Anxiety disorders: matching evidence to clinical and population decisions

Abstract

Background. Anxiety disorders are treatable, yet substantial unmet need persists alongside reports of rising burden. Translation is hindered when evidence about prevalence, susceptibility, symptom change, and disorder prevention is used interchangeably.

Objective. To examine which clinical and population decisions are supported by selected evidence on anxiety disorders, and identify the additional evidence needed for consequential claims.

Approach. This critical narrative review uses purposive selection and targeted source verification completed on October 10, 2026. It integrates repeated diagnostic surveys, genetic and quasi-experimental studies, intervention trials, selected evidence syntheses, and clinical guidance. It is not a systematic review; no new pooled analysis or formal certainty grading was undertaken.

Findings. Repeated surveys support increasing anxiety prevalence in some populations without identifying a single explanation. Genetic discovery has advanced biological investigation but has not established routine treatment assignment. Environmental and maintaining processes require different causal tests. Prevention evidence includes reduced onset in selected high-risk families, but symptom benefits do not consistently translate into fewer diagnostic cases. A large 2026 school-based trial illustrates this divergence. Cognitive behavioural therapy and several medications are effective in defined populations; comparative choice depends on diagnosis, developmental stage, comparator, harms, and durability. Treatment efficacy does not establish effective or equitable coverage.

Implications. Evidence should be matched to the population, outcome, comparison, time horizon, and setting of the proposed decision. Improving access to established care is justified now. Prevention and service innovations should be evaluated using independently assessed disorder outcomes, sustained functioning, and reach among all eligible people.

Keywords: anxiety disorders; causal inference; epidemiology; prevention; cognitive behavioural therapy; implementation

Introduction

The central problem in anxiety research is the distance between what a study establishes and the decision it is used to justify. An increase in symptom reporting can be interpreted as an increase in disorder incidence; a genetic association can be treated as a treatment target; and a reduction in symptoms after an intervention can be described as prevention. Each inference may be plausible, but each requires additional evidence. Keeping these distinctions explicit is essential when translating a large and heterogeneous literature into clinical and population responses.

The treatment gap illustrates why this matters. Across the World Mental Health Surveys in 21 countries, 27.6% of respondents with a 12-month anxiety disorder received any treatment and 9.8% received care meeting the study's definition of possibly adequate treatment.[1] These historical survey estimates are not current global coverage rates, and service contacts do not guarantee effective care. Nevertheless, they establish that demonstrating efficacy and delivering sufficient treatment are separate achievements (Figure 1).

Figure 1. Historical treatment coverage among people with a 12-month anxiety disorder in 21-country World Mental Health Surveys.[1] The bars share the same denominator and are nested: possibly adequate treatment is a subset of any treatment. “Possibly adequate” is the survey definition, not a measured recovery outcome. Estimates are not current global coverage rates.

This review develops an interpretive framework for linking burden, causal evidence, and intervention. It focuses on generalised anxiety disorder (GAD), panic disorder, agoraphobia, social anxiety disorder, and specific phobia. Childhood separation anxiety is discussed where included in pivotal trials. Obsessive–compulsive disorder (OCD) and post-traumatic stress disorder (PTSD) are distinguished from anxiety disorders, although older syntheses sometimes combine these conditions. The organising question is how evidence changes a decision, rather than how many findings can be catalogued. Five features govern that judgement: population, outcome, comparator, time horizon, and setting. Table 1 applies this approach to selected consequential findings; Table 2 identifies the additional evidence needed when moving between domains. This is an interpretive framework, not a validated prediction model or formal grading system.

Review approach

This is a critical narrative review. Targeted web searches of scholarly literature and retrieval of publisher, PubMed, and PubMed Central records were completed on October 10, 2026. Searches combined anxiety-related terms with prevalence trends, genome-wide association, childhood adversity, prevention, cognitive behavioural therapy, medication, relapse, and treatment delivery; exact-title searches checked pivotal publications. Selection favoured studies that addressed a consequential inferential contrast: anxiety-specific versus combined outcomes, association versus intervention effects, symptom change versus disorder onset, and efficacy versus service delivery. Recent studies were added when they challenged or refined an interpretation; selection was not based on statistical significance. Recent syntheses were considered alongside older trials whose designs addressed a specific question.

