Drawing and visual art interventions for anxiety in children aged 5–12: a systematic review and meta-analysis of randomised controlled trials Authors: Yu Jia, Shaojie Zhang, Diying Liao Published: 2026 ABSTRACT AbstractObjectiveTo evaluate the effects of drawing- and visual art-based interventions on anxiety in children aged 5–12 years and to examine variation across comparator structures and settings.MethodsPubMed, Web of Science, Cochrane Library, ERIC, DOAJ and China National Knowledge Infrastructure were searched for randomised controlled trials (RCTs). Postintervention anxiety was the primary outcome and was pooled as standardised mean differences. Trials using non-art or usual-care comparators and those using active art comparators were synthesised separately. Risk of bias was assessed using RoB 2.Results10 RCTs involving 816 children were included. The primary meta-analysis included eight trials with non-art or usual-care comparators conducted in medical or procedure-related contexts. Drawing- and visual art-based interventions were associated with lower short-term anxiety compared with these controls (SMD=−2.70, 95% CI −3.73 to −1.68; I²=96%). Leave-one-out sensitivity analyses and exclusion of studies at high risk of bias did not materially change the direction of the effect although heterogeneity remained substantial. Two school-based trials used active art-based comparators and were synthesised separately.ConclusionDrawing- and visual art-based interventions may reduce short-term anxiety mainly in acute medical or procedure-related contexts. Because heterogeneity was substantial and school-based evidence used different comparator structures, the current evidence does not support a single generalisable effect across paediatric settings.PROSPERO registration numberCRD420261288781.Keywords: Child, Child Health, Child Psychiatry, Therapeutics Introduction Anxiety symptoms are among the earliest-emerging and most prevalent emotional difficulties in childhood, often becoming evident during the early school years.1 2 Epidemiological evidence suggests that a substantial proportion of children aged 5–12 years experience clinically meaningful anxiety symptoms even without meeting formal diagnostic criteria for an anxiety disorder.3 These symptoms are associated with poorer academic functioning, impaired social relationships and an increased risk of persistent mental health difficulties extending into adolescence and adulthood.4 5 Yet in routine settings where time, personnel and specialist resources are limited, it remains unclear which early interventions can produce measurable and reproducible anxiety reduction at developmentally sensitive stages.6 This uncertainty constrains the practical reach of early child mental health support. Cognitive behavioural therapy is widely regarded as a first-line intervention for childhood anxiety.7 8 However, in younger children, implementation in routine care is often limited by restricted access to specialist services, variable engagement and adherence, and developmental constraints in verbal and cognitive capacities.9–11 There is therefore continued interest in low-barrier, developmentally appropriate approaches that can be delivered in settings such as schools and healthcare environments, where children commonly encounter stress, but specialist mental health provision may be limited.12 Drawing- and visual-art-based activities are increasingly used as part of emotional support and psychosocial interventions for children.13–15 These approaches include free or guided drawing, colouring, drawing-based storytelling and other structured forms of visual expression delivered individually or in groups. Compared with interventions that rely more heavily on verbal articulation or abstract cognitive processing, visual-art activities may offer a more developmentally accessible route to emotional expression and regulation, particularly for children whose language-based self-expression is still emerging.16–19 Their feasibility and adaptability have also made them attractive in school and healthcare settings. However, plausibility and practicality should not be mistaken for evidential clarity. Although previous meta-analytic work has suggested that art-based interventions may reduce anxiety symptoms in children and adolescents overall, the interpretability of that evidence remains limited when substantively different trial structures are synthesised together. Studies vary not only in age range, intervention format, setting and outcome timing but also in comparator structure. This distinction is important because trials comparing drawing- or visual-art-based interventions with non-art or usual-care controls address a different practical and inferential question from trials comparing one art-based activity with another. When