The present study is a secondary analysis of data collected during a school-based health education program (Balassagyarmat Health Education Program [BEP]) that aimed to improve health literacy and focused on sexual health, substance use, basic life support, infection control, nutrition, and physical activity but did not cover screen time as a topic. The program was described in detail previously (15). In short, consecutive ninth graders were recruited in BEP from all five secondary schools (three grammar schools and two vocational secondary schools) in a northern Hungarian town (Balassagyarmat, approximately 15,000 inhabitants) in 2018–2020. Participants first completed a baseline survey, then participated in the health education program that spanned over a year. After completing the program, students filled in a follow-up survey, approximately one year after the baseline survey in 10th grade.

For the present analysis, we selected nineth-grade participants who completed their baseline assessment in either 2018 or 2020 (not 2019). For controls (baseline in 2018, unaffected by the COVID-19 pandemic), both baseline and follow-up surveys were filled in online under the supervision of research assistants (teachers were not present during the survey process) during school hours. For cases (baseline in 2020, affected by the COVID-19 pandemic during follow-up), baseline assessment took place in circumstances identical to controls, while the follow-up survey was filled in at home because of the school closures. Students were asked to complete the survey during specified school hours, and they could contact a research assistant online in case of any questions. Students recruited in 2019 were excluded from the current analysis because their follow-up survey was conducted during the early, transitional phase of school lockdown. The questionnaire and methodology for this study was approved by the Institutional Review Board of Semmelweis University (SE TUKEB: 276/2017). Parental approval of the participants was sought with an opt-out procedure.

All ninth-grader students were invited to participate. In 2018, out of the 454 ninth-grade students of the ‘pre-COVID classes’ (defined as controls), 332 agreed to participate at baseline resulting in a 0.73 sampling fraction. The baseline survey was completed in February 2018, while the follow-up survey was completed in March 2019. In the control group, 98 students were lost to follow-up. In 2020, out of the 446 ninth-grade students in the ‘COVID classes’ (defined as cases), 334 agreed to participate at baseline, resulting in a 0.75 sampling fraction. Cases completed the baseline survey in February 2020 (approximately 1 month before the COVID-19-related school lockdowns in Hungary) and the follow-up survey in March 2021. In the case group, 83 students were lost to follow-up. Flowchart of participants is presented in Figure 1. During the study period, school lockdowns affected students from 16 March 2020 to 2 June 2020 and then from 11 November 2020 to the end of study. According to the Organization for Economic Co-operation and Development (OECD) report, Hungarian secondary schools were fully closed for 164 days (excluding school holidays, public holidays and weekend) between 1 January 2020 and 20 May 2021 (16). Overaged students (mean age + 3 SD) and those living without a biological or stepparent were excluded.

To evaluate screen time, we used four questions based on the Hungarian version of the Health Behavior in School-aged Children study (HBSC 2014) (17, 18): (1) ‘How many hours a day, in your free time, do you usually spend watching TV, videos (including YouTube or similar services), DVDs, and other entertainment on a screen?’ (2) ‘How many hours a day, in your free time, do you usually spend playing games on a computer, games console, tablet (like iPad), smartphone or other electronic device (not including moving or fitness games)?’ (3) ‘How many hours a day, in your free time, do you usually spend using electronic devices such as computers, tablets (like iPad) or smart phones for other purposes, for example, homework, emailing, tweeting, Facebook, chatting, surfing the internet?’ (4) ‘How many hours of this do you spend on doing homework?’. Students were required to express the amount of time spent on each activity as hours and fractions of an hour. The items resulted in five outcome variables: overall screen time (sum of Question 1, 2 and 3), screen time for watching videos (Question 1), playing games (Question 2), doing homework (Question 4), and screen time for other purposes (derived as the difference between data provided in Questions 3 and 4) representing time spent mainly on social media activities and communication).

Since studies indicate that family support is an important protective factor against unhealthy behaviors among adolescents (13), we adjusted our results for family structure (living with two parents, with one parent, or in a stepfamily) and family communication [short version of the Clear Communication Scale from Family Dynamics Measure II [FDMII] (18, 19)] in our analyses. To assess family communication, we used the Hungarian version of the FDMII implemented in the Hungarian HSBC study 2014 (Cronbach-alfa: 0.74) (17). The questionnaire consists of four Likert items with a maximum score of 20. A higher score represents more positive judgement on family communication (17).