Selection was purposive and limited to accessible English-language material. This process did not include a registered protocol, exhaustive bibliographic database search, independent duplicate screening, or formal study-level risk-of-bias assessment. Full texts were consulted where accessible; some records could be verified only through abstracts or bibliographic metadata. No pooled estimate was calculated. These limitations preclude claims of systematic completeness or a formal certainty assessment.

Interpretation considered the population, diagnostic definition, comparator, outcome, and follow-up of each study. Experimental evidence was prioritised for intervention effects, repeated representative surveys for temporal trends, and genetically informative or quasi-experimental studies for causal questions. Published meta-analytic results were reported with their original scope. A statistically nonsignificant difference was not interpreted as equivalence, and symptom change was not treated as interchangeable with diagnostic remission, incidence reduction, or functional recovery.

Population trends require comparable denominators

What repeated surveys establish

The NEMESIS programme provides evidence of increasing prevalence in Dutch adults using repeated population samples and structured diagnostic assessment.[2,3] In the harmonised samples aged 18–64 years, 12-month prevalence of GAD, panic disorder, agoraphobia, or social anxiety disorder increased from 6.3% in 2007–2009 to 9.8% in 2019–2022; specific phobia was not included in this composite. The often-cited change from 10.2% to 16.7% refers to the combined anxiety-or-mood outcome, not anxiety alone.[3] These figures cannot be substituted for estimates based on a broader set of anxiety diagnoses.

The later analysis found no consistent explanation across 14 measured risk factors, and clinical profiles did not indicate a shift towards uniformly milder cases.[3] This weakens a simple account based exclusively on less severe reporting. It does not eliminate changing disclosure, unmeasured exposures, or measurement effects. Response rates also differed between waves, making residual selection bias a relevant alternative explanation. Failure of a measured risk-factor set to explain a trend also does not identify the residual explanation.

The earlier NEMESIS report compared interviews before and during the pandemic and found no significant difference in 12-month disorder prevalence within that comparison.[2] Thus, the long-term Dutch increase should not simply be attributed to pandemic timing. Conversely, a null comparison within one national survey cannot establish that the pandemic had no mental health effects elsewhere, at other times, or in particular subgroups.

Global models and clinical records answer different questions

The COVID-19 Mental Disorders Collaborators estimated a 25.6% increase in global anxiety disorder prevalence in 2020, with a 95% uncertainty interval of 23.2%–28.0%, corresponding to 76.2 million additional prevalent cases.[4] The estimates combine observed information with modelling assumptions and extrapolation. They are not a count of newly diagnosed individuals, and the additional prevalent cases should not be described as directly observed incident disorders. Sparse data in some regions and the conversion of symptom changes into disorder estimates limit geographical precision.

There is no necessary contradiction between a modelled global increase and a national survey without a significant pandemic contrast. Their populations, periods, measurements, and counterfactuals differ. Administrative diagnoses introduce further dependencies on access, clinician coding, service capacity, and willingness to seek help. Surveillance should therefore present symptom distributions, structured diagnoses, impairment, and service use separately before examining their convergence.

Prevalence also depends on persistence and recurrence. A population can have more prevalent disorder because onset increases, recovery slows, relapse increases, or ascertainment changes. Treatment demand alone cannot discriminate among these pathways. The resulting research priority is repeated, harmonised measurement with longitudinal follow-up, rather than a search for one exposure that correlates with a rising aggregate curve.

Etiology and maintenance operate at different levels

Genetic susceptibility does not provide a clinical assignment rule

Twin and molecular studies support a substantial genetic contribution to anxiety, but heritability is a population-specific variance estimate rather than the fraction of an individual's illness caused by genes.[5] It also does not indicate how modifiable a condition is. Genetic and environmental contributions can coexist without proving a particular statistical gene–environment interaction. Such an interaction requires a measured exposure, a specified scale, and replicated evidence beyond the separate associations.

Friligkou and colleagues analysed more than 1.2 million participants, including 97,383 cases, across five ancestry groups and identified 51 anxiety-associated loci.[6] Strom and colleagues subsequently reported 58 loci in a discovery sample of 122,341 European-ancestry cases and 729,881 controls.[7] Differences in phenotype, ascertainment, ancestry, and sample overlap mean that locus counts should not be read as directly comparable measures of progress or independent replications of a single experiment.

