Abstract

Background:

Adolescent depression is linked to daily activities like screen use, physical activity, and sleep. Most studies examine these factors separately. This study investigates the relationship between screen time, physical activity, sleep, and depressive symptoms in adolescents using compositional isotemporal substitution models (CISM).

Methods:

A cross-sectional study collected data from 6,666 adolescents in Hefei, China, using self-administered questionnaires. Compositional Data Analysis was used to examine 24-h activity patterns and their association with depression. A compositional linear regression model was constructed to explore the relationships between screen time, low-intensity (LPA) and moderate-to-vigorous physical activity (MVPA), sleep (SLP), and adolescent depression. The CISM was then applied to analyze the effects of reallocating and substituting time between different activities on depression scores.

Results:

The compositional linear regression model revealed a significant positive correlation between screen time and depression relative to the remaining activities (βST = 0.693, P < 0.001), while LPA, MVPA and SLP showed significant negative correlations with depression relative to the remaining activities (βLPA = −0.132, P < 0.05, βMVPA = −0.293, P < 0.001, βSLP = −0.981, P < 0.001). Using a 10-min substitution as an example, replacing MVPA, LPA, or SLP with ST increased depression scores by 0.09, 0.06 and 0.03 points, respectively. Conversely, replacing ST with MVPA, LPA, or SLP decreased scores by 0.07, 0.05 and 0.03 points, respectively. Longer substitution durations amplified these effects.

Conclusion:

The study highlights that reducing screen time and increasing physical activity or sleep can help alleviate depressive symptoms in adolescents.

1 Introduction

Depressive disorder (also known as depression) is a common mental disorder characterized by persistent low mood and a loss of interest or pleasure in activities (1). Adolescence is a transitional period between childhood and adulthood, during which physiological, psychosocial, and cognitive changes make adolescents more susceptible to psychological problems (2), with depression being particularly common. Research has found that some cases of depression can be traced back to childhood and increase dramatically during adolescence (3). It is estimated that nearly 14% of adolescents meet the diagnostic criteria for depression before the age of 18 (4). By the age of 19, approximately 25% of adolescents have experienced at least one depressive episode (5). In the United States, suicide has become the second leading cause of death among adolescents aged 12 to 17. Epidemiological studies show that 43% to 90% of adolescent suicide victims had at least one mental disorder at the time of death, with depression being the most common (6). Furthermore, depression is not only a major cause of disability and impairment worldwide (7), but also a major contributor to the global disease burden (8).

Depression results from multiple factors, including genetic, biological, environmental, and behavioral influences. Adolescents' daily behaviors, such as physical activity (PA), sedentary behavior (SB), and sleep (SLP), are potential depression factors. SB, including screen use, leisure-time sitting, and work-related sitting (9), is especially notable. The rise in screen time (ST) is a key feature of modern society, making its impacts a crucial issue in adolescent health research.

Regarding the definition of ST, most scholars believe that it is a metric for measuring the time individuals spend using electronic screens. Tang et al. (10), in their study on the relationship between ST and mental health in young people, stated that “ST” is a general term encompassing various devices (such as computers, televisions, and mobile phones) and their uses (such as gaming and social communication). Based on previous research, this study defines ST as the total duration spent watching or using electronic devices (such as televisions, computers, mobile phones, tablets, video game consoles, e-readers, and other electronic devices).

Although new electronic devices may offer potential benefits, children and adolescents are spending more time on screen-based activities than ever before (11). The World Health Organization (WHO) recommends that children and young people's recreational screen time be limited to no more than 2 h per day, but the majority not adhere to this guideline (12). A large-scale longitudinal study in the United States on brain development and child well-being found that increased ST is closely associated with depression and anxiety (13). Furthermore, adolescents who engage in more than 4 h of passive ST per day are more likely to meet the diagnostic criteria for major depressive episodes, social phobia, and generalized anxiety disorder (14). In the course of an adolescent's daily activities, other behavioral patterns, such as PA and SLP, also have significant impacts on depression. These behavioral patterns are interconnected and mutually influential, collectively shaping the adolescent's mental health status.

