Effectiveness and acceptability of exercise treatments for adults with anxiety disorders: a systematic review and network meta-analysis

Abstract

Background

Anxiety disorders are persistent, functionally impairing conditions with high societal burden. In contrast, anxiety symptoms are elevated anxiety meeting scale thresholds (not formal diagnoses). While exercise is recognised as a complementary intervention for anxiety, the most effective mode of exercise for anxiety disorders remains unclear.

Objective

To evaluate the effectiveness and acceptability of exercise treatments for adults with anxiety disorders via network meta-analysis.

Methods

Systematic review and frequentist network meta-analysis were conducted, searching five databases (MEDLINE, EMBASE, PsycINFO, CENTRAL, SPORTDiscus) up to 1 March 2025. Treatments assessed included aerobic exercise, mind-body exercise (MBE), resistance training (RT), stretching, multicomponent exercise, cognitive-behavioral therapy, psychoeducation, waitlist control, usual care and placebo. Risk of bias was assessed using the Cochrane Risk of Bias Tool 2.0 (RoB 2.0). Outcomes included anxiety severity reduction (for effectiveness) and all-cause discontinuation rates (for acceptability).

Results

30 RCTs (1421 participants) were included: four had low risk, 19 some concerns and seven high risk. For combined anxiety conditions (anxiety disorders and symptoms), RT was most effective (standardised mean difference (SMD) −0.80, 95% CI −1.24 to −0.36, SUCRA 78.7%), followed by MBE (SMD −0.78, 95% CI −1.12 to −0.44, SUCRA 77.8%). For anxiety disorders, RT was most effective (SMD −0.79, 95% CI −1.18 to −0.40, SUCRA 78.5%) and MBE for anxiety symptoms (SMD −0.84, 95% CI −1.48 to −0.20, SUCRA 77.9%). No acceptability differences across treatments; network meta-regression showed diagnosis status did not alter effects.

Conclusions

Both resistance exercise and MBE may be effective for anxiety disorders, with moderate certainty evidence. Acceptability remains unclear.

Prospero registration number

CRD42024561238.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Exercise is recognised as a potential anxiety intervention, with guidelines (eg, WHO, Royal Australian and New Zealand College of Psychiatrists) supporting its use.

• Prior meta-analyses often fail to distinguish between anxiety disorders (Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5)/ International Classification of Diseases (ICD-11) diagnosed with functional impairment) and anxiety symptoms (elevated scale scores without formal diagnosis), leading to generalised conclusions.

• Many existing studies have methodological flaws: mixing control group types (eg, waitlist, treatment as usual), using outdated diagnostic criteria, incorrectly including post-traumatic stress disorder/obsessive-compulsive disorder (no longer anxiety disorders in DSM-5/ICD-11), or incorporating healthy individuals/those with other conditions—diluting findings’ specificity.

WHAT ARE THE NEW FINDINGS

• This review only included parallel-arm randomised controlled trials to focus on real-world effectiveness (rather than efficacy in idealised settings), with eligible participants being adults with diagnosed anxiety disorders (per DSM-5/ICD-11) or anxiety symptoms.

• For combined anxiety conditions (anxiety disorders + symptoms), resistance training (RT) showed the highest effectiveness (standardised mean difference (SMD) −0.80, surface under the cumulative ranking curve (SUCRA) 78.7%).

• For anxiety symptoms alone, mind-body exercise (MBE) was the most effective intervention (SMD −0.84, SUCRA 77.9%); RT did not demonstrate significant effectiveness for anxiety symptoms.

• No significant differences in acceptability (measured by all-cause discontinuation rates) were found across exercise treatments, controls or psychological therapies.

Introduction

Anxiety disorders are common mental health issues marked by excessive worry and fear, accompanied by persistent functional impairment and broader clinical manifestations (eg, impaired social, occupational or daily functioning).¹ According to the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) and the International Classification of Diseases, 11th Revision (ICD-11),^{2 3} they include Social Anxiety Disorder (SAD), generalised anxiety disorder, panic disorder (PD), agoraphobia and others—listed in descending order of global prevalence in adult populations.^{4 5} The Global Burden of Disease 2019 study showed their significant impact, particularly in individuals aged 10–24 years, contributing 2868 disability-adjusted life years per 100 000 population.^{5 6} Globally, anxiety disorder prevalence rose by 25.6% during the 2020 COVID-19 pandemic,⁷ with data spanning 204 countries/territories. A later meta-analysis validated this upward trend, highlighting the need for effective treatments.⁸

Current treatment guidelines recommend both medication and psychological therapies as first-line treatments for anxiety disorders.⁹ Medications like selective serotonin reuptake inhibitors are effective for anxiety disorders,¹⁰ but they face challenges such as long treatment durations and high costs.¹¹ Psychological therapies like cognitive behavioural therapy (CBT) are also effective,¹² though many patients do not respond to initial CBT or experience high relapse rates. CBT further has unique limitations, including incomplete remission in many cases and high dropout rates.¹³ Additionally, both medications and CBT often fail to address physical comorbidities, and certain pharmacological options may even increase cardiovascular risks.¹⁴ Stigma and low acceptance rates further hinder treatment.¹⁵