such contrasts are pooled without sufficient distinction, overall effect estimates may suggest a favourable direction while remaining less informative about what, specifically, the evidence supports for younger children. This issue is especially important because the pathways through which visual-art interventions may influence anxiety are likely to differ by context. In medical or procedure-related settings, brief drawing or colouring tasks may reduce acute anxiety through attentional engagement and competition with threat-related cues.20–22 In school-based, non-clinical contexts, multi-session programmes may instead operate through repeated opportunities for emotional expression, regulation practice and social participation.23 24 These are not interchangeable pathways, and they should not automatically be treated as evidence of the same kind. The present review therefore focused specifically on children aged 5–12 years, a developmental period chosen to capture school-aged childhood while excluding adolescence, during which anxiety presentation, self-report capacity and intervention responsiveness may differ in important ways. We synthesised randomised controlled trial evidence on drawing- and visual-art-based interventions using a comparator-structured framework, treating trials with non-art or usual-care controls as the primary comparison and trials with active art comparators as supplementary evidence. Rather than merely updating pooled estimates, this review aimed to clarify the scope of inference that current evidence can support for paediatric visual-art interventions across different settings, intervention doses and outcome time points. Materials and methods The protocol was registered prospectively in PROSPERO (CRD420261288781), and the conduct and reporting followed the Cochrane Handbook for Systematic Reviews of Interventions and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Search strategy We conducted systematic database searches complemented by supplementary searching to identify randomised controlled trials evaluating drawing- or visual arts-based interventions for anxiety in children. PubMed, Web of Science Core Collection, the Cochrane Library, ERIC, DOAJ and the China National Knowledge Infrastructure (CNKI) were searched from inception to 20 December 2025, with no restrictions on language or publication year. Search terms combined controlled vocabulary (where available) with free-text keywords and were structured around three concepts: intervention, outcome and population. Intervention terms included “drawing”, “painting”, “coloring”, “art therapy” and “visual art”; outcome terms included “anxiety” and “anxious symptoms”; and population terms included “child” and “school-aged”. Database-specific syntax was adapted as needed, and concepts were combined using Boolean operators (AND/OR). Full strategies for each database are provided in the online supplemental materials. Inclusion criteria We included randomised controlled trials (individual or cluster) in children aged 5–12 years. Studies with wider age ranges were eligible only if 5–12-year data were reported separately or could be derived. Eligible interventions had drawing/visual art as the primary active component (any format, setting or delivery). Comparators were non-art conditions or active art controls differing from the intervention in prespecified elements. Anxiety (continuous scale) was the primary outcome; postintervention endpoints were prioritised (change scores used if endpoints were unavailable), with follow-up outcomes analysed secondarily. Quantitative synthesis required extractable/convertible effect-size data. To avoid double counting, one effect size per study per analysis was selected using an a priori hierarchy. We excluded non-randomised or uncontrolled designs, interventions where the independent visual-art effect was not isolatable, studies without usable anxiety data, protocols/abstracts-only reports, unavailable full texts and duplicate publications (retaining the most complete report). Study selection One reviewer conducted the initial database search, with search terms cross-checked by a second team member. All retrieved records were imported into EndNote V.21 and deduplicated. Two reviewers independently screened titles and abstracts, followed by full-text assessment, against prespecified inclusion and exclusion criteria. Discrepancies were resolved through discussion; if consensus could not be reached, a third reviewer adjudicated. Data extraction Two reviewers independently extracted data using a prespecified, standardised form. We recorded study characteristics (author, year, country/setting, design, sample size, age or grade eligibility and group-specific mean age), intervention and comparator details, and outcome measurement information (anxiety instrument and assessment time points). For quantitative synthesis, we preferentially extracted postintervention means and SD when available. When continuous outcomes were reported as medians and ranges or IQRs, means and SD were estimated using the validated Wan et al25 algorithm. Extracted entries were cross-checked independently to minimise transcription errors. Quality assessment Two reviewers independently assessed risk of bias for all included trials using the Cochrane Risk of Bias 2 tool (RoB 2)26 across five domains: randomisation process, deviations from intended interventions, missing outcome data, outcome measurement and selection of the reported result. Overall risk of bias was judged as low risk, some concerns or high risk according to RoB 2 guidance. Disagreements were resolved by discussion with third-reviewer adjudication when required. Certainty of evidence for the primary outcome was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach and summarised in a Summary of findings table.27 Statistical analysis Statistical analyses were conducted using Review Manager V.5.4 and Stata V.18.0. Given differences across studies in the structure and scoring ranges of anxiety assessment instruments, effects were pooled using standardised mean differences (SMDs), with Hedges’ g applied to correct for small-sample bias. Heterogeneity was quantified using the I² statistic; where heterogeneity was at least moderate (I²≥50%), pooled estimates were obtained using random-effects models to accommodate anticipated between-study variability. When I² exceeded 50%, subgroup analyses by intervention dose were performed to explore potential sources of heterogeneity. Sensitivity analyses were performed to assess the robustness of the pooled estimates. Results Search results The search identified 1258 records: PubMed (n=117), Web of Science Core Collection (n=42), Cochrane Library (n=832), ERIC (n=244), DOAJ (n=10) and CNKI (n=13). After removal of 61 duplicates, 1197 records underwent title and abstract screening, and 1158 were excluded. Thirty-nine reports were sought for retrieval and four could not be obtained. 35 full-text reports were assessed for eligibility and 25 were excluded for predefined reasons. 10 randomised controlled trials met the inclusion criteria and were included in the review and quantitative synthesis. The selection process is shown in the PRISMA flow diagram (figure 1). Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA; 2020) diagram of study screening and selection. CNKI, China National Knowledge Infrastructure. Basic characteristics of the included studies We included 10 randomised controlled trials (RCTs) in this systematic review,28–37 conducted in Iran,28 37 Canada,29 35 Iraq,30 31 India32 34 and Turkey33 36 (two trials per country). The evidence base was uneven in both setting and comparator: eight trials28 30–34 36 37 were delivered in medical/procedural contexts and used non-art activities or usual care as controls, whereas only two29 35 were conducted in school/classroom settings and used active art-based activities as comparators. To maintain comparability of control conditions, the prespecified primary meta-analysis was restricted to medical/procedural trials with non-art or usual-care controls (k=8); the two school-based trials are presented separately as supplementary evidence. Across the 10 included trials, 816 children were enrolled (409 allocated to intervention groups and 407 to control groups). The effect estimates included in the quantitative synthesis were derived from age-eligible samples consistent with the protocol-specified population of children aged 5–12 years; mean ages ranged from 7.16 to 11.5 years. Most trials used individually randomised, parallel-group designs; one was a cluster-randomised pilot trial.29 Only a minority of studies reported blinding (single-33 or double-blind30). Although trial designs were broadly similar, important differences remained in the unit of randomisation, blinding practices and the reporting of implementation details (table 1). Table 1Characteristics and age eligibility of randomised controlled trials evaluating drawing and visual art interventions for anxiety in childrenStudy (year)Country/SettingDesignParticipants(EG/CG)Age eligibility and mean ageIntervention (EG)Control (CG)Intervention exposureOutcome included in meta-analysisRezazadeh et al28Iran; burn unit (hospitalised burned children, 6–12 years)RCT (three-group randomised clinical trial)20/206–12 years; 9.9±1.4/9.6±1.2Painting-based art therapy (15 min drawing+30 min discussion around the drawing; guardian present)Hospital services/routine hospital care10 sessions; 45 min/session; daily (total 7.5 hours)Postintervention, total anxiety scoreMalboeuf-Hurtubise et al29Canada; elementary school (online)Cluster