The latter study highlighted GABAergic signalling and reported substantial genetic overlap with internalising phenotypes.[7] Pathway enrichment supports functional investigation; it neither establishes an individual neurotransmitter deficit nor demonstrates preferential response to a drug acting on that pathway. Clinical use of a genetic score would require external validation, calibration across populations, incremental value beyond clinical information, and evidence that score-guided decisions improve outcomes. The cited discoveries do not supply that chain of validation.[5,6,7]

Environmental evidence benefits from triangulation

Associations between childhood adversity and later psychopathology can reflect exposure effects, familial liabilities, and correlated disadvantage. A meta-analysis of 34 quasi-experimental studies involving 54,646 participants found that the association between maltreatment and mental health problems attenuated from Cohen's d=0.56 to d=0.31 after adjustment using quasi-experimental designs.[8] The adjusted estimate had a 95% confidence interval of 0.24–0.37. This pattern is consistent with a causal contribution and substantial confounding; it is not an anxiety-specific effect estimate.

Different designs address different biases. Sibling comparisons reduce shared familial confounding but remain vulnerable to nonshared confounding and measurement error. Natural experiments can strengthen causal inference while estimating effects in a restricted context. Agreement across designs with different assumptions is more informative than treating any one method as definitive. Intervention on a harmful exposure can be justified by its broader consequences even when its precise contribution to anxiety incidence is unresolved.

Avoidance is a treatment target without being a universal origin story

Avoidance can maintain threat expectations by preventing corrective learning. Exposure, an important component of many anxiety-focused CBT protocols, creates opportunities to test feared predictions and learn that distress can be tolerated or an anticipated outcome does not occur. It is not the sole component of all effective psychological treatments, and the appropriateness of confronting a situation depends on distinguishing feared from actual danger. Inhibitory-learning models emphasise how this learning is retrieved across contexts.[9] They provide a treatment rationale; improvement after exposure does not, by itself, prove that an experimentally isolated learning mechanism mediated the benefit or originally caused the disorder.

Family accommodation provides a clinically useful example. In a randomised noninferiority trial of 124 children aged 7–14 years, a parent-based programme focused on accommodation, SPACE, was noninferior to child CBT on the specified anxiety outcomes.[10] The result supports an additional treatment route for a defined population. It does not establish that parents caused the child's disorder, that all supportive behaviour is accommodation, or that abrupt withdrawal of support is appropriate.

This distinction between onset and maintenance prevents misplaced blame and improves research design. An exposure that initiates vulnerability may no longer be modifiable, whereas a maintaining process may remain treatable. Mediation analyses, repeated measurements preceding symptom change, and experimental manipulation of candidate processes are needed to determine how a treatment works. Mechanistic explanations should remain proportional to those designs.

Modifiable environments require specific causal questions

Digital exposure illustrates the limitations of broad labels. In a specification-curve analysis of 355,358 adolescents, digital technology use explained at most 0.4% of variation in well-being.[11] That finding concerns a broad exposure and outcome, not the causal effect of a particular platform on incident anxiety disorder. An umbrella review likewise identified heterogeneous interpretations across reviews of social media and adolescent mental health.[12] Average associations can obscure differences by content, context, vulnerability, and pattern of use.

The appropriate causal question specifies the intervention: reducing night-time use, preventing harassment, altering a recommendation system, or removing access are different policies. Their effects cannot be inferred from the same screen-time coefficient. Reverse causation and time-varying confounding remain concerns even in longitudinal data. Randomised changes and credible natural experiments should measure both benefits and harms, including loss of valued social connection, with diagnostic and functional outcomes where feasible. Claims that a single digital exposure explains a population trend exceed the cited evidence.

Work research has a similar specificity problem. High demands, low control, insecurity, and low support are associated with common mental health problems, but the relevant meta-review pools a broader outcome domain than anxiety disorders alone.[13] Organisational interventions, manager training, and support for workers are addressed in WHO guidance.[14] Their rationale extends beyond an anxiety-specific causal effect. Policies should be evaluated using exposure change, participation, functioning, and clinical outcomes, while accounting for workers leaving employment and other selection processes.

Structural adversity should likewise be treated as a set of concrete exposures rather than an all-purpose explanation. Housing security, financial strain, and access barriers are plausible intervention targets, but the magnitude and distribution of anxiety-specific benefits require evaluation. A population strategy should combine protection from harmful conditions with accessible care; it should not require individuals to compensate for every environmental risk through coping skills.

