A systematic review and meta-analysis of acceptance- and mindfulness-based interventions for DSM-5 anxiety disorders

This meta-analysis systematically reviewed the evidence on standardized acceptance-/mindfulness-based interventions in DSM-5 anxiety disorders. Randomized controlled trials examining Acceptance and Commitment Therapy (ACT), Mindfulness-Based Cognitive Therapy (MBCT), and Mindfulness-Based Stress Reduction (MBSR) were searched via PubMed, Central, PsycInfo, and Scopus until June 2021. Standardized mean differences (SMD) and 95% confidence intervals (CI) were calculated for primary outcomes (anxiety) and secondary ones (depression and quality of life). Risk of bias was assessed using the Cochrane tool. We found 23 studies, mostly of unclear risk of bias, including 1815 adults with different DSM-5 anxiety disorders. ACT, MBCT and MBSR led to short-term effects on clinician- and patient-rated anxiety in addition to treatment as usual (TAU) versus TAU alone. In comparison to Cognitive Behavioral Therapy (CBT), ACT and MBCT showed comparable effects on both anxiety outcomes, while MBSR showed significantly lower effects. Analyses up to 6 and 12 months did not reveal significant differences compared to TAU or CBT. Effects on depression and quality of life showed similar trends. Statistical heterogeneity was moderate to considerable. Adverse events were reported insufficiently. The evidence suggests short-term anxiolytic effects of acceptance- and mindfulness-based interventions. Specific treatment effects exceeding those of placebo mechanisms remain unclear. Protocol registry: Registered at Prospero on November 3rd, 2017 (CRD42017076810).

Patients who were diagnosed by prior versions of the DSM/ICD were also eligible, if their diagnosis is listed as a DSM-5 anxiety disorder as well. Studies involving patients with anxiety comorbid with other physical/mental disorders were eligible, if the comorbidity was not examined as the primary study outcome. Studies including heterogeneous psychiatric populations such as patients with anxiety disorders as well as those with depression or obsessive-compulsive disorder were excluded, while studies assessing mixed anxiety diagnoses (as defined above) were considered as well.
Types of interventions: Standardized acceptance-or mindfulness-based interventions like MBSR, MBCT, ACT, and variations of these programs (regardless of program length, frequency, or setting) were eligible. Studies allowing individual co-interventions were eligible as long as all participants in all groups received the same co-interventions. Acceptable control interventions were no treatment/wait-list or treatment as usual (TAU) as well as any other active treatments.
Types of outcomes: Improvement in the severity of anxiety symptoms measured by validated clinician-and/ or patient-rated scales closest to 2 months after randomization (short-term) were defined as primary outcomes. Secondary outcomes included anxiety symptoms closest to 6 months and 12 months after randomization and safety. The severity of depressive symptoms and health-related quality of life were included as secondary outcomes as well, as anxiety disorders have high rates of comorbidity with depressive symptoms and reduced quality of life. If an outcomes was assessed by more than one instrument, primary endpoints were preferred over secondary ones, disease-specific instruments over generic ones and multi-item over single-item ones. Safety was assessed as the number of adverse events (AE) or study withdrawals due to AEs. AEs were defined as any untoward medical occurrence in a patient, without necessarily having a causal relationship with the study treatment. Cases of any untoward medical occurrence that had resulted in death, was life-threatening, required inpatient hospitalization, or caused persistent or significant disability were rated as serious AEs 29 . Data sources. PubMed, Central, PsycInfo, and Scopus were searched from their inception to June 22nd, 2020. An update was executed in PubMed until June 14th, 2021. Table 1 shows the search strategy in PubMed, which was adapted for each database as necessary. No language restrictions were applied. Moreover, we manu- www.nature.com/scientificreports/ ally searched reference lists of previous reviews. For ongoing and unpublished studies, we searched international trial registries of the WHO and the NIH applying identical search terms. Two reviewers (HH and PB) independently screened titles and abstracts and assessed full-texts for eligibility using EndNote. Any disagreement were rechecked with a third reviewer (HC) until consensus was achieved.
Data extraction. Two reviewers (PB and MS) independently extracted predefined data on the study setting and characteristics of the included studies. All Discrepancies were rechecked with a third reviewer (HH) until consensus was achieved. For each study, the risk of selection bias, performance bias, detection bias, attrition bias, reporting bias, and other source of bias were independently assessed by two reviewers (PB and MS) using the Cochrane risk of bias tool 28 . Each domain was assessed as either, 'low risk of bias' if all requirements were adequately fulfilled, 'high risk of bias' if the requirements were not adequately fulfilled, and as 'unclear risk of bias' if insufficient data for a judgment was provided. Divergent assessments were rechecked with a third reviewer (HC) until consensus was achieved.
Data synthesis. Overall analyses. If at least two studies had assessed the same outcome to the same defined time point with the same type of control intervention, pairwise meta-analyses using random-effects models (inverse variance method) with Hedges' correction for small samples 28 were conducted using Review Manager Software (RevMan, Version 5.3, The Nordic Cochrane Centre, Copenhagen). For continuous outcomes, standardized mean differences (SMDs) with 95% confidence intervals (CIs) were calculated indicating the difference in means between groups divided by the pooled standard deviation (SD). In cases where no SDs were published, they were calculated from standard errors, CIs or t-values 28 , or were requested from trial authors by email. A negative SMD indicated greater effects of the experimental intervention over the respective control condition, except for quality of life. In accordance with Cohen's categories, SMDs of 0.2-0.5 were interpreted as small effects, SMDs of 0.5-0.8 as medium effects, and SMDs > 0.8 as large effects 30 . For binary outcomes such as AEs, risk ratio analyses were planned. However, as most studies did not report AEs systematically, AEs were analyzed descriptively.
Assessment of statistical heterogeneity. Statistical heterogeneity between the study effects was analyzed by the Chi 2 statistics with a p-value of ≤ 0.10 indicating significant heterogeneity. The magnitude of heterogeneity was categorized by the I 2 with I 2 > 25%, I 2 > 50%, and I 2 > 75% representing moderate, substantial, and considerable heterogeneity, respectively 28,31 .
Subgroup analyses. Subgroup analyses were planned for patients with different anxiety diagnoses and different mindfulness interventions but could only be realized for the latter, as the number of studies for comparisons was too small.