Recently, exercise has gained attention as a potential alternative therapy. The WHO and the Royal Australian and New Zealand College of Psychiatrists recommend exercise for anxiety management.^{16 17} Although studies suggest that resistance training (RT), mind-body exercises (MBE) and aerobic exercises (AEs) can alleviate anxiety symptoms,¹⁸,²⁰ not all meta-analyses differentiate between anxiety disorders and symptoms. Specifically, anxiety symptoms refer to elevated anxiety intensity measured via self-reported scales (eg, State-Trait Anxiety Inventory (STAI)) that meet validated thresholds but do not confirm a clinical syndrome; anxiety disorders, by contrast, require the presence of persistent functional impairment and broader clinical manifestations (per DSM-5/ICD-11). This lack of differentiation leads to generalised conclusions about exercise effectiveness.²¹,²⁷ Although statistical power was enhanced, specificity was compromised. Anxiety disorders may exist on a continuum, suggesting there are shared mechanisms for symptom relief across different severities.²⁸ Some studies attempt to address this through subgroup analyses but face challenges due to outdated diagnostic criteria and sample selection issues.²⁹,³¹ Post-traumatic stress disorder (PTSD) and obsessive-compulsive disorder (OCD) are no longer classified as anxiety disorders in DSM-5 and ICD-11, which affects treatment evaluations.^{29 32 33} Additionally, including healthy individuals or those with other conditions can dilute results specific to anxiety disorders.²¹,^{2434 35} Methodologically, many analyses mix different types of control groups, which can reduce the clarity of findings.²¹,^{2629 30}

This study will tackle existing literature gaps by clearly distinguishing between anxiety symptoms and disorders and using updated diagnostic criteria.^{2 3} To this end, we focus on well-defined comparators and employ rigorous methodologies—prioritising parallel-arm randomised controlled trials (RCTs), which align with evaluating real-world effectiveness by avoiding residual intervention effects of crossover designs and mirroring routine clinical care.³⁶ It aims to deliver robust evidence on both the effectiveness of exercise for anxiety treatment and its acceptability among participants—a key factor for translating findings into real-world practice—with this dual focus strengthening the rationale for exercise’s potential inclusion in clinical guidelines, ensuring recommendations are evidence-based for effectiveness and feasible for broad patient adherence.

Method

We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement items for systematic reviews (SRs) in the fields of sports and exercise medicine, musculoskeletal rehabilitation and sports science,³⁷ as well as the PRISMA extension statement for reporting SRs that incorporate network meta-analysis (NMA) of healthcare interventions.^{38 39} The study protocol is registered with the International Prospective Register of Systematic Reviews (CRD42024561238).

Information sources and search strategy

We conducted a comprehensive search of five international electronic databases—MEDLINE, EMBASE, PsycINFO, CENTRAL and SPORTDiscus—for potential RCTs from their inception to March 2025. To ensure we did not miss relevant studies, we also performed a supplementary search for potential SRs. We applied validated specificity-maximised search filters for SRs to the MEDLINE, EMBASE and PsycINFO databases.⁴⁰,⁴² The searches were restricted to studies published in English and involving human participants. The search strategies are detailed in online supplemental appendix A. This thorough search process is part of a broader initiative referred to as the anxiety meta-analytical research domain.⁴³

Eligibility criteria

Types of studies

Only parallel-arm trials were included in this review, while cluster, crossover or within-person trials were excluded. This design choice aligns with our focus on evaluating real-world effectiveness: parallel-arm RCTs eliminate crossover designs’ residual intervention effects (a critical limitation) and mirror routine clinical practice’s standard delivery model.³⁶

Types of participants

We included adults aged 18 years or older, of any sex, who had been diagnosed with an anxiety disorder using validated tools, such as the DSM (DSM-III–DSM-5-Text Revision²⁴⁴,⁴⁸ or the ICD (ICD-10, ICD-11).^{3 49} Participants could also be selected based on a cut-off score for anxiety symptoms (eg, Hamilton Anxiety Rating Scale (HARS) >14) or diagnosed by a clinician in an inpatient setting. We excluded RCTs requiring participants to have a diagnosis of PTSD or OCD and with physical comorbidities (eg, diabetes), other mental disorders. This step reduces potential confounding, as such physical conditions could impact exercise tolerance or independently influence mental health outcomes, ensuring study results more accurately reflect exercise’s specific effect on anxiety.

Types of interventions

Exercise interventions could include any structured physical activities, such as AE, RT, MBE, stretching and multicomponent exercise (MCE).⁵⁰ We excluded interventions that combined exercise with non-exercise components (eg, exercise combined with CBT) and those involving single bouts of exercise.

Types of comparators

For passive control groups, we included those without active components, such as waitlist or no treatment groups. For active control groups, we considered treatments as usual, inactive placebo controls and RCTs comparing different forms of exercise or psychotherapy (eg, CBT).⁵¹

Types of outcome measures

We selected one primary and one secondary outcome measure to evaluate the effectiveness and acceptability of the intervention, respectively.

Primary outcome (effectiveness): The primary outcome is the severity of anxiety symptoms at the study endpoint, assessed using standardised tools. To ensure clinical precision, we will conduct separate analyses for:

1. Anxiety disorders: defined by studies where anxiety is diagnosed using authoritative criteria (eg, DSM or ICD).

2. Anxiety symptoms: defined by studies where anxiety is measured via self-report questionnaires or clinician-administered tools without formal diagnosis.

