Introduction

Approximately 8% of children aged between 5 and 10 years experience mental disorders, and around 20% report subthreshold mental health difficulties (Green, McGinnity, Meltzer, Ford, & Goodman, 2005). Neuropsychiatric disorders contribute substantially to childhood disability (WHO, 2013) and frequently persist into adulthood (Kessler et al., 2005; Patton et al., 2014). Intervening early is a key focus of child mental healthcare internationally, but outcomes to date have been mixed, with suggestions that more vulnerable children show least improvement, and insufficient assessment of mechanisms maintaining vulnerability and increasing resilience (Dray, 2021; Dray et al., 2017).

Psychotic disorders are typically diagnosed in late adolescence/early adulthood (Kirkbride et al., 2013), but there is considerable evidence that symptoms develop gradually and can originate in childhood (Fusar-Poli et al., 2012). A distinction is often made between clinically relevant psychotic symptoms associated with a diagnosed severe mental health condition, and milder, unusual or ‘psychotic-like’ experiences (UEs). UEs can include perceptions in any modality (e.g. auditory, visual, or tactile) not experienced by others, as well as thoughts and ideas that others may find bizarre. Despite being considered unusual in their nature or content, epidemiological studies suggest that UEs are a relatively common occurrence in the general population (van Os & Reininghaus, 2016). In a sample of 8000 children aged 9 to 11 years, two thirds reported at least one UE (Laurens, Hobbs, Sunderland, Green, & Mould, 2012). Rather than being a reliable marker of illness, UEs appear to lie on a continuum overlapping with normality, with the vast majority (80%) being transient in nature (van Os & Reininghaus, 2016). However, for the minority of young people whose UEs are persistent, distressing, and disabling, there is a higher likelihood of transition to clinically relevant psychosis in later life (Kaymaz et al., 2012).

Appraisals of UEs (for example, as uncontrollable, unacceptable to oneself and others, external, and threatening) differentiate individuals with and without a need for clinical care (Lovatt, Mason, Brett, & Peters, 2010; Peters et al., 2017; Underwood, Kumari, & Peters, 2016). Such appraisals are hypothesized to arise from a range of cognitive biases. Chief amongst these is a jumping to conclusions bias (JTC), wherein individuals with psychosis, and particularly delusions, reach certainty after limited data gathering, compared with nonclinical and nonpsychotic clinical groups (Freeman, Garety, Kuipers, Fowler, & Bebbington, 2002; McLean, Mattiske, & Balzan, 2017).

Cognitive behavioural therapy for psychosis (CBTp) targets unhelpful appraisals, and early application (from mid-adolescence) is recommended by international treatment guidelines to potentially avert the development of clinical psychosis (e.g. NICE, 2013). As both clinical and functional outcomes tend to be poor once young people have developed an at-risk mental state (Fusar-Poli et al., 2012; McGorry et al., 2002), there is a strong rationale for implementing interventions for young people with subthreshold experiences at an early stage, in transdiagnostic youth mental health services, or even before they come into contact with mental health services, as public health or educational initiatives (Guloksuz & van Os, 2018).

CUES-Ed is a universal, classroom-based early intervention cognitive behavioural therapy (CBT) programme for primary school children. CUES-Ed combines CBT techniques and strategies to promote physical wellbeing, emotional literacy, and regulation skills whilst also trying to promote adaptive, nonstigmatising appraisals of mental experiences, to encourage appropriate sharing and help-seeking when needed. Unique to CUES-Ed, drawing on cognitive models of psychosis vulnerability and early interventions is an additional focus on better understanding and coping with UEs, and reducing the tendency to JTC.

