The present study is a secondary analysis of data derived from the Balassagyarmat Health Education Program (BEP) (6, 18). BEP was a school-based health education project that focused on measuring and improving various health aspects, including sexual health, substance use, basic life support, infection control, nutrition, physical activity, and mental health of ninth-grade students from all secondary schools (three grammar schools and two vocational schools) in Balassagyarmat. Balassagyarmat is the capital of a northern Hungarian district with around 40,000 inhabitants. BEP operated between 2018 and 2021. Ninth-grade students underwent an online baseline survey before engaging in the one-year long health education program. Following the program, students completed an online follow-up survey in their 10th grade, approximately 1 year after the baseline survey.

For the current analysis, we specifically chose students with a baseline assessment in 2018 or 2020 (excluding 2019). Students with a baseline assessment in 2018 were considered the pre-pandemic group (controls) because they were not affected by the COVID-19 pandemic. This group filled in both baseline and follow-up surveys at school during teaching hours under the supervision of a research assistant without the presence of teachers. Students with a baseline assessment in 2020 were considered the lockdown group because they experienced school lockdowns due to the COVID-19 pandemic. Their baseline assessment happened similarly to the pre-pandemic group before the pandemic, the follow-up survey, however, was filled out at home during the lockdowns. Students were instructed to complete the survey during designated school hours, with the option to contact a research assistant online for any queries. We decided to exclude students enrolled in 2019 because school lockdown was undergoing an early adaptation period in Hungary at the time of their follow-up survey (March 2020). Ethical approval was obtained from the Institutional Review Board of Semmelweis University (SE TUKEB: 276/2017). Passive parental approval (opt-out) consent was sought for every participant.

All ninth-graders in Balassagyarmat were invited to participate. Out of 454 eligible ninth-grade students in pre-pandemic classes, 332 (73.1%) completed the baseline survey in February 2018. Of these 98 students were lost to follow-up, thus 234 completed the follow-up survey in March 2019. As for the lockdown group, out of 446 ninth-grade students, 334 participated (74.9%) at the baseline investigation in February 2020, approximately 1 month before the COVID-19-related school lockdowns. The follow-up survey was completed by 251 students in March 2021 (Figure 1). It should be noted that all items on the full questionnaire were compulsory to fill in, so no missing individual outcomes or covariates were in the database. We further excluded overage students (mean + 3SD, n = 7) and participants reporting living with no parents from the analyses (n = 11) leaving to a final analytical sample of 227 students in the pre-pandemic group (128 girls and 99 boys) and 240 students in the lockdown group (118 girls and 122 boys) (Figure 1).

Secondary schools in Hungary were affected by the lockdowns twice during the pandemic. The first period began on 16 March 2020 and ended on 2 June 2020, while the second began on 11 November 2020 and was still ongoing at the end of our study. Excluding school holidays, public holidays, and weekends, Hungarian schools were closed for a total of 164 days between 1 January 2020 and 20 May 2021 (19).

Well-being was measured by the validated Hungarian version of the WHO-5 Well-being Index (WBI) (20, 21). This instrument consists of five items, each with a Likert scale ranging from 0 (at no time) to 3 points (all of the time). The items assess whether the participants (1) ‘have felt cheerful in good spirits’, (2) ‘have felt calm and relaxed’, (3) ‘have felt active and vigorous’, (4) ‘woke up feeling fresh and rested’ or (5) daily life has been filled with things that interest them in the last 2 weeks. The maximal score is 15. A higher score indicates better well-being.

Self-esteem was measured by the validated Hungarian version of the Rosenberg Self-esteem Scale (RSS) (14, 22), which has 10 items (5 of these with reverse scoring) with 4-point Likert scales (ranging from strongly agree to strongly disagree). The maximal score is 30. RSS is a unidimensional construct reflecting global self-esteem. A higher score is associated with higher self-esteem. The items read (1) ‘On the whole, I am satisfied with myself’; (2) ‘At times, I think I am no good at all’; (3) ‘I feel that I have a number of good qualities’; (4) ‘I am able to do things as well as most other people’; (5) ‘I feel I do not have much to be proud of’; (6) ‘I certainly feel useless at times’; (7) ‘I feel that I’m a person of worth, at least on an equal plane with others’; (8) ‘I wish I could have more respect for myself’; (9) ‘All in all, I am inclined to feel that I am a failure’; (10) ‘I take a positive attitude toward myself’.