All analyses were conducted stratified by sex. Descriptive baseline data of pre-COVID vs. COVID classes and follow-up data of pre-COVID vs. COVID classes were compared with Chi-squared tests for categorical variables and independent samples t-tests for continuous variables. Descriptive baseline vs. follow-up data within pre-COVID and COVID classes were compared with marginal homogeneity tests for categorical variables and paired t-tests for continuous variables. Linear mixed models were built to assess the effect of COVID-19-related school lockdowns on overall screen time and on its individual components. Model 1 includes case/control status as predictor, while Model 2 is adjusted for time-varying family structure and family communication. In our models, only family communication was treated as a continuous variable. To exclude the potential for a non-linear relationship, we tested whether adding a quadratic term of family communication would improve our models. Given that these quadratic terms were non-significant, we removed them from the final model to achieve parsimony. All other variables were categorical variables, and thus non-linearity was not investigated. Given that we had a sufficient number of cases (>200 individuals for each analysis) and only 2 time-points in a random slope, random intercept model, we decided to use the unstructured covariance matrix with the least number of assumptions, as it only increased the number of parameters in the model minimally (by one) compared to other frequently used covariance structures (e.g., variance component or autoregressive). Furthermore, we also tested the information criteria (AIC, BIC) of the above covariance structures, and the unstructured covariance structure had the lowest values. All statistical analyses were performed using IBM SPSS Statistics version 28.0.0.0. Statistical significance was set at p < 0.05.

According to Model 1, cases’ overall screen time was higher by 1.35 (95% CI: 0.12–2.57) hours compared to controls. This difference became non-significant after adjustment for family structure and family communication (MD: 1.17 [−0.06–2.40]; 9.06 [95% CI: 8.03–10.08] vs. 10.23 [95% CI: 9.35–11.09] hours) (Table 2; Figure 2). As Model 1 and Model 2 yielded similar results for the individual components of screen time, we only provide detailed description of Model 2 in the following. According to Model 2, at baseline cases spent 0.79 (95% CI: 0.13–1.45) hours more on watching videos compared to controls (3.49 [95% CI: 2.93–4.04] vs. 4.28 [95% CI: 3.80–4.76] hours), while screen time of controls and cases was similar for playing games (3.08 [95% CI: 2.59–3.58] vs. 3.11 [95% CI: 2.68–3.54] hours), doing homework (0.79 [95% CI: 0.53–1.04] vs. 0.91 [95% CI: 0.70–1.12] hours), and other purposes (3.00 [95% CI: 2.41–3.59] vs. 3.63 [95% CI: 3.13–4.14] hours) (Table 3; Figure 3).

In controls, with aging overall screen time did not show a significant change (9.06 [95% CI: 8.03–10.08] vs. 9.59 [95% CI: 8.55–10.63] hours) (Figure 2), however the pattern of its components was altered. Screen time for watching videos remained stable (3.49 [95% CI: 2.93–4.04] vs. 3.46 [95% CI: 2.81–4.11] hours), playing games decreased by 0.63 (95% CI: −1.12 – −0.14) hours (3.08 [95% CI: 2.59–3.59] vs. 2.45 [95% CI: 1.87–3.04] hours), doing homework also remained stable (0.79 [95% CI: 0.53–1.04] vs. 0.84 [95% CI: 0.46–1.22] hour), while screen time for other purposes increased by 1.11 (95% CI: 0.25–1.97) hours (3.00 [95% CI: 2.41–3.59] vs. 4.11 [95% CI: 3.36–4.87] hours) (Table 3; Figure 3).

In the COVID classes overall screen time increased by 3.29 (95% CI: 2.29–4.29) hours (10.23 [95% CI: 9.35–11.09] vs. 13.51 [95% CI: 12.49–14.53] hours) during the follow-up period (Table 2; Figure 2). All components of screen time also increased. Screen time spent on watching videos increased by 1.56 (95% CI: 1.05–2.08) hours (4.28 [95% CI: 3.80–4.76] vs. 5.84 [95% CI: 5.25–6.44] hours), playing games – as opposed to controls – increased by 1.44 (95% CI: 1.02–1.85) hours (3.11 [95% CI: 2.68–3.54] vs. 4.55 [95% CI: 4.01–5.09] hours), homework increased by 1.38 (95% CI: 1.02–1.73) hours (0.91 [95% CI: 0.70–1.12] vs. 2.29 [95% CI: 1.94–2.62] hours), and other purposes increased by 2.24 (95% CI: 1.49–2.99) hours (3.63 [95% CI: 3.13–4.14] vs. 5.87 [95% CI: 5.18–6.56] hours) (Table 3; Figure 3).

Furthermore, we found that boys living with only one parent spent more screen time on watching videos and playing games. We also found that boys who scored higher on family communication had lower overall screen time and screen time for watching videos and playing games (Table 2; Table 3).