We have previously reported an evaluation of standardised outcomes following CUES-Ed that showed improvements, particularly in those children whose baseline scores suggested emotional and behavioural vulnerability (Redfern, Jolley, Bracegirdle, Browning, & Plant, 2019). Here, in an overlapping sample, we use novel assessments to report on the mechanisms of cognitive vulnerability targeted by CUES-Ed, namely stigmatising appraisals of emotional expression that may discourage sharing and help-seeking, stigmatising appraisals of UEs that may increase distress and the likelihood of future mental health problems, and the tendency to JTC that may increase the likelihood of unhelpful appraisals, particularly of UEs. We first considered fitness for purpose of our routine assessments of cognitive vulnerability and our methods for identifying vulnerable children, including differences according to gender and year-group. Second, we evaluated change in cognitive vulnerability outcomes for the whole class and for vulnerable children. We expected a reduction in cognitive vulnerability from before to after CUES-Ed and compared with a naturalistic control group, across all children, with larger changes, and a shift from vulnerable status, for those children identified pre-intervention as cognitively vulnerable.

Method Intervention

CUES-Ed is a manualised intervention delivered over eight sessions by clinical psychologists and cognitive behavioural therapists to whole classes of 7–10 year olds as part of their normal school day. The intervention is based around a character named Ed and his friend Chloe (Redfern et al., 2019) and includes skills for promoting physical self-care (eating well, sleeping well, relaxing, and being active), emotional literacy (spotting physical and behavioural ‘cues’ to emotions in self and others), and coping with internal and external difficulties using behavioural (help-seeking, relating, and distraction) and cognitive (noticing, identifying, and changing unhelpful thoughts) regulation skills. The current study concerns cognitive change strategies: teaching children less stigmatising appraisals of emotional expression and unusual experiences, and belief flexibility/reduced JTC. We will present the remaining intervention components in subsequent studies.

Audit approval

Outcomes were collected as part of a routine service evaluation, with approval given by the South London and Maudsley National Health Service Foundation Trust Child and Adolescent Mental Health Services Clinical Academic Group Audit Committee (Ref: #2014-08). Permission was given by head teachers (or nominated representatives) of individual schools commissioned to receive CUES-Ed. Schools ensured that parents were given an information sheet explaining that CUES-Ed was being offered as part of the curriculum, and offered the opportunity for their child to opt out.

Procedure

Assessments were completed as a whole class during the first and last CUES-Ed sessions with questionnaires read out by CUES-Ed clinicians. Teaching and CUES-Ed staff supported children to complete questionnaires as needed. For the naturalistic waitlist, assessments were completed 8 weeks before and during the first CUES-Ed session. Assessment booklets were identified by pseudonymised letter–number codes, generated by the research team from the class register, and distributed in class to ensure correctly matched pre-post outcomes. Data were fully anonymised after entry, by confidentially destroying the code list. Children were told this and the purpose of evaluation. Individuals were able to opt out of outcome collection if they wished.

Service recipients

Delivery of CUES-Ed was evaluated during the 2016/17 and 2017/18 academic years (May to November 2017) in the London Borough of Southwark. Data were collected from 33 classes across 14 primary schools, with 20–30 children in each form (n = 900). Additionally, two classes (n = 60), who were waiting to receive CUES-Ed, completed the measures before and after an 8-week period, forming a naturalistic control group. The current report includes all children completing at least one assessment of cognitive vulnerability to the end of 2017. The current sample incorporates delivery from May to July 2017 for which standardised, but not cognitive vulnerability, outcomes have previously been reported (n = 437; Redfern et al., 2019). This previous report also included an earlier cohort who did not complete cognitive vulnerability assessments and thus could not be included in the current sample.

Measures

Demographic data other than self-reported gender and year-group were not collected, but publicly reported data for participating schools were representative of the demographics of South London: Southwark is densely populated and relatively socio-economically deprived, with a young population, around half belonging to Black, Asian, and Minority ethnic groups (Southwark Council, 2015). ‘Vulnerable’ status was defined by negative and/or inflexible responses, according to the scoring for each cognitive vulnerability measure, detailed below. To suit the classroom context, measures were presented in a developmentally tailored ‘workbook’ format (Figures 1-3).

