therapies; k, number of effects; QOL, quality of life; and RCT, randomized clinical trial.
Figure 2. Forest plots of the unweighted distribution of Hedges d effect sizes (95% confidence intervals) for studies assessing anxiety (A) and depression (B).
Figure 3. Forest plots of the unweighted distribution of Hedges d effect sizes (95% confidence intervals) for studies assessing pain (A) and fatigue (B). NA indicates not applicable.
Figure 4. Interaction of cancer group (homogeneous vs heterogeneous) and intervention setting. Error bars indicate standard error.
Figure 5. Forest plot of the unweighted distribution of Hedges d effect sizes (95% confidence intervals) for studies related to quality of life.
Figure 6. Mean delta effect size for posttreatment and follow-up assessments across psychological symptoms and quality of life. * P < .05.
Author Timothy W. Puetz, PhD, MPH, discusses Effects of Creative Arts Therapies on Psychological Symptoms and Quality of Life in Patients With Cancer
Puetz TW, Morley CA, Herring MP. Effects of creative arts therapies on psychological symptoms and quality of life in patients with cancer. JAMA Internal Medicine.. Published online May 13, 2013. doi:10.1001/jamainternmed.2013.836.
eReferences. Bibliography of included trials
eTable 1. Definitions for levels of moderators
eTable 2. Statistical tests of publication bias
eTable 3. Summary of anxiety univariate moderator analysis
eTable 4. Summary of depression univariate moderator analysis
eTable 5. Summary of fatigue univariate moderator analysis
eTable 6. Summary of pain univariate moderator analysis
eTable 7. Summary of QOL univariate moderator analysis
eFigure. Funnel plots
Customize your JAMA Network experience by selecting one or more topics from the list below.
Puetz TW, Morley CA, Herring MP. Effects of Creative Arts Therapies on Psychological Symptoms and Quality of Life in Patients With Cancer. JAMA Intern Med. 2013;173(11):960–969. doi:10.1001/jamainternmed.2013.836
Author Affiliations: Office of the Director, National Institutes of Health, Bethesda, Maryland (Dr Puetz); The ArtReach Foundation, Atlanta, Georgia (Mr Morley); and Department of Epidemiology, University of Alabama at Birmingham (Dr Herring).
Importance Creative arts therapies (CATs) can reduce anxiety, depression, pain, and fatigue and increase quality of life (QOL) in patients with cancer. However, no systematic review of randomized clinical trials (RCTs) examining the effects of CAT on psychological symptoms among cancer patients has been conducted.
Objectives To estimate the effect of CAT on psychological symptoms and QOL in cancer patients during treatment and follow-up and to determine whether the effect varied according to patient, intervention, and design characteristics.
Evidence Review We searched ERIC, Google Scholar, MEDLINE, PsycInfo, PubMed, and Web of Science from database inception to January 2012. Studies included RCTs in which cancer patients were randomized to a CAT or control condition and anxiety, depression, pain, fatigue and/or QOL were measured pre- and post-intervention. Twenty-seven studies involving 1576 patients were included. We extracted data on effect sizes, moderators, and study quality. Hedges d effect sizes were computed, and random-effects models were used to estimate sampling error and population variance.
Findings During treatment, CAT significantly reduced anxiety (Δ = 0.28 [95% CI, 0.11-0.44]), depression (Δ = 0.23 [0.05-0.40]), and pain (Δ = 0.54 [0.33-0.75]) and increased QOL (Δ = 0.50 [0.25-0.74]). Pain was significantly reduced during follow-up (Δ = 0.59 [95% CI, 0.42-0.77]). Anxiety reductions were strongest for studies in which (1) a non-CAT therapist administered the intervention compared with studies that used a creative arts therapist and (2) a waiting-list or usual-care comparison was used. Pain reductions were largest during inpatient treatment and for homogeneous cancer groups in outpatient settings; significantly smaller reductions occurred in heterogeneous groups in outpatient settings.