Similarly to boys, Model 1 and Model 2 yielded similar results in girls. According to Model 2, we detected that cases’ overall screen time at baseline was higher by 1.53 (0.44–2.62) hours than that of controls (8.50 [95% CI: 7.62–9.38] vs. 10.03 [95% CI: 9.19–10.87] hours) (Table 2; Figure 2). As for the individual components, screen time spent on watching videos was higher in cases by 0.62 (95% CI: 0.06–1.17) hours than in controls (3.51 [95% CI: 3.05–3.96] vs. 4.13 [95% CI: 3.69–4.55] hours], there was no significant difference in screen time for playing games (1.51 [95% CI: 1.10–1.93] vs. 1.79 [95% CI: 1.40–2.19] hours) and doing homework (1.08 [95% CI: 0.92–1.24] vs. 1.03 [95% CI: 0.88–1.17] hours), but screen time for other purposes was also higher by 1.28 (95% CI: 0.57–2.00) hours in cases compared to controls (3.71 [95% CI: 3.14–4.29] vs. 4.99 [95% CI: 4.45–5.54] hours) (Table 4; Figure 4).

During follow-up, control girls’ overall screen time remained stable (8.50 [95% CI: 7.62–9.38] vs. 8.47 [95% CI: 7.56–9.41] hours) (Table 2; Figure 2). Unlike boys’, control girls’ screen time did not show any significant change in the individual components with aging. They spent 3.45 h (95% CI: 2.85–4.04) on watching videos, 1.14 h (95% CI: 0.70–1.59) on playing games, 1.02 h (95% CI: 0.71–1.32) on doing homework and 4.18 h (95% CI: 3.56–4.80) for other screen time purposes (Table 4; Figure 4).

On the other hand, classes affected by COVID increased their overall screen time by 2.27 (95% CI: 1.37–3.18) hours (10.03 [95% CI: 9.19–10.87] vs. 12.30 [95% CI: 11.35–13.25] hours) (Table 2; Figure 2). As for the individual components, time spent on watching videos increased by 1.66 (95% CI: 1.13–2.19) hours (4.13 [95% CI: 3.69–4.55] vs. 5.79 [95% CI: 5.18–6.37] hours), playing games remained stable (1.80 [95% CI: 1.40–2.19] vs. 1.76 [95% CI: 1.32–2.19] hours), doing homework increased by 2.10 (95% CI: 1.80–2.40) hours (1.03 [95% CI: 0.88–1.17] vs. 3.13 [95% CI: 2.81–3.43] hours), and screen time spent on other activities increased by 1.49 (95% CI: 0.87–2.12) hours (4.99 [95% CI: 4.45–5.54] vs. 6.48 [95% CI: 5.87–7.10] hours) during follow-up (Table 4; Figure 4).

Living in a single-parent household was associated with higher overall screen time and screen time spent on social media. Higher score on family communication was associated with lower overall screen time in girls, however, it showed no association with the individual components of screen time (Table 2, 4).

A major strength of our study is its longitudinal design, which enabled us to observe within participant trends of screen time. Studies indicate that longitudinal studies have much more statistical power and thus result in a more precise assessment of screen time, as shown by a study in which retrospective assessment was compared to longitudinal assessment (21). As observed by the authors, retrospective assessment tended to overestimate the true amount of screen time (21). Furthermore, our design allowed us to differentiate and compare the effect of aging and the pandemic. Another strength is that we examined the different dimensions of screen time, and thus were able to analyze how engagement in different screen time categories changed as a result of lockdown. A subsequent strength is that both case and control students came from the same settlement, and controls preceded cases by only two years reducing the potential of time period effects. Moreover, our study is one of the first to offer a deeper insight in screen time in Balassagyarmat, a typical city of a deprived region in central Europe.

A limitation of our study is that answers of student were not cross-referenced by parents. A study similar to ours found, however, that the estimation of screen time by students tends to be similar to the assessment of parents (6). Another limitation of our study is that we were not able to account for double screen time, for instance watching TV while scrolling social media. Our questions also did not focus exclusively on social media and had to be calculated post hoc from the HBSC 2014 questions. We also experienced baseline difference in males’ screen time on watching videos and females’ screen time on watching videos and other activities, which may be explained by the different time periods of the baseline and follow-up surveys (February vs. March) with different weather patterns and daylight hours. This, however, is less likely to have severely biased the observed trends between baseline and follow-up. Finally, our study did not contain information on other sedentary behaviors and level of physical activity.