RCT (pilot)14/8Grades 4–5; 11.2/11.5Emotion-based directed drawing (online group-based) with group discussion (incl. COVID-19 themed activities)Mandala drawing (non-directed; online group-based; active art comparator)Five sessions; 45 min/session; weekly (total 3.75 hours)Postintervention, child-reported anxietySuleman et al30Iraq (Kurdistan Region); hospital procedure (PIVC)Parallel double-blind RCT48/526–12 years; 7.79±1.78/7.98±1.96TICK-B tracing and colouring during peripheral venous cannulation (PIVC)Routine hospital care (no intervention/distraction)Single session during procedure (~2–3 min; converted to ~0.033–0.05 hours)During-procedure anxiety scoreSuleman et al31Iraq (Kurdistan Region); hospital procedure (venipuncture)Parallel RCT73/716–12 years; 7.75±1.84/7.96±2.00TICK-B tracing and colouring during venipunctureRoutine venipuncture care (no intervention/distraction)Single session during procedure (~2–3 min; converted to ~0.033–0.05 hours)During-procedure anxiety scorePatil et al32India; dental clinicRCT (lottery)60/606–12 years; 7.16±1.36/7.16±1.36Art therapy (colouring pages, connecting numbers)Routine dental treatmentThree consecutive appointments; duration NRPostintervention anxiety scoreKöksal et al33Turkey; PICUSingle-blinded parallel-group RCT35/357–12 years; 10.3±1.68/9.8±2.01Therapeutic communication: mutual storytelling (day 1)+spontaneous drawing (day 2)Standard PICU nursing careTwo sessions over two consecutive days; ~40 min total (~0.67 hours)Postintervention anxiety scoreHambire et al34India; paediatric dental clinicRandomised controlled clinical trial75/756–12 years; 8.5±1.8/8.3±1.9Structured art therapy (clay modelling, free drawing, colouring pre-drawn templates, collage)Routine tell-show-do behaviour managementThree sessions (three consecutive dental visits); 15 min/session (total 0.75 hours)Postintervention dental anxiety scoreCarsley et al35Canada; classroom (private elementary school, grades 4–6, Montreal)Classroom-based school study RCT26/26Grades 4–6; 10.84±0.80/11.00±0.84Structured mandala colouringFree drawing/colouring (blank paper; active art comparator)Single session; 15 min (total 0.25 hours)Postsession test anxiety score (pre–post immediately, before spelling test)Akay et al36Turkey; preoperativeRCT38/406–12 years; 7.55±1.57/8.55±1.64Structured mandala colouring (pre-designed templates+coloured pencils)Routine preoperative careSingle session; ~30 min (0.5 hours)T2 (30 min after surgery), researcher reportZamanifard et al37Iran; online (WhatsApp)Randomised controlled clinical trial20/208–12 years; 9.65±1.26/10.20±1.50Virtual directed painting therapy+usual care (WhatsApp group)Usual careSix sessions over 6 weeks; 120 min/session (total 12 hours)Postintervention total anxiety scorePICU, paediatric intensive care unit; RCT, randomised controlled trial. Characteristics and age eligibility of randomised controlled trials evaluating drawing and visual art interventions for anxiety in children PICU, paediatric intensive care unit; RCT, randomised controlled trial. Medical/procedural studies spanned burn units,28 paediatric intensive care units,33 perioperative care,36 invasive procedures such as venipuncture or blood sampling30 31 and dental treatment.32 34 Accordingly, the primary pooled estimate should be interpreted primarily as evidence from acute medical, procedural or clinical-care contexts under non-art or usual-care comparisons, rather than as an overall estimate for drawing- and visual art-based interventions across all paediatric settings. All interventions centred on drawing and visual art activities, including guided therapeutic drawing with discussion,28 37 structured art tasks (eg, combined drawing and collage)34 and visually focused activities such as colouring or tracing (including mandala colouring, dot-to-dot drawing and guided tracing).30–32 36 Intervention dose varied substantially among trials contributing to the primary meta-analysis: three used single-session formats in time-constrained procedural or preoperative contexts, lasting approximately 2–3 min to 30 min30 31 36; five delivered multi-session or course-based programmes (2–10 sessions or a 6-week course),28 32–34 37 with session durations of 15–120 min and an estimated total exposure of ~0.67–12 hours.28 33 34 37 Several trials incompletely reported session duration and/or total exposure, indicating limited transparency in dose reporting across the included literature.32 Comparators ranged from usual care, routine procedures or other non-art controls to active art-based comparators, suggesting systematic differences in comparator intensity across studies. Anxiety outcomes were assessed using heterogeneous instruments and timepoints, spanning immediate state anxiety around procedures or preoperative periods30 31 35 36 and symptom-oriented