In contrast to ST, PA offers a positive behavioral choice, and its relationship with depression has become a focal point of research. PA is typically defined as any type of bodily movement produced by skeletal muscle tissue. PA not only benefits physical health but also promotes mental health through various pathways. A meta-analysis of a prospective study reported that individuals with higher levels of PA were 17% less likely to develop depression compared to those with lower PA levels (15). Additionally, an 11-year follow-up study found that exercising for an hour a week can reduce the risk of depression by 12% (16).

When studying the association between adolescent behaviors and depression, SLP also plays a crucial role as an important health behavior pattern. Healthy sleep patterns help maintain good physical health, immune function, mental health, and academic performance (17), whereas poor sleep is a direct catalyst for the development of emotional difficulties and affective disorders (18). A meta-analysis revealed a close relationship between depressive symptoms and sleep quality in children and adolescents (19). Furthermore, reports suggest that adolescents with sleep disorders are at a higher risk of developing depression later in life (20).

Although numerous studies have revealed the individual effects of ST, PA and SLP on adolescent depression, these studies either focus solely on ST or PA, without considering the intrinsic relationships between these behaviors (21). Due to the fixed total of 24 h per day, a competitive and trade-off relationship exists between screen time, physical activity, and sleep. An increase in time devoted to one behavior will inevitably reduce time available for the others, and such interactions cannot be ignored (22). To address issues related to time reallocation and behavioral substitution, the compositional isotemporal substitution model (CISM) is adopted in this study. CISM addresses the constrained nature of 24-h time-use data by treating 24-h activity behaviors as compositional data and using Compositional Data Analysis (CoDA) methods, which allow for the transformation from simplex to Euclidean space, addressing the constrained nature of 24-h time use data. It the estimation of changes in health indicators after reallocating time between behaviors, rather than merely exploring the effect of changing a single activity on health outcomes (22).

To date, many studies have used CISM to explore the relationship between activity-behavior time reallocation and adolescent health outcomes. While CISM has been widely used to investigate the effects of time allocation between PA, SB, and SLP on health outcomes, existing studies often categorize screen-based behaviors simply as sedentary behaviors, overlooking their unique impact on health as a distinct behavioral pattern. The unique value of this study lies in separating screen time from the broader category of sedentary behavior and treating it as an independent behavioral component. Using the CISM model, this study can precisely quantify the specific impact on depressive symptoms when screen time is substituted with physical activity of different intensities and sleep. The increase in adolescent ST may not only directly affect depressive symptoms but may also indirectly alter health outcomes by compressing PA and SLP time. However, few studies have used CISM to separately explore the potential impact of reallocating ST, PA, and SLP on adolescent depression. This study adopts this perspective, aiming to use CISM to evaluate the impact of the mutual substitution of screen use, PA, and SLP on adolescent depression using CISM. It further seeks to reveal the unique role of ST and its potential pathways, providing scientific evidence for adolescent mental health interventions and behavioral pattern optimization, while also offering guidance for public health practices.

2 Materials and methods2.1 Study participants

In this study, from April to June in 2024, four districts (Yaohai District, Luyang District, Shushan District, and Baohe District) and Chaohu City (a county-level city) under the jurisdiction of Hefei City were selected. Four schools were randomly selected from each district (two middle schools and two high schools), totaling 20 schools. Using a stratified cluster sampling method, four classes from each of the first and second grades were randomly selected from each school, with a total of 7,081 students included in the study. Due to missing or invalid data, the final number of valid questionnaires included in the analysis was 6,666, yielding an effective response rate of 94.14%, as shown in Figure 1.

Research participant flowchart.

2.2 Study content and methods

In this study, paper questionnaires were used for on-site investigation. The questionnaires were distributed by the class teachers or teachers by the uniformly trained investigators after obtaining the informed consent of the subjects, and the students filled in the questionnaires on the spot in the school classroom. In the process of filling out, students can raise their hands if they have any questions, and the investigators will answer them according to the uniform standards. After completion, the investigator checked the completeness of the questionnaire on the spot and promptly withdrew it. Quality control measures included: in the questionnaire design stage, the scale suitable for the study population was selected by referring to domestic and foreign literature, and epidemiological experts were invited to review, and the formal questionnaire was perfected after pre-investigation. During the survey implementation stage, all investigators received unified training, including the purpose of the survey, procedures, precautions and emergency handling. The standardized language was used to introduce the requirements to ensure information consistency. EpiData 3.1 was used to establish the database in the stage of data entry and verification. Two people independently entered the data and checked the consistency.