Prevention must distinguish symptom reduction from avoided onset

Universal prevention is offered irrespective of individual risk; selective prevention targets elevated risk; indicated prevention targets subthreshold symptoms. These categories should remain distinct from early treatment of an established disorder. A prevention study needs a baseline definition that establishes who is free of disorder and an outcome that determines whether onset occurred during follow-up.

A meta-analysis of 29 randomised trials involving 10,430 participants combined studies measuring symptoms, incident disorder, or both and reported a pooled standardised mean difference of −0.31 (95% CI −0.40 to −0.21), with substantial heterogeneity (I²=61.1%).[15] The common effect metric facilitates synthesis but is not an observed absolute reduction in disorder onset across all participants. Converting that pooled estimate into cases prevented requires assumptions about baseline risk and the relationship between continuous and diagnostic outcomes. Programme-specific diagnostic effects are therefore more useful for estimating absolute public health benefit.

Direct evidence for onset prevention does exist. In a trial of 136 families with an anxious parent and a child initially without an anxiety disorder, a family intervention was associated with cumulative onset of approximately 5% versus 31% in the control condition during one year.[16] This is important evidence for selective prevention. The size of the benefit in a selected family sample should not be projected onto all schoolchildren or assumed to persist indefinitely. Replication, longer follow-up, recruitment feasibility, and the resources required per family determine its population relevance.

For FRIENDS, the 2024 Campbell review found favourable average child-reported anxiety symptom outcomes after intervention and at 12 months, with low to very low certainty across the reviewed outcomes and inconclusive evidence beyond 12 months.[17] Parent reports did not show the same clear benefit. This difference should prompt attention to what each rater measures, rather than an assumption that one informant is inherently correct.

The 2026 MYCATS cluster trial extends this question to 865 children aged 4–7 years across 95 English schools.[18] At 12 months, anxiety diagnoses occurred in 6.8% versus 11.5% of assessed children, but the adjusted comparison was inconclusive (odds ratio 0.67, 95% CI 0.37–1.21). Parent-reported secondary outcomes favoured intervention. Diagnostic data were available for 622 children; baseline diagnostic absence was not established. This was therefore a test of later diagnostic status, not cleanly of incident disorder. The result supports symptom benefits without establishing a diagnostic prevention effect.

Together, these trials show why an intervention can be useful without satisfying every prevention claim. A selected family trial with baseline diagnostic exclusion and a school-screening pathway with a later diagnostic endpoint answer different questions. Neither an encouraging symptom result nor an inconclusive diagnostic result should erase the other. The next generation of trials should prespecify incident disorder assessed independently, symptom trajectories, functioning, adverse effects, and costs. At least 24 months of follow-up would help distinguish delayed onset from sustained prevention, although feasibility and developmental timing must guide the interval. Universal and selective strategies should be compared on reach and absolute benefit as well as average efficacy. Targeting can concentrate resources but may miss children whose risk is unrecognised or whose families cannot participate.

Treatment effects depend on the clinical question

Psychological treatment

The 2025 synthesis by Cuijpers and colleagues included 375 trials and 32,968 participants across multiple mental disorders, not 375 anxiety trials.[19] Anxiety-specific results supported CBT, but the total sample and effect-size range cannot be attributed exclusively to anxiety. Waitlist, usual care, pill placebo, and credible psychological comparators estimate different contrasts. A large waitlist effect does not establish superiority over another available treatment.

For adult GAD, the corrected network meta-analysis by Papola and colleagues included 65 trials and 5,048 participants. CBT reduced acute symptoms relative to treatment as usual (standardised mean difference −0.74, 95% CI −1.09 to −0.38); third-wave CBT approaches also showed benefit.[20,21] Network comparisons depend on whether trials are sufficiently comparable in effect modifiers and design. Treatment rankings alone should not determine care when estimates are imprecise or based on indirect comparisons. For an individual patient, functioning, burden of attendance, preference, and prior response are also relevant.

Longer-term evidence is encouraging but less complete. A review of 69 trials involving 4,118 adults found benefits of CBT across anxiety-related conditions during follow-up, with limitations from study quality and sparse evidence at longer intervals.[22] That review also included OCD and PTSD. Maintenance of a mean group difference does not mean every patient remains in remission, and follow-up comparisons may be affected by subsequent treatment and attrition. Recovery and relapse need explicit definitions and continued measurement.

