Published on 16 September 2026
Anxiety screening in primary schools: parents see a benefit, children report none
In brief
Eighty-four primary and junior schools in England, 5,335 children aged eight to nine in the participating classes, 4,473 of them in the total population, and screening questionnaires completed by a parent for only 1,459, that is 27%. Among them, 409 children screened positive and were followed for two years. The pathway combines a brief parent-completed screen, feedback on the result, digital parent-led cognitive behavioural therapy supported by telephone, and a single whole-class session. At twelve months, in this target population, 61% of children in the intervention arm screened negative again against 38% in the control arm, an adjusted odds ratio of 2.32 (95% CI 1.41 to 3.81; p = 0.0009), and every parent-reported secondary outcome points the same way. Two reservations weigh heavily on this reading. In the target population, no outcome reported by the child reaches statistical significance. And in the total population, the only significant differences on child-reported outcomes favour the control group, which is an unfavourable signal and not a neutral result.
The context
Anxiety disorders are the commonest psychiatric disorders of childhood, and the most often missed. An anxious child causes little disruption, often does well at school, and avoidance passes for shyness. The authors point out that families run into two successive obstacles: recognising the difficulty, then finding their way through a complex help-seeking process. Hence the idea, an old one but rarely tested with this degree of rigour, of looking for these children where they all are, that is at school, rather than waiting for a family to come through the door of a clinic.
The question here is not only whether a psychotherapy works. It is whether a complete pathway works: screen, feed back, refer, treat. Each of these links can break, and the trial assesses them together. The authors state that, to their knowledge, no previous randomised trial had evaluated an integrated screening and intervention approach for child anxiety in primary schools.
The study at a glance
| Item | |
|---|---|
| Population | |
| Children in Year 4 (aged eight to nine) in 84 primary and junior schools in England. Total population 4,473 children, that is 84% of the 5,335 children in participating classes. Target population made up of the 409 children who screened positive, 205 allocated to the intervention and 204 to the control. Mean age 8.8 years (SD 0.3). On parent report, 222 girls (54%) out of 408 and 185 boys (45%). On school-reported data, 325 children (85%) out of 384 are described as White in the target population, against 2,970 (72%) out of 4,142 in the total population. | |
| Intervention | |
| Screening with the parent-completed iCATS-2 (two items scored 0 to 3, total score 0 to 6), offered to all parents in the classes concerned. Written feedback on the result, then a feedback call with a children’s wellbeing practitioner in the event of a positive screen, with an offer of the digital parent-led cognitive behavioural therapy OSI: seven online modules of twenty to thirty minutes, each followed by a call of about twenty minutes. A single whole-class session was delivered in each participating class of the intervention arm. | |
| Comparator | |
| Assessment only, with no feedback on screening, and usual school practice. Schools that had a mental health support team at the time of recruitment were excluded. | |
| Primary outcome | |
| Screen-negative status on the iCATS-2 (score 0 to 2) versus screen-positive (score 3 to 6), parent-reported, in the target population, at twelve months after randomisation. Secondary outcomes: the same status at four and twenty-four months, and in the total population at all three time points; anxiety symptoms (SCAS-8 completed by parent, child and teacher; RCADS anxiety, parent and child); interference from anxiety (child-reported and teacher-reported items); depressive symptoms (RCADS depression, parent and child); conduct problems and hyperactivity-inattention (SDQ, parent and child). Educational, health economic and qualitative outcomes are reported separately. | |
| Design | |
| Pragmatic, parallel-group, superiority, cluster-randomised controlled trial with an internal pilot. The unit of randomisation is the school, 1:1 allocation stratified by school-level pupil deprivation, blocks of two and four. Two recruitment waves, 28 schools in phase 1 and 56 in phase 2, recruitment and baseline assessments from 6 January to 30 November 2022. Follow-up at four, twelve and twenty-four months. Registration ISRCTN76119074. Oxford level of evidence 1b. | |
The quality check
| Criterion | Judgement |
|---|---|
| Registration and protocol | Sound |
| FindingProspective registration with the ISRCTN registry under number ISRCTN76119074. Protocol and statistical analysis plan published, the latter signed off before database lock. Internal pilot of 28 schools with progression criteria judged by an independent steering committee. Ethical approval from the University of Oxford, reference R66068. | |