The questionnaire used in this study was independently designed, integrating scales with high reliability and validity that are widely used both domestically and internationally. The content of the questionnaire covered adolescents' general information, PA levels, SLP, screen use, and depression status.

2.2.1 General information

The general survey covered information including their grade, gender, date of birth, height, weight, academic performance, family composition, whether they are an only child, whether they are a day student or boarder, parental education level, and household per capita annual income.

2.2.2 Physical activity level

The physical activity levels of the participants were assessed using the Chinese version of the International Physical Activity Questionnaire Short Form (IPAQ-SF) (23). This questionnaire is a self-reported measure based on recall of activity levels over the past 7 days, and it has been comprehensively validated in 12 countries/regions (24). The questionnaire contains seven questions aimed at collecting the daily time (in minutes) and weekly frequency (in days) of walking, moderate-intensity, and vigorous-intensity activities, as well as daily sedentary time. Since the WHO and most epidemiological studies recommend combining MPA with VPA, as they have similar positive effects on mental health, and in terms of statistical stability and interpretation, this combination can reduce the number of variables to avoid multicollinearity issues and make the model simpler and the results more comparable, therefore in this study, MPA and VPA were combined into MVPA.

2.2.3 Sleep time

Sleep status was measured using the Pittsburgh Sleep Quality Index (PSQI), which was proposed by Buysse et al. (25) in 1989 and is one of the most widely used scales for measuring sleep quality. The Chinese version of the scale, translated and validated by Liu et al. (26), was used in this study to assess individuals' sleep quality over the past month, with a primary focus on the dimension of sleep duration. We selected the dimension of sleep duration as the quantitative indicator for sleep (SLP) because it is highly consistent with the 24-h time-use analytical framework adopted in this study. With the compositional isotemporal substitution model (CISM) as the core analytical method, this study focuses on the time allocation and mutual substitution effects of different activities (MVPA, LPA, ST, NSST, SLP) within a 24-h period. All behavioral indicators included in the analysis are quantified in minutes. Therefore, extracting the “sleep duration” dimension from the PSQI ensures the unified dimensionality of all behavioral indicators, meets the data format requirements of compositional data analysis (CoDA), and enables the direct comparison of time spent on different activities and the analysis of their substitution effects.

2.2.4 Screen time

The screen Time section of the questionnaire was adapted from the study on screen time by Tang et al. (27). The reliability and validity of this part of the questionnaire were tested by Tang using Cronbach's α coefficient, with an internal consistency α coefficient of 0.847, indicating high reliability. To collect information on participants' daily screen time over the past week (including time spent on entertainment, social interaction, education, and other purposes), we asked them to report their screen time on school days and weekends, involving the use of televisions, computers, mobile phones, tablets, video game consoles, e-readers, and other electronic devices. The calculation method was as follows: screen time on school days was multiplied by five, and screen time on weekends was multiplied by 2. The total was then summed and divided by seven to calculate the average daily screen time over the past week.

2.2.5 Depression status

The depression status of the participants was assessed using the Patient Health Questionnaire-9 (PHQ-9). This scale consists of nine items, each rated on a 4-point scale, ranging from 0 to 3 (0 = Not at all; 1 = Several days; 2 = More than half of the days; 3 = Nearly every day). The total score ranges from 0 to 27, with higher scores indicating more severe depression. Total scores of 0–4, 5–9, 10–14, and 15–27 represent no depression, mild depression, moderate depression, and severe depression, respectively. In this study, the Cronbach's α coefficient for the PHQ-9 was 0.904.

2.3 Statistical analysis

In the statistical analysis part, descriptive statistical analysis was used in order of application. The continuous variables conforming to normal distribution were described by mean and standard deviation, the continuous variables not conforming to normal distribution were described by median and interquartile range, and the categorical variables were described by frequency and constituent ratio. Then non-parametric tests were used to compare the differences in depression scores, screen time and physical activity among different demographic characteristics. Secondly, the component data analysis method was used to calculate the geometric mean and variation matrix of 24-h activity behavior to describe the central tendency and discrete tendency. After that, the component data were transformed from simplex space to Euclidean space by the isometric log ratio transformation for the application of traditional statistical methods. Then, a component linear regression model was constructed to analyze the effects of each activity relative to other behaviors with ILR coordinates as the independent variable and depression score as the dependent variable.