severity scales.28 32–34 37 Where multiple timepoints or informants were reported, data extraction followed the prespecified hierarchy and included a single timepoint and a single source per study. For example, in Akay et al,36 child-reported anxiety at 30 min postoperatively (T2) was prioritised for pooling; where outcomes were reported as medians (IQR) (eg, Zamanifard et al37), values were converted to means and SD using the prespecified approach. Differences in scale structure, scoring and assessment windows reduced direct comparability across studies and likely contributed to indirectness and heterogeneity. Risk of bias assessment As shown in figure 2a,b, risk of bias was assessed for the 10 included randomised clinical trials using the RoB 2 tool across five domains: randomisation process (D1), deviations from intended interventions (D2), missing outcome data (D3), measurement of the outcome (D4) and selection of the reported result (D5). Two reviewers assessed risk independently, with disagreements resolved by consensus. Risk-of-bias assessment for the included randomised controlled trials. (a) Summary of judgements across risk-of-bias domains. (b) Judgements for each study across risk-of-bias domains. Overall, one trial was judged at low risk of bias,34 six were judged as having some concerns30 31 33 35–37 and three were judged at high risk.28 29 32 Concerns were concentrated primarily in the randomisation process and outcome measurement domains, indicating that confidence in internal validity was limited mainly by incomplete reporting of allocation procedures and by potential vulnerability of outcome assessment to lack of blinding or insufficient procedural control. For the randomisation process (D1), five trials were judged at low risk of bias,30 31 34 36 37 three as having some concerns32 33 35 and two at high risk.28 29 Judgments in this domain were driven mainly by insufficient detail on sequence generation and allocation concealment. For D2, four trials were at low risk,30 31 33 34 five had some concerns28 29 32 36 37 and one was at high risk35 largely because reporting was insufficient to determine whether adherence, cointerventions or contamination were adequately monitored. For D3, eight trials were at low risk28 29 31 32 34–37 and two had some concerns,30 33 with no trial judged at high risk, suggesting that missing outcome data were unlikely to have materially influenced the pooled estimates. Risk of bias was most pronounced in D4. Two trials were judged at low risk,34 36 six had some concerns29–31 33 35 37 and two were at high risk.28 32 Most studies used subjective anxiety measures, and masking of participants and intervention personnel was inherently difficult in drawing- and visual art-based interventions, increasing the possibility of bias in outcome assessment when blinding or measurement procedures were insufficiently reported. For D5, five trials were judged at low risk31 34–37 and five as having some concerns,28–30 32 33 mainly because prespecified analysis plans or trial registrations were not clearly reported. Given that overall high risk of bias was observed in three trials,28 29 32 sensitivity analyses excluding these studies were conducted to examine the robustness of the primary pooled estimate under stricter methodological criteria. Meta-analysis A total of 10 randomised controlled trials, comprising 816 children, were included. Random-effects models were used to pool the effects of drawing- and visual art-based interventions on anxiety outcomes, with results presented as forest plots. Primary meta-analysis In the primary analysis restricted to studies using non-art or usual-care control conditions (k=8),28 30–34 36 37 drawing- and visual art-based interventions were associated with lower postintervention anxiety levels in children (SMD=−2.70, 95% CI −3.73 to −1.68; p<0.00001; figure 3). The result was highly heterogeneous (τ²=2.09; Q=170.55, df=7, p<0.00001; I²=96%), indicating substantial variation in effect estimates across studies. Primary meta-analysis of drawing- and visual art-based interventions on anxiety outcomes in children. Studies with active art controls Two studies conducted in school-based settings compared the intervention against active art comparator conditions rather than non-art or usual-care controls.29 35 Because these trials addressed a different comparative question, they were not included in the primary analysis. When synthesised separately, the pooled estimate favoured the active art comparator group (SMD=0.52, 95% CI 0.05 to 0.99; p=0.03), with no substantial heterogeneity (I²=0%). This finding is more appropriately interpreted as a comparison between alternative art-based formats than as a test of visual art intervention versus non-art control (online supplemental figure 1). Subgroup