Follow-up effects of mindfulness-based interventions on anxiety in adolescents: a systematic review and meta-analysis of randomized controlled trials

Abstract

Background

Anxiety is one of the most prevalent mental health problems in adolescence, and its onset often coincides with the secondary school years. Mindfulness-based interventions (MBIs) have shown short-term benefits for reducing anxiety in young people, but the durability of these benefits after treatment remains uncertain. To inform clinical and school-based practice, a quantitative synthesis focused specifically on follow-up effects is needed.

Methods

We conducted a systematic review and meta-analysis of randomized controlled trials (RCTs) evaluating MBIs in adolescents aged 10–19 years and reporting anxiety outcomes at follow-up assessments conducted at least 3 months after the intervention. The review followed PRISMA 2020 guidelines and was prospectively registered in PROSPERO (CRD420251234254). Major databases (PubMed, Embase, Web of Science, Cochrane Library, Scopus and CNKI) were searched from inception to November 2025. Two reviewers independently screened studies, extracted data and assessed risk of bias using the Cochrane RoB 2 tool. Standardized mean differences (SMDs; Hedges’ g) with 95% confidence intervals (CIs) were calculated under a random-effects model. Pre-specified subgroup analyses were conducted by age group (10–14 vs. 15–19 years), intervention type (MBSR/MBCT vs. mindfulness-based stress prevention, MBSP), and follow-up interval (3–6 vs. ≥6 months). Univariable meta-regression explored the impact of mean age and total intervention duration on effect sizes. Certainty of evidence for key outcomes was rated using GRADE.

Results

Eleven randomized controlled trials (n = 9,856 adolescents) were included. Overall, MBIs were associated with a modest reduction in anxiety symptoms at follow-up assessments ≥ 3 months after treatment (SMD = − 0.30, 95% CI − 0.44 to − 0.16; I² = 79%). Subgroup analyses indicated a moderate effect at 3–6 months (SMD = − 0.40, 95% CI − 0.56 to − 0.25; I² = 52%) but a very small effect at ≥ 6 months (SMD = − 0.09, 95% CI − 0.14 to − 0.05; I² = 0%), with a significant difference between subgroups (p < 0.001).

Conclusions

MBIs may confer a modest follow-up benefit on adolescent anxiety, driven mainly by outcomes measured at 3–6 months. Evidence at ≥ 6 months is limited and of low certainty, and does not support a firm conclusion regarding maintenance of effects. Future trials should test whether maintenance-focused components, such as booster sessions or ongoing practice support, improve durability.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13034-026-01134-0.

Introduction

Adolescence is a high-risk period for the onset of anxiety disorders, with epidemiological studies indicating that around 10–20% of adolescents experience clinically relevant anxiety and that prevalence has been rising globally [1]. This burden is not limited to transient distress; adolescent anxiety is associated with pervasive worry, avoidance behaviours, and heightened physiological arousal, and is strongly linked to later adverse outcomes, including other psychiatric disorders (such as depression), and long-term functional impairment [2–4]. Identifying interventions that can produce not only immediate but also sustained reductions in anxiety symptoms is a major public-health and clinical priority.

In recent years, mindfulness-based interventions (MBIs) such as Mindfulness-Based Stress Reduction (MBSR) and Mindfulness-Based Cognitive Therapy (MBCT) have emerged as promising approaches for young people. These programmes use systematic mindfulness practice to cultivate a deliberate, non-judgemental awareness of present-moment experience, and have shown beneficial effects on a range of mental-health outcomes in both adults and youth [5]. Building on this evidence, MBIs have increasingly been introduced into schools, clinics, and community settings as a potential strategy to prevent or alleviate anxiety in children and adolescents.