CUES-Ed cognitive vulnerability workbook: appraisals of emotional expression

CUES-Ed cognitive vulnerability workbook: appraisals of unusual experiences

CUES-Ed cognitive vulnerability workbook: jumping to conclusions

Standardised outcomes

The Child Outcome Rating Scale (Duncan et al., 2003) measures wellbeing/distress across four items rated 0 (worst) to 10 (best); scores of 32 or more are in the clinical range. Me and My Feelings is a school-based measure of childhood emotional/behavioural difficulties (M&MF E/B; Deighton et al., 2012). Ratings from 0 (best) to 2 (worst) are totalled across 16 items, 10 emotional (M&MF-E, clinical cut-off > 9) and 6 behavioural (M&MF-B, clinical cut-off >5).

Child workbooks

Workbooks were designed by the CUES-Ed team in consultation with education representatives. Published measures were adapted into brief, engaging, developmentally appropriate assessments of the specific intervention targets of CUES-Ed, suitable for individual completion in a classroom setting. Measures were piloted and refined through iterative completion and feedback from the earliest classes receiving CUES-Ed. The current study focused on the three cognitive vulnerability components (Figures 1-3); the four remaining assessments (of physical self-care skills; ‘cues’ to emotional states; identifying thoughts, feelings, and behaviours; and regulation/coping skills) will be presented in subsequent studies.

Appraisals of emotional expression (stigma-self, stigma-others, and stigma-language)

Adapted from previously validated measures of stigma in children (Cameron & Rutland, 2006; Doyle & Aboud, 1995; Greenwood et al., 2016). Two Likert scales assessed children's stigmatising views of their own emotional expression and that of others, respectively (stigma-self: ‘When YOU feel sad, worried or angry, is it okay to show it and talk about it?’; and stigma-others: ‘When SOMEONE ELSE feels sad, worried or angry, is it okay for them to show it and talk about it?’). Children were instructed to ‘tick one of the boxes below’, and given four options: ‘It is never ok’, ‘It is usually not ok’, ‘It is sometimes ok’, and ‘It is mostly ok’, scored from 0 to 3, respectively. The response option ‘always okay’ was specifically not included, as this is not necessarily desirable or helpful. It was expected that scores would shift from lower (i.e. more stigmatising) pre-intervention to higher (less stigmatising) post-intervention. For each scale, answering ‘it is never ok’ to show or talk about feelings indicated vulnerability. In addition, children were asked to finish a sentence (‘When someone shows or talks about their feelings when they are sad, worried or angry, I think that it is….’) by circling as many as they wanted from a list of 14 words comprising seven positive (understandable, brave, strong, normal, okay, fine, and good) and seven negative (weak, rude, crazy, weird, bad, scary, and annoying). The total number of positive words circled was added to the number of negative words left uncircled and divided by seven, to give a final ‘stigma-language’ score, with higher scores indicating less stigmatising views. Circling more negative than positive words indicated vulnerability for this component.

Appraisals of unusual experiences (UE-appraisals)

A vignette, featuring Ed and Chloe, described an anomalous experience: (a) ‘One day, Ed hears Chloe call his name. But…no one is around. Ed is totally by himself….’ Children then chose one of several possible appraisals (‘Why might Ed have heard somebody that isn't really there? Tick ONE box that you think might explain what is happening to Ed’), comprising three normalising (Ed is really tired; Ed's mind is playing tricks on him; or Ed is stressed), scored 0, and three potentially stigmatising options, drawing on validated child stigma assessments (Ed is weird; Ed is losing his mind; or Ed is crazy), scored 1 (Cameron & Rutland, 2006; Doyle & Aboud, 1995; Greenwood et al., 2016). The wording aimed to avoid obvious alternative explanations (e.g. someone else is calling his name). Scoring 1 (i.e. a stigmatising appraisal, UE-SA) indicated vulnerability.