3. All anxiety: defined by studies including both participants with anxiety disorders and those with anxiety symptoms.

When multiple standardised tools are available for measuring anxiety severity, we will apply a predefined hierarchical algorithm (see online supplemental appendix B) based on psychometric properties and consistency across trials to ensure measurement validity. This algorithm prioritises clinician-administered tools (eg, the HARS) over self-report measures (eg, the Beck Anxiety Inventory), due to the former’s greater objectivity and reliability in assessing anxiety symptoms.

Secondary outcome (acceptability): the secondary outcome is all-cause trial discontinuation, defined as the proportion of participants who withdrew prior to the end-of-treatment assessment for any reason, serving as a proxy for acceptability. This outcome will be analysed across:

1. Anxiety disorders: to align with the study’s focus on clinically diagnosed conditions.

2. Anxiety symptoms: to specifically address the anxiety symptom network analysis, focusing on subclinical anxiety populations

3. All anxiety: encompassing both anxiety disorders and anxiety symptoms, to provide a comprehensive perspective consistent with our original design.

Additionally, for detailed information on which standardised tools were used, please refer to the instrument column in online supplemental appendix E: characteristics of the included studies in the supplementary materials. Their use strictly adheres to our predefined hierarchical algorithm.

Selection of literature and data extraction

Two reviewers (EF-CL and K-wW) independently screened the titles and abstracts of the retrieved citations from the electronic databases, removed duplicates and determined the eligibility of RCTs based on our inclusion criteria. We obtained the full texts of the eligible citations for further review. The search results were imported into EndNote V.20 for deduplication.⁵² Given the variability in false negatives and positives depending on the searcher’s skill level,⁵³ we selected the most recent RCTs among duplicate citations for data extraction, using older versions as supplementary information when necessary. Any conflicts were resolved through discussion with senior authors (ZD and BTT).

Risk of bias

The risk of bias assessment for the included RCTs was performed independently by two reviewers (EF-CL and K-wW), with disagreements resolved through discussion and arbitration by senior review authors (ZD and BTT). We employed the Cochrane risk-of-bias tool for randomized trials V.2 (RoB 2) to assess all included RCTs—covering both the primary outcome (effectiveness) and secondary outcome (acceptability).⁵⁴

Data synthesis

We conducted a series of head-to-head pairwise meta-analyses for all direct comparisons, followed by a frequentist NMA,^{55 56} both using a random-effects model, to evaluate the comparative effectiveness of different treatments of exercise for alleviating anxiety.⁵⁷ Data extracted from the included RCTs for each outcome were analysed using STATA V.14.0.⁵⁸

For continuous outcomes, we pooled the standardised mean differences (SMDs) with corrections for small studies (Hedges’ g). When intention-to-treat (ITT) data were available, we prioritised them over per-protocol analyses. If necessary, we contacted study authors for additional data or clarification. For dichotomous outcomes, we calculated relative risks with 95% CIs) based strictly on ITT principles.

When studies included multiple groups with slightly different versions of the same intervention, we combined these groups into a single category for analysis. We classified effect sizes using the following criteria: small (SMD 0.20–<0.50), moderate (SMD 0.50–<0.80) and large (SMD≥0.80). We established a threshold of SMD 0.20 for clinical effectiveness,⁵⁹ and we created a treatment rank using the surface under the cumulative ranking curve SUCRA and mean ranks, with ‘waitlist control’ serving as the reference. To assess the transitivity assumption,⁶⁰ we evaluated the distribution of key variables—such as the percentage of females, mean age,^{31 61} session frequency per week,³¹ session duration and trial duration^{26 27}—across different comparisons using box plots and error bar plots in R V.4.4.0.⁶² Furthermore, we conducted a Kruskal-Wallis test on the same variables to assess differences across groups. Heterogeneity was evaluated by examining the heterogeneity variance (τ²), categorised as follows: low (<0.04), low-moderate (0.04–0.16), moderate-high (0.16–0.36) and high (>0.36), to quantify variability.^{63 64} In addition, we assessed incoherence using two methods: the global design-by-treatment interaction model and the node-splitting method.^{65 66} Comparisons demonstrating significant inconsistency (p<0.10) were downgraded accordingly.

Additionally, we used the gemtc package in R V.4.4.0⁶⁷ for NMR in NMA to assess exercise effectiveness across anxiety conditions, examining variation by condition type (disorders vs symptoms) using an unrelated parameter model.⁶⁸

For the effectiveness outcome, we conducted two sensitivity analyses to determine whether excluding specific studies would alter the results:

1. Prespecified sensitivity analysis: excluding RCTs with unclear diagnostic tools (a priori planned in the study protocol).

2. Post-hoc sensitivity analysis: excluding RCTs with high RoB 2 risk (exploratory analysis added to assess the association between study quality and effectiveness outcomes).

Publication bias assessment

We assessed publication bias by visually inspecting the comparison-adjusted funnel plot and using Egger’s regression intercept if there were 10 or more studies in a direct comparison.^{69 70} Additionally, we searched ClinicalTrials.gov to identify completed but unpublished RCTs with available results relevant to our study.

Credibility of evidence assessment

To assess the certainty of each comparison against a control group, we employed the Confidence in Network Meta-Analysis (CINeMA) tool.⁷¹ CINeMA is an adaptation of the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework, specifically tailored for NMA and encompasses six domains of assessment, namely: within-study bias, reporting bias, indirectness, imprecision, heterogeneity and incoherence. The overall certainty for each comparison was rated as high, moderate, low or very low.