Conclusions and Relevance Exposure to CAT can improve anxiety, depression, and pain symptoms and QOL among cancer patients, but this effect is reduced during follow-up.
Approximately 40% of adults in the United States report using at least 1 complementary and alternative medicine (CAM) therapy, with prevalence estimates among patients with cancer ranging from 18% to 91%.1-7 Various CAM therapies have improved psychological symptoms frequently associated with cancer and cancer treatment, including cancer-related fatigue,8 pain,9 and symptoms of anxiety and depression.10,11
Creative arts therapies (CATs), including music therapy,12-14 dance/movement therapy,15-18 and various forms of art therapy,19-21 have received less empirical attention than other CAM therapies more commonly used among adults, such as vitamin and nonvitamin supplements and mind-body therapies.1 Although CAT research has been predominately qualitative, clinical research on CAT has expanded from purely observational science to a wider, cross-disciplinary approach that includes fields such as neuropsychiatry.16,22 Prior reviews have suggested that CAT may be a useful adjuvant therapy to improve cancer- and treatment-related symptoms during and after treatment.23,24 For example, recent systematic reviews of CAT among cancer patients concluded that music interventions may have beneficial effects on anxiety, pain, and mood, whereas music and dance therapies may improve quality of life (QOL).15,25 However, no systematic review of randomized clinical trials (RCTs) examining the effects of CAT on psychological symptoms among cancer patients has been conducted.
This systematic review used the results of RCTs to evaluate the effect of exposure to CAT on psychological symptoms and QOL among patients with cancer. The aims of this meta-analysis were to estimate the effect size of exposure to CAT on psychological symptoms (ie, anxiety, depression, pain, cancer-related fatigue) and on QOL among cancer patients during and after treatment and to determine how potential moderators may influence the efficacy of CAT during and after cancer treatment.
The review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.26 Electronic searches of databases were conducted via ERIC (Educational Resource Information Center), Google Scholar, MEDLINE, PsycInfo, PubMed, and Web of Science from database inception to January 2012 using the search terms cancer and (anxiety or depression or pain or fatigue or quality of life) and (art or art therapy or creative arts therapy or dance or drama or music or writing). Searches were restricted to English-language RCTs. Supplemental searches of reference lists from retrieved articles were performed manually.
Included studies compared CAT with no treatment, waiting list, usual care, or placebo control in cancer patients regardless of age, sex, cancer type, cancer stage, or treatment type. Patients could have been receiving treatment, in long-term follow-up, or receiving palliative care. Interventions could take place in an inpatient or outpatient setting and be group or individual based. The authors considered art, dance, drama, music, writing, or combined creative arts modalities. Outcomes included measures of anxiety, depression, pain, fatigue, and/or QOL assessed before and during and/or after exposure to CAT.
Excluded studies explicitly examined mind-body techniques (eg, yoga, meditation, qigong) without including additional features of CAT and/or compared CAT only with an active therapy (eg, pharmacotherapy, counseling). A flowchart of study selection is presented in Figure 1.
The authors independently extracted data and resolved discrepancies by consensus judgment. Effect sizes were calculated by subtracting the mean change in the control condition from the mean change in the treatment condition and dividing the difference by the pooled standard deviation of preintervention scores.27 Effect sizes were adjusted using the Hedges small-sample bias correction and calculated so that decreased anxiety, depression, pain, and fatigue and increased QOL resulted in positive effect sizes.26 When precise means were not reported, effect sizes were estimated28 from F tests29 or Figures.30,31 When precise standard deviations were not reported,31-34 the standard deviation was drawn from published norms or the largest other study using the same measure.