Jumping to conclusions (JTC)

Two scenarios were created, involving Ed (JTC-Ed) or Chloe (JTC-Chloe) jumping to a negative conclusion regarding a social experience, based on limited evidence (e.g. ‘Chloe has been calling her friend Ed and he isn't answering. Chloe thinks that Ed is ignoring her’; ‘Ed walks over to his seat in class and all the children on his table go quiet. Ed thinks that they are talking about him’). Children rated how certain the character can be of their conclusion on a 5-point Likert scale rated from 1 (‘Ed/Chloe can't know for sure’) to 5 (‘Ed/Chloe is definitely right’). Reduced bias is indicated by lower scores, i.e. shifts away from ‘definitely right’ towards ‘can't know for sure’. Scoring 5 (absolute certainty) on either version indicated vulnerability. The measure was adapted from a developmentally adjusted version of the ‘beads’ task (Ames et al., 2014; Garety et al., 2005).

Analyses

Data were entered into a spreadsheet, checked, and cleaned against paper copies, and analysed using SPSS version 27 (IBM Corp, 2020). Standardised M&MF outcomes were prorated if two or fewer (M&MF-E) or one (M&MF-B) items were missing; otherwise, measures with missing data were excluded on a pairwise basis, with numbers reported for each analysis. To assess fitness for purpose of our measures, we examined all available baseline data for the spread of scores on each measure, and relationships between items and measures, employing reliability, correlational, and principal components analyses as indicated to evaluate consistency with our aim of assessing three differing cognitive vulnerability constructs. Vulnerability criteria were evaluated by comparing cognitive vulnerability and standardised outcome scores between children identified as vulnerable or not, using multivariate and post hoc one-way analyses of variance (ANOVA). Gender and year-group differences in mean scores and vulnerability criteria were assessed using ANOVA and Chi-square, respectively. Change from before to after CUES-Ed, across measures and compared with the naturalistic waitlist, was evaluated using repeated measures ANOVA across time and measure, with group (CUES-Ed/naturalistic waitlist) as a between subjects factor, with vulnerability as a second between subjects factor, covarying for gender and year-group as required. Pre-post effect sizes for change from before to after CUES-Ed intervention/waitlist were calculated as the difference between pre and post scores divided by the common standard deviation, adjusted for a set correlation between scores (.5). Between group effect sizes were calculated as the difference between means for the CUES-Ed and naturalistic waitlist groups divided by the common standard deviation (d; Cohen, 1988). Effect sizes for changes in vulnerable status were converted to d-values using the probit method.

Results Participants

No parents opted out of CUES-Ed or the evaluation; one child declined to complete assessments. Numbers completing each measure at baseline are shown in Table S1; of 960 children participating, 851 completed all measures, 902 completed at least one. At follow-up, 855 completed at least one measure, 816 of these had also completed at least one baseline measure, and 732 completed all measures at baseline and follow-up. Some data were missing, therefore, for 227 children (excluding the child declining to complete measures), due to either absence from the session or poor completion. Half of the participants reported themselves to be boys (50%, n = 477), 47% as girls (n = 452), and 31 young people did not report their gender. Year-groups represented were 3 (ages 7–8 years, n = 275, 29%), 4 (ages 8–9, n = 453, 47%), and 5 (ages 9–10, n = 232, 24%).

Fitness for purpose of measures

Table S1 shows the spread of scores, means, and standard deviations for each of the six cognitive vulnerability measures; absolute values of skewness and kurtosis were all below 1.5; therefore, parametric analyses were deemed acceptable. For stigma-language (the only multi-item measure), reliability analysis indicated that Cronbach's α was acceptable at .8 for the combined scale (n = 902). Inter-item correlations ranged from 0 to .5, item-total correlations from .2 to .5, and α ≤ .8 irrespective of item removal.

Of the 15 possible inter-measure correlations, 11 were significant and of small to moderate size (.1 to .4); a lack of association between JTC (both items) and stigma (self and others) accounted for the four nonsignificant associations (r values < .1, p values > .05).