As family plays a significant role in children’s development, distress, well-being (23, 24), and self-esteem (16), our analysis was adjusted for family structure and family communication measures based on the recommendations from the Health Behavior in School-aged Children Study (25). Family structure was characterized by the participants’ response to the question with whom they lived together. Answers were collapsed into three categories: living with both parents, with one parent, or in a stepfamily (one biological parent and his/her new partner who lives in the same household).

Family communication was assessed by the Hungarian short version of the Clear Communication Scale from Family Dynamics Measure II (FDMII) (25, 26). The questionnaire has four items with a Likert scale ranging from 1 to 5 (strongly disagree to strongly agree) with a maximum score of 20. A higher score indicates a more positive assessment of family communication. The items assess whether in the family (1) ‘I think the important things are talked about’; (2) ‘When I speak someone listens to what I say’; (3) ‘We ask questions when we do not understand each other’; (4) ‘When there is misunderstanding, we talk it over until it’s clear’.

All analyses used a ‘full case’ design and were stratified by sex. Chi-squared tests for the categorical variable family structure and independent samples t-tests for the continuous variables (age and family communication) were used to compare baseline data of pre-pandemic vs. lockdown classes as well as follow-up data of pre-pandemic vs. lockdown classes, respectively.

Marginal homogeneity tests for the categorical variable family structure and paired t-tests for continuous variables (age and family communication) were used to compare baseline vs. follow-up data within pre-pandemic and lockdown groups.

Linear mixed models were used to assess changes associated with COVID-19-related school lockdowns in well-being (WHO-5 WBI) and self-esteem (RSS). Model 1 includes group status (pre-pandemic/lockdown) as the sole predictor, while Model 2 is adjusted for time-varying family structure and family communication. Both models include a random slope and a random intercept with an unstructured covariance matrix.

To further clarify the possible causes of the unexpected improvement in well-being during the pandemic, we run separate models for each question of the WHO-5 WBI using similar linear mixed models with group status as the only predictor (Model 1). These models showed that the only significant difference between the pre-pandemic and the lockdown groups was in the change in the score on Item 4 (waking up feeling fresh and rested) over follow-up. Thus, we hypothesized that those adolescents who struggled with waking up early profited the most from the lockdowns. To test this hypothesis, first we divided the students into two subgroups: those with a low score on Item 4 (0 and 1) vs. those with a high score (2 or 3) at baseline. Then we extended Model 1 with the previous grouping variable (including the main effect, its interaction with pandemic status, with follow-up, and a 3-way interaction with pandemic status and follow-up). This parameterization allowed us to test whether the subgroups based on Item 4 showed different behaviors before and during the pandemic.

Given that RSS has a unidimensional structure (14, 27), we decided against conducting any post-hoc analyses on RSS questions.

All statistical analyses were performed using IBM SPSS Statistics version 29.0.1.0. Statistical significance was set at 2-tailed p < 0.05.

Table 1 presents the descriptives for the study sample by sex, group status (pre-pandemic/lockdown), and timepoint (baseline/follow-up).

Regarding boys, age, family structure, and family communication results were similar in the pre-pandemic and the lockdown classes at both timepoints (all p > 0.05).

Girls in lockdown classes were 0.22 years older compared to the pre-pandemic group. Furthermore, the baseline family structure of the pre-pandemic and lockdown classes differed significantly for girls with a higher proportion of stepfamilies in the lockdown group. However, we found no significant differences in family structure or family communication at follow-up.

When comparing baseline and follow-up characteristics within groups, we found that the distribution of family structure changed (less two-parent families at follow-up) and family communication worsened significantly from baseline to follow-up in the pre-pandemic group for boys. No other significant changes were found in any of the groups from baseline to follow-up (Table 1).

According to the unadjusted model (Model 1), WHO-5 well-being scores were similar in the pre-pandemic and lockdown groups of boys at baseline. The pre-pandemic boys’ score declined by 1.05 (95% confidence intervals [CI]: −1.72 to −0.38) points over 1 year of follow-up. In contrast, the lockdown group showed a significantly smaller decline during follow-up, leading to a non-significant but positive point estimate of change. Adjustment for family structure and family communication (Model 2) did not materially change this pattern. Furthermore, family communication showed a significant positive association with the WHO well-being score cross-sectionally: better communication within the family was associated with higher score on well-being of boys (Table 2; Figure 2).