Study quality was assessed using a 15-item scale35 and addressed randomization, sample selection, quality of outcome measures, and statistical analysis. Quality assessment was independently performed by the authors (T.W.P. and M.P.H.) and showed high concordance between raters (intraclass correlation coefficient [3,2], 0.94 [95% CI, 0.87-0.99]).36 According to the Bland-Altman method for limits-of-agreement,37,38 the mean disagreement was close to zero (−0.22 [95% CI, −0.05 to −0.40]), suggesting no evidence for systematic disagreement bias. Quality scores were not used as weights or moderators because of the potential disparity in results that depends on the specific quality scale used.39
To better understand the effect of exposure to CAT on psychological symptoms and QOL during the course of treatment and recovery, separate analyses were performed for investigations assessing anxiety, depression, fatigue, pain, and QOL. Analyses were further subdivided by posttreatment and follow-up outcomes.
The MeanES macro (SPSS, version 19.0; SPSS, Inc) was used to calculate the aggregated mean effect size delta value (Δ), associated 95% confidence interval, and sampling error variance according to a random-effects model.40 We used random-effects models to account for between-studies heterogeneity associated with study-level sampling error and population variance.40 Each effect was weighted by the inverse of its variance and reestimated after the random-effects variance component was added.31 Heterogeneity and consistency were evaluated with the Q statistic and the I2 statistic, respectively.41 Heterogeneity also was examined relative to observed variance and was indicated if the sampling error accounted for less than 75% of the observed variance.27 Publication bias was addressed by inspection of a funnel plot42 and quantified with rank correlation and regression methods.42,43
To provide focused research hypotheses about the effects of exposure to CAT on treatment symptoms among cancer patients,44 primary moderators were selected for each model that met criteria for heterogeneity of effects. Variable selection was based on logical, theoretical, or prior empirical relation to CAT and outcomes. Two variables were selected for the anxiety model (ie, therapeutic monitoring and comparison type). Three variables were selected for the pain model (ie, intervention setting, homogeneity of the cancer group, and the intervention setting × homogeneity of group interaction). Variable definitions are provided in eTable 1.
Using the MetaReg macro (SPSS, version 19.0), primary moderator variables were included in a weighted, least-squares, multiple regression analysis with maximum-likelihood estimation27,40 adjusted for nonindependence of multiple effects contributed by single studies.45 Test results of the regression model (the QR statistic) and its residual error (the QE statistic) are reported. Significant categorical moderators were decomposed using a random-effects model to compute mean effect sizes and 95% confidence intervals.40
Secondary moderators were selected for descriptive, univariate analyses based on logical, theoretical, or prior empirical relation with CAT and/or outcomes and grouped into participant characteristics, intervention characteristics, and study design characteristics. Variable definitions are provided in eTable 1. We computed mean effect sizes and 95% confidence intervals for continuous and categorical variables using a random-effects model.40
Twenty-seven trials of 1576 patients were included in the meta-analysis and are presented in the eReferences. Characteristics of these trials and study quality assessment are presented in Table 1. Variables of CAT are provided in Table 2. Funnel plots for all models were inspected and found to be roughly symmetrical (eFigure). The Begg rank correlation and Egger regression analyses were not statistically significant for any model, suggesting absence of publication bias (eTable 2).
Anxiety was significantly reduced after exposure to CAT interventions (Δ = 0.28 [95% CI, 0.11-0.44]; z = 3.26 [P = .001]). Distribution of the 25 effects is presented in Figure 2A. The effect was heterogeneous (QTotal(24) = 56.65 [P < .001]). Sampling error accounted for 44.0% of the observed variance. The effect was moderately consistent across studies (I2 = 59.4% [95% CI, 49.2%-67.6%]).