The pattern of inter-relationships indicated an investigation of components: scores for the six measures (stigma-self, stigma-others, stigma-language, UE-SA, JTC-Ed, and JTC-Chloe) were subjected to a principal components analysis with an oblique rotation (direct oblimin), as the components were not expected to be fully orthogonal. Two components were extracted with eigenvalues >1, together accounting for 50% of the variance in cognitive vulnerability, apparently corresponding to stigma (self, others, or language) and JTC, with UE-SA loading on both factors, with a small inter-factor correlation (r = −.15). UE-SA loaded highly on the third component, with an eigenvalue of .9, accounting for a further 15% of the variance. In the three-factor model, the correlation between factors corresponding to stigma and JTC reduced, and both showed small correlations with the factor corresponding to UE-SA (r = −.2 and r = .1, respectively). The three stigma items were, therefore, combined to create a stigma-total score (Cronbach's α: .8), and the two JTC items were combined to give an overall JTC score. Structure matrices for the two-factor and three-factor solutions are shown in Table S2.

Identifying vulnerable children

Children scoring in the vulnerable range on each measure are shown in Table S1. Informed by the factor analysis, three cognitive vulnerability indexes were formed from the sum of the three stigma vulnerability indicators (stigma-vulnerability: scoring in the vulnerable range on stigma-self and/or stigma-others, and/or stigma-language, range 0–3), choosing a stigmatising UE-appraisal (range 0–1), and the sum of the two JTC vulnerability indicators (JTC-vulnerability: scoring in the vulnerable range on JTC-Ed and/or JTC-Chloe, range 0–2). The distribution of scores on stigma-vulnerability (n = 883) was as follows: 0 (n = 705, 80%); 1 [n = 123, 14% (33 stigma-self, 18 stigma-others, and 72 stigma-language)]; 2 [n = 47, 5% (10 stigma-self and stigma-others; 21 stigma-self and stigma-language; 16 stigma-others and stigma-language)]; and 3 (n = 8, 1%). The distribution of scores on JTC-vulnerability (n = 883) was as follows: 0 (n = 602, 68%); 1 [n = 207, 23% (101 JTC-Ed; 106 JTC-Chloe)] and 2 (n = 74, 8%). Inter-relationships between vulnerability scores are shown in Table S3. The sum of these scores for each child formed a ‘total-vulnerability’ score, ranging from 0 to 6, with overall vulnerability indicated by scoring in the vulnerable range on any measure (‘any-vulnerability’). Total-vulnerability covaried significantly with all three cognitive vulnerability mean scores [MANOVA F(3, 847) = 638.1, p < .001], with significant one-way associations with each separate measure [sigma-total: F(1, 849) = 328.8, p < .001; UE-SA: F(1, 849) = 376.9, p < .001; JTC: F(1, 849) = 288.4, p < .001]. Total-vulnerability also covaried with standardised wellbeing and emotional/behavioural difficulties, such that outcomes worsened with increasing vulnerability [MANOVA F(3, 790) = 11.2, p < .001], with significant one-way associations with wellbeing [F(1, 792) = 14.9, p < .001] and behavioural problems [F(1, 792) = 29.9, p < .001] and a trend towards an association with emotional problems [F(1, 792) = 3.5, p = .06]. Any-vulnerability was also associated with all cognitive vulnerability mean scores [MANOVA F(3, 847) = 251.2, p < .001; one-way ANOVAs: stigma-total F(1, 849) = 122.9, p < .001; UE-SA: F(1, 849) = 282.4, p < .001; JTC: F(1, 849) = 203.2, p < .001], and all standardised outcomes [MANOVA F(3, 791) = 9.2, p < .001; one-way ANOVAs: wellbeing: F(1, 793) = 12.0, p = .001; emotional problems: F(1, 793) = 5.1, p = .02; behavioural problems: F(1, 793) = 24.7, p < .001]. Mean cognitive vulnerability and standardised outcome measure scores for each vulnerability indicator, any combination of indicators, and combinations of different indicators (i.e. stigma, UE-SA, and JTC) are shown in Table S3, with one-way ANOVAs. Being identified as vulnerable on any index was generally associated with greater cognitive vulnerability across measures, lower wellbeing, and more emotional/behavioural problems, with poorer outcomes as vulnerability increased but with stronger associations of emotional problems with stigma-vulnerability than with UE-SA or JTC-vulnerability (Table S3).