Equity, diversity and inclusion statement

Our study recruited a diverse group of healthy adults from varied genders, socioeconomic backgrounds and cultural contexts via global RCTs, ensuring balanced demographic representation. The author team comprises junior and senior researchers, including a Master’s student, a PhD candidate and an assistant professor, spanning multiple disciplines and based in Hong Kong and Canada. Although no authors are from marginalised groups, we emphasised diverse perspectives through interdisciplinary collaboration. Data collection was inclusive, addressing participants’ accessibility needs, regional differences and socioeconomic and educational levels. Analysis considered potential inequities tied to gender, socioeconomic status and other social factors.

The Consensus on Exercise Reporting Template assessment

We assessed the quality of exercise intervention reporting in 30 RCTs using the Consensus on Exercise Reporting Template (CERT).⁷²

Protocol deviations

All deviations from the study protocol are outlined in online supplemental appendix M.

Results

Selection of studies

The study selection process is illustrated in figure 1. A total of 9670 citations were identified from databases and registers. After removing 2603 duplicates and screening titles and abstracts, 149 full-text articles were assessed, leading to the exclusion of 123 trials. Finally, 26 RCTs met the inclusion criteria.

Figure 1. Flowchart of eligible study selection. PTSD, post-traumatic stress disorder; RCT, randomised controlled trial; SR, systematic review.

From an initial pool of 41 eligible SRs (online supplemental appendix D), we identified 697 citations. After excluding 472 duplicates and 219 based on title and abstract screening, we proceeded with full-text evaluation of six RCTs; two were subsequently excluded due to an excessively high proportion of participants with chronic diseases and PTSD diagnoses, respectively. The supplementary search for eligible RCTs conducted between June 2024 and March 2025 did not identify any new studies that met the inclusion criteria. In total, 29 RCTs met the inclusion criteria for data extraction in the effectiveness network, while 30 eligible RCTs were included in the acceptability network. Inter-rater agreement was consistently high across all stages of the review process. For study selection, agreement reached 94% with a Kappa value of 0.87 (online supplemental appendix O).

Characteristics of included RCTs

The characteristics of the included RCTs are summarised in table 1 and online supplemental appendix E. For data extraction, agreement was 96% with a Kappa of 0.89.

Table 1. Characteristics of randomised clinical trials included in the network meta-analysis.

Characteristic	Effectiveness network for anxiety disorder	Acceptability network for anxiety disorder	Effectiveness network for anxiety symptom	Acceptability network for anxiety symptom	Effectiveness network for anxiety	Acceptability network for anxiety
Number of studies	17	18	12	12	29	30
Number of patients recruited	927	1028	657	657	1587	1685
Number of patients included	760	824	578	548	1372	1421
Age (years)	31.7	32.2	34.8	34.8	35.8	36.3
Women	73.1%	72.7%	71.3%	71.3%	72.6%	72.8%
Year of publication
1989–2000	2	2	1	1	3	3
2001–2011	2	2	0	0	2	2
2012–2024	13	14	11	11	24	25
Study duration, weeks
2–6	5	5	4	4	9	9
7–12	12	13	7	7	19	20
>12	0	0	1	1	1	1
Follow-up duration, weeks*
1–12	4	5	4	4	8	9
13–26	1	1	0	0	1	1
27–52	3	3	0	0	3	3
Number of sessions per week
1–3	12	12	10	10	22	22
4–7	3	4	2	2	5	6
Unclear	2	2	0	0	2	2
Risk of bias
Low risk	4	4	0	0	4	4
Some concerns	9	10	9	9	18	19
High risk	4	4	3	3	7	7
Type of analysis
Intention to treat	8	8	5	5	13	13
Per protocol	9	10	7	7	16	17

^*Follow-up data was provided by 12 studies.

^†For the hierarchy of outcomes, see online supplemental appendix B.

Participants

A total of 1421 participants from 30 RCTs were included, covering both effectiveness and acceptability analyses. For effectiveness outcomes: 760 participants were in the anxiety disorder group, and 578 participants were in the anxiety symptom group. For acceptability outcomes: 824 participants were included in the anxiety disorder group, and 548 participants were in the anxiety symptom group. Across all participants, the average age was 36.3 years, and 72.3% were female.

Interventions and comparators

Definitions and descriptions of all interventions and comparators are presented in table 2. Among the 30 RCTs, AE was the most frequently investigated intervention (n=10), followed by MBE (n=9). The average session duration was 43.2 min, with participants engaging in sessions three times weekly over an 8-week period. Only 12 trials reported follow-up data, which ranged from 1 week to 52 weeks. The most common comparator was a waitlist (n=10), followed by treatment as usual (n=9) and other comparators.

Table 2. Definitions and descriptions of exercise and psychological treatments and comparators.

