Prevention programmes

Eight RCTs met inclusion criteria, evaluating 8 prevention programmes: 5 delivered in schools and 3 in other community settings.27–36 One programme—Coping and Promoting Strength—was evaluated in 2 RCTs while the others were evaluated in single RCTs. One RCT also evaluated 2 interventions: Cognitive Bias Modification and a CBT programme. One programme was universal while 7 focused on at-risk children, that is, with temperamental inhibition, anxiety symptoms or a parent with an anxiety disorder.

Six programmes used CBT techniques including psychoeducation, relaxation, cognitive restructuring and exposure exercises.28–30 34–36 Four CBT programmes were delivered to groups of children in schools, with 2 also providing parent sessions.28 34–36 The 2 other CBT programmes focused on parents; 1 was delivered to individual families and 1 trained parents to provide CBT to their young children.29 30 The 2 non-CBT programmes used computer delivery. Cognitive Bias Modification aimed to reduce negative assumptions in noticing and interpreting social situations.36 Meanwhile, Mindset taught teens to cope with stress by viewing personality traits as being modifiable.33 All 8 programmes were relatively brief, ranging from 30 min to 5.5 months.

Three prevention programmes—all using CBT and all focusing on at-risk children—significantly reduced anxiety diagnoses and/or symptoms. Coping and Promoting Strength was notable for reducing anxiety diagnoses across 2 RCTs, with a large effect size for the second (OR=8.54).30 31 This included only 0%–5% of intervention children developing disorders between post-test and 9-month follow-up, compared with 30%–31% of controls.30–32 Coping and Promoting Strength also significantly reduced symptoms: on 2 measures with large effect sizes (Cohen’s d [d]=0.82 and 1.99) in the first RCT and on 3 measures with medium effect sizes (d=0.54–0.74) in the second.30 31 Cool Little Kids and Friends also significantly reduced symptoms on one measure but did not reduce diagnoses.29 35

Five prevention programmes failed to show benefits pertaining to anxiety, including 3 CBT programmes (Aussie Optimism, Feelings Club and Generic CBT) and both non-CBT programmes (Cognitive Bias Modification and Mindset).27 28 33 34 36 Adverse events were only assessed for Feelings Club; none were reported by children or parents.34 Table 2 summarises the 8 prevention evaluations.

Psychosocial treatments

Fourteen RCTs met inclusion criteria, evaluating 12 psychosocial treatments: 1 delivered in homes, 8 in clinics and 3 in schools. 37–50 ^w51 Two treatments—Coping Cat and Skills for Academic and Social Success (SASS)—were evaluated in 2 RCTs while the others were evaluated in single RCTs.42 45 50 ^w51 As well, 3 RCTs compared different formats including individual child versus individual family delivery42 and psychologist versus counsellor delivery.^w51 Most studies included children with a variety of anxiety disorders.

Eleven treatments used CBT, delivered to children and parents individually and in groups in a variety of settings. The only non-CBT treatment, Attention Bias Modification Training (ABMT), used computers to teach children to reduce their focus on socially threatening situations.41 All treatments were relatively brief, ranging from 3 hours to 6 months.

Ten psychosocial treatments—all CBT—significantly reduced anxiety diagnoses and/or symptoms by final follow-ups. The home-based Strongest Families reduced diagnoses (approximately 25% for intervention children vs 50% for controls) with a large effect size (OR=2.51).37 Of the clinic-based treatments, Cool Little Kids Plus Social Skills reduced diagnoses (66% vs 100%) and number of disorders per child (d=1.76); it also reduced symptoms on 3 measures, with large effect sizes (d=0.89–2.11).38  Cool Kids also reduced primary diagnoses (31% vs 55%), any anxiety diagnoses (51% vs 70%) and symptoms on 3 measures.43  Parent Education Program reduced diagnoses (40% vs 69%) and symptoms on 2 measures.39  Timid to Tiger reduced primary diagnoses (46% vs 76%) and any anxiety diagnoses (54% vs 91%), both with large effect sizes (OR=3.68 and 8.50, respectively), but had no impact on symptoms.40  One-Session Treatment reduced specific phobia diagnoses (51% vs 65%) as well as symptoms on 1 measure.44 Meanwhile, Coping Cat reduced anxiety diagnoses (18% vs 35%) and symptoms on 1 measure (OR=3.29 and 2.56, respectively) in 1 trial.45 In the other Coping Cat trial, primary anxiety diagnoses were not significantly reduced for either child or family formats; however, the child version did reduce symptoms on 1 measure.42

