1. Study design
Enabee is a repeated national cross-sectional study conducted in schools in France using self-administrated questionnaires. The protocol for the first edition, Enabee 2022 - which was conducted in the same year exclusively in metropolitan France (i.e., no French overseas territory was involved) - is described in detail elsewhere (21). Enabee 2022 aimed to establish a comprehensive set of indicators concerning children’s mental health from 3 to 11 years old. In France, elementary school for pupils aged 6-11 comprises five levels called CP, CE1, CE2, CM1, CM2, which correspond to 1st to 5th grades in the US system. As part of Enabee 2022, children from these five grades had to fill in a self-administered questionnaire which included the DI.
2. Data collection
a) Procedure
A probabilistic multistage stratified sampling plan was used to constitute the Enabee 2022 study sample. First, schools were randomly selected. Second, up to four classes per school were randomly selected. All pupils within each selected class were eligible. In schools with four classes or fewer, all classes were selected. School principals, teachers and parents received an information letter about the study. A school’s participation was contingent upon the agreement of the school principal; parents could disagree to their child’s participation. Children whose parents agreed to their participation could independently refuse to answer the questionnaire or to interrupt answering at any time.
Data from the child questionnaire were collected during dedicated sessions held in participating schools between 2 May and 5 July 2022. Survey facilitators visited each class in pairs, equipped with a tablet and headphones for each participating child.
During the answering sessions, children were first provided with general instructions explaining the purpose of the study and the rules. In particular, the survey facilitators emphasized that all answers would be confidential, and assured the children that there were no wrong answers. Children could ask questions before and during the session, but the survey facilitators were not allowed to interfere with the children’s answers.
b) Child self-reported mental health using the DI
Self-reported mental health was assessed using the DI (13). The original version of the DI comprised 91 questions related to the symptoms of seven common mental disorders observed in children as described in the DSM-5: attention-deficit/hyperactivity disorder (ADHD), conduct disorder (CD), oppositional defiant disorder (ODD), specific phobias (SPh), separation anxiety disorder (SAD), generalized anxiety disorder (GAD), and major depressive disorder (MDD). Ten questions focus on the child’s strengths and competencies, unrelated to any of the seven disorders. Moreover, three components of ADHD can be distinguished: inattention, hyperactivity and impulsiveness; accordingly, DI items relative to ADHD can be separately analyzed (22). In order to improve the acceptability of the tool by school authorities, the steering committee of the Enabee 2022 study suggested to remove all 13 questions of the DI specifically related to CD. Accordingly, the study only investigated six of the DI’s seven mental health disorders.
At the beginning of the DI, children were asked to choose an avatar from many options. The aim of proposing different avatars to choose from was to foster self-identification with the avatar finally chosen, with a view to improving the quality of the child’s answers. For all children, the chosen avatar was called Dominic (note: Dominic is a unisex name in France). A tutorial was provided to ensure that children understood how to answer the various questions asked. Each question had to be answered before proceeding to the next one. A short scene, illustrating the emotional and behavioral symptoms being described, accompanied each query. Children responded “yes” or “no” to the 78 questions in a variety of pictorial scenarios, according to whether they thought, felt, or acted like Dominic.
Symptom scores were obtained by adding 1 point for each reported symptom of the evaluated disorder. For each disorder, thresholds were used to classify children into three groups: no disorder, possible disorder, probable disorder. These thresholds were based on the mean and standard deviations collected from the sample of 585 children living in Montreal in 1992 who were the first children to complete the DI in 1999 (15). Specifically, children whose number of reported symptoms was between one or two exclusive standard deviations above the mean in the Montreal sample were categorized as ‘possibly’ presenting a disorder; those with two standard deviations or more above the mean were categorized as ‘probably’ presenting a disorder. Generally, in epidemiological studies, the latter threshold is used to define the prevalence of mental health disorders (16,18,20). In the present analysis, (and all Enabee 2022 analyses), we followed suit in order to ensure comparison with other studies.
c) Covariates
We used data from the databases of the French Ministry of Education to collect the following information: child’s gender, grade (i.e., from 1st to 5th), aged-related school type (i.e., elementary – 6 to 11 years old, primary – 3 to 11 years old), type of school (i.e., public or private under contract with the French Ministry of Education), and whether the participating schools were officially recognized as being located in a socially deprived (Réseau d’éducation prioritaire, REP hereafter) or highly socially deprived (REP+ hereafter) area. The REP and REP+ programs are systems of educational support implemented in French public schools, aimed at addressing educational inequalities by providing additional resources and support to schools in socially deprived areas.