analyses To explore potential sources of heterogeneity, prespecified exploratory subgroup analyses were conducted according to intervention dose, classifying studies as single-session or multi-session/course-based interventions. Among single-session studies (k=3)30 31 36 the pooled estimate was larger in magnitude (SMD=−3.41, 95% CI −3.78 to −3.04) and between-study heterogeneity was low (I²=12%). Multi-session or course-based studies (k=5)28 32–34 37 also favoured the intervention (SMD=−2.32, 95% CI −3.78 to −0.86), although heterogeneity remained substantial (I²=97%) (figure 4). However, formal tests for subgroup differences were not statistically significant (χ²=2.01, df=1, p=0.16). Accordingly, the observed difference in point estimates should not be interpreted as evidence of a dose–response relationship. Random-effects meta-analysis of drawing/visual art interventions vs non-art/usual-care controls, stratified by intervention dose. Sensitivity analyses Leave-one-out sensitivity analyses showed that the primary result was not materially affected by the removal of any single study (table 2; online supplemental figure 2). Across all iterations, the pooled estimate remained in favour of drawing- and visual art-based interventions compared with non-art or usual-care comparators. When the two studies judged to be at overall high risk of bias in the primary comparison were excluded (Rezazadeh et al28 and Patil et al32), the pooled effect remained statistically significant and in the same direction (SMD=−2.31, 95% CI −3.33 to −1.29; I²=95%). Substantial heterogeneity persisted, limiting inference about a single common effect magnitude. Table 2Leave-one-out sensitivity analysis for the primary meta-analysisOutcomePooled SMD (95% CI)Primary analysisRange of leave-one-out estimatesInterpretationAnxiety(medical/procedural)−2.70 (−3.73 to −1.68) −2.99 (−4.17 to −1.80)−2.26 (−3.13 to −1.38) Direction of effect remained unchanged across all iterationsSMD, standardised mean difference. Leave-one-out sensitivity analysis for the primary meta-analysis −2.99 (−4.17 to −1.80) −2.99 (−4.17 to −1.80) −2.26 (−3.13 to −1.38) −2.26 (−3.13 to −1.38) SMD, standardised mean difference. Reporting bias To assess potential small-study effects, we examined contour-enhanced funnel plots and conducted Egger’s regression test (figure 5).38 Egger’s regression test did not indicate statistical evidence of small-study effects or funnel plot asymmetry (z=−1.43; p=0.154). However, because the primary analysis included only eight studies and between-study heterogeneity was substantial (I²=96%), the ability of funnel plot-based methods to detect or distinguish publication bias from other sources of asymmetry was limited. Accordingly, the absence of statistical evidence of asymmetry should be interpreted cautiously and does not rule out publication bias or other small-study effects. Contour-enhanced funnel plot for the primary meta-analysis. Certainty of the evidence Using the GRADE framework, certainty of the evidence for the primary anxiety outcome was judged to be low (table 3). The evidence was downgraded one level for risk of bias and one level for inconsistency, because only one of the eight studies in the primary meta-analysis was rated as overall low risk of bias and between-study heterogeneity was substantial (I²=96%). Indirectness was not judged serious for the prespecified primary comparison because this analysis was limited to trials with non-art or usual-care comparators and did not combine school-based active-comparator trials with clinical comparisons. This judgement applies to the primary comparison and should not be taken to imply generalisability across school-based or other non-acute paediatric settings. Imprecision was not downgraded for the direction of effect because the CI did not cross the line of no effect; however, the small evidence base and wide interval limited confidence in the precise magnitude of the effect. Publication bias could not be excluded, as formal assessment of small-study effects was limited by the small number of studies. Detailed rationale for each GRADE domain judgement is provided in online supplemental materials. Table 3Summary of findings for the primary outcomeOutcomeIllustrative effectsRelative effect (95% CI)Participants (studies)Certainty of the evidence(GRADE)CommentsAnxiety (short-term)Lower anxiety in the intervention groupSMD −2.70(−3.73 to −1.68)742 (eight RCTs)⊕⊕◯◯*†LOWDowngraded for risk of bias and inconsistency; heterogeneity was substantial (I²=96%). Outcome timing and measurement instruments varied across studiesData are presented as standardised mean difference (SMD) with 95% CIs.*Downgraded one level for risk of bias.