In general, girls had lower WHO-5 well-being scores than boys. The overall pattern of change in girls was similar to boys in the unadjusted model (Model 1): similar scores in the pre-pandemic and the lockdown groups at baseline with a significant decline in the pre-pandemic group (mean difference [MD]: 0.71, 95%CI: −1.31 to −0.11), and a significantly different change leading to a non-significant improvement in the lockdown group. After adjustment for time-varying family communication and structure (Model 2), the decline in the pre-pandemic group hugely attenuated and became non-significant, while the difference in the change between the groups remained leading to an actual improvement in well-being in the lockdown group (MD: 0.64, 95%CI: 0.01 to 1.72). Living in a single-parent family was associated with a worse well-being score (−0.77 [95%CI: −1.52 to −0.01]) compared to two-parent families and stepfamilies cross-sectionally. Family communication had a positive effect of on well-being with a similar effect size as it did in boys (Table 2; Figure 2).

After experiencing an unexpected improvement in well-being during the pandemic, we decided to clarify the reason by running separate models for the individual items of the WHO-5 WBI using similar linear mixed models with group status as the only predictor. When analyzing individual items of the WHO-5 WBI questionnaire, the boys’ lockdown group had slightly higher scores at baseline for Item 1 (feeling cheerful in good spirits; 0.32 [95%CI: 0.09 to 0.54]) and Item 3 (feeling active and vigorous; 0.27 [95%CI: 0.03 to 0.50]) compared to the pre-pandemic group. No significant change for the group of pre-pandemic boys was found during the follow-up period. As for the lockdown group, no change in Items 1, 2, 3, and 5 was found, however, the change over time was significantly different for Item 4 (waking up feeling fresh and relaxed) between the pre-pandemic and lockdown groups, showing an improvement in the lockdown group from baseline to follow-up (Table 3).

As for girls, baseline differences were detected only for Item 4 with lockdown girls having a lower score (−0.31 95%CI: −0.53 to −0.10). The score of Item 4 declined in pre-pandemic girls during the follow-up period, while (similarly to boys) the lockdown girls’ score changed in the opposite direction with a similar effect size observed in boys (Table 3).

Since the only significant difference between the pre-pandemic and the lockdown groups was in the change in the score on Item 4 (waking up feeling fresh and rested), we hypothesized that adolescents who struggled with waking up early may have profited the most from the lockdowns. To test this, we compared the changes in the WHO-5 well-being score in subgroups based on Item 4 at baseline. The WHO-5 WBI score of boys waking up feeling fresh and relaxed (Item 4 score ≥ 2) declined similarly in the pre-pandemic and the lockdown classes (MD of changes: 0.33 [95%CI: −1.10 to 1.77]). In contrast, boys with lower scores on Item 4 at baseline behaved differently in pre-pandemic and lockdown groups: total WHO-5 WBI score of the pre-pandemic group declined, while the score of lockdown group increased (MD of changes: 1.20 [95%CI: 0.15 to 2.25]) (Figure 3).

The changes observed in girls were similar to boys. The WHO-5 WBI score of girls waking up feeling fresh and relaxed (Item 4 score ≥ 2) declined similarly in the pre-pandemic and the lockdown classes (MD of changes: -0.31 [95%CI: −2.02 to 1.41]). In contrast, girls with lower scores on Item 4 at baseline behaved differently in the pre-pandemic and the lockdown groups: total WHO-5 WBI score of the pre-pandemic group declined, while the score of the lockdown group increased (MD of changes: 1.20 [95%CI: 0.23 to 2.17]) (Figure 3).

The average score on the Rosenberg self-esteem scale was similar for boys in the pre-pandemic and the lockdown groups without a significant change in unadjusted models (Model 1) during the one-year follow-up. The model adjusted for family structure and communication (Model 2) yielded similar results. Furthermore, this model also found that living in a stepfamily was associated with a 2.26 (95%CI: 0.35 to 4.16) point lower score on the self-esteem scale, as well as better family communication was associated with significantly higher self-esteem (Table 4; Figure 4).