Several systematic reviews and meta-analyses of randomized controlled trials (RCTs) suggest that MBIs can yield small-to-moderate short-term reductions in anxiety symptoms among young people [6–10]. However, the current evidence base has important limitations that hinder clinical translation. First, most existing quantitative syntheses focus on post-intervention outcomes and provide little dedicated evaluation of follow-up effects after treatment, typically defined as outcomes assessed ≥ 3 months after the end of the intervention [6–9, 11]. As a result, clinicians and policy-makers still lack robust evidence on whether benefits remain detectable after treatment in anxious adolescents. Second, many reviews pool heterogeneous outcomes (e.g. depression, stress, general psychological distress, and broad mental-health composites) alongside anxiety, which dilutes inferences about MBIs’ specific effects on this highly prevalent and clinically meaningful target [7, 8, 10]. Third, relatively few studies have systematically examined potential moderators of follow-up effects, such as age group or intervention type, and almost none have analysed how effect sizes change across different follow-up durations once the active training has ceased [12]. Consequently, it remains unclear whether the benefits of MBIs persist, attenuate, or disappear over time in adolescent populations, and whether these trajectories differ by developmental stage or programme characteristics.

Taken together, these gaps highlight the need for a focused synthesis of follow-up outcomes. Therefore, this systematic review and meta-analysis had three aims: (1) to quantify follow-up effects of mindfulness-based interventions on anxiety in adolescents at ≥ 3 months after treatment; (2) to examine whether these effects change as follow-up duration increases; and (3) to explore whether age and intervention type moderate follow-up outcomes.

Methods

Study design and registration

This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guideline and was prospectively registered in PROSPERO (CRD420251234254) [13].

Use of a large language model (ChatGPT, OpenAI) was limited to language polishing and structural editing of the manuscript draft and did not involve generation of original data or independent scientific conclusions, in accordance with the journal’s policy on LLMs.

Eligibility and exclusion criteria

Using the PICOS framework, we defined adolescents as individuals aged 10–19 years, consistent with the World Health Organization and commonly used age ranges in adolescent health research [14]. We included adolescents aged 10–19 years regardless of diagnostic status (clinical, high-risk, or community/school-based samples). Most trials recruited community or school-based participants and assessed outcomes using validated symptom scales rather than DSM/ICD-based diagnostic interviews. Where reported, we extracted baseline anxiety severity (mean and SD) and any information on comorbidities; however, anxiety subtype distributions and severity-based classifications (e.g., mild/moderate/severe) were rarely available and therefore could not be synthesized.

Eligible studies were required to meet all of the following criteria: (1) participants were adolescents aged 10–19 years; (2) the intervention was a mindfulness-based intervention (MBI), operationalised as a programme in which systematic mindfulness practice was a core component and which was explicitly described as MBSR, MBCT, MBSP or another structured mindfulness course; (3) the comparison condition was no intervention, waitlist control, usual curriculum, or another active control (e.g. health education or regular classes); (4) the primary outcome included at least one validated measure of anxiety symptoms; (5) at least one follow-up assessment was reported ≥ 3 months after the end of the intervention. We note that follow-up time points were heterogeneous across studies (approximately 3–12 months post-intervention) and therefore do not represent a uniform durability interval; (6) the study design was a randomized controlled trial; and (7) the full text was available in either English or Chinese.

Exclusion criteria were: (1) duplicate publications; (2) incomplete or irretrievable data; (3) absence of relevant outcome measures; (4) non-randomized or quasi-experimental designs; (5) interventions in which mindfulness was not the primary or core component; (6) no follow-up assessment reported; and (7) samples not falling within the 10–19-year adolescent age range.

Search strategy and information sources

We conducted a comprehensive literature search in the following electronic databases: PubMed, Embase, Cochrane Library, Web of Science, Scopus, and CNKI. Searches covered the period from database inception to November 2025. The search strategy was developed based on the PICOS framework, combining controlled vocabulary (e.g. MeSH/Emtree terms) and free-text terms, and using the Boolean operators AND and OR to construct search strings in line with established methodological guidance [15]. In addition, we screened reference lists of included trials and relevant prior reviews, and performed citation tracking for key studies to identify additional eligible RCTs. The detailed search strategy for PubMed is presented in Table 1, and search strategies for the other databases were adapted accordingly. Detailed search frameworks and retrieval information for all databases are provided in Supplementary Table S1.

Table 1.