Category	Items	Definition/description
Exercise treatments	Resistance training	A form of physical activity designed to improve muscular strength and endurance by exercising a muscle or a muscle group against external resistance.
	Aerobic exercise	Continuous, rhythmic physical activity that primarily uses large muscle groups to improve cardiovascular endurance.
	Mind-body exercise	Physical activities that combine movement with mental focus and controlled breathing to enhance physical and mental well-being, such as yoga and tai chi.
	Multicomponent exercise	An exercise regimen that includes a combination of aerobic, resistance, balance and flexibility exercises to improve overall physical fitness.
	Stretching	Exercises aimed at improving flexibility and range of motion by lengthening muscles and tendons.
Psychological treatments	cognitive-behavioral therapy (CBT)	A structured, time-limited psychological treatment that aims to alter maladaptive thinking patterns to change behaviour and emotional states.
	Third-wave CBTs	Includes therapies like acceptance and commitment therapy, dialectical behaviour therapy and mindfulness-based cognitive therapy, which focus on context and function of psychological phenomena rather than their form.
	Psychoeducation	Interventions that provide information and support to help individuals understand and manage their mental health conditions more effectively.
Control	Waitlist control	Participants are placed on a waiting list and receive no intervention during the study period but are offered the intervention after the study concludes.
	Treatment as usual	Participants continue to receive the standard care or routine practice that they would typically receive if they were not part of the study.
	Placebo control	A treatment designed to mimic the external sensory characteristics of the experimental treatment, such as appearance, smell, taste and texture, but not to replicate its psychotropic effects or adverse reactions. It may also be referred to as an ‘inert placebo’, ‘true placebo’, or ‘sham’.

Outcomes

Self-report scales were predominantly used (n=22, 73%), with the STAI-Y being the most common. Clinician-administered tools (n=8) were used less frequently, with the HARS being the most common (n=5, 63%).

Risk of bias of included RCTs

Risk of bias ratings were identical for both primary and secondary outcomes across all included studies. Among the evaluated studies, seven studies (23%) were classified as having a high risk of bias, while 19 studies (63%) were rated as having ‘some concerns’. Only four studies (13%) were deemed to have a low risk of bias. In the ‘selection of the reported result’ domain, several RCTs did not disclose information regarding the study protocol and preplanned analyses, resulting in a ‘some concerns’ rating for seven studies (23%). Similarly, 12 studies (40%) inadequately reported on the randomisation process, often omitting details regarding allocation concealment. A per-protocol approach was employed in the statistical analyses of 17 studies (56%), impacting the ‘deviations from intended interventions’ domain, with over half of the studies categorised as having either some concerns (10 studies, 33%) or high risk (seven studies, 23%). In the ‘measurement of the outcome’ domain, most trials (23 studies, 77%) were rated as having ‘some concerns,’ although participants were likely aware of the type of exercise and allocation concealment often mitigated bias risk. Regarding missing outcomes, five studies (17%) were considered high risk, while three studies (10%) had some concerns. For the ROB2 assessments, agreement was 90% with a Kappa of 0.82. According to the interpretation guidelines, these Kappa values (ranging from 0.81 to 1.00) indicate ‘almost perfect’ agreement.

Results of NMA geometry

Comprehensive details of standard pairwise meta-analyses, NMA, assessments of transitivity, heterogeneity, incoherence and certainty of evidence for both primary and secondary outcomes are provided in table 3, figure 2, online supplemental appendices F–H and N.

Table 3. Results of six NMAs.