The overall multiple regression model for anxiety was significantly related to effect size (QR(3) = 17.44 [P < .001]; R2 = 0.41; QE(21) = 24.76 [P = .26]). Therapeutic monitoring (β = 0.46; z = 2.99 [P = .003]) and type of comparison (β = 0.51; z = 3.12 [P = .002]) were independently related to effect size. When the number of effects allowed decomposition of these variables, larger improvements were found for studies in which (1) the intervention was administered by a non-CAT therapist (Δ = 0.32 [95% CI, 0.13-0.51]) compared with those delivered by a CAT therapist (Δ = 0.17 [−0.12 to 0.46]) and (2) a waiting-list or usual-care comparison (Δ = 0.37 [0.20-0.54]) compared with a placebo condition was used (Δ = −0.04 [−0.35 to 0.28]).
Anxiety was not significantly reduced during the period after exposure to CAT interventions (Δ = 0.08 [95% CI, −0.26 to 0.42]; z = 0.46 [P = .64]). Distribution of the 5 effects is presented inFigure 2A. The effect was homogeneous (QTotal (4)= 6.92 [P = .14]). Sampling error accounted for 58.3% of the observed variance. The effect was moderately consistent across studies (I2 = 56.7% [95% CI, 24.8%-75.0%]).
Depression was significantly reduced after exposure to CAT interventions (Δ = 0.23 [95% CI, 0.05-0.40]; z = 2.49 [P = .01]). Distribution of the 11 effects is presented inFigure 2B. The effect was homogeneous (QTotal(10) = 11.47 [P = .32]). Sampling error accounted for 87.3% of the observed variance. The effect was consistent across studies (I2 = 21.5% [95% CI, 0.0%-45.1%]).
Depression was not significantly reduced during the period after exposure to CAT interventions (Δ = −0.09 [95% CI, −0.42 to 0.22]; z = 0.61 [P = .54]). Distribution of the 5 effects is presented in Figure 2B. The effect was homogeneous (QTotal(4) = 6.16 [P = .19]). Sampling error accounted for 65.2% of the observed variance. The effect was moderately consistent across studies (I2 = 51.2% [95% CI, 15.6%-72.0%]).
Pain was significantly reduced after exposure to CAT interventions (Δ = 0.54 [95% CI, 0.33-0.75]; z = 5.04 [P < .001]). Distribution of the 18 effects is presented in Figure 3A. The effect was heterogeneous (QTotal(17) = 52.15 [P < .001]). Sampling error accounted for 36.5% of the observed variance. The effect was moderately consistent across studies (I2 = 69.3% [95% CI, 60.6%-76.1%]).
The overall multiple regression model for pain was significantly related to effect size (QR(4) = 33.98 [P < .001]; R2 = 0.65; QE(13) = 18.07 [P = .16]). The intervention setting × homogeneity of patient interaction was independently related to effect size (β = 0.71; z = 1.30 [P = .02]). Significantly smaller effects were found in studies with heterogeneous cancer groups exposed to CAT in outpatient settings (Δ = 0.10 [95% CI, −0.12 to 0.31]) compared with the average effect for all other groups (Δ = 0.81 [0.65-0.96]; QBetween(1) = 27.95 [P < .001]) (Figure 4).
Pain was significantly reduced during the period after exposure to CAT interventions (Δ = 0.59 [95% CI, 0.42-0.77]; z = 6.51 [P < .001]). Distribution of the 7 effects is presented inFigure 3A. The effect was homogeneous (QTotal(6) = 8.59; P = .20). Sampling error accounted for 70.3% of the observed variance. The effect was consistent across studies (I2 = 41.8% [95% CI, 7.5%-63.4%]).
Fatigue was not significantly reduced after exposure to CAT interventions (Δ = 0.16 [95% CI, −0.04 to 0.37]; z = 1.54 [P = .12]). Distribution of the 7 effects is presented inFigure 3B. The effect was homogeneous (QTotal(6) = 2.22 [P = .90]). Sampling error accounted for 99.9% of the observed variance. The effect was consistent across studies (I2 = 0.0% [95% CI, 0.0%-7.6%]).