| Randomisation | Sound |
| FindingAlgorithm built by the trial statistician, allocation run by an independent statistician. Stratification on the proportion of pupils eligible for free school meals, above or below the national median of 21.6%. Schools ordered before allocation by the size of their target population, blocks of two and four. Intraclass correlation coefficients documented outcome by outcome in the appendix. | |
| Masking | Absent |
| FindingOpen trial. Neither schools, participants, clinicians nor the study team were masked. Only the trial statisticians remained masked until the final datasets were prepared. With self-reported outcomes completed by parents, this limitation is not a minor one, and the authors acknowledge it as such. | |
| Uptake of screening | Low |
| Finding1,459 parents completed the screening questionnaire, that is 27% of the 5,335 children in participating classes and 33% of the 4,473 children in the total population. The pathway was therefore evaluated only in the minority of families who responded. | |
| Missing data | Differential |
| FindingPrimary outcome collected for 145 children out of 205 (71%) in the intervention arm against 163 out of 204 (80%) in the control arm, 75% across the whole target population, slightly below the target of 80%. Multiple imputation on 50 datasets, complete case analysis, analysis including data collected outside the window and tipping point analysis all conducted, with consistent results. | |
| Representativeness | Limited |
| FindingUnder-representation, among screened children, of minoritised ethnic groups and economically disadvantaged families: 85% of children described as White in the target population against 72% in the total population. A single country, England, a single school system, and schools selected for having no mental health support team. | |
| Funding and competing interests | Public, but ties to the intervention |
| FindingPublic funding, National Institute for Health and Care Research, Programme Grants for Applied Research, reference RP-PG-0218-20010, with additional support from the same body and a Prudence Trust fellowship. The funder is stated to have had no role in design, data collection, analysis, interpretation or writing. Competing interests, on the other hand, are not nil: the last author and a co-author developed the OSI platform; the University of Oxford receives consultancy and royalty payments linked to distribution of the intervention outside research, with no personal payment to the last author, while the co-author who developed it personally receives consultancy and royalty payments from the commercialisation of OSI. Another author owns the copyright to the SCAS items and states that she may benefit financially from large-scale commercial licensing. The research group of another author receives research methods consultancy funding from Place2Be, an organisation that provides mental health training and interventions in schools. The other authors declare no competing interests. | |
The findings
| Outcome | Result |
|---|---|
| Screen-negative status at twelve months, target population | Adjusted OR 2.32 (95% CI 1.41 to 3.81), p = 0.0009 |
| ReadingPrimary outcome. Marginal logistic regression by generalised estimating equations on imputed data, adjusted for baseline score, trial phase, size of the school’s target population, school-level deprivation and the child’s free school meal status. The sensitivity analyses all point the same way: complete cases, independent correlation structure, complier average causal effect, data collected outside the window. | |
| Screen-negative status, four and twenty-four months | Significant difference at both time points |
| ReadingObserved proportions: 77 out of 135 (57%) against 51 out of 158 (32%) at four months, 93 out of 150 (62%) against 74 out of 161 (46%) at twenty-four months. The adjusted odds ratios of the main analysis at these two time points cannot be read in Table 3 as it appears in the document consulted; the sensitivity analysis including data outside the window gives 2.67 (1.60 to 4.47) at four months and 1.78 (1.10 to 2.86) at twenty-four months. In the control arm, the proportion of screen-negative children rises from 38% at twelve months to 46% at twenty-four months, which the authors relate to the natural remission described in the literature. | |
| Parent-reported anxiety (SCAS-8-P) | Favours the intervention at all three time points |
| ReadingObserved means at twelve months of 7.3 (SD 4.4) against 9.4 (3.9). The adjusted estimate of the main analysis cannot be read in Table 3 of the document consulted; the complete case analysis gives an adjusted difference of 2.1 points in favour of the intervention (95% confidence interval 1.3 to 2.9; p < 0.0001) and the out-of-window analysis 1.6 points (0.9 to 2.3; p < 0.0001). No standardised mean difference appears in the sources consulted. An effect consistent with the primary outcome, which relies on the same informant. | |