Medication and diagnosis-specific choice

In adult GAD, a network meta-analysis of 89 trials involving 25,441 participants identified several effective pharmacological options, including duloxetine, pregabalin, venlafaxine, and escitalopram.[23] The endpoint was change on an anxiety symptom scale, while acceptability was assessed through all-cause discontinuation. Neither outcome fully describes sexual adverse effects, withdrawal burden, everyday functioning, or long-term preferences. Evidence from acute outpatient GAD trials should not be generalised automatically to other diagnoses or ages.

NICE recommends a choice of high-intensity psychological intervention or medication for GAD with marked impairment or inadequate response to lower-intensity care. When medication is chosen, an SSRI is recommended; alternative options depend on response and tolerability. For panic disorder, benzodiazepines are discouraged, while their use in GAD is restricted to short-term crisis management.[24] For adult social anxiety disorder, NICE prioritises individual disorder-specific CBT.[25] These distinctions argue against one universal prescription for all anxiety disorders. Findings from GAD medication trials do not establish a first-line drug strategy for specific phobia, and adult recommendations should not be transferred to children without age-specific evidence.

Comparative and combined treatment

The frequently cited psychotherapy–medication difference of Hedges' g=0.02 (95% CI −0.07 to 0.10) comes from a synthesis of 67 trials across depressive and anxiety disorders, of which 27 involved anxiety disorders.[26] It is neither an anxiety-only effect nor proof of equivalence for each diagnosis. Failure to reject a difference in a superiority analysis does not establish noninferiority, equal harms, or equal durability.

Combination evidence also depends on the population. In CAMS, 488 children aged 7–17 years with separation anxiety, GAD, or social phobia were randomised to CBT, sertraline, their combination, or placebo. At 12 weeks, clinical improvement was reported in 80.7%, 59.7%, 54.9%, and 23.7%, respectively.[27] These are response rates, not remission rates. The trial establishes a meaningful acute advantage for combined treatment in that population; it does not justify extrapolation to all adult anxiety presentations.

Selection and sequencing should therefore specify diagnosis, developmental stage, comorbidity, severity, patient preference, and available expertise. Nonresponse should trigger reassessment of the diagnosis, treatment exposure, adherence, adverse effects, and barriers to participation. Repeated outcome measurement can guide adjustment without implying that a validated biological matching algorithm already exists.

Sustained benefit and treatment burden

Among antidepressant responders in 28 relapse-prevention trials, discontinuation was associated with greater relapse odds than continuation (odds ratio 3.11, 95% CI 2.48–3.89).[28] The synthesis included anxiety disorders, OCD, and PTSD and followed participants for up to one year. Enrichment for initial responders limits generalisability; withdrawal symptoms can also complicate relapse classification. The finding informs maintenance discussions without defining an optimal stopping time for every patient.

NICE advises discussing activation, adverse effects, withdrawal, and continuation after remission when prescribing for GAD or panic disorder.[24] Psychological treatment also carries burdens, including distress during exposure and time commitments. A useful comparative study should record deterioration and harms in both modalities rather than equating absence of pharmacological exposure with absence of burden. Shared decisions need absolute outcomes and uncertainties that matter to the patient.

Population benefit requires effective delivery

Internet-delivered CBT can extend treatment options, and controlled evidence supports benefit for several anxiety and depressive disorders.[29] Results for tested programmes should not be transferred to untested applications or automated conversational systems. The relevant implementation questions include completion, clinician support, privacy, deterioration monitoring, and access for people with limited connectivity or literacy. A digital pathway should have a clear route to more intensive care when necessary.

The CALM trial offers experimental evidence at the service level. In 1,004 patients across 17 US primary-care clinics, a programme combining patient choice, care management, medication support, and computer-assisted CBT improved anxiety outcomes compared with usual care.[30] The diagnostic scope included PTSD as well as anxiety disorders. Because this was a multicomponent intervention, the result cannot isolate the effect of software or any single staffing component. It shows that delivery organisation can itself be an intervention worth testing.