Quality of life was significantly increased after exposure to CAT interventions (Δ = 0.50 [95% CI, 0.25-0.74]; z = 3.98 [P < .001]). Distribution of the 6 effects is presented in Figure 5. The effect was homogeneous (QTotal(5) = 5.09 [P = .41]). Sampling error accounted for 98.1% of the observed variance. The effect was consistent across studies (I2 = 21.4% [95% CI, 0.0%-49.1%]).
Quality of life was not significantly increased during the period after exposure to CAT interventions (Δ = 0.22 [95% CI, −0.09 to 0.54]; z = 1.40 [P = .16]). Distribution of 6 of the effects are presented inFigure 5. The effect was homogeneous (QTotal(5) = 10.11; P = .16). Sampling error accounted for 52.7% of the observed variance. The effect was moderately consistent across studies (I2 = 60.4% [95% CI, 35.4%-75.8%]).
The number of effects (k), mean effect size (Δ), 95% confidence interval, P value, and I2 value for each level of each moderator for the anxiety, depression, pain, fatigue, and QOL models are presented in eTable 3 through 7, respectively. These results represent descriptive, univariate analyses and should be interpreted accordingly.
The cumulative evidence summarized in this review indicates that exposure to CAT reduces symptoms of anxiety, depression, and pain and improves QOL among cancer patients after treatment. The magnitude of the effects is generally diminished during follow-up. Exposure to CAT did not significantly reduce symptoms of fatigue after treatment or during follow-up (Figure 6). These findings are consistent with the findings of previous reviews of the positive effects of CAT on anxiety, pain, mood, and QOL among cancer patients.15,25 The magnitude of the overall effects of CAT exposure on symptoms of anxiety, depression, pain, and QOL was small but similar to improvements reported for other CAM therapies among cancer patients, including (1) mindfulness-based therapy on anxiety and depression62; (2) acupuncture and massage therapy on pain intensity63; (3) yoga on anxiety, depression, and QOL46,64; and (4) exercise on anxiety,65 depression,66,67 and QOL.68 The effects of CAT exposure on cancer-related fatigue is more difficult to interpret. These effects may be modality dependent such that movement-based creative expression has effects more comparable to those seen in exercise studies than other CAT modalities.15,69,70 Although exposure to CAT elicited significant improvements in depression and QOL, the mean effects were found to be homogeneous and therefore were not subjected to moderator analyses. Significant findings for anxiety and pain are discussed in the following sections.
Anxiety reduction was strongest for studies in which (1) the intervention was administered by a non-CAT therapist and (2) a waiting-list or usual-care comparison was used. Why larger anxiety reductions resulted from CAT interventions not administered by CAT therapists is uncertain. Treatment differences may result from interventions administered by CAT therapists who have undergone the rigorous training and credentialing expected of nationally certified CAT therapists compared with interventions administered by non-CAT therapists. Another possibility is related to the need to reduce tension in the perceived polarization of CAM and biomedicine in medical settings. Such tension can be an important factor in shaping cancer patients' first impressions and influencing their confidence in CAM practices.71-73 Although openness to experience may predicate the use of provider-directed CAM, clinical distress has predicated the use of self-directed CAM.74 This perception of openness is likely related to a need for better integration of CAM and conventional medicine in medical settings. Patients do not necessarily expect clinicians to believe in the philosophy of CAM, but they do want medical approval and to know that their CAM choices are reasonable and safe.72,73 Therefore, material used by practitioners to explain CAM to potential patients should avoid challenging patients' beliefs about the perceived disadvantages and instead focus on the positive and preventive effects of CAM.75
Larger anxiety reductions also resulted from investigations that used a waiting-list or usual-care control condition rather than a placebo control. Our analysis did not permit a rigorous decomposition of this effect. However, these findings suggest that future well-designed trials may benefit from the use of a waiting-list or usual-care comparison in addition to intervention and placebo conditions to control for differences in expectancy, conditioning, and meaning.76,77 Previous studies have highlighted the need to examine the placebo effect in alternative medicine.77-79 Researchers need to focus on factors that influence expectancy and possible mediators of the placebo effect. For example, younger women with a higher level of education and patients with greater clinical distress or a longer duration after cancer diagnosis are more likely to use CAM and critically engage clinicians regarding CAM and biomedical care.71,80 These issues of expectancy are at least partially independent of known direct biological effects of interventions and require improving clinical trial design and interpretation of nonspecific healing responses that constitute the placebo effect.
