We conducted a randomized controlled multicenter trial in the Sri Jayewardenepura Education Zone in the Colombo district, Sri Lanka. The target population was children in grade nine (aged 13–15 years). Each school was considered a cluster, and a randomly selected class of nine grades was selected from each school for the study. The average number of students per group was 20. The sample size and number of groups for each arm were decided using a formula proposed by Hayes Moulton (2017) for an individually randomized trial with inflation for the design effect for a cluster randomized trial [12]. The values for the true mean and standard deviation of the outcome variable in the presence (µ1 = 7.35, σ1 = 6.93) and absence (µ0 = 9.58, σ0 = 6.44) of intervention were informed by Barrett and Turner (2001) in their study on reducing anxiety using a CBT-based package. For alpha (Zα) and beta (Zβ) errors, the values used were Zα = 1.96 for α = 0.05 and Zβ = 0.84 for β = 0.80 [13]. According to Shackleton et al. (2016), the ICC for psychological outcomes ranges from 0.01 to 0.07; we used the higher end of this range, setting ICC (ρ) at 0.07 to maximize sample size [14]. All Sinhala medium schools with year nine classes were invited to participate in the study; 37 schools agreed upon the study, and one school withdrew prior to randomization. A total of 720 students from 36 schools and their parents provided assent and informed written consent, respectively.

The trial was registered in the Sri Lanka Clinical Trial Register (SLCTR/2018/018) and was approved by the Ethics Review Committee of the Faculty of Medicine, University of Kelaniya (P/19/01/2018). Administrative approval was obtained from the Ministry of Education, Sri Lanka.

The consented schools were stratified by school gender type, i.e., male, female, or mixed. Within each stratum, schools were randomized to the universal intervention group based on CBT or the control group using the block randomization method to ensure allocation of schools in a 1:1 ratio to two groups (Fig. 1). Schools, students, and the research team were masked in the allocation.

Student progress through the trial– CONSORT flow chart

Schools that were randomized into the intervention arm received a universal CBT-based intervention package that was developed based on the UK Medical Research Council guidance by De Silva et al. [15]. The intervention was delivered by a teacher who underwent a comprehensive training program covering CBT principles, specific techniques, and strategies for engaging students. To ensure consistency, the training included role-playing exercises, peer reviews, and feedback sessions. Teachers were provided with a detailed handbook outlining session plans and activities. The delivery of the intervention was monitored using a fidelity monitoring framework involving regular supervision by experienced mental health professionals. Teachers submitted weekly reports, and random observations were conducted to assess adherence to the intervention protocol. Any deviations were addressed through additional support and training, aiming to minimize variations and maintain the integrity of the intervention across different schools. Each teacher delivered the intervention package to randomly selected classes in the intervention arm in a 40-minute session per week for eight consecutive weeks and for a subsequent one-month duration to practice what they learned. An additional session was arranged for students who missed a session prior to the next session. During these eight sessions, the children were trained in CBT-based skills that were aimed at reducing anxiety and relaxing. The outline of the intervention package is shown in Table 1, and its details are published elsewhere by De Silva et al. (2023) [15]. They were instructed to practice these skills under the close and distant supervision of the teacher.

A 40-minute session was held for parents by the same teachers, overlapping with the routine parent meeting. Parents were educated about anxiety, anxiety disorders, how to identify them, and how to adapt to reduce anxiety. This session aimed to support parents in practicing the anxiety reduction skills learned by their children at home. Materials included the Teachers’ Handbook, the Workbook for Children, and a monitoring tool for students’ practice, along with a leaflet for parents. The intervention was monitored using the process evaluation framework (PEF) at the school level [15]. The PEF of the intervention group is provided in Additional File 1 Table S1.

We acknowledge the potential for performance bias due to the lack of blinding among intervention deliverers and participants. While blinding presents challenges in school-based interventions, several measures were taken to minimize bias and ensure the study’s internal validity. The rigorous training, monitoring procedures, and use of validated assessment tools, such as the SCARED tool for anxiety and the Rosenberg self-esteem scale, helped reduce subjective bias. Despite the difficulties in implementing blinding, these steps significantly contributed to the robustness of our findings. Future studies should explore feasible blinding strategies to further strengthen internal validity.

Routine schoolwork was carried out in randomly selected classes from the control arm. The participants were assessed at baseline, after the intervention, and at follow-up.

Both groups were assessed at baseline using a self-administered questionnaire consisting of sociodemographic questions, a tool to measure the level of anxiety, the state of depression, and the level of self-esteem. The participants’ level of anxiety, depression status and self-esteem were assessed postintervention, as well as at three months of follow-up, using the same tools as follows.

The child version of the Screen for Child Anxiety Related Disorders (SCARED) is a 41-item instrument rated on a 3-point Likert scale. This instrument has been validated for assessing children aged 8–18 years [16, 17]. The translated and validated questionnaire, which has high reliability according to Cronbach’s alpha (0.87) and test-retest correlation coefficient (0.74), was used to assess the level of anxiety [18]. A higher score indicates a greater level of anxiety.

Depression scales of the validated DAS-21 questionnaire were used in the local setting among adolescents by Weerasinghe in 2012. This approach has been validated for use in Sinhalese adolescents, as it has good psychometric properties. All the subscales had good reliability (Cronbach’s alpha > 0.7). Its cut-off value for the depression subscale was 19, with a sensitivity = 80% and specificity = 83%. [19]. The state of depression was determined using this tool.

The Rosenburg self-esteem instrument was validated for use among adolescents internationally. The NRS-2002 is a 10-item scale that measures global self-worth by measuring both positive and negative feelings about the self. The scale is believed to be unidimensional. All the items are answered using a four-point Likert scale ranging from strongly agreeing to strongly disagreeing. The scores were on a continuous scale ranging from 10 to 40. A higher score indicates greater self-esteem [20, 21].

All the analyses were based on intention-to-treat principles. All baseline categorical variables are presented as numbers and percentages, and all numerical variables are presented as the means and standard deviations (Table 2). As the first step of the pre specified statistical analysis plan (SAP), all the outcome variables were compared between two groups using conventional statistics with their effect sizes. Those results are presented in the (Additional File 1 Table S2 to Table S4).

The outcome variables were analysed to determine the effectiveness of the intervention using marginal linear regression with coefficients estimated by the generalized estimating equation (GEE). This approach allowed us to control for the effect of clustering and adjust for imbalanced covariates and probable confounders identified at the beginning of the study. Specifically, we considered several covariates and confounding variables, including sex, age, ethnicity, religion, permanent and current residence, school functional type, attendance at tuition classes, engagement in extracurricular activities, having siblings, anxiety sensitivity, behavioral inhibition, mother’s level of education, mother’s and father’s occupation, and perceived parenting style of the mother. By adjusting for these covariates, we aimed to isolate the effect of the intervention from other factors that might affect anxiety, depression, and self-esteem levels among the participants, thereby minimizing bias and improving the internal validity of our findings. The marginal models allow the effect of the explanatory variables on the outcome, and the correlation between observations is modelled separately. Each outcome variable was analysed after the intervention and at the follow-up; in other words, separate marginal models were used for each outcome variable after the intervention and at the follow-up. Only the relevant anxiety levels from the outcome tables are shown here (Tables 3 and 4), and the anxiety levels for depression status and self-esteem are shown in the supplementary material. In each marginal model, the scores of the outcome variable were used as the dependent variable, the arm of the study (intervention or control), the baseline value of the outcome variable and the baseline covariates that needed to be adjusted were used as predictors. Clusters were used as subjects to control for the nested/cluster effect. SPSS 22 was used for the data analysis.