Interventions	k	SUCRA (%)	Reference	SMD with 95% CI	Design-by-treatment	τ²	Certainty rating (CINeMA)
Effectiveness NMA: anxiety Disorders (n=760)
Resistance training	2	78.5	Waitlist	−0.79 (−1.18 to −0.40)	0.95	0.00	Moderate
Third-wave CBTs	1	76.4	Waitlist	−0.75 (−1.15 to −0.36)			Moderate
Mind-body exercise	4	71.3	Waitlist	−0.71 (−1.23 to −0.19)			Moderate
CBT	0	67.5	Waitlist	−0.70 (−1.42 to 0.02)			Low
Aerobic exercise	1	63.6	Waitlist	−0.65 (−0.99 to −0.31)			Moderate
Multicomponent exercise	1	56	Waitlist	−0.58 (−0.88 to −0.29)			Low
Treatment as usual	0	41.8	Waitlist	−0.40 (−1.11 to 0.31)			Low
Stretching	0	29.5	Waitlist	−0.26 (−0.79 to 0.26)			Low
Inactive placebo controls	0	1.4	Waitlist	0.55 (−0.17 to 1.28)			Very low
Acceptability NMA: anxiety Disorders (n=824)
Treatment as usual	0	83.1	Waitlist	0.55 (0.13 to 2.26)	0.97	0.00	Low
Resistance training	4	73.8	Waitlist	0.80 (0.41 to 1.54)			Low
Mind-body exercise	1	67	Waitlist	0.91 (0.38 to 2.14)			Low
Multicomponent exercise	1	55.1	Waitlist	1.07 (0.71 to 1.60)			Low
Third-wave CBTs	1	49.9	Waitlist	1.18 (0.58 to 2.38)			Low
Aerobic exercise	2	36.1	Waitlist	1.42 (0.66 to 3.05)			Very low
Inactive placebo controls	0	33.1	Waitlist	1.55 (0.58 to 4.13)			Low
CBT	0	29.7	Waitlist	2.89 (0.12 to 68.23)			Low
Stretching	0	10	Waitlist	2.97 (0.91 to 9.72)			Moderate
Effectiveness NMA: anxiety Symptoms (n=578)
Mind-body exercise	4	77.9	Waitlist	−0.84 (−1.48 to −0.20)	0.93	0.36	Low
Psychoeducation	0	62	Waitlist	−0.65 (−2.10 to 0.81)			Very low
Resistance training	0	62	Waitlist	−0.64 (−2.34 to 1.05)			Very low
Aerobic exercise	0	63.5	Waitlist	−0.62 (−1.82 to 0.58)			Very low
Multicomponent exercise	2	50.7	Waitlist	−0.40 (−1.20 to 0.41)			Very low
Treatment as usual	0	9.2	Waitlist	0.30 (−0.62 to 1.23)			Very low
Acceptability NMA: anxiety Symptoms (n=657)
Psychoeducation	0	88.9	Waitlist	0.35 (0.08 to 1.55)	0.47	0.00	Very low
Treatment as usual	0	54.2	Waitlist	0.98 (0.36 to 2.68)			Very low
Resistance training	0	47.9	Waitlist	1.05 (0.02 to 56.23)			Very low
Aerobic exercise	0	40.2	Waitlist	1.17 (0.43 to 3.18)			Very low
Multicomponent exercise	4	38.4	Waitlist	1.19 (0.54 to 2.65)			Very low
Mind-body exercise	4	27.0	Waitlist	1.37 (0.73 to 2.56)			Very low
Effectiveness NMA: combined conditions (n=1372)
Resistance training	4	78.7	Waitlist	−0.80 (−1.24 to −0.36)	0.99	0.08	Moderate
Mind-body exercise	5	77.8	Waitlist	−0.78 (−1.12 to −0.44)			Moderate
Third-wave CBTs	1	65.8	Waitlist	−0.77 (−1.27 to −0.27)			low
Aerobic exercise	2	63.5	Waitlist	−0.65 (−1.05 to −0.26)			low
CBT	0	59.6	Waitlist	−0.64 (−1.58 to 0.31)			Very low
Psychoeducation	0	53.5	Waitlist	−0.58 (−1.47 to 0.31)			Very low
Multicomponent exercise	3	34.4	Waitlist	−0.52 (−0.88 to −0.16)			low
Stretching	0	20.3	Waitlist	−0.23 (−0.91 to 0.46)			Very low
Treatment as usual	0	15.2	Waitlist	0.09 (−0.34 to 0.52)			Very low
Inactive placebo controls	0	4.1	Waitlist	0.54 (−0.33 to 1.40)			Very low
Acceptability NMA: combined conditions (n=1421)
Psychoeducation	0	94.6	Waitlist	0.31 (0.07 to 1.27)	0.717	0.00	Low
Resistance training	4	76.4	Waitlist	0.75 (0.39 to 1.45)			Low
Treatment as usual	0	63.1	Waitlist	0.94 (0.46 to 1.93)			Very low
Multicomponent exercise	3	45.6	Waitlist	1.14 (0.80 to 1.62)			Very low
Third-wave CBTs	1	47.2	Waitlist	1.14 (0.63 to 2.06)			Very low
Mind-body exercise	5	42.1	Waitlist	1.18 (0.73 to 1.91)			Low
Aerobic exercise	2	41.1	Waitlist	1.21 (0.69 to 2.10)			Very low
Inactive placebo controls	0	36.1	Waitlist	1.32 (0.58 to 3.01)			Very low
Stretching	0	16.7	Waitlist	1.91 (0.72 to 5.06)			Low
CBT	0	26.8	Waitlist	3.07 (0.13 to 72.13)			Very low

Note: k represents the number of direct comparisons and n represents the total number of participants.

CBT, cognitive-behavioural therapy; CINeMA, Confidence in Network Meta-Analysis; NMA, network meta-analysis; SMD, standardised mean difference; SUCRA, surface under the cumulative ranking curve.

Figure 2. Network plots. AE, aerobic exercise; CBT, cognitive-behavioral therapy; EDU, psychoeducation; MCE, multicomponent exercise; MBE, mind-body exercise; PLA, placebo control; RT, resistance training; STR, stretching; TAU, treatments as usual; TWT, third-wave CBTs; WL, waitlist.

For the primary outcomes of effectiveness, first, in the NMA targeting anxiety disorders, RT again showed the highest effectiveness compared with the waitlist group (SMD −0.79, 95% CI −1.18 to −0.40, SUCRA 78.5%). This was closely followed by CBT (SMD −0.75, 95% CI −1.15 to −0.36, SUCRA 76.4%) and MBE (SMD −0.71, 95% CI −1.23 to −0.19, SUCRA 71.3%). The network analysis revealed low heterogeneity (τ² = 0.00).

Second, in the NMA for anxiety symptoms, MBE ranked first compared with the waitlist group (SMD −0.84, 95% CI −1.48 to −0.20, SUCRA 77.9%). The network displayed high heterogeneity (τ² =0.362).

Lastly, in the NMA for the combined anxiety condition, RT again demonstrated the highest effectiveness (SMD −0.80, 95% CI −1.24 to −0.36, SUCRA 78.7%), with MBE closely following (SMD −0.78, 95% CI −1.12 to −0.44, SUCRA 77.8%). The network analysis indicated low heterogeneity (τ² =0.08).

For the secondary outcomes of acceptability, in the NMA for anxiety disorders, no significant differences in acceptability were observed across various comparisons. The network analysis indicated low heterogeneity (τ² =0.00). In the NMA for the combined anxiety condition and anxiety symptom, apart from psychoeducation, which demonstrated slightly higher acceptability than stretching and MBE, respectively, no significant differences in acceptability outcomes were observed across various comparisons. These two networks exhibited low heterogeneity (τ² =0.00).

No evidence of incoherence was detected in either the primary or secondary outcome networks (table 3 and online supplemental appendices G and H).

Additional NMR analysis indicates that the diagnosis status does not alter the effects, with an I² value of 2% indicating low heterogeneity (figure 3 and online supplemental appendix I).