| Parent-reported RCADS, anxiety and depression | Favours the intervention at all three time points |
| ReadingAt twelve months, in the sensitivity analysis including data outside the window, adjusted difference of 6.0 points on anxiety (3.5 to 8.4; p < 0.0001) and of 1.2 points on depression (0.5 to 1.9; p = 0.0005), both in favour of the intervention. No adjustment was made for multiple comparisons and the authors state that they treat all secondary outcomes as exploratory. | |
| Child-reported outcomes, target population | No significant difference |
| ReadingThis is the turning point of the paper. No self-reported outcome differs between groups in the target population. That does not demonstrate an absence of benefit for the child, but it does not establish one either. The authors recall that they chose a parent-reported outcome because, in their preparatory work, child-completed questionnaires did not discriminate accurately between children with and without anxiety disorders at this age. Teacher-reported outcomes, for their part, favour the intervention on interference from anxiety at all three time points, with no significant difference on the SCAS-8-T. | |
| Total population | Unfavourable signal on self-report |
| ReadingAt this scale, screen-negative status is only a secondary outcome, and it does not differ at any time point. The only significant differences on child-reported outcomes favour the control group: slightly higher self-reported anxiety in the intervention arm at four months on the SCAS-8-C and on RCADS-C anxiety, and at twelve months on RCADS-C anxiety only. The adjusted differences found in the sensitivity analyses are of the order of 0.3 to 0.4 points on the SCAS-8-C and 1.5 to 2.0 points on RCADS-C anxiety. Exploratory secondary outcomes, with no correction for multiplicity. | |
| Safety | No serious trial-related event |
| ReadingNo serious adverse event related to the trial procedures or to the intervention, and no adverse event related to the intervention. Six adverse events related to trial procedures were recorded, five of them involving distress while completing child-reported questionnaires, two in the intervention arm and three in the control arm. | |
Critical appraisal
| Domain | Risk of bias |
|---|---|
| Randomisation process | Low |
| FindingAllocation run by an independent statistician, stratification by school-level deprivation, blocks of two and four, intraclass correlation documented. The sample size calculation aimed at 80 schools and 398 children to detect a difference of 20 percentage points with power slightly above 90%; 84 schools and 409 children were included. | |
| Deviations from intended interventions | Some concerns |
| FindingOpen, pragmatic trial, analysed by intention to treat according to the school of allocation. A family that knows it has been screened and then supported does not fill in an anxiety questionnaire the way a family left without feedback does. | |
| Missing outcome data | Some concerns |
| FindingPrimary outcome missing for 29% of the intervention arm against 20% of the control arm, in the direction that favours the intervention if the least improved drop out more. Parents without primary outcome data more often had no partner and a low income. The tipping point analysis shows that the effect would remain significant even if the proportion of children screening negative among missing data in the intervention arm fell to 23%, against 50% estimated by imputation. This limits the concern without removing it. | |
| Measurement of the outcome | High |
| FindingThe primary outcome is reported by the parent, who is also the person who received the intervention, the feedback and the practitioner’s calls. The authors acknowledge this risk of bias and rely on the concordance of teacher-reported outcomes to temper it. The discordance with the child’s self-report remains the most interesting result of the trial, and the least comfortable. | |
| Selection of the reported result | Low |
| FindingProtocol and statistical analysis plan published, the latter signed off before database lock. The unfavourable signal on the universal component is reported by the authors themselves, who conclude that the whole-class session should not be integrated. That is to their credit. | |
Level of evidence
Oxford level of evidence 1b. Confidence is high on one point: in families who actually enter the pathway, the programme reduces the child’s anxiety as the parent describes it, and this effect holds at one year, with an echo in the interference observed by the teacher. Confidence is low on three others. We do not know whether the child is actually better, since the child’s own answers do not move. We do not know what the programme would achieve if participation in screening exceeded 27%. And in the total population, the only significant signal coming from the children runs against the intervention: the authors consider it reasonable to attribute it to the whole-class session, or to that session combined with the assessment procedures, without any analysis allowing a choice between these explanations.
