Physical, mental and behavioral health indicators in relation to academic performance in European boys and girls: the I.Family study

Table of Contents

Data and study design

The I.Family study is based on a cohort from nine European countries: Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain, Sweden and Poland [30]. This cohort was initially recruited from eight countries during the IDEFICS study from 2007 to 2008, when the children were two to ten years old [31]. The I.Family study subsequently expanded this cohort to include a new country (Poland), as well as including children’s siblings and parents, which allows for a more comprehensive analysis of family health dynamics. In all survey waves, data were gathered through physical examinations, including anthropometric measures, and self-administered questionnaires completed by adolescents aged 12 or older [32]. Questionnaires were identical across the nine countries, with slight adaptions for language and cultural contexts. A more detailed description about the cohort, including recruitment strategies and representativeness, can be found in Ahrens et al. [30, 31].

As described earlier, our theoretical conceptualization assumes that various health indicators may influence each other and jointly impact academic performance (e.g., screen time or sleep may impact academic performance indirectly through physical and mental health indicators). The present analysis is based on data from the third examination wave (I.Family, 2013–2014) when academic performance outcomes could first be reported by adolescents in each survey country. Although earlier-life exposure data is available in many of the original participants, the present sample size was optimized with data from siblings who were recruited at the third examination. Consequently, we opted for a cross-sectional design, positioning this study as an exploratory analysis of how different health indicators collectively associate with academic performance.

Figure 1 illustrates our analytical approach, in which we identified eight health indicators based on their relevance to physical, mental, behavioral and other sleep-related aspects of adolescent health, ensuring a comprehensive assessment of available data. These indicators include mental well-being (comprising the KINDL health-related quality of life [HRQoL] score and stressful life events), physical fitness (represented by BMI), sleep (with separate indicators for duration and quality), and behavioral indicators (including healthy diet score, media use, and sports club participation). Sleep is treated as a separate domain due to its dual role as both a physical and behavioral factor, influencing both daily functioning and overall well-being. This distinction allows for a more nuanced understanding of sleep’s unique contribution to academic performance.

Rather than modeling complex interdependences among these indicators (e.g., through path analysis), we use multivariable regression models to assess the effect of each health indicator while accounting for the presence of others, or more formally, conditional on other indicators. Given our study design and available data, this approach allows us to explore key associations while recognizing the need for further research to establish more intricate relationships. Adolescents aged 12 to 16.9 years with non-missing academic performance data were included in the analysis (N = 3,388).

The IDEFICS and I.Family study protocols adhered to the Declaration of Helsinki ethics principles for research that include human subjects. Participant inclusion required that parents give written informed consent for their children, and young children also gave verbal consent. From the age of 12 years onwards, children gave their own written informed consent in plain language. We followed the general data protection regulation (GDPR). Data are stored on a secure central server with password encryption, and samples are stored in secure biobanks. This study was coordinated by the Leibniz Institute for Prevention Research and Epidemiology – BIPS (Germany), and survey centers obtained ethics approval from respective local institutions in each country. Below, we provide detailed information on all the measures.

Measures

Outcomes: academic performance

To measure academic performance, we used self-reported grades in mathematics and their own country’s native language, as these two subjects are most consistently graded across curricula in all participating countries. Grades were harmonized using the five-point European Credit Transfer System (ECTS) [33] to standardize them into five levels: excellent, good, satisfactory or mediocre, pass or sufficient, and fail or insufficient.

For this study, we dichotomized these levels into a binary indicator for each subject, categorizing grades as either higher performance (i.e., excellent and good) or all other levels. This resulted in one indicator per subject (mathematics and language) representing whether the participant reported higher performance in that subject. This decision was based on both theoretical and empirical considerations. Conceptually, the middle and lower performance categories may reflect shared risk factors for underachievement and combining them supports a clearer interpretation focused on academic success. Empirically, the number of participants in the lowest category (“fail or insufficient”) was small across subgroups (≤ 6%), limiting model stability. In follow-up sensitivity analyses, we also explored ordinal logistic models but found that the proportional odds assumption was significantly violated in all sex- and subject-stratified models, with extensive data sparsity and poor model fit. These findings reinforced the use of a dichotomized outcome to ensure robust and interpretable comparisons.