Figure 3. Network meta-regression of exercise interventions versus waitlist: treatment effects across diagnosis levels. Note: (a) regression plot for mind-body exercise versus waitlist; (b) regression plot for multicomponent exercise versus waitlist; (c) regression plot for stretching versus waitlist; (d) regression plot for aerobic exercise versus waitlist; (e) regression plot for resistance training versus waitlist.

Sensitivity analysis

In the prespecified sensitivity analysis: excluding RCTs with unclear diagnostic tools (online supplemental appendix), RT ranked second (SMD −0.95, 95% CI −1.52 to −0.39), while MBE rose to the top position (SMD −1.04, 95% CI −1.51 to −0.56). In the post-hoc sensitivity analysis: Excluding RCTs with high RoB 2 risk (online supplemental appendix), RT’s effect size and ranking declined further (SMD −0.84, 95% CI −1.53 to −0.14), while MBE maintained the top position (SMD −0.89, 95% CI −1.41 to −0.37).

Publication bias

The comparison-adjusted funnel plot revealed no significant asymmetry (online supplemental appendices G and H). Furthermore, no completed but unpublished RCTs meeting our inclusion criteria were identified through the ClinicalTrials.gov search.

Certainty of evidence

All detailed data for the primary outcome of effectiveness (across anxiety disorders, anxiety symptoms and combined conditions) are presented in table 3.

For anxiety disorders: RT was most effective (SMD −0.79, 95% CI −1.18 to −0.40; SUCRA 78.5%, moderate certainty), followed by MBE (SMD −0.71, 95% CI −1.23 to −0.19; SUCRA 71.3%, moderate certainty); aerobic exercise and stretching had lower effectiveness.

In anxiety symptoms: MBE ranked top (SMD −0.84, 95% CI −1.48 to −0.20; SUCRA 77.9%, low certainty), while RT showed no significant effect (SMD −0.64, 95% CI −2.34 to 1.05; SUCRA 62.0%, very low certainty).

For combined conditions: RT (SMD −0.80, 95% CI −1.24 to −0.36; SUCRA 78.7%) and MBE (SMD −0.78, 95% CI −1.12 to −0.44; SUCRA 77.8%) both placed top 2, with moderate certainty for both.

Quality of intervention reporting

The CERT evaluation indicates that the reporting quality of exercise interventions across 30 RCTs varies, with an average score of 10.3 (ranging from 6 to 19). For detailed information, please refer to online supplemental appendix N.

Discussion

In this NMA, our principal findings reveal that for individuals with anxiety disorders, moderate certainty evidence suggests RT and MBE are the most effective interventions. Among populations presenting with anxiety symptoms, low-certainty evidence indicates that only MBE demonstrates significant effectiveness. Analysing the data across the clinical spectrum of anxiety disorders yields consistent results, with RT and MBE identified as the most and second-most effective interventions, respectively. Our NMR indirectly corroborates these findings, indicating that the effectiveness of various exercise regimens is not contingent on diagnostic status. Nonetheless, secondary outcomes concerning acceptability show no significant differences across the different exercise modalities.

Our findings align with a recent NMA identifying RT as the most effective intervention for anxiety disorders.³³ Unlike the prior analysis based on five RCTs, our review incorporates a larger sample size, enhancing the robustness of evidence supporting RT’s anxiolytic effects and providing stronger implications for clinical practice. The substantial female participation (88.6%) suggests RT’s anxiolytic effects may be particularly relevant for women. A psychodynamic perspective suggests RT necessitates high levels of focus and vigilance, potentially fostering escapism among women.⁷³ This state may facilitate the redirection of attention from external stressors towards the exercise itself, thereby mitigating anxiety. Additionally, the hypermasculine culture often associated with weightlifting may deter some women from participating,⁷⁴ as it conflicts with prevailing societal gender norms. However, as women progressively lift heavier weights, they not only challenge these stereotypes but also enhance their self-efficacy and self-esteem.⁷³ From a neurobiological perspective, recent meta-analyses indicate that following RT, serum levels of insulin-like growth factor 1 (IGF-1) significantly increase in women but not in men.⁷⁵ IGF-1 is known to mediate skeletal muscle growth post exercise and possibly exert anxiolytic effects by regulating FKBP5 expression in the brain, promoting hippocampal neurogenesis and influencing growth factor responses.^{76 77}

80% of participants in our study were untrained individuals, a group potentially more responsive to muscle hypertrophy and increases in cross-sectional area,^{78 79} which can enhance body image and serve as a motivational factor. Notably, neural and connective tissue adaptations are responsible for most strength gains within the initial 4–8 weeks,^{80 81} coinciding with our RT duration of 6.4 weeks. This aligns with findings from a recent cohort study showing a negative association between strength gains and anxiety,⁸² suggesting that increased strength may support hippocampal volume growth and reduce white matter hyperintensities, both linked to improved mental health outcomes.⁸³ Emerging hypotheses propose that parasympathetic stimulation through techniques like diaphragmatic breathing and reduced brain pulsatility may further mitigate anxiety, although these areas of research remain in early stages.⁸⁴

We evaluated the appropriate metabolic equivalent (MET) for exercise intensity using original reports or the 2024 compendium, discovering that included RCTs predominantly employed moderate intensity (3.5–6.5 METs).⁸⁵ Recognising that METs may not fully encapsulate the complexity of RT, we prioritised intensity ranges from the original texts, generally 60%–70% of one-repetition maximum (1RM). Contrary to previous studies with broader populations,^{22 86} our findings indicate that moderate-intensity RT may be particularly effective in alleviating anxiety disorders. Additionally, the role of rest intervals in RT should be considered; prior research suggests that 50–55% 1RM with 90 s rest intervals optimises state anxiety.⁸⁷ However, only two studies in our review reported rest durations, which ranged from 80 s to 90 s. This highlights an important avenue for future research: identifying the most effective combinations of intensity and rest intervals for anxiety reduction. While acute exercise research often focuses on these aspects, long-term training protocols may overlook them, underscoring the need for comprehensive reporting of intervention characteristics.