3. Study Sample
As shown in Figure 1, of a total of 554 primary/elementary schools randomly selected, 326 agreed to participate (participation rate 58.8%). Class participation rate was 93.5%, representing 19,725 children. The following children were not eligible to participate: those whose parents objected to the study, those whose parents did not receive information about the study at home, those who moved house, and those whose teacher considered that they were not able to properly complete the child questionnaire due to a disability or too poor a French linguistic level.
For the statistical analysis, we further excluded children with incomplete DI and those with inconsistent responses. Finally, a total of 15,206 children were included in the final analysis.
To enable statistical inference of the results, we calculated weights for the sample of children who fully completed the child questionnaire, taking into account the observed total non-response at each sampling stage (i.e., school, class, and individual). A calibration step was then performed using margins derived from the reference sample (children in 1st to 5th grade in public schools or private schools under contract) to ensure the study population could be compared with the reference sample in terms of the individual child’s gender, the type of school (public or private under contract), and grade (i.e., 1st to 5th grade). We subsequently applied a truncation step to avoid excessive dispersion of the weights.
4. Statistical analysis
a) Descriptive analysis
The statistical descriptive analysis described the main characteristics of the study population, taking into account the sampling design (i.e., weights, stratifications and sampling stage).
b) Assessment of DI psychometric properties
We first examined the ceiling and floor effects for each of the 78 questions in the DI. For each of the six mental health disorders examined, we estimated the internal consistency using Cronbach’s alpha and Mc Donald’s Omega. Cronbach alphas were also assessed for gender and for grade. To confirm the factorial structure, we reproduced Kuijpers’ approach by performing a categorical confirmatory factor analysis (CCFA) for each subscale (19). We used conventional limits for fit indices as follows: a CFI greater than 0.90 and an RMSEA less than 0.08 were considered acceptable while a CFI greater than 0.95 and an RMSEA less than 0.05 were considered good (23).
R software was used to perform this assessment, specifically the survey (24) and EFAtool packages (25), to assess Cronbach alphas and Mc Donald Omegas with a categorical CFA, and the lavaan package for the CCFA (26). Weights were used for the analysis with the WLMSV estimator. It was not possible to specify a complex sampling plan with categorical data in the calculation of the McDonalds omega with the lavaan package. In contrast, we were able to specify a complex sampling plan for the calculation of Cronbach’s alpha.
c) Prevalence assessment
We computed the prevalence of ‘probable’ cases and their 95% confidence intervals (CI) for each ‘probable’ mental health disorder assessed by the DI. We also measured the percentages and their 95% CI for probable cases of the following four indicators: at least one probable internalizing disorder (i.e., MDD, SAD, GAD, or SPh), at least one probable externalizing disorder (i.e., ADHD or ODD), at least one probable mental health disorder (i.e., any one of the six studied), and at least one probable internalizing and one probable externalizing disorder.
We also described the statistical distribution of each mental health disorders and each three components of ADHD (inattention, hyperactivity and impulsiveness) independently.
We assessed prevalence according to gender, grade, and type of school (private school under contract or public school in an non-REP or REP+ area, public school located in REP area, public school located in REP+ area), and compared them using Pearson chi-square tests.
Finally, we examined co-occurrences between each probable mental health disorder assessed by the DI and identified the most frequent associations. We computed the Pearson correlation of the sum of the item scores for each of the six diagnostic-specific subscales of the DI.
These various statistical analyses were performed taking into account the sampling design of the study including non-response adjusted weights, strata and the finite population correction factor. They were performed using SAS Enterprise Guide version 7.11.
5. Ethical considerations
The study was approved by the French authority for data protection (Commission nationale informatique et libertés, CNIL, authorization demand n° 921423, deliberation DR-2022-009, 7 January 2022), as well as a French ethics committee (Comité éthique et scientifique pour les recherches, les études et les évaluations dans le domaine de la santé, CESREES, decision of october 14, 2021, n° 5268850). The study also received approval from the French council for statistical information (CNIS, avis n° 85/H030) and the Committee of Public Statistics (n°2022_11193_DG75-L002).