Predictors of interest: mental, physical, sleep-related, and behavioral health indicators

As described earlier, we identified eight indicators encompassing physical, mental, sleep-related, and behavioral aspects that may relate to self-reported academic performance (Fig. 1).

First, the mental well-being aspect include the KINDL HRQoL score and an assessment of stressful life events. The KINDL questionnaire, a widely used instrument for measuring health-related quality of life (HRQoL) in children and adolescents [34, 35], serves as the basis for the mental well-being aspect in this study. Since HRQoL captures individuals’ subjective perception of well-being, the KINDL score is considered a valid and meaningful indicator in this context. The questionnaire consists of 12 questions across four domains: emotional well-being, self-esteem, family, and social contacts (see Appendix for questions). Respondents were asked to report on how they had been feeling during the past week. Each domain was scored on a scale from 0 to 12, with an overall score ranging from 0 to 48, where higher scores indicate better well-being.

For stressful life events, participants reported whether they had experienced any of 13 potentially stressful events ever, such as parental divorce, the death of a parent, or the death of a family member (see Appendix for the complete list) [32]. We acknowledge that these events may have occurred at different points in time. However, previous research has shown that both long-term and recent stressful experiences contribute to mental well-being, with potential cumulative effects over time [36,37,38]. If they had experienced an event, they indicated the level of distress it caused, either “rather strongly” (score = 2) or “rather little” (score = 1). We then calculated a total score by summing these responses across all 13 events, yielding a possible range from 0 to 26, with higher scores indicating greater distress from these life events.

Second, body mass index (BMI) was used as an indicator of body composition, a key dimension of physical health. Trained personnel measured participants’ weight to the nearest 0.1 kg using a TANITA BC 420 MA scale and height to the nearest 0.1 cm using a SECA 225 stadiometer [30, 39]. Participants were measured early in the day wearing only light clothing. Age- and sex- specific BMI z-scores were then constructed based on an external standard population to standardize participants’ weight status [40].

Third, sleep indicators included nocturnal sleep duration and sleep quality. Participants reported their usual bedtime and wake-up time on both weekdays and on weekends [32]. We calculated nocturnal sleep duration as the average hours per night across both weekdays and weekends. For sleep quality, participants evaluated their sleep during the past month on a four-point scale: very good, fairly good, fairly bad, and very bad [32]. We then created a binary indicator, with “good” sleep quality defined as either very or fairly good (1) and “not good” as fairly bad or very bad (0).

Lastly, behavioral indicators include a healthy diet adherence score, media use, and involvement in physical activities in leisure time. The healthy diet adherence score was derived from frequency data in the validated Food Frequency Questionnaire, where participants reported how many times they had consumed specific food items in the past month. The score comprised five sub-scores reflecting adherence to nutrition guidelines on consumption for fruit and vegetables, fish, whole grains, fat quantity, and sugar. The overall score ranged from 0 to 50, with higher values indicating better adherence to these guidelines [32, 39]. This questionnaire has been previously assessed against 24-h dietary recall data in validation studies, demonstrating its reliability in estimating dietary intake patterns [32].

Sedentary media use was measured by the self-reported number of hours spent with audiovisual media, including personal computers, game consoles, and televisions, over an entire week. Participants were asked, “How long do you usually watch TV and/or video/DVD per day?” and “How long do you usually sit at a computer/game console per day?” They provide answers separately for weekdays and weekends. We then calculated the total hours spent on these activities over the entire week.