Sensitivity analyses.
We conducted sensitivity analyses, where studies with high or unclear risk of bias were compared with those of low risk of bias. If substantial or considerable statistical heterogeneity was present in a meta-analysis, we used sensitivity analyses to explore heterogeneity in effect estimates.
Risk of bias across studies. The assessment of publication bias was planned by visual analysis of funnel plots and Egger's test, if more than 10 studies were included in a meta-analysis 32 .

Results
Literature search. The electronic search revealed 7651 articles, a PubMed update additional 362 ones ( Fig. 1). After the exclusion of supplicates and non-eligible abstracts, 49 articles were red in full. We excluded further 9 ones as they reported data of an already included sample 33-41 , 8 as they included mixed anxiety disorders containing patients with other than the defined DSM-5 anxiety diagnoses 42-49 , 2 articles because of younger aged samples 50,51 , 2 as they did not include a predefined outcome 52,53 , and 4 articles as they did not contain a predefined control group [54][55][56][57] . One additional study did not investigate a standardized ACT, MBCT, or MBSR program 58 . Thus, 23 RCTs published between 2007 and 2021 including 1815 patients were included in the metaanalysis  . Searching of trial registries revealed no additional unpublished or ongoing studies. Study characteristics. The characteristics of the included RCTs are presented in detail in Table 2. The RCTs were conducted in the US, Sweden, Canada, Germany, Norway, Brazil, Japan, Iran, Romania, and China and investigated patients diagnosed with Generalized Anxiety Disorder, Social Anxiety Disorder, and mixed anxiety diagnoses. Samples randomized ranged from 24 to 182 with a median N of 81 containing a median of 67.1% of women. The participants' median age was 35.3 years.
Twelve RCTs investigated ACT interventions, 3 ones MBCT, and 8 ones MBSR. Individual-and group-based approaches varied as well as online and offline/in-person settings. Control interventions included TAU/waitlist, individualized or group-based CBT, psychoeducation, and relaxation. The median duration of the study treatments was 10 (4 to 16) weeks. Concurrent psychotropic drug use was reported by a median of 26.1% of the participants.
All studies provided data directly after the end of the intervention. Seven RCTs assessed a follow-up closest to 6 months after randomization, 4 RCTs additionally closest to 12 months after randomization. Outcomes www.nature.com/scientificreports/ included clinician-rated severity of anxiety, patient-rated severity of anxiety, patient-rated severity of depression, and patient-rated health-related quality of life. Funding was reported by all but 3 RCTs and contained no specific funding (3 RCTs), university grants (4 RCTs), governmental grants (8 RCTs), university and governmental funding (1 RCT), private foundations (3 RCTs), and industrial funding (1 RCT).