Search strategy on PubMed

Query number	Search terms
#1	Mindfulness [MeSH Terms] OR Meditation [MeSH Terms]
#2	mindfulness [Title/Abstract] OR “mindfulness-based” [Title/Abstract] OR “mindfulness based” [Title/Abstract] OR “mindfulness-based intervention*” [Title/Abstract] OR “mindfulness based intervention*” [Title/Abstract] OR “mindfulness-based program*” [Title/Abstract] OR “mindfulness based program*” [Title/Abstract] OR “mindfulness-based stress reduction” [Title/Abstract] OR MBSR [Title/Abstract] OR “mindfulness-based cognitive therap*” [Title/Abstract] OR MBCT [Title/Abstract] OR “mindfulness training” [Title/Abstract] OR “school-based mindfulness” [Title/Abstract]
#3	#1 OR #2
#4	Adolescent [MeSH Terms]
#5	adolescen* [Title/Abstract] OR teen* [Title/Abstract] OR youth* [Title/Abstract] OR “middle school” [Title/Abstract] OR “high school” [Title/Abstract]
#6	#4 OR #5
#7	Anxiety [MeSH Terms] OR “Stress, Psychological” [MeSH Terms]
#8	anxi* [Title/Abstract] OR stress* [Title/Abstract] OR “psychological distress” [Title/Abstract] OR PSS [Title/Abstract] OR STAI [Title/Abstract] OR RCADS [Title/Abstract] OR DASS [Title/Abstract] OR SDQ [Title/Abstract] OR GHQ [Title/Abstract] OR K10 [Title/Abstract] OR K6 [Title/Abstract]
#9	#7 OR #8
#10	Randomized Controlled Trial [Publication Type]
#11	randomized [Title/Abstract] OR randomised [Title/Abstract] OR random* [Title/Abstract] OR “cluster random*” [Title/Abstract] OR placebo [Title/Abstract]
#12	#3 AND #6 AND #9 AND (#10 OR #11)

The asterisk (*) was used as a wildcard to retrieve different word endings (e.g., anxi for anxiety, anxious)

Study selection and data extraction

Two reviewers (HB and BE) independently screened records in two stages using a piloted screening form: (1) title and abstract screening, (2) followed by full-text eligibility assessment. At each stage, studies were assessed against the predefined inclusion and exclusion criteria, and potentially eligible reports were preliminarily coded in a standardized data extraction sheet. Any disagreements between the two reviewers were resolved through discussion with a third reviewer (JY) until consensus was reached.

Data extraction was also conducted independently by two reviewers using a predesigned form. Extracted information included general study characteristics (first author, publication year, country), participant characteristics (sample size, age), intervention characteristics (type of MBI, total duration), comparator type (e.g. usual care, waitlist, active control), follow-up duration (3–6 months vs. ≥6 months), and anxiety outcomes. Anxiety measures were grouped into two categories based on their content: Anx-SPEC, referring to anxiety-specific scales (e.g. RCADS [16], STAI [17]), and Distress-COM, referring to composite measures of general psychological distress (e.g. DASS-Total [18], GHQ-12 [19], K10/K6 [20]). When a study reported multiple eligible anxiety-related measures, we followed a pre-specified priority order (RCADS > STAI > DASS-Anxiety > composite distress measures) and selected a single outcome per study to avoid double counting. In addition, we extracted key implementation characteristics, including dose (sessions × minutes; total contact hours), delivery setting, and whether home practice was explicitly required or encouraged; these are summarised in Supplementary Table S2.

Because included trials used different validated instruments, we synthesized continuous outcomes using standardized mean differences (SMDs; Hedges’ g), calculated as the between-group mean difference divided by the pooled within-study SD. To enhance construct interpretability, we pre-specified two outcome groups based on instrument content (Anx-SPEC vs. Distress-COM), prioritized anxiety-specific measures when multiple eligible outcomes were reported, and conducted sensitivity analyses restricted to Anx-SPEC outcomes.

For quantitative synthesis, we preferentially extracted means and standard deviations at the latest follow-up time point. When multiple follow-up time points were reported within a trial, we used the latest eligible follow-up for the primary analysis; for time-window subgroup analyses (3–6 vs. ≥ 6 months), we extracted the time point that fell within the corresponding window. When only standard errors or 95% confidence intervals were reported, standard deviations were derived using formulas recommended in the Cochrane Handbook [21]. If data were presented solely in graphical form, we used digital extraction software to estimate the corresponding numerical values. Studies were excluded from the meta-analysis if the required data could not be obtained or reliably approximated using these methods. Any discrepancies arising during data extraction were resolved through discussion with a third reviewer until agreement was reached.