Innovation and implementation should be assessed together. A 2026 state-of-the-art review describes emerging approaches while noting inconsistent results for some proposed treatment enhancements, including D-cycloserine augmentation of exposure.[31] Mechanistic appeal is insufficient grounds for displacing established care. Comparative trials should test whether an innovation adds sustained, patient-valued benefit and whether its staffing and cost requirements permit equitable use.

For service evaluation, the denominator should include all people eligible for care, not only those completing treatment. A proposed sequence of measures is recognition, offer, initiation, adequate delivery, response, and sustained recovery. This is an evaluative framework rather than a validated prediction model. Reporting each stage by deprivation, language, geography, and other relevant characteristics would distinguish failures of reach from failures of treatment and identify where improvement is needed.

Implications for clinical and public health decisions

Clinical assessment should connect symptoms to impairment, diagnosis, comorbidity, developmental stage, and the person's priorities. Treatment discussions can then use evidence from the closest relevant population, distinguishing acute response from remission and sustained recovery. Choice is not resolved by an average effect size alone: adverse effects, withdrawal concerns, time demands, access, and preference can change which effective option is suitable. Guidance supports diagnosis-specific decisions, with reassessment when an adequate treatment course has not helped.[24,25]

For schools and commissioners, the decision is which programme to offer to which population for which outcome. A programme supported for symptom improvement should be described and commissioned on that basis unless it also demonstrates reduced disorder onset. Screening is part of a care pathway, not an outcome in itself; capacity to assess and support those identified must accompany expansion. Recruitment, uptake, attrition, and missing outcome data should be reported alongside treatment effects so that apparent scalability is not inferred from outcomes among participants alone.

For research funders, the highest-value next study may close a decision gap rather than add another association. Examples include testing whether biomarker-guided selection improves outcomes over good clinical assessment, whether a prevention programme reduces independently assessed onset over developmentally meaningful follow-up, and whether an accessible care pathway improves sustained functioning among all eligible people. Each requires a comparator that represents the decision actually facing patients or services.

Research priorities and limitations

Three linked programmes follow from the evidence. First, surveillance should estimate onset, persistence, recurrence, and functional impairment using harmonised instruments and explicit diagnostic coverage. Linking repeated surveys to service records can reveal discrepancies while preserving the distinction between clinical recognition and underlying disorder. Measurement changes and differential nonresponse require analysis, not merely acknowledgement.

Second, causal studies should define a modifiable exposure and a plausible intervention before estimating its effect. Genetically informative designs, natural experiments, and randomised interventions can be triangulated to assess different sources of bias. Mechanistic treatment studies should establish temporal ordering and test whether changing a candidate process improves outcomes beyond a credible comparison. These requirements are particularly important for claims about digital environments and personalised care.

Third, prevention and service trials should measure outcomes beyond short-term symptom scales. Incident disorder, sustained remission, educational or occupational participation, harms, costs, and treatment coverage should be prespecified. Pragmatic trials can examine whether gains survive routine delivery and whether expanding one pathway displaces care for other patients. Replication across settings is necessary before transferring a selective prevention effect or specialist treatment result to population policy.

This review's purposive selection and reliance on accessible material introduce selection and language bias. No independent screening, formal certainty grading, or de novo quantitative synthesis was conducted. Some important domains, including older adulthood, perinatal care, and low-resource delivery, receive insufficient coverage for comprehensive recommendations. The inferential framework is a proposed synthesis, not an empirically validated classification. Its value should be judged by whether it clarifies research questions and prevents unsupported translation across populations and outcomes.

Conclusion

Anxiety research supports effective treatment and promising selective prevention, while leaving major uncertainties about population trends and mechanisms. Progress depends on preserving the distinction between susceptibility, causal exposure, maintaining process, and intervention effect. A defensible response combines better access to established care with causal research and prevention trials that measure onset and sustained functioning. The appropriate standard for translation is whether evidence supports a specified decision for a specified population, rather than the apparent size or novelty of a finding.

Table 1. Selected findings and their decision limits

Illustrative evidence selected for the narrative argument; this is not a systematic evidence inventory or certainty assessment.

Table 2. Evidence needed when translating between domains

Proposed interpretive framework. The additional evidence depends on the population, outcome, comparison, time horizon, and setting of the proposed decision.