For school-based treatments, Coping Koala reduced diagnoses (20% vs 39%) and symptoms on 2 measures.49  SASS (delivered by psychologists and psychology graduate students) and Counsellor-Delivered SASS both reduced social anxiety diagnoses (27% vs 93% and 61% vs 88% [OR=4.89], respectively, although Psychologist-Delivered SASS did not).^50–w51 All SASS versions also reduced symptoms on 2-to-4 measures, with moderate-to-large effect sizes for 2 versions (OR=16.21 and d=0.38–0.93 for Counsellor-Delivered and OR=7.61 and d=0.34–0.83 for Psychologist-Delivered). In contrast, Friends improved 1 symptom measure but failed to reduce diagnoses.47 48

The remaining 2 treatments—both clinic-based—showed no benefits: Generic CBT and ABMT (the only non-CBT treatment).41 46 Adverse events were assessed for 2 treatments. Strongest Families participants reported no adverse events, and none were observed for One-Session Treatment participants.37 44 Table 3 summarises the 14 psychosocial treatment evaluations.

Pharmacological treatments

Eleven RCTs met inclusion criteria, evaluating 7 medications: SSRIs (fluoxetine, fluvoxamine, paroxetine and sertraline); a tricyclic (imipramine); a selective-noradrenergic-reuptake-inhibitor (venlafaxine); and an N-methyl-D-aspartate-partial-agonist (D-cycloserine).^w52–w62 Fluoxetine was evaluated in 3 RCTs while imipramine and sertraline were each evaluated in 2.^{w54–w58 w60–w61} The other medications were evaluated in single RCTs.^{w52 w53 w59 w62} In addition to placebos, fluoxetine was compared with behaviour treatment in 1 RCT^w54 and with CBT in another,^w56 while sertraline was compared with CBT in 1 RCT.^w61 All medications were assessed at post-test only, except D-cycloserine which was assessed at 1-week follow-up.^w52 Some studies included children with a variety of anxiety disorders,^{w53 w55 w56 w61} while others focused on single disorders.^{w52 w54 w58–w60 w62}

For fluoxetine, 2 RCTs assessed diagnoses and symptoms. In 1 trial (fluoxetine I; see table 4), the medication significantly reduced social anxiety diagnoses (79% for intervention children vs 97% for placebo controls) and reduced symptoms on 4 measures.^w54 However, when compared with a 12-week behaviour treatment, fluoxetine was less effective at reducing social anxiety diagnoses (79% for fluoxetine vs 47% for behaviour treatment) as well as at reducing symptoms on 4 measures.^w54 In another trial (fluoxetine III), the medication was given with CBT but failed to reduce anxiety diagnoses compared with CBT alone or compared with CBT plus placebo (68% for fluoxetine plus CBT vs 65% for CBT alone vs 77% for CBT plus placebo).^w56 Similarly, there was no difference among the 3 conditions on symptoms on 4 measures.^w56 A third trial (fluoxetine II) did not assess diagnoses but reduced symptoms on 1 measure.^w55

For the other 6 medications, RCTs assessed anxiety symptoms but not diagnoses. D-cycloserine reduced symptoms with medium effect sizes on 2 measures (r=0.35 and 0.37).^w52 Fluvoxamine reduced symptoms on 2 measures.^w53 The first imipramine RCT showed 8 symptom reductions, with a medium effect size for the 1 outcome where this was calculated (Cohen’s h=0.73).^w57 As well, despite initially refusing to attend school or only attending with marked distress, by the end of the trial, 81% of intervention children were regularly attending compared with only 47% of controls.^w57 In contrast, the second imipramine RCT found the medication no better than placebo on 28 of 29 measures and worse on 1.^w58 Paroxetine reduced symptoms on 6 social anxiety measures, with large effect sizes for the 2 outcomes where these were calculated (OR=5.44 and 6.05).^w59 Sertraline also reduced symptoms in 2 RCTs. In the first, it reduced symptoms on 6 measures.^w60 In the second, it reduced symptoms on 2 measures compared with placebo with moderate-to-large effect sizes (Hedges’ g [g]=0.45; OR=3.9); however, outcomes were not significant when compared with CBT.^w61 Meanwhile, venlafaxine reduced symptoms on 2 measures, with moderate effect sizes (g=0.46 and number needed to treat=5).^w62