Risk of bias assessment

Risk of bias was assessed independently by two reviewers using the Cochrane Risk of Bias 2 (RoB 2) tool for randomized trials [22]. The tool covers five domains: (1) bias arising from the randomization process; (2) bias due to deviations from intended interventions; (3) bias due to missing outcome data; (4) bias in measurement of the outcome; and (5) bias in selection of the reported result. Each trial was rated as having low risk of bias, some concerns, or high risk of bias.

Statistical analysis

We calculated standardized mean differences (SMDs; Hedges’ g) with 95% CIs for continuous anxiety outcomes [23] and pooled them using random-effects models (DerSimonian–Laird) [24]. Heterogeneity was assessed using τ² and I², with I² values of approximately 25%, 50% and 75% indicating low, moderate and high heterogeneity, respectively [25]. When standard deviations were missing, we derived them from standard errors, confidence intervals, t/z statistics or p values where possible, or approximated means and SDs from medians and interquartile ranges following published methods [26]. Subgroup analyses were pre-specified by age group (10–14 vs. 15–19 years), intervention type (MBSR/MBCT vs. MBSP) and follow-up interval (3–6 vs. ≥6 months). Univariable random-effects meta-regression was conducted to explore whether mean age and total intervention duration (in weeks or hours) were associated with effect size. Publication bias was examined using visual inspection of funnel plots and Egger’s regression test [27]. Sensitivity analyses included switching to fixed-effects models and leave-one-out analyses to assess the influence of individual studies [28]. All analyses were conducted using Review Manager (RevMan) version 5.4 and Stata version 17.0.

Several included trials used cluster randomisation (e.g., by school or class). For these studies, we preferentially extracted effect estimates from the original authors’ analyses that explicitly accounted for clustering (e.g., multilevel/mixed-effects models or GEE with robust standard errors). When cluster adjustment was not reported, we adjusted the effective sample size using a design-effect approach (DE = 1 + (m − 1)×ICC), where m is the average cluster size and ICC was taken from the trial report when available. To assess the impact of cluster designs on pooled estimates, we conducted (i) a subgroup analysis comparing cluster-randomised versus individually randomised trials and (ii) a sensitivity analysis excluding cluster-randomised trials.

To improve readability and reduce redundancy, only key figures are presented in the main text (PRISMA flow diagram, overall pooled effect forest plot, and the RoB 2 summary). Additional subgroup, sensitivity, and diagnostic plots (e.g., design-stratified analyses, exclusion sensitivity analyses, meta-regression bubble plots, leave-one-out plots, and funnel plot) are provided in the Supplementary Material (Supplementary Figures S1–S13).

GRADE evidence assessment

We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to rate the certainty of evidence for the primary outcome (follow-up anxiety outcomes assessed at ≥ 3 months after treatment) [21]. Starting from “high” certainty for RCTs, we considered downgrading by one or two levels for risk of bias, inconsistency, indirectness, imprecision and publication bias. A summary of findings table was generated to present pooled effect sizes and GRADE ratings.

Results

Study selection

After database and other-source searching, we identified 932 records (929 from databases and 3 from other sources). After removing 590 duplicates, 342 records remained for title and abstract screening. Of these, 264 were excluded. We assessed 78 full-text reports for eligibility and excluded 67 for reasons including ineligible study design, ineligible population/age range, lack of eligible anxiety-related outcomes at ≥ 3-month follow-up, or insufficient data to compute effect sizes. Eleven RCTs were included in the final meta-analysis (Fig. 1).

Fig. 1.

PRISMA flow diagram of study selection

Study characteristics

A total of 9,856 participants aged 10–19 years were included across 11 RCTs. Six trials evaluated “classical” MBSR/MBCT-type programmes, and five evaluated mindfulness-based stress prevention programmes (MBSP). Follow-up time points ranged from 3 to 12 months post-intervention: eight studies reported outcomes at 3–6 months and three reported outcomes at ≥ 6 months. With respect to age strata, five studies primarily involved participants aged 10–14 years and six involved participants aged 15–19 years. Comparators included both passive controls (e.g., waitlist, TAU/usual curriculum) and active controls. Detailed study characteristics are presented in Table 2.

Table 2.