The colleague test
What a child psychiatrist would say if you put this trial to them in two minutes, between two consultations.
“ Screening at school plus remote parent-led therapy takes the proportion of children screening negative at one year from 38 to 61%, for about two and a half hours of practitioner time per family. Very good. But the children themselves say nothing different. And across the whole cohort, their self-report tends to go the wrong way. So before rolling this out, I want to know what has really improved: the child’s anxiety, or the parent’s worry. ”
What this means in practice: the model deserves to be adapted and tested elsewhere, the universal whole-class session is not to be rolled out as it stands, and discordance between informants is to be handled as clinical information, not as a measurement flaw.
What you can do with this
- When a parent describes a clear improvement and the child does not confirm it, the divergence is not an artefact: the two informants are not measuring the same thing. One observes behaviour, the other lives an inner experience.
- A brief parent-completed screen can identify anxious children who would never have reached you. Two items are enough to open the conversation.
- Parent-led cognitive behavioural therapy, guided remotely, is a credible option for the school-age child: in this trial, 165 of the 205 targeted families (80%) started the intervention and 127 (62%) completed the core content. It is not a second best for the waiting list.
- When asked about a universal school-based prevention programme, this trial provides an argument for caution: the targeted intervention shows a benefit, the universal whole-class session a contrary signal that the authors consider sufficient not to keep it.
- What to watch if you support a project of this kind: the uptake of screening, on which everything else depends, and the gap between what the parent reports and what the child reports.
- The course of action is set out in the NICE decision tree for generalised anxiety and panic disorder.
Frequently asked questions
Can screening for anxiety at school create anxiety where there was none?
That is exactly the question raised by the signal observed in the total population. On children’s self-report, anxiety is slightly higher in the intervention arm at four months on two scales and at twelve months on one only. These are exploratory secondary outcomes, with no correction for multiplicity, so this is a signal and not a demonstration. The authors nevertheless consider it sufficient to recommend not integrating the whole-class session, and they also note that five of the six adverse events recorded involved distress in a child completing the questionnaires.
Why did children not self-report any improvement after parent-led CBT?
The trial cannot decide between several explanations: self-report scales at the age of eight are not very sensitive to change and discriminate poorly, which the authors had already observed before the trial; the parent may overestimate the improvement because they were supported; or the benefit genuinely bears on observable behaviour more than on inner experience. None of these hypotheses is tested here.
Can this school screening-to-intervention model be used outside England?
The concept can: a brief screen, feedback, remotely guided parent-led therapy. The organisation cannot be carried over as it stands, since it was built for the English school system and its workforce, and the authors themselves call for work on adapting the approach to other countries. The publication does not document where the OSI programme is available outside the trial.
Does an odds ratio of 2.32 mean the intervention works twice as well?
No. It is an odds ratio, not a risk ratio, and it concerns a switch in a binary screening status. In concrete terms, 61% of children in the intervention arm are screen-negative at twelve months against 38% in the control arm. It is this gap of 23 points that speaks, not the factor of two.
Does 27% uptake of screening invalidate the trial?
It does not invalidate the result obtained in participants, who remain randomised by school. It limits what can be said at the scale of an entire school population, all the more so as the least advantaged families participated less. At that scale, screen-negative status is in any case only a secondary outcome, and it does not differ at any time point.