Sports club participation was used as a proxy for physical activity. Participants were first asked whether they were members of a sports club. If they answered “no,” their weekly time spent engaging in sports at a sports club was recorded as zero. If they answered “yes,” they were then asked, “How much time do you usually spend doing sport in a sports club per week?”. This measure was used as a proxy for their level of organized sports in leisure time (leisure time physical activity, LTPA). While objective accelerometry data was not available for all countries, and was missing in a substantial number of cases, sports club participation was considered a valid, subjective proxy for physical activity. Among participants with available data on moderate-to-vigorous physical activity (MVPA), a small but highly significant correlation was found between sports club participation and MVPA (0.19, p < 0.001). Given its availability of data across all countries, we used sports club participation as a comprehensive measure to preserve sample size and ensure consistency across countries in the analysis.

Control variables/covariates

In all multivariable models, we included parents’ education and income, as well as adolescents’ age, as covariates. The parent completing the questionnaire reported both their own and their partner’s educational attainment. The highest level of education between the two was categorized according to International Standard Classification of Education (ISCED) [41] to ensure comparability across countries. Parental education was further categorized into three levels: low, medium, and high. Reported family income was standardized across participating countries by linking each household’s income to the average country-specific net household income. This income was then categorized into four levels: low/medium, medium, medium/high, and high [31].

Statistical analysis

We started with descriptive analyses, comparing academic performance and all mental, physical, sleep-related, and behavioral indicators with stratification on sex. Statistical comparisons were conducted using Wilcoxon rank sum tests and Pearson’s chi-square tests.

To examine the associations between these indicators and academic performance, we employed binary logistic regression models. To account for potential country-specific effects, we used logistic regression models with country fixed effects, with eight dummy variables to represent the nine countries. In all models, parents’ education, income, and adolescents’ age were included as covariates. Additionally, to account for potential sex-related heterogeneity and difference between mathematics and language subjects, we conducted separate analyses for girls and boys, as well as for language and mathematics subjects. This approach allows us to explore how the identified factors may influence academic performance differently across sex and academic subject categories, providing a more nuanced understanding of these associations.

Given the potential collinearity among the eight mental, physical, sleep, and behavioral indicators, we started our multivariable analyses with optimized sample sizes by including each indicator in a separate model, with parental education, income and adolescents’ age as covariates. We then compared these single-indicator models to models including all eight indicators, using pairwise deletion. Since the results were similar, we report the findings from the models with all indicators in the results section, while the single-indicator models are presented in the Appendix as sensitivity analyses. As an additional check for multicollinearity, we calculated Spearman correlation among all eight health indicators, and we found that all correlations were below 0.19.

For the models with all indicators, we carried out further analyses to assess the effect of each specific dimension of the HRQoL score (i.e., emotional well-being, self-esteem, family, and social contacts). Again, considering the collinearity among these dimensions, we assessed each dimension separately, with each model including one dimension of the HRQoL score and the other seven health indicators as covariates. Similarly, stressful life events were initially analyzed as a composite score, summing the total scores of thirteen events. However, given unexpected findings in the primary analysis, we conducted additional exploratory analyses to examine the individual events contributing to the observed results.

We used pairwise deletion to handle missing data throughout the analyses, but multiple imputation was also performed as part of sensitivity analyses. Additional sensitivity analyses were conducted to assess the robustness of our findings concerning model specification (i.e., treating country as random effects rather than fixed effects). We also compared the single-indicator models with the models including all indicators. Furthermore, we examined the consistency of results when academic performance outcomes were defined using a reverse dichotomization of the original outcome. We also explored ordinal logistic regression using the original five-category outcome, but all sex- and subject-stratified models violated the proportional odds assumption and exhibited sparse data structures, reinforcing the choice of binary logistic regression.

All statistical tests were two-sided with a significance level set at 5%. In binary logistic regression analyses, we reported the results of all pre-planned comparisons with p-values without adjustment for multiplicity. Thus, marginally significant results must be interpreted with caution. All statistical analyses were performed using the R software environment (version 4.4.1) and STATA (version 18.0).

link