†Downgraded one level due to considerable statistical heterogeneity, which may reflect clinical diversity in the types of visual art activities and variation in session duration across studies.GRADE, Grading of Recommendations Assessment, Development and Evaluation; RCT, randomised controlled trial; SMD, standardised mean difference. Summary of findings for the primary outcome Data are presented as standardised mean difference (SMD) with 95% CIs. Downgraded one level for risk of bias. Downgraded one level due to considerable statistical heterogeneity, which may reflect clinical diversity in the types of visual art activities and variation in session duration across studies. GRADE, Grading of Recommendations Assessment, Development and Evaluation; RCT, randomised controlled trial; SMD, standardised mean difference. Discussion Principal findings This systematic review and meta-analysis suggests that drawing- and visual art-based interventions may reduce short-term anxiety in children, primarily in acute medical, procedural or selected clinical-care contexts and when compared with usual care or non-art controls. In the primary analysis restricted to trials using non-art or usual-care comparators, the pooled effect was large (SMD=−2.70), but between-study heterogeneity was very high (I²=96%) and the certainty of evidence was low. Accordingly, the pooled estimate should be interpreted as evidence of a consistent direction of short-term benefit within the primary comparison, rather than as a precise or readily generalisable estimate of effect magnitude across paediatric settings. The magnitude of the pooled effect may partly reflect the structure of the available evidence. Most contributing trials were conducted around acute medical or procedural stressors, assessed anxiety over short time windows and used inactive or usual-care comparators. Under these conditions, between-group effects may be amplified when outcomes are measured close to the stressor and comparator conditions provide little attentional or regulatory engagement. Sensitivity analyses indicated that the direction of effect was not dependent on any single study or on the inclusion of studies at high risk of bias; however, the magnitude of the pooled estimate remained uncertain because heterogeneity persisted. Comparator structure further shapes the interpretation of the findings. Trials using non-art or usual-care controls estimate the incremental value of visual art activities relative to routine or inactive conditions, whereas active art comparator trials address a different question: whether one art-based format differs from another. The two school-based active-comparator trials should therefore not be interpreted as evidence against visual art interventions, but as evidence that between-group contrasts become narrower and less directly comparable when both study arms include art-based engagement. Potential pathways and sources of heterogeneity The substantial heterogeneity observed in the primary synthesis is more plausibly explained by structural differences across studies than by any single methodological feature. In particular, setting, comparator structure, intervention delivery and outcome timing appeared to cluster in ways that shaped the observed effect sizes. In medical and procedural contexts, interventions were typically delivered close to an acute stressor, and anxiety was often assessed during or immediately after the target event. In such settings, observed effects are likely to depend not only on intervention content but also on the timing of measurement and the intensity of the comparator condition. Where usual-care or non-art controls offered limited structured distraction or emotional engagement, short-term between-group contrasts may have been amplified. The resulting effects are therefore better interpreted as situational reductions in acute anxiety than as evidence of durable benefit extending beyond the immediate context.39 40 By contrast, school-based or repeated-session interventions may operate through different pathways. In these contexts, observed effects may reflect structured attention, guided engagement, emotional expression, group discussion or repeated exposure, depending on the intervention design, rather than immediate modulation of anxiety under acute demand. This may help explain why multi-session studies remained highly heterogeneous despite dose stratification, suggesting that repeated delivery alone does not define a single, coherent intervention effect.41 42 Consistent with this interpretation, the subgroup findings are better understood as reflecting the joint influence of intervention dose, procedural context, comparator structure and outcome timing than as evidence of a simple dose-based gradient.43 These considerations also constrain the clinical implications of the review. The current evidence does not support broad implementation of drawing- and