Gender and year-group differences

Differences in cognitive vulnerability according to gender and year-group are shown in Table S4. MANOVA across all three cognitive vulnerability measures showed a significant effect of both gender [F(3, 827) = 11.6, p < .001] and year-group [F(6, 1656) = 0.009], but no interaction effect [F(6, 1656) = 0.5, p > .8]. Boys were significantly more vulnerable on every measure than girls, and older children made less stigmatising appraisals of unusual experiences [F(2, 829) = 6.0, p = .003], with no significant differences for stigma-total or JTC (F values < 1.7, p values > .1). Boys were also more likely, and older (9–10 year old) children less likely, to be identified as vulnerable, for each indicator and across indicators (χ2 values > 4.0, p values < .03), with the exception of JTC, where vulnerability was nonsignificantly lower for older children.

Change in cognitive vulnerability Mean cognitive vulnerability scores

Repeated measures ANOVA on the three cognitive vulnerability scores covarying for gender and year-group, with group and any-vulnerability as between subjects variables, revealed no overall significant effect of time [F(1, 716) = 1.3, p = .2], a time × group interaction [F(1, 716) = 22.7, p < .001] that also differed by measure [F(2, 715) = 15.5, p < .001] and a time × group × vulnerability interaction [F(1, 716) = 4.7, p = .03] that did not differ by measure [F(2, 715) = 2.4, p = .09]. Not controlling for gender and year-group changed the pattern of results only in that the effect of time reached significance [F(1, 728) = 19.8, p < .001]; time × group [F(1, 728) = 23.5, p < .001] and time × group × vulnerability [F(1, 728) = 4.4, p = .04] remained unchanged. Exactly the same pattern of results was found for a three-group coding of vulnerability (0, 1, and 2+ vulnerability indicators), irrespective of controlling for gender and year-group, with a slightly larger interaction effect [time × group × vulnerability F(2, 714) = 4.8, p = .008] but with a significant interaction with measure [F(4, 1428) = 4.0, p = .003]. Analysis of higher levels of vulnerability was not possible due to the small size of groups in the naturalistic waitlist condition. Change in mean scores for any-vulnerability and increasing levels of vulnerability are shown in Table S5; change was generally greater for more vulnerable children, with small between-group effects for stigma, small to medium for UE-SA, and large for JTC. All between-group effects favoured CUES-Ed except for children with a single stigma vulnerability indicator for whom there was a very small difference favouring waitlist. Further analysis of stigma scores suggested the limited effects were primarily for stigma-other (ES: −0.09; 95% CI: −0.72, 0.54): stigma-self and stigma-language showed small to medium between-group effects (0.45, 95% CI: −0.18, 1.08; 0.33, 95% CI: −0.3, 0.96) respectively. There were no overall effects of year-group or gender, or significant main interactions with time (F values < 3, p values > .05). However, there was a tendency towards a time × gender interaction, such that boys tended to show more change than girls [F(1, 714) = 2.9, p = .09], and a time × year-group × measure interaction, whereby 9–10 year olds changed more on stigma and JTC, but not UE-SA [F(2, 713) = 3.3, p = .04].

Change in vulnerable status

Of 323 children with any vulnerability pre-CUES-Ed, 209 (65%) were no longer vulnerable after CUES-Ed, compared with 10/27 (37%) of naturalistic waitlist children (d = .7). Although there were some children switching from no vulnerability to vulnerable, these were fewer in the CUES-Ed group compared with the waitlist (42/360, 12%; 5/27, 19%, d = .3). In children identified as vulnerable, total-vulnerability reduced significantly more over time for the CUES-Ed group (pre mean = 1.7, SD = 1.0; post mean = 0.5, SD = 0.9) compared with the waitlist group (pre mean = 1.7, SD = 0.7; post mean = 1.0, SD = 1.1), F(1, 338) = 4.2, p = .04, d = −.6. Results were the same when gender and year-group were not controlled [F(1, 345) = 4.4, p = .04]. Change across vulnerability indicators by group is shown in Tables S5 and S6.