Characteristics of the included randomized controlled trials

Author	Year	Country	Design (0/1)	Sample size	Mean age (years)	Intervention type	Intervention duration (weeks)	Control group	Follow-up time point (months)	Outcome measure
Basch M. et al. [29]	2024	USA	0	N = 42	14.8	MBSP	6	AC	3	Anx-SPEC
Lahtinen and Salmivalli [30]	2020	Finland	0	N = 1349	17	MBSP	8	PC	3	Anx-SPEC
Johnson et al. [31]	2017	Australia	1	N = 555	13.44	MBSR	9	PC	12	Distress-COM
Weintraub M. J. et [32]	2023	USA	0	N = 66	15.1	MBSP	9	AC	3	Distress-COM
Biegel G. M. et al. [33]	2009	USA	0	N = 102	15.35	MBSR	8	AC	3	Anx-SPEC
Abrkar Z. et al. [34]	2023	Iran	0	N = 100	14	MBCT	6	PC	3	Anx-SPEC
Tan & Martin [35]	2015	Australia	0	N = 108	15.4	MBSP	5	AC	3	Distress-COM
Lan et al. [36]	2024	China	0	N = 653	13	MBCT	8	PC	3	Anx-SPEC
Kuyken et al. [37]	2022	UK	1	N = 8376	12.2	MBCT	10	AC	12	Anx-SPEC
Scafuto et al. [38]	2024	Italy	0	N = 361	14	MBSP	12	PC	3	Distress-COM
Hilt et al. [39]	2025	USA	0	N = 152	13.71	MBCT	3	AC	6	Anx-SPEC

MBSP Mindfulness-Based Stress Prevention; MBSR Mindfulness-Based Stress Reduction; MBCT Mindfulness-Based Cognitive Therapy; PC Passive Control; AC Active Control; Anx-SPEC anxiety-specific measures; Distress-COM = measures of general psychological distress and comorbidity. Design: 0 = individually randomised; 1 = cluster-randomised (school/class)

Risk of bias

Risk of bias was assessed for all 11 RCTs using the RoB 2 tool (intention-to-treat effect). Overall, 10 trials were judged as having some concerns and one trial was judged as high risk of bias (Fig. 2). As shown in the domain-level summary (Fig. 2), “some concerns” were primarily driven by the randomization process, whereas most trials were rated low risk for deviations from intended interventions and missing outcome data; risk related to measurement of the outcome was generally low. Study-level RoB 2 judgements (traffic-light plot) and supporting rationales are provided in Supplementary Figure S1.

Fig. 2.

Risk-of-bias summary (RoB 2)

Overall follow-up effects on anxiety

Across 11 RCTs, MBIs were associated with a statistically significant reduction in anxiety at follow-up assessments ≥ 3 months after treatment (SMD = − 0.30, 95% CI [− 0.44, − 0.16]). Heterogeneity was substantial (I² = 79%, p < 0.001), indicating considerable between-study variability. We therefore conducted pre-specified subgroup analyses and meta-regression to explore potential moderators (Fig. 3).

Fig. 3.

Overall follow-up effects of mindfulness-based interventions on adolescent anxiety

Subgroup analyses

Age 10–14 vs. 15–19

In the 10–14 years subgroup, six trials (n = 8,899) were included. Heterogeneity was substantial (I² = 85%, p < 0.001). Using a random-effects model, MBIs were associated with a statistically significant reduction in anxiety at follow-up (SMD = − 0.26, 95% CI [− 0.45, − 0.08]). In the 15–19 years subgroup, five trials (n = 957) were included, with low heterogeneity (I² = 10%, p = 0.35). The pooled effect was larger (SMD = − 0.36, 95% CI [− 0.51, − 0.20]). The test for subgroup differences by age was not statistically significant (χ² = 0.60, df = 1, p = 0.44; Supplementary Figure S2), indicating no clear evidence that follow-up effects differed between the two age groups.

MBSR/MBCT vs. MBSP

Six trials were classified as MBSR/MBCT. Heterogeneity was low (I² = 2%, p = 0.40), and the pooled effect indicated a small but statistically significant reduction in anxiety at follow-up (SMD = − 0.11, 95% CI [− 0.16, − 0.06]). Five trials were classified as MBSP. Heterogeneity was moderate (I² = 66%, p = 0.02), and the pooled effect was larger (SMD = − 0.46, 95% CI [− 0.71, − 0.21]). The test for subgroup differences was statistically significant (χ² = 7.41, df = 1, p = 0.006; Supplementary Figure S3).