Insights on MBEs

Our findings align with recent studies,^{24 25 34 88 89} supporting the effectiveness of MBE as a treatment for anxiety symptoms and disorders. However, we observed that MBE appears more effective for anxiety symptoms than for anxiety disorders.^{34 89} Although our regression analysis showed no significant differences, most included studies used self-reported or clinician-administered tools instead of clinical diagnostic criteria (eg, DSM). Unlike traditional exercise, MBE involves low-intensity activity combined with mindfulness and meditation,⁹⁰,⁹² which may help reduce common side effects of traditional exercise, such as muscle soreness, shortness of breath and nausea. Given that 70% of participants were female, and women are more prone to anxiety-related physical symptoms,⁹³ MBE may be especially significant for addressing gender-related differences in anxiety. Neurobiologically, MBE may enhance parasympathetic nervous system activity and increase GABA levels in the thalamus, potentially improving mood. Another hypothesis is that MBE could reduce hypothalamic-pituitary-adrenal axis activation to alleviate anxiety, though evidence for this remains inconsistent and requires further exploration.⁹⁴,⁹⁶ While MBE’s effectiveness is well-supported, there are limitations in defining exercise intensity and reporting adverse events. Our calculation of MET revealed that most studies (except two RCTs) employed low-intensity MBE (MET values between 2.3 and 3.3),⁸⁵ suggesting its greater effectiveness for anxiety. Additionally, long-duration meditation may pose risks, including negative psychological reactions (eg, flashbacks or hallucinations), which need further documentation in future studies.⁹⁷

Strengths and limitations

This NMA provides a comprehensive evaluation of exercise effectiveness in patients with anxiety disorders. Key strengths include adherence to diagnostic criteria by excluding PTSD and OCD, enhancing the applicability to clinical practice. Our dual search strategy included SRs and RCTs, reducing the risk of missing relevant data, including unpublished data. Unlike previous reviews focusing on passive controls, we included comparisons between exercise and psychological interventions, allowing us to categorise exercise treatments and therapies for future practice.

However, our study has limitations. Only four studies had a low risk of bias, with some showing flaws like poor allocation concealment. Nearly half of the RCTs did not use ITT analyses, potentially inflating intervention effectiveness. The lack of RCT evidence for disorders like SAD and PD limits our ability to compare RT and MBE. Only 12 RCTs reported follow-up data (1–12 weeks), hindering our exploration of exercise’s sustained effects. A study found a 12-week intervention remained effective at the 1-year follow-up. Future RCTs should collect follow-up data to strengthen this effect.

We used all-cause discontinuation as an acceptability proxy due to limited RCT data (only 8/30 noted withdrawal reasons, 2/30 distinguished intervention-related causes), which cannot capture multidimensional acceptability.⁹⁸ Future RCTs should collect subjective data per PRISMA 2020 to improve rigour.

Additionally, the anxiety symptoms network (12 RCTs) showed high heterogeneity (τ² = 0.362), potentially reducing this subgroup’s result robustness. Key drivers include drastic exercise intensity variation (MET 1.5–7.0, eg, qigong vs high-intensity interval training), seven distinct anxiety assessment tools (eg, STAI series, HARS) and intervention durations of 4–16 weeks. Thus, findings for anxiety symptoms should be interpreted cautiously.

Notably, the predominantly female (72.8%) and untrained participant composition of included RCTs may restrict generalisability to male or athletic populations. Reliance on aggregate-level data limits generalisability to individual profiles. Future research should use individual data to explore patient-specific factors influencing outcomes. Additionally, combining similar exercise intervention arms to optimise NMA feasibility may mask subtle differences in intervention dose, modality, or intensity—slightly constraining conclusions about specific exercise subtypes’ effectiveness. Though the combined risk of small-study bias and publication bias is low, conclusions should still be interpreted with caution, given the inherent limitations of statistical methods in fully distinguishing between the two biases. While NMA is a powerful tool for evidence synthesis, it has challenges, as statistical assumptions and issues of inconsistency may affect result reliability.⁹⁹

Conclusion

This SR and NMA provide evidence supporting the effectiveness of both RT and MBE as effective interventions for adults with anxiety disorders. However, the certainty of the evidence ranges from very low to moderate. RT shows promise as a viable option for the long-term management of anxiety disorders, with consistent evidence supporting its effectiveness. While MBE also demonstrates robust effectiveness in treating anxiety disorders, its priority for treating anxiety disorders is less pronounced compared with RT. Given the variability in the quality and scope of the available evidence, further clinical research is required to better understand the specific effects and underlying mechanisms of these two exercise interventions.

Data availability statement

Data are available in a public, open access repository. Data are available upon reasonable request.