Discussion

We set out to evaluate, in a service setting, fitness for purpose of our assessments of cognitive vulnerability and our methods for identifying vulnerable children, as well as change following our CUES-Ed intervention. We expected cognitive vulnerability to reduce following CUES-Ed, and compared with a naturalistic waitlist, with greater change for vulnerable children. We found that assessments had an adequate structure, and vulnerability indicators generally identified young people reporting more difficulties on standardised outcomes, as well as greater cognitive vulnerability. Girls and older children tended to have lower cognitive vulnerability scores. Cognitive vulnerability reduced from before to after CUES-Ed, with very small to large pre-post and between-group effects, depending on the measure. Effects were generally greater for vulnerable children. We conclude that CUES-Ed may have some promise in changing factors associated with future mental illness and thus may have potential to reduce vulnerability. Our preliminary service findings support further controlled evaluation.

Assessment of cognitive vulnerability

We designed age-appropriate, engaging assessments of the cognitive vulnerability mechanisms targeted in the CUES-Ed intervention and presented these to children in a ‘workbook’ format, in keeping with the classroom setting. We measured appraisals of emotional expression and UEs that may be linked with reluctance to share experiences with others and thus limit access to help, as well as the JTC reasoning bias thought to underlie unhelpful appraisals. Children appeared to be able to respond meaningfully to the workbook questions; responses approximated a normal distribution. Reliability and factor analysis suggested the items assessed the intended constructs and related to each other as expected. Mechanistic change has been insufficiently addressed in school-based interventions to date, in part due to a lack of appropriate measures (Dray, 2021). Our work goes some way towards addressing this need. Around half of children had at least one vulnerability, with decreasing frequency of cumulative vulnerabilities. Prevalence of vulnerabilities ranged from 6% for stigma-others to 20% for UE-SA and JTC, although around 20% showed any stigma vulnerability, suggesting a similar rate across the three factors. Rates concord with adult population estimates for JTC (Freeman, Pugh, & Garety, 2008). Children identified as vulnerable showed a range of other vulnerabilities and scored more highly on standardised outcomes, particularly behavioural problems, with associations between vulnerability and wellbeing present but less robust, and associations with emotional problems primarily for stigma. The associations are consistent with a degree of validity of the vulnerability indicators. Boys and younger children generally scored as more vulnerable than girls and older children. Research into later, high risk trajectories in adolescence and young adulthood highlights individual variability: behavioural and functional predictors are important; self-reporting distress is not a reliable predictor of future problems and may indicate a level of recognition and communication of difficulties that is uncharacteristic of early presentations of conditions such as psychosis (Ajnakina, David, & Murray, 2019; McGorry, Hartmann, Spooner, & Nelson, 2018; Montemagni, Bellino, Bracale, Bozzatello, & Rocca, 2020; Power, Polari, Yung, McGorry, & Nelson, 2016). Gender differences in at-risk presentations have also been found, with young men showing more behavioural and communication difficulties (Rietschel et al., 2017). A recent review has confirmed that cognitive biases are common and are linked with severity of presentation (Livet, Navarri, Potvin, & Conrod, 2020). Our service data may therefore be consistent with early, population level manifestations of vulnerability that, within a continuum model, may combine with other vulnerabilities to cumulatively increase risk (van Os & Linscott, 2012).

Change after CUES-Ed

We examined in-service change from before to after CUES-Ed, and compared with a naturalistic waitlist group, in mean cognitive vulnerability scores and vulnerability indicators, for whole classes and for children showing different levels of cumulative vulnerability. Generally, outcomes favoured CUES-Ed, with large pre-post and between-group effects for JTC, small to medium effects for stigmatising appraisals of UEs, and very small to small effects for stigma, with one outcome favouring the waitlist by a very small margin. Overall findings are consistent with change in targeted processes following CUES-Ed. Importantly, changes were greater for vulnerable children. While this might be expected in pre-post outcomes, as the baseline variance is reduced by selection, this does not apply to the between-group findings that still evidence change. As school-based interventions have sometimes improved outcomes for less vulnerable children, whilst increasing risk for the most vulnerable (Dray et al., 2017), this is a key potential strength of CUES-Ed and is consistent with our earlier report of small but consistent improvements in emotional/behavioural difficulties and wellbeing for vulnerable children following CUES-Ed (Redfern et al., 2019). Moreover, our data strongly suggest vulnerability status change following CUES-Ed, such that children move from a vulnerable to nonvulnerable category. While there is noise in the data, and children also moving from nonvulnerable to vulnerable, advantages were evident for CUES-Ed, with large effects in changing overall vulnerability status compared with the waitlist group. The findings are promising and support future controlled evaluation.