Risk of bias of individual studies.
Risk of selection bias was assessed as low in 21.7% of the included studies. Additional 30.4% reported adequate random sequence generation but did not provide information or reported in-adequate information about allocation concealment. No study had low risk of performance bias. The risk of detection bias was low in 60.9% of the studies. Here, a low risk of bias judgement was also given in cases, when the outcome was clinician-reported and the clinician was blinded to group allocation but also when the staff who handed out the patient questionnaires was blinded, even if patients cannot be blinded. The same amount of the studies (60.9%) were assessed as low risk of attrition bias as they recorded less than 10% drop-outs per group or/and performed intention-to treat analysis. The risk of reporting bias and possible risks of other sources of bias were low in 21.7% and 78.3% of the studies, respectively (Fig. 2).

Pooling of effects.
Clinician-rated anxiety. ACT showed a significantly larger short-term effect than TAU  (Fig. 3). However, for both analyses, only the risk of detection bias and other bias was low. Up to 12 months, again no significant differences were found between ACT and CBT 61 . In comparison to less complex psychological interventions such as psychoeducation or relaxation, ACT did not show a significant larger effect (2 RCTs, SMD = −0.18, 95%CI = [−1.00, 0.64], I 2 = 77%, N = 107) (Fig. 3). Because of the considerable heterogeneity, the pooled effect should be interpreted with restraint, also because one of the RCTs 63 reported significant medium-term effects of ACT against psychoeducation. Detailed analyses can be found in the Supplementary Fig. 1.  www.nature.com/scientificreports/ One further RCT investigated MBSR against TAU and showed significantly higher improvements on clinicianrated anxiety up to 2 months post randomization 75 , while as compared to CBT, MBSR showed significantly lower short-term improvements (2 RCTs, SMD = 0.50, 95%CI = [0.17, 0.83], I 2 = 0%, N = 147) (Fig. 1).       (Fig. 3) or longer-term differences 61 between the two therapies. In comparison to less complex psychological interventions, ACT led to no significantly different short-term effect (2 RCTs, SMD = −0.16, 95%CI = [−0.54, 0.22], I 2 = 0%, N = 107) (Fig. 3).
For MBSR, the meta-analysis revealed significantly larger short-term anti-depressive effects against TAU (4 RCTs, SMD = −1.13, 95%CI = [−1.96, −0.31], I 2 = 86%, N = 218). The considerable heterogeneity could be reduced by excluding one low-quality RCT with the largest effect that compared MBSR to TAU without using a waiting list 59 . This resulted in still significantly higher short-term effects of MBSR compared to TAU (3 RCTs, SMD = −0.58, 95%CI = [−0.87, −0.28], I 2 = 0%, N = 187) (Fig. 3). The effect was robust against risk of detection bias, attrition bias, and other bias. In comparison to CBT, no superiority of MBSR on depressive symptoms could be detected (3 RCTs, SMD = 0.07, 95%CI = [−0.24, 0.37], I 2 = 23%, N = 222) (Fig. 3). One additional RCT showed that MBSR was superior to psychoeducation in reducing short-term depressive symptoms 60 . Detailed analyses on depression can be found in the Supplementary Fig. 3.  (Fig. 3). Longer-term effects of ACT did not significantly differ from those of CBT at 6 and 12 months 61 , were reported as significantly higher as those of psychoeducation at 6 months 63 , but not significantly different from those of relaxation at 12 months 67 .
Further RCTs reported significantly higher short-term effects on quality of life in favor of MBSR compared to TAU 80 but not compared to CBT 73 or psychoeducation 60 . Safety. Safety data were reported insufficiently (Tab. 2). Fourteen RCTs did not report any information on AEs or reasons for study withdrawal. Serious AEs were reported by one RCT in the ACT group (bypass surgery), which was highly likely not related to the study intervention 67 . Minor AEs were equally distributed between experimental and control groups 65,69,[73][74][75][76][77]80 .