References

1. Alonso J, Liu Z, Evans-Lacko S, et al. Treatment gap for anxiety disorders is global: results of the World Mental Health Surveys in 21 countries. Depression and Anxiety. 2018;35:195–208. https://doi.org/10.1002/da.22711

2. Ten Have M, Tuithof M, van Dorsselaer S, et al. Prevalence and trends of common mental disorders from 2007–2009 to 2019–2022: results from the Netherlands Mental Health Survey and Incidence Studies (NEMESIS), including comparison of prevalence rates before vs. during the COVID-19 pandemic. World Psychiatry. 2023;22:275–285. https://pmc.ncbi.nlm.nih.gov/articles/PMC10168151/

3. Gkitakou A, Ten Have M, Batelaan NM, Luik AI, Penninx BWJH. Anxiety and mood disorders on the rise: exploring clinical profiles and risk factors. British Journal of Psychiatry. 2026;228:513–520. https://doi.org/10.1192/bjp.2025.10454

4. COVID-19 Mental Disorders Collaborators. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic. Lancet. 2021;398:1700–1712. https://pmc.ncbi.nlm.nih.gov/articles/PMC8500697/

5. Ask H, Cheesman R, Jami ES, Levey DF, Purves KL, Weber H. Genetic contributions to anxiety disorders: where we are and where we are heading. Psychological Medicine. 2021;51:2231–2246. https://doi.org/10.1017/S0033291720005486

6. Friligkou E, Løkhammer S, Cabrera-Mendoza B, et al. Gene discovery and biological insights into anxiety disorders from a large-scale multi-ancestry genome-wide association study. Nature Genetics. 2024;56:2036–2045. https://doi.org/10.1038/s41588-024-01908-2

7. Strom NI, Verhulst B, Bacanu SA, et al. Genome-wide association study of major anxiety disorders in 122,341 European-ancestry cases identifies 58 loci and highlights GABAergic signaling. Nature Genetics. 2026;58:275–288. https://doi.org/10.1038/s41588-025-02485-8

8. Baldwin JR, Wang B, Karwatowska L, et al. Childhood maltreatment and mental health problems: a systematic review and meta-analysis of quasi-experimental studies. American Journal of Psychiatry. 2023;180:117–126. https://doi.org/10.1176/appi.ajp.20220174

9. Craske MG, Treanor M, Conway CC, Zbozinek T, Vervliet B. Maximizing exposure therapy: an inhibitory learning approach. Behaviour Research and Therapy. 2014;58:10–23. https://doi.org/10.1016/j.brat.2014.04.006

10. Lebowitz ER, Marin C, Martino A, Shimshoni Y, Silverman WK. Parent-based treatment as efficacious as cognitive behavioral therapy for childhood anxiety: a randomized noninferiority study of Supportive Parenting for Anxious Childhood Emotions. Journal of the American Academy of Child and Adolescent Psychiatry. 2020;59:362–372. https://doi.org/10.1016/j.jaac.2019.02.014

11. Orben A, Przybylski AK. The association between adolescent well-being and digital technology use. Nature Human Behaviour. 2019;3:173–182. https://doi.org/10.1038/s41562-018-0506-1

12. Valkenburg PM, Meier A, Beyens I. Social media use and its impact on adolescent mental health: an umbrella review of the evidence. Current Opinion in Psychology. 2022;44:58–68. https://doi.org/10.1016/j.copsyc.2021.08.017

13. Harvey SB, Modini M, Joyce S, et al. Can work make you mentally ill? A systematic meta-review of work-related risk factors for common mental health problems. Occupational and Environmental Medicine. 2017;74:301–310. https://doi.org/10.1136/oemed-2016-104015

14. World Health Organization. WHO guidelines on mental health at work. Geneva: World Health Organization; 2022. https://www.who.int/publications/i/item/9789240053052

15. Moreno-Peral P, Conejo-Cerón S, Rubio-Valera M, et al. Effectiveness of psychological and/or educational interventions in the prevention of anxiety: a systematic review, meta-analysis, and meta-regression. JAMA Psychiatry. 2017;74:1021–1029. https://pubmed.ncbi.nlm.nih.gov/28877316/

16. Ginsburg GS, Drake KL, Tein JY, Teetsel R, Riddle MA. Preventing onset of anxiety disorders in offspring of anxious parents: a randomized controlled trial of a family-based intervention. American Journal of Psychiatry. 2015;172:1207–1214. https://doi.org/10.1176/appi.ajp.2015.14091178