3–6 months vs. ≥6 months

Eight trials contributed data to the 3–6 months follow-up subgroup. Heterogeneity was substantial (I² = 52%, p = 0.04). The pooled effect indicated a moderate reduction in anxiety (SMD = − 0.40, 95% CI [− 0.56, − 0.25]). Three trials contributed data to the ≥ 6 months subgroup. Heterogeneity was negligible (I² = 0%, p = 0.81). The pooled effect was very small (SMD = − 0.09, 95% CI [− 0.14, − 0.05]). The test for subgroup differences was statistically significant (χ² = 14.19, df = 1, p < 0.001; Supplementary Figure S4), indicating that effects attenuated over longer follow-up (Table 3).

Table 3.

Subgroup analyses of follow-up effects of mindfulness-based interventions on adolescent anxiety

Subgroup variable	Subgroup	No. of studies (N)	Pooled SMD (95% CI)	I² within subgroup (%)	P value for subgroup difference
Age	10–14 years	6	−0.26 [− 0.45, − 0.08]	85%	p = 0.440
	15–19 years	5	−0.36 [− 0.51, − 0.20]	10%
Intervention type	MBSR/MBCT	6	−0.11 [− 0.16, − 0.06]	2%	p = 0.006
	MBSP	5	−0.46 [− 0.71, − 0.21]	66%
Follow-up duration	3–6 months	8	−0.40 [− 0.56, − 0.25]	52%	p < 0.001
	≥ 6 months	3	−0.09 [− 0.14, − 0.05]	0%

Impact of cluster-randomised trials

Design-stratified subgroup analysis. In analyses stratified by trial design (Supplementary Figure S5), cluster-randomised trials showed a very small pooled effect (SMD = − 0.09, 95% CI [− 0.14, − 0.05]; I² = 0%), whereas individually randomised trials showed a larger pooled effect (SMD = − 0.38, 95% CI [− 0.52, − 0.24]; I² = 50%). The between-subgroup difference was statistically significant (χ² = 14.31, df = 1, p = 0.0002).

Sensitivity analysis excluding cluster-randomised trials. Excluding cluster-randomised trials (Supplementary Figure S6), the pooled effect remained statistically significant (SMD = − 0.38, 95% CI [− 0.52, − 0.24]; I² = 50%), indicating that the overall direction and statistical significance were consistent.

Meta-regression

In this study, we used univariable random-effects meta-regression to explore whether study-level continuous variables were associated with follow-up effect size. Standardized mean differences (SMDs) were entered as the dependent variable, and mean age and total intervention duration were used as covariates. When mean age was entered as a covariate, the meta-regression including all 11 studies showed that age was significantly associated with follow-up efficacy (β = −0.116, 95% CI [− 0.260, − 0.0271], p < 0.05). Given the small number of studies, this finding should be interpreted as exploratory and requires confirmation in future research (Supplementary Figure S7).

When total intervention duration (in weeks) was entered as a covariate, the meta-regression indicated no significant association with follow-up efficacy (β = −0.012, 95% CI [− 0.0815, 0.0574], p = 0.705), suggesting that longer programmes are not necessarily linked to larger follow-up effects (Supplementary Figure S8 and Table 4).

Table 4.

Meta-regression of study-level covariates on follow-up intervention effects

Covariate	β	SE	95% CI	P value	Model R²	Significance
Mean age (years)	−0.116	0.063	[− 0.260, − 0.0271]	< 0.05	0.33	Significant
Total intervention duration (weeks)	−0.012	0.030	[− 0.0815, 0.0574]	0.705	−0.12	Not significant

Sensitivity analyses and publication bias

Results were similar under a fixed-effect model (SMD = − 0.28, 95% CI [− 0.38, − 0.18]) compared with the primary random-effects model (SMD = − 0.30, 95% CI [− 0.44, − 0.16]), indicating that conclusions were not sensitive to the choice of model.