Pain reduction was largest for studies conducting interventions during inpatient treatment and with homogeneous cancer groups in outpatient settings. Significantly smaller reductions occurred in heterogeneous cancer groups in outpatient settings. We are uncertain why less pain reduction resulted from CAT interventions in heterogeneous cancer groups in outpatient settings. One possibility is that the openness to experience predicated the use of provider-directed CAM.74 For example, inpatient and outpatient consultation services have shown success in addressing questions raised by the possible integration of CAM therapies with conventional care, particularly among patients who have severe, chronic, or incurable conditions and likely need inpatient facilities.81 This type of integration may explain the similar effects found in inpatient groups. The differential effect of cancer groups in outpatient settings is more difficult to interpret. Complementary and alternative medicine therapies might be more useful in augmenting traditional analgesic therapy in certain cancer outpatient groups who cannot tolerate or may be reluctant to take pain medications.63 Integrated services may be highly valued by these types of cancer patients who have traditionally preferred their complementary health care to be provided in a nonmedicalized environment.72 Prior reviews have provided striking observations about the paucity of well-designed trials evaluating CAM interventions for cancer-related pain.9 Because the available literature suggests a large degree of heterogeneity regarding the design and administration of CAT related to cancer-related pain, the present findings may be particularly noteworthy with regard to methodological issues within CAT research.
The included trials had notable limitations. Many lacked well-validated symptom assessments among cancer patients82 and adequate information regarding features of the intervention, appropriateness of comparisons, adherence rates, and medication use. These limitations emphasize the importance of adoption of and adherence to reporting guidelines to improve the quality of future trials. To inform the design of appropriate CAT interventions and offer insight into putative biopsychosocial mechanisms of symptom reduction during and after cancer treatment, well-designed RCTs should (1) seek to better characterize the features of the CAT intervention (ie, certified instruction, frequency, session duration, program length, and modality); (2) examine CAT exposure on the interrelationship between neurobiological and psychological measures of cancer symptoms; and (3) investigate the mechanistic similarities, differences, and interactions among various CAT modalities, psychosocial interventions, and pharmacologic treatments used to improve psychological symptoms in patients with cancer.
This systematic review offers a unique look into the potential benefits of CAT that may guide further hypothesis-driven investigation into adjuvant treatments to improve conventional disease management. The cumulative evidence indicates that CAT can decrease symptoms of anxiety, depression, and pain and increase QOL among cancer patients after treatment. The effects are greatly diminished during follow-up. Future well-designed RCTs are needed to address the methodological heterogeneity found within this field of research.
Correspondence: Matthew P. Herring, PhD, Department of Epidemiology, University of Alabama at Birmingham, 1530 Third Ave S, Ryals Public Health Bldg 210E, Birmingham, AL 35294 (firstname.lastname@example.org).
Accepted for Publication: January 26, 2013.
Published Online: May 13, 2013. doi:10.1001/jamainternmed.2013.836
Author Contributions: Dr Puetz had full access to all the data and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Puetz. Acquisition of data: Puetz and Herring. Analysis and interpretation of data: Puetz, Morley, and Herring. Drafting of the manuscript: Puetz, Morley, and Herring. Critical revision of the manuscript for important intellectual content: Puetz, Morley, and Herring. Statistical analysis: Puetz and Herring. Administrative, technical, or material support: Puetz and Morley. Supervision: Puetz and Herring.
Conflict of Interest Disclosures: None reported.