Discussion
Summary of evidence. The literature search revealed 23 RCTs investigating the effectiveness of ACT, MBCT, and MBSR in patients with DSM-5 anxiety disorders such as Generalized Anxiety Disorder, Social Anxiety Disorder, and mixed samples including different anxiety diagnoses. The meta-analyses revealed at least short-term effects on clinician-and patient-rated anxiety for ACT, MBCT and MBSR in addition to TAU versus TAU alone. In comparison to CBT, ACT and MBCT showed comparable effects on both anxiety outcomes, while MBSR showed significantly lower effects. Pooled effects up to 6 months post randomization can only be calculated for ACT but did not show any significant differences compared to TAU or CBT. Short-term effects of ACT, MBCT and MBSR on secondary outcomes were superior against TAU but not against CBT or less complex psychotherapeutic interventions such as psychoeducation or applied relaxation. Most effects were robust against most risk of bias domains. However, the risk of selection bias and performance bias remains unclear or high for almost all meta-analyses. Statistical heterogeneity could be reduced in several meta-analyses by excluding lowquality studies with extreme values. A correlation of those extreme values with funding concerns from industrial www.nature.com/scientificreports/ companies or private foundations could not be determined. Moreover, comparisons to CBT should be interpreted with restraint as the number of included trials was very low and none of the studies tested non-inferiority. Adverse events were reported insufficiently. If safety issues were reported, ACT, MBCT, and MBSR did not lead to more adverse events than usual care and comparable adverse events to CBT. In contrast to prior systematic reviews, which found promising effects of mindfulness on anxiety symptoms in the general population 13,14,23 and somatically ill samples 15,17 , the evidence for patients with manifest DSM-5 anxiety disorders should be interpreted with restraint. Reasons include the low number of studies investigating longer-term effects against eligible control conditions such as CBT, the overall unclear study quality, and the often missed systematic assessment and reporting of adverse events.
Limitations. This meta-analysis has further limitations. We often were not able to pool data of RCTs because of non-comparable controls, or could perform sensitivity analyses of comparisons including only 2 RCTs. In addition, we were not able to calculate subgroup analyses of different anxiety disorders. Thus, the present metaanalysis does not allow to draw conclusions for a specific diagnosis. Since risk of publication bias could not be assessed, the effects might also be overestimated due to unpublished studies, even if the search of trial registries did not result in registered but unpublished studies 82 .
Implications for further research. Further higher-quality trials on mindfulness-based interventions are needed to verify the effects in patients with manifest anxiety disorders. Authors should ensure rigorous methodology and reporting according to CONSORT 83 and choose adequate control conditions. Recent meta-analyses on pharmacological and psychological intervention for anxiety disorders conclude that wait-list control groups may produce nocebo effects in trials of psychotherapy 84 . TAU, in addition, was also found to be a very heterogeneous control condition and anything but usual or standard care 85,86 . Psychological placebo effects were estimated on average 0.83 in patients with anxiety disorders 87 . Thus, adequate control groups for trials on manifest disorders rather than subclinical symptoms should be designed considering nocebo as well as placebo effects. Non-inferiority trials against standard psychotherapies such as CBT are missing completely. To enhance study quality and reduce the risk of performance bias, even if patients and therapists cannot be blinded, controlling for patients' treatment expectations and their perceptions of quality of the alliance towards the treating therapist would be feasible. Blinding of outcome assessors, especially for clinician-rated outcomes as well as more adequate statistics (including intention-to-treat analyses) should be standard. Future studies are requested to strictly report AEs and reasons of drop-out.

Conclusions
The evidence suggests clinically relevant short-term anxiolytic effects of acceptance-based and with less extent of mindfulness-based interventions when added to usual care that, however, might be a result of nocebo-and/ or placebo effects. The relevance of longer-term effects as well as the comparability to standard CBT remain preliminary until further high-quality studies will be published.

Data availability
The data and material analyzed during the current study are available from the corresponding author on request.

Code availability
The RevMan code used for the current study is available from the corresponding author on request.