Leave-one-out analyses showed that omitting any single study did not materially change the pooled estimate (range: SMD − 0.37 to − 0.29), and all re-estimated effects remained statistically significant (Supplementary Figure S9). The study with the greatest impact on heterogeneity was Johnson et al. [31], which produced the largest reduction in I² when omitted, but the pooled SMD remained stable. These analyses support statistical stability of the pooled estimate, although they do not resolve conceptual heterogeneity arising from differences in outcomes, contexts, and intervention implementation.

Additional sensitivity analyses restricted to anxiety-specific outcomes, including exclusion of the two largest trials, are presented in Supplementary Figures S10–S12.

Visual inspection of the funnel plot did not suggest marked asymmetry. Egger’s test was also non-significant (p = 0.353), indicating no statistically detectable small-study effects or publication bias in the overall meta-analysis (Supplementary Figure S13).

GRADE summary

The certainty of evidence for the primary outcome was assessed using the GRADE approach. For change in anxiety symptoms at 3–6 months of follow-up, certainty was downgraded from high to moderate because of risk of bias (most trials had some concerns, particularly related to the randomization process and lack of blinding) and inconsistency across studies (I² ≈ 52%). For change in anxiety symptoms at ≥ 6 months of follow-up, certainty was downgraded to low because of risk of bias and indirectness. Indirectness reflects that the ≥ 6-month evidence base was small and largely derived from school-based designs with limited diagnostic characterization (e.g., DSM/ICD status, subtypes, severity, and comorbidity) and heterogeneous outcome constructs (anxiety-specific versus broader distress composites), which limits generalisability to clinically diagnosed anxiety and to a uniform durability claim. We did not detect important publication bias for either outcome and therefore did not downgrade further (Table 5).

Table 5.

Summary of findings for follow-up effects of mindfulness-based interventions on anxiety

Outcome (anxiety)	Study design	Downgrading factors	Participants (No. of studies)	Effect size (SMD, 95% CI)	Quality of evidence
follow-up 3–6 months	RCT	Risk of bias (-1); inconsistency (-1)	1953 participants (8 RCTs)	−0.40 [− 0.56, − 0.25]	Moderate
follow-up ≥ 6 months	RCT	Risk of bias (-1); indirectness (-1)	7903 participants (3 RCTs)	−0.09 [− 0.14, − 0.05]	Low

Overall, evidence for follow-up effects of MBIs on adolescent anxiety at ≥ 3 months should be interpreted cautiously. The 3–6-month outcome was supported by moderate-certainty evidence, whereas the ≥ 6-month outcome was based on only three studies and rated as low certainty. These data do not support a firm conclusion regarding maintenance of effects beyond 6 months.

Discussion

Main findings and interpretation

This meta-analysis suggests that mindfulness-based interventions (MBIs) are associated with a modest reduction in adolescent anxiety at follow-up assessments conducted at least 3 months after treatment. However, this overall estimate should be interpreted cautiously because heterogeneity was substantial and the pooled result combined studies that differed in setting, comparator type, intervention format, and outcome construct.

Most importantly, the apparent follow-up benefit was not uniform across time windows. The pooled effect was moderate at 3–6 months but very small at ≥ 6 months, and the certainty of evidence for the ≥ 6-month subgroup was low. Accordingly, our findings do not support a firm conclusion that MBIs produce durable maintenance effects beyond 6 months in adolescents. The most defensible interpretation is attenuation over time rather than confirmed long-term maintenance.

Exploratory subgroup and meta-regression findings should also be interpreted cautiously. Although larger effects were observed in some subgroups, including 3–6-month follow-up and MBSP-type programmes, these analyses were based on a small number of studies and may reflect differences in design, population, or comparator intensity rather than true intervention superiority. These results should therefore be regarded as hypothesis-generating only.

Context and implications

The attenuation pattern observed here is broadly consistent with literature cautioning against assuming that brief mindfulness programmes necessarily produce stable trait-level change [40, 41]. It is also compatible with broader intervention research suggesting that benefits from time-limited psychological programmes may weaken over longer follow-up when structured reinforcement is absent [42, 43, 46]. Related adult MBCT literature has reported a similar need for cautious interpretation of durability beyond the early follow-up period [44, 45].