Changes were smallest for stigma, but on further analysis, this seemed to be driven by a lack of change in stigma-other, with small to medium changes for the other two components. There is a dearth of research into what is communicated to children about mental illness (Mueller, Callanan, & Greenwood, 2016) and the development of mental illness stigma during childhood (Hinshaw, 2005), with targeted interventions typically demonstrating limited impact (Mueller et al., 2016). Our findings may indicate a need to review CUES-Ed content to clarify the intended learning for children. It may also be that stigmatising views of others is less well related to the other items than the baseline associations suggest. The intention of this content is to improve the classroom environment in relation to emotional expression and support an ethos of talking about feelings – and this tolerance of others may not be so well associated with change to acceptance in oneself.

The large effect sizes for JTC suggest that CUES-Ed achieves its aim of promoting cognitive flexibility, both for the whole class and for vulnerable children. JTC has been linked with co-occurring affective problems and unusual experiences, but otherwise appears to be a specific psychosis risk factor in adults (Reininghaus et al., 2019; So, Siu, Wong, Chan, & Garety, 2016), although its significance in children clearly requires further research to determine.

Limitations

The study has many limitations. First, the service context means participants were limited to self-selecting schools opting into the intervention. The naturalistic waitlist data were acquired opportunistically with no safeguard against selection bias. There are large amounts of missing data due to children's absence and poor completion, and this is a potential source of bias. Children were not deliberately sampled across gender and year-group, which may be another source of bias for these analyses. Assessments were bespoke, designed by the CUES-Ed team, for the purpose of capturing intervention outcomes. While not our intention, there may be concordance between intervention and assessment that inflates effects, although the previously reported positive outcomes on standardised measures are consistent with our cognitive vulnerability findings. Completion post CUES-Ed may be also be influenced by familiarity with the characters. Aside from the standardised outcomes, there was no other validity criterion for the assessment of vulnerability, and this will be required in future research. For analyses involving children identified as vulnerable, the effect of limiting the variance in outcomes though selection should be considered, and between-group analyses give a better indication of effects. Furthermore, the restricted range of vulnerability scores in the waitlist group (perhaps due to small size or opportunistic selection) may have limited the degree of change, further biasing outcomes.

Conclusions

We evaluated our measurement of cognitive vulnerability and change after CUES-Ed, our universal, classroom-based early intervention CBT programme for primary school children. Findings support the appropriateness of our developmentally adapted and contextualised assessments of cognitive vulnerability, and the potential of the intervention to reduce vulnerability in targeted mechanisms, which may in turn improve future mental health. Controlled evaluation is indicated, not only of standardised outcomes but also of targeted cognitive vulnerabilities. Longitudinal research is required to clarify the significance of cognitive vulnerability in childhood and the trajectories of those identified as vulnerable.

Acknowledgements

CUES-Ed was established using routine Trust funding; delivery was commissioned by schools and the London Borough of Southwark. Funders have not influenced the design or interpretation of this evaluation, which is solely the work of the authors. We thank children, parents, and staff at participating primary schools. The data that support the findings of this study are available from the corresponding author R.U., upon reasonable request. The code is available from the corresponding author by request. The authors have declared that they have no competing or potential conflicts of interest.

Ethical information

Service evaluation approval, and confirmation that ethical approval was not required, was given by the South London and Maudsley NHS Foundation Trust CAMHS Clinical Academic Group Audit Committee reference #2014-08.