Annotated bibliography
Source study. Reardon T, Ukoumunne OC, Taylor L, Ball S, Violato M, Larkin M, Fisk J, Halliday G, Yu S, Williamson V, Ford T, Gray A, Hill C, Jasper B, Macdonald I, Morgan F, Pollard J, Sancho M, Sniehotta FF, Spence SH, Stainer J, Stallard P, Aitken E, Baker J, Calladine H, Chessell C, Green I, Hennigan C, Hooper C, Koenig A, Manley H, Martineau L, Njuguna R, Pall ND, Pearcey S, Potts R, Robertson O, Scott-Day E, Skjerdingstad N, Talbot-Ponsonby W, Vesey T, Watson R, Weisser T, Wright J, Young P, Creswell C. Screening-to-intervention pathway for child anxiety problems alongside usual school practice versus usual school practice only (iCATSi2i): a cluster-randomised, controlled trial in primary schools in England. The Lancet Psychiatry. 2026; 13 (5): 396-412, published online 9 April 2026. DOI 10.1016/S2215-0366(26)00064-7. PMID 41969016. Trial registered under number ISRCTN76119074. Funding: National Institute for Health and Care Research, Programme Grants for Applied Research, reference RP-PG-0218-20010. Declared competing interests: development of the OSI platform by two authors, consultancy and royalty payments received by the University of Oxford and, personally, by one co-author; copyright to the SCAS items held by another author; research methods consultancy funding paid by Place2Be to the research group of another author; no competing interests declared by the other authors. The author list is truncated after Susan H Spence in the document consulted; it is completed here with the two missing authors, Jason Stainer and Paul Stallard, and with the named members of the iCATS team, as recorded in bibliographic databases.
Context references, taken from the reference list of the source publication. They appear there without identifiers. Their existence, journal, year, volume and pagination were confirmed against bibliographic databases on 2 September 2026.
James AC, Reardon T, Soler A, James G, Creswell C. Cognitive behavioural therapy for anxiety disorders in children and adolescents. Cochrane Database of Systematic Reviews. 2020; 11: CD013162. DOI 10.1002/14651858.CD013162.pub2. The reference review on the efficacy of CBT for child anxiety, and the authors’ source on the lower sensitivity of self-reported outcomes before the age of twelve.
Creswell C, Taylor L, Giles S, et al. Digitally augmented, parent-led CBT versus treatment as usual for child anxiety problems in child mental health services in England and Northern Ireland: a pragmatic, non-inferiority, clinical effectiveness and cost-effectiveness randomised controlled trial. The Lancet Psychiatry. 2024; 11: 193-209. The trial that established the non-inferiority of OSI in clinical services, the direct forerunner of the present work.
Reardon T, Ukoumunne OC, Ball S, et al. Development of a brief assessment tool to identify children with probable anxiety disorders. JCPP Advances. 2024; 5: e12265. The development of the iCATS-2, with a cut-off of 3 associated with sensitivity and specificity above 75% in children aged eight to eleven.
Stallard P, Skryabina E, Taylor G, et al. Classroom-based cognitive behaviour therapy (FRIENDS): a cluster randomised controlled trial to Prevent Anxiety in Children through Education in Schools (PACES). The Lancet Psychiatry. 2014; 1: 185-192. One of the studies cited in support of the inconsistency of universal classroom interventions.
Guzman-Holst C, Streckfuss Davis R, Andrews JL, Foulkes L. Scoping review: potential harm from school-based group mental health interventions. Child and Adolescent Mental Health. 2025; 30: 208-222. The scoping review on which the authors draw to discuss the possible harmful effects of group programmes.
Reardon T, Ukoumunne OC, Violato M, et al. Identifying Child Anxiety Through Schools-identification to intervention (iCATS-i2i): protocol for a cluster randomised controlled trial to compare screening, feedback and intervention for child anxiety problems to usual school practice. Trials. 2022; 23: 896. The published protocol.
Ball S, Reardon T, Creswell C, et al. Statistical analysis plan for a cluster randomised controlled trial to compare screening, feedback and intervention for child anxiety problems to usual school practice: identifying Child Anxiety Through Schools-identification to intervention (iCATS-i2i). Trials. 2024; 25: 62. The published statistical analysis plan.