visual art-based interventions as stand-alone treatments for paediatric anxiety across settings. At present, these activities may be considered low-burden adjuncts in selected paediatric procedural, inpatient or clinical-care contexts, particularly when incorporated into existing nursing, child-life or psychosocial support practices. They should not replace established approaches to procedural preparation, pain control, psychological support or clinically indicated anxiety care. Wider implementation requires adequately powered trials with clearly specified comparators, prespecified outcome windows, standardised anxiety measures, and follow-up beyond the immediate procedural period. Strengths and limitations This review provides a quantitative synthesis of randomised controlled trial evidence on drawing- and visual art-based interventions for paediatric anxiety outcomes. By prespecifying comparator categories and restricting the primary meta-analysis to trials with non-art or usual-care comparators, we aimed to improve interpretability and define the scope of inference more clearly. Exploratory stratified analyses also helped identify study features that may contribute to heterogeneity, including setting, comparator intensity, intervention structure and outcome timing. This review also has several limitations. First, although ten trials were included, only eight contributed to the primary meta-analysis and only two used active art-based comparators. This limited the power of subgroup analyses and constrained assessment of small-study effects. Second, heterogeneity in the primary synthesis was very high. Although sensitivity analyses showed that the direction of the primary effect was not dependent on any single study or on the inclusion of studies at high risk of bias, the included trials differed in setting, comparator intensity, intervention dose, outcome instruments and timing of assessment. The pooled estimate should therefore not be interpreted as a single stable effect applicable across paediatric contexts. Third, the evidence was unevenly distributed across settings. Medical and procedural trials generally evaluated visual art activities delivered around bounded acute stressors, whereas school-based trials used active art-based comparators and different outcome windows. Although the primary meta-analysis was restricted to non-art or usual-care comparisons and school-based active-comparator trials were analysed separately, the findings should not be generalised beyond acute medical, procedural, or clinical-care contexts without further evidence. Fourth, most outcomes were measured using subjective rating scales. Creative and behavioural interventions are difficult to blind, and reliance on subjective anxiety measures may increase susceptibility to performance and measurement bias.44 45 Incomplete reporting of blinding and outcome-assessment procedures further reduced comparability across studies. Together with modest sample sizes and short assessment windows, these factors lowered confidence in the precise magnitude of the pooled effect. Future trials should use adequately powered designs, clearly specified comparators, prespecified outcome windows, standardised anxiety measures or sufficient data for conversion and transparent reporting of blinding and outcome-assessment procedures. They should also assess whether effects persist beyond the immediate procedural or intervention period and identify which children are most likely to benefit. Conclusions This review synthesised randomised controlled trial evidence on drawing- and visual-art-based interventions for anxiety in children aged 5–12 years. Across studies, effects were generally in the direction of lower postintervention anxiety, but the evidence does not support a single stable effect size transferable across contexts. Instead, the findings are better interpreted as suggesting short-term anxiety reduction mainly in acute medical/procedural settings than as demonstrating a generalised therapeutic effect for childhood anxiety. The evidence base is concentrated in medical and procedural settings with short assessment windows and non-art or usual-care comparators, whereas evidence from non-clinical contexts remains limited. Future research should prioritise clearer comparator specification, more explicit alignment of outcome windows with target stressors, reproducible reporting of intervention delivery and dose, and outcome measures that better support cross-study synthesis. Supplementary material Footnotes Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors. Provenance and peer review: Not commissioned; externally peer reviewed. Ethics approval: This review was based on previously published studies and did not require ethics approval. References Associated Data Supplementary Materials