Our results also suggest several priorities for future research. Developmental stage may influence durability, given ongoing maturation of cognitive control and self-regulation across adolescence [47, 48], but the age signal in our meta-regression was exploratory and should not be overinterpreted. Likewise, nominal programme duration was not associated with larger follow-up effects, which is consistent with evidence that adherence, practice quality, and treatment fidelity may matter more than calendar length alone [49, 50, 52]. Future trials should therefore measure post-intervention practice more carefully and directly test maintenance-oriented strategies such as booster sessions, digital supports, and embedding mindfulness practice into routine school delivery [42, 43, 53, 54].

Strengths and limitations

A key strength of this review is its focus and analytic depth. To our knowledge, it is among the first meta-analyses to quantify follow-up effects of mindfulness-based interventions (MBIs) on anxiety in adolescents at ≥ 3 months after treatment and to examine attenuation over time. We also applied pre-specified subgroup analyses and exploratory meta-regression to investigate potential sources of heterogeneity and conducted multiple sensitivity analyses to assess robustness.

Several limitations should be acknowledged. First, overall heterogeneity was substantial (I² = 79%), indicating considerable variability across trials in populations, implementation features, and outcome measures. In addition, psychometric properties of the anxiety/distress instruments (e.g., internal consistency such as Cronbach’s α) were not consistently reported across included trials, which may have introduced measurement uncertainty and contributed to between-study variability. Second, what we defined as follow-up varied widely (3–12 months), limiting the granularity with which we could characterise time-dependent trajectories.

Third, almost none of the included RCTs reported adherence to home mindfulness practice during the follow-up period, even though continued practice is often considered important for maintaining treatment gains [51, 52]. The absence of such data limited our ability to test mechanistic explanations for the observed attenuation in effects.

Fourth, some included trials used cluster randomisation at the class or school level. We preferentially extracted effect estimates that accounted for clustering when reported by trial authors; however, for trials where clustering adjustment was unclear, additional ICC-based corrections were not feasible. Design-stratified analyses and exclusion sensitivity analyses did not materially change pooled effect estimates, suggesting that any residual unit-of-analysis bias is more likely to affect precision than direction.

Finally, because most included trials did not apply standardized DSM-5/ICD-11 diagnostic interviews or report anxiety subtypes, comorbidity profiles, or severity strata, our findings primarily reflect changes in symptom scale scores and cannot be generalized to specific anxiety disorders or severity-defined clinical populations. In addition, because this review synthesized follow-up outcomes rather than change from post-treatment to follow-up, it cannot establish whether gains were fully maintained, partially lost, or newly emerged after treatment; it can only describe the magnitude of between-group differences at follow-up time points. Trials also spanned heterogeneous contexts, from universal school prevention to targeted or clinical interventions, which may moderate durability and generalisability.

Clinical implications and future directions

From a clinical and educational perspective, current evidence suggests that MBIs may offer short-to-medium-term benefit for adolescent anxiety, but their durability remains uncertain. The marked reduction in pooled effect from 3 to 6 months to ≥ 6 months suggests that follow-up benefits may erode without ongoing reinforcement. These implications are hypothesis-generating rather than prescriptive, given the limited ≥ 6-month evidence.

Future interventions could test brief booster sessions, digital support for home practice, and integration of mindfulness skills into routine school contexts [53, 54]. However, the effectiveness and feasibility of such maintenance-oriented approaches must be established in well-designed trials with longer follow-up, anxiety-specific outcomes, and better reporting of adherence.

In summary, MBIs were associated with a modest follow-up benefit on adolescent anxiety, driven mainly by outcomes measured at 3–6 months. Evidence at ≥ 6 months was very small and low certainty, and does not support a firm conclusion regarding maintenance of effects. Future research should focus on whether maintenance-phase components can improve the durability of benefits.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Hongbo Zhao is responsible for conceptualizing, reviewing, writing and editing the initial draft.Hongxin Huang is responsible for data analysis, methodology and funding acquisition.All authors have read and approved the final manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Data availability

All data extracted and analysed in this review are available in the published articles included in the study. Additional extraction sheets are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. This study is a systematic review and meta-analysis of published randomized controlled trials and did not involve any new data collection from human participants.

Consent for publication

Not applicable. This manuscript does not contain any individual person’s data in any form.

Competing interests

The authors declare no competing interests.