17. Filges T, Smedslund G, Eriksen T, Birkefoss K, Kildemoes MW. The FRIENDS preventive programme for reducing anxiety symptoms in children and adolescents: a systematic review and meta-analysis. Campbell Systematic Reviews. 2024;20:e1443. https://doi.org/10.1002/cl2.1443

18. Reardon T, Ukoumunne OC, Dodd H, et al. Parent-led CBT delivered via online and telephone support alongside usual school practice versus usual school practice only for young children identified as at risk for anxiety disorders through screening in schools: a cluster randomised controlled trial. Journal of Child Psychology and Psychiatry. 2026;67:1280–1296. https://doi.org/10.1111/jcpp.70119

19. Cuijpers P, Harrer M, Miguel C, et al. Cognitive behavior therapy for mental disorders in adults: a unified series of meta-analyses. JAMA Psychiatry. 2025;82:563–571. https://doi.org/10.1001/jamapsychiatry.2025.0482

20. Papola D, Miguel C, Mazzaglia M, et al. Psychotherapies for generalized anxiety disorder in adults: a systematic review and network meta-analysis of randomized clinical trials. JAMA Psychiatry. 2024;81:250–259. https://doi.org/10.1001/jamapsychiatry.2023.3971

21. Papola D. Errors in network meta-analysis of generalized anxiety disorder psychotherapies. JAMA Psychiatry. 2024;81:320. https://doi.org/10.1001/jamapsychiatry.2023.5364

22. van Dis EAM, van Veen SC, Hagenaars MA, et al. Long-term outcomes of cognitive behavioral therapy for anxiety-related disorders: a systematic review and meta-analysis. JAMA Psychiatry. 2020;77:265–273. https://doi.org/10.1001/jamapsychiatry.2019.3986

23. Slee A, Nazareth I, Bondaronek P, Liu Y, Cheng Z, Freemantle N. Pharmacological treatments for generalised anxiety disorder: a systematic review and network meta-analysis. Lancet. 2019;393:768–777. https://doi.org/10.1016/S0140-6736(18)31793-8

24. National Institute for Health and Care Excellence. Generalised anxiety disorder and panic disorder in adults: management. Clinical guideline CG113. 2011, updated 2020. Accessed October 10, 2026. https://www.nice.org.uk/guidance/cg113/chapter/recommendations

25. National Institute for Health and Care Excellence. Social anxiety disorder: recognition, assessment and treatment. Clinical guideline CG159. 2013. Accessed October 10, 2026. https://www.nice.org.uk/guidance/cg159/chapter/recommendations

26. Cuijpers P, Sijbrandij M, Koole SL, Andersson G, Beekman AT, Reynolds CF III. The efficacy of psychotherapy and pharmacotherapy in treating depressive and anxiety disorders: a meta-analysis of direct comparisons. World Psychiatry. 2013;12:137–148. https://doi.org/10.1002/wps.20038

27. Walkup JT, Albano AM, Piacentini J, et al. Cognitive behavioral therapy, sertraline, or a combination in childhood anxiety. New England Journal of Medicine. 2008;359:2753–2766. https://doi.org/10.1056/NEJMoa0804633

28. Batelaan NM, Bosman RC, Muntingh A, Scholten WD, Huijbregts KM, van Balkom AJLM. Risk of relapse after antidepressant discontinuation in anxiety disorders, obsessive-compulsive disorder, and post-traumatic stress disorder: systematic review and meta-analysis of relapse prevention trials. BMJ. 2017;358:j3927. https://doi.org/10.1136/bmj.j3927

29. Andrews G, Basu A, Cuijpers P, et al. Computer therapy for the anxiety and depression disorders is effective, acceptable and practical health care: an updated meta-analysis. Journal of Anxiety Disorders. 2018;55:70–78. https://doi.org/10.1016/j.janxdis.2018.01.001

30. Roy-Byrne P, Craske MG, Sullivan G, et al. Delivery of evidence-based treatment for multiple anxiety disorders in primary care: a randomized controlled trial. JAMA. 2010;303:1921–1928. https://doi.org/10.1001/jama.2010.608

31. Zbozinek TD, Khalsa SS, Craske MG. New and emerging treatments for anxiety disorders. BMJ. 2026;392:e083441. https://doi.org/10.1136/bmj-2024-083441