Girls had a lower score on the Rosenberg self-esteem scale compared to boys. There was no difference in the self-esteem scores either in the unadjusted or the adjusted models between pre-pandemic and lockdown groups at baseline. We found no significant change in the self-esteem score in the pre-pandemic classes during the one-year follow-up, while the change in the lockdown classes was significantly larger, leading to an increase over time. Model 2 confirmed this finding. Furthermore, we found that better family communication was associated with higher self-esteem score in girls, while family structure was not related to self-esteem (Table 4; Figure 4).

In a secondary epidemiological analysis of an intervention program of 9^th grade secondary school students in a Hungarian city, we compared changes in well-being and self-esteem before and during the COVID-19 pandemic over a one-year period.

Our results showed that adolescents’ well-being declined under ordinary circumstances over 1 year, but this decline was significantly smaller (even showing positive point estimates) during the COVID-19 pandemic. This finding disproved our hypothesis. The differences between the changes in the pre-pandemic vs. the lockdown cohorts remained significant even after adjusting for family communication and structure. When we looked for a potential explanation of this observation, we found that 4 of the 5 WHO-5 WBI items changed similarly before and during the pandemic, while there was a significant difference in the change of Item 4 (waking up feeling fresh and relaxed) between the cohorts over follow-up. Based on this, we tested for an interaction between item 4 scores at baseline and lockdown status over time. This analysis showed that the total WHO-5 WBI score declined in those with a high baseline score on item 4 irrespective of sex and lockdown status over time. For those with a low baseline score on item 4, it remained stable in those not affected by the lockdown, while it improved significantly in those adolescents affected by the pandemic.

While the self-esteem of boys changed similarly in the cohorts over follow-up, self-esteem of girls affected by the pandemic increased compared to the pre-pandemic classes over follow-up. Again, adjustment for family structure and communication had no major effect on these observations.

Finally, our study also revealed that better family communication was consistently associated with higher well-being and higher self-esteem for both boys and girls.

Our study benefits from several strengths. First, our participants from a deprived Central-European region well represent adolescents from similar circumstances giving some external validity to our findings. Given that we collected data on within person changes over time, we have sufficient power to investigate relatively small absolute changes in well-being and self-esteem. Furthermore, the fact that the pre-pandemic and the lockdown cohorts came from the same source population and had similar baseline characteristics, the bias resulting from cohort effect may be minimal. The longitudinal design of our study is also notable because most studies investigating similar questions had a cross-sectional design or did not have adequate control groups.

Our study has some limitations that has to be acknowledged. First, neither the participation rate, nor the capture of participants at follow-up were perfect that limits external validity. Similarly, our participants represent an ethnically homogenous group, thus extrapolation to non-white adolescents is limited. Our outcomes are based on short versions of the well-being and the self-esteem questionnaires that may limit their sensitivity. Furthermore, the WHO-5 questionnaire may be prone to an ‘influential question’: given that students were not required to get up early during the pandemic, Item 4 of the WHO-5 WBI questionnaire showed a significant improvement. However, we think that this does not necessarily reflect an improvement in overall well-being. This observation highlights the importance to investigate the effect of an intervention or changing circumstance on each item of the questionnaire and shows limited generalizability of this measure. Furthermore, even though the WHO-5 WBI is widely used to measure well-being in different phases of life, the definition of well-being and the factors affecting it may differ between adults and adolescents (53). For instance, while the WHO-5 well-being index was shown to be associated with psychosocial working factors among adults (54), its association with school environment or academic pressure have not been investigated as extensively, which factors can be relevant regarding the well-being of adolescents. Moreover, our questionnaire did not include questions and instruments that measured certain feelings related to the pandemic either, such as fear and anxiety of the pandemic (55), that may also have a great impact on the well-being of adolescents. The main reason for this is that our questionnaire was developed before the pandemic. Similarly, other determinants of well-being and self-esteem (such as, social connections (56), teacher-student relationship (57), or environmental factors (56)) were not captured in our study, and these and other unmeasured confounders could have biased our findings. Given this and the fact that ours is a post hoc analysis of health promotion program precludes determining causal relationships and our results are only for hypothesis generation.