Adverse events were common for most medications. These included more than 25% of children experiencing: abdominal pain/nausea and drowsiness with fluoxetine; headaches with D-cycloserine; abdominal discomfort and headaches with fluvoxamine; drowsiness, dry mouth, constipation, irritability and dizziness with imipramine; and drowsiness, dry mouth, restlessness and leg spasms with sertraline.^{w52 w53 w55 w57 w58 w60} Table 4 summarises the 11 medication evaluations, including adverse events. (We reported adverse events where at least 25% of children were affected or where significantly more children on medication versus placebo were affected; however, not all studies tested the statistical significance of adverse events.)

Risk of bias in included studies

Based on data provided in each RCT, we evaluated 5 risk indicators using the Cochrane risk-of-bias tool.24 25 For most prevention studies, bias risks were low, with the exception of performance bias. For most psychosocial treatment studies, selection bias was unclear, while performance and detection biases were high; however, attrition and reporting biases were all low. For most medication studies, bias risks were low. Overall risk-of-bias profiles favoured medication over psychosocial studies. (Online supplementary appendix B gives individual RCT risk-of-bias assessments; online supplementary appendix C gives aggregated risk-of-bias by intervention category.) We also augmented our risk-of-bias assessment to address concerns not covered in the Cochrane tool.24 25 Specifically, we identified conflicts-of-interests for 6 of 11 medication RCTs—with author(s) declaring ties to pharmaceutical companies including receiving honouraria, owning stock and/or being company employees.^{w52 w54 w59 w60 w62} In contrast, conflict-of-interest was reported for only 1 psychosocial study, with an author declaring the intervention may be commercialised.37

Meta-analysis

Beyond identifying specific interventions for preventing and treating childhood anxiety to guide policy and practice, we undertook meta-analyses to determine the common effects of similar interventions where there were sufficient RCTs. This included CBT prevention programmes, CBT treatment programmes and SSRIs. For CBT prevention and treatment programmes, ORs for diagnoses prevented or remitted were either extracted or calculated. Because diagnostic outcomes were not available for most SSRI RCTs, we used ORs for symptom improvement.

CBT prevention programmes did not significantly outperform comparison conditions for preventing diagnoses (Log OR=0.50; 95% CI, −0.14 to 1.13). However, study populations showed significant heterogeneity (Cochran’s Q=16.1, p=0.01). In contrast, the CBT treatment programmes significantly outperformed comparison conditions for reducing diagnoses (Log OR=0.95; 95% CI, 0.69 to 1.21) with acceptable levels of heterogeneity (Cochran’s Q=15.5, p=0.16). SSRIs resulted in significantly more symptom improvement than placebos at post-test (Log OR=1.57; 95% CI, 1.09 to 2.06), also with acceptable levels of heterogeneity (Cochran’s Q=10.56, p=0.06). While the Log OR was greater for SSRIs than CBT treatment programmes, the latter used a more robust outcome measure—diagnostic versus symptom improvement. As well, CBT treatments were assessed at follow-ups averaging 10 months, while SSRIs were assessed at post-test. (Online supplementary appendices D and E summarise the meta-analyses.)

Regarding publication bias, Egger’s tests showed asymmetry in CBT prevention and CBT treatment studies (p<0.01 for both) and symmetry in SSRI studies (p=0.21). However, the publication bias found for CBT treatment studies had minimal impact on our main findings. When we limited the analysis to studies with sample sizes greater than 100, asymmetry disappeared (p=0.38) and CBT treatment programmes still effectively reduced diagnoses (Log OR=0.92; 95% CI, 0.63 to 1.22). (Online supplementary appendix F gives funnel plots assessing publication bias.)