Five studies examined the effects of music listening on psychological distress by use of the Profile of Mood States (POMS) (Cadigan 2001; Hermele 2005, Leist 2011; Schou 2008; Stein 2010) and one study by means of the Brief Symptom Inventory (BSI) (Mandel 2007a). The pooled estimate of those studies that used the POMS (N = 228) indicated a small beneficial effect of music interventions on distress, and this result was consistent across studies (mean difference (MD) = -1.26, 95% confidence interval (CI) -2.30 to -0.22, P = 0.02, I² = 0%) (Analysis 1.1). Mandel 2007a (N = 68) did not find statistically significant differences for psychological distress between the music therapy group and the standard care control group at posttest but did find an effect size of 0.54 at four-month follow-up.
Sixteen studies examined the effects of music on anxiety, seven of which reported mean state anxiety as measured by the Spielberger State-Trait Anxiety Inventory (STAI). Nine studies reported mean anxiety measured by other scales such as numeric rating scale and visual analogue scale.
We first conducted an overall analysis of studies that used the STAI to measure anxiety as well as those that used other scales. In order to pool the effect sizes of studies that use different scales, we only included studies that reported post-intervention scores in the analysis.The standardized mean difference (SMD) of those studies (10 studies, N = 353) that reported post-intervention anxiety scores, regardless of the scale used, revealed a moderate (Cohen 1988) effect favoring music interventions (SMD = -0.70, 95% CI -1.17 to -0.22, P = 0.004), but results were inconsistent across studies (I² = 77%). Grouping the studies by participant type (myocardial infarction (MI) patients, surgical/procedural patients, rehabilitation patients) (Analysis 1.2) did not resolve heterogeneity. However, grouping the studies by music preference led to homogeneous results for the four studies (N = 144) that used participant-selected music (SMD = -0.89, 95% CI -1.42 to -0.36, P = 0.001, I² = 48%) but not for the five that used researcher-selected music (N = 179) (SMD = -0.74, 95% CI -1.55 to 0.08, P = 0.08, I² = 85%) (Analysis 1.3).
When pooling studies that only used STAI State Anxiety form (STAI-S) to measure state anxiety (seven studies, N = 310), significantly lower state anxiety was found in participants who received standard care combined with music interventions than those who received standard care alone (MD = -4.58, 95% CI -7.78 to -1.39; P = 0.005). However, considerable statistical heterogeneity remained (I² = 88%). Six out of these seven studies included participants with MI and one study included surgical patients. Pooling the results of only the MI studies (N = 243) resulted in a larger effect size that was more homogeneous (MD = -5.87, 95% CI -7.99 to -.3.75, P < 0.00001, I² = 53%) (Analysis 1.4.1).
We then explored whether music preference affected the outcome of state anxiety as measured by STAI-S. Grouping the studies by whether participant-preferred music (three studies, N = 167) or researcher-selected music (four studies, N = 143) was used did not reduce heterogeneity (Analysis 1.5). However, as noted above, all studies in this subgroup analysis were MI studies except for one (Barnason 1995). A subsequent analysis from which we excluded this surgical study suggested a greater anxiety-reducing effect and homogeneity for studies that used participant-preferred music. Studies of people with MI (two studies, N = 100) that used participant-preferred music resulted in an average anxiety reduction of 7.36 units on the STAI (95% CI -9.45 to -5.27, P < 0.00001, I² = 0%) compared to a reduction of 4.68 units for those studies that used researcher-selected music (four studies, N = 143) (95% CI -8.27 to -1.10, P = 0.01, I² = 66%) (Analysis 1.6).
The pooled estimate for studies that measured anxiety by scales other than the STAI-S (seven studies, N = 248) suggested no strong evidence of an effect (SMD = -0.43, 95% CI -0.93 to 0.06, P = 0.09). Here too, results were statistically heterogeneous (I² = 70%). Two studies (Cutshall 2011; Stein 2010) could not be included in the meta-analysis because change scores were used. Cutshall 2011 reported a small effect size of -0.12 (95% CI -0.51 to 0.27) in 100 participants, whereas Stein 2010 (N = 36) resulted in an effect size of -0.35 (95% CI -1.01 to 0.31). Neither effect size was statistically significant. Grouping the studies by type of participant resulted in a larger pooled estimate for surgical/procedural patients (four studies, N = 171) but heterogeneity remained (SMD = -0.63, 95% CI -1.25 to -.0.01, P = 0.05, I² = 73%). Three studies with MI and rehabilitation patients (N = 77) did obtain a homogeneous effect but this effect was very small and did not reach statistical significance (SMD = -0.03, 95% CI -0.61 to 0.56, P = 0.93; I² = 31%) (Analysis 1.7). A subgroup analysis on the impact of music preference suggested that the use of participant-preferred music (four studies, N = 144) resulted in a large anxiety reduction that was statistically significant and consistent across studies (MD = -0.89, 95% CI -1.42 to -0.36, P = 0.001, I² = 48%). In contrast,researcher-selected music (in this case, classical music was used) appeared to slightly increase anxiety although this was not statistically significant (three studies, N = 104) (SMD = 0.11, 95% CI -0.28 to 0.49, P = 0.58, I² = 0%) (Analysis 1.8).
Six studies (N = 217) included depression as an outcome. Their pooled estimate indicated that participants who listened to music did not significantly differ in their reported levels of depression from those participants who received standard care (SMD = -0.11, 95% CI -0.38 to 0.16, P = 0.42, I² = 0%) (Analysis 1.9).
Two studies (N = 97) used a numeric rating scale to measure the effects of music on mood. Their pooled estimate indicated that participants who listened to music reported greater mood enhancement than those receiving standard care (SMD = 1.08, 95% CI -0.02 to 2.17, P = 0.05); however, there was disagreement between the two studies about the size of the effect (I² = 80%) (Analysis 1.10). The results are therefore inconclusive.
Quality of life
Only one study (Mandel 2007a) considered quality of life as an outcome. However, significant data loss greatly reduces the usefulness of the data from this study.