Finally, this study used the compositional isotemporal substitution model (CoDA) to process compositional data and analyze the effects of behavioral components on depression scores. Compositional data refers to data where the sum of components is a fixed total (e.g., 24 h) and typically includes three types of activities: PA, SB, and SLP. In this study, PA was divided into moderate-to-vigorous physical activity (MVPA) and light physical activity (LPA), and SB was reallocated into ST and non-screen-based sedentary time (NSST), with a particular focus on screen-based behaviors. The specific operation was based on the geometric mean of each activity as the baseline, a series of new activity combinations were systematically created, and the reallocation of time among all pairs of activities was simulated, such as 5, 10, 15, and 20 min. The new combination was substituted into the component linear regression model by Isometric Log Ratio (ILR) transformation to obtain the new predicted value, and the predicted value of the pre-replacement combination was subtracted to obtain the mean change in depression scores by time reallocation.

Existing statistical methods are mostly applicable to unconstrained data, so the Isometric Log Ratio (ILR) transformation in CoDA was used to convert the data from the simplex space to the Euclidean space, making it easier to apply traditional statistical analysis methods. Using the sequential binary partitioning method, activity behaviors were converted into ILR coordinates, as follows:

In the regression model, the transformed ILR coordinates were used as independent variables, with the depression score as the dependent variable. The compositional linear regression model is constructed as: y = β0+β1z1+β2z2+β3z3+β4z4+ε. In this equation, β1β2β3, and β4 are regression coefficients, and ε is the random error term. Finally, the CISM was used to investigate the effect of reallocating the time of five compositional variables on the outcome variable (PHQ-9 score). For example, to evaluate the effect of replacing 10 min of ST with MVPA, a new activity combination was created: (GMVPA − 10, GLPA, GST+10, GNSST, GSLP). The ILR transformation was applied to these combinations, and the regression model was used to calculate the change in PHQ-9 scores after the substitution.

4 Results4.1 Depression status of study participants

A total of 6,666 adolescents were included in this study, of whom 53.81% were boys and 46.19% were girls. Grades were evenly distributed, accounting for about 25% each from grade 1 to grade 2. In terms of body mass index, 53.44% of the participants were normal weight, 28.95% were underweight, 12.27% were overweight, and 5.34% were obese. 67.21% of the children were not the only child, and 81.23% did not live in school. Their academic scores were mainly middle and upper middle, totaling 61.78%. The father's education level was junior high school or below (35.49%), and the mother's education level was junior high school or below (44.19%).

The average depression score of the participants was 6.18 ± 5.79. The distribution of depression severity was as follows: no depression (3,134 participants, 47%), mild depression (2,109 participants, 31.6%), moderate depression (794 participants, 11.9%), and severe depression (629 participants, 9.4%), as shown in Table 1. In Table 2, non-parametric tests (Mann–Whitney U-test or Kruskal–Wallis H-test) were used to compare the differences in screen time among different demographic characteristics, and the statistic Z or H value and the corresponding P value were used to determine the significance of the differences between groups. The analysis results indicated that variables such as gender, grade, BMI, academic performance, and family background were significantly associated with depression scores (P < 0.001). Specifically, female participants had significantly higher depression scores than male participants. Higher grades and higher BMI levels were associated with increased depression scores, and there was a significant difference in depression scores between boarding and non-boarding students. Additionally, adolescents from families with lower incomes had higher depression scores, as shown in Table 2.

GenderNo depressionMild depressionModerate depressionSevere depressionMale1,9311,045354257Female1,2031,064440372Total number of people3,1342,109794629

Depression score rating levels.

CharacteristicsDepression scoreZ/H valuePGenderMale4.00 (1.00, 8.00)−13.588< 0.001Female6.00 (3.00, 10.00)GradeGrade 73.00 (1.00, 7.00)207.865< 0.001Grade 84.00 (1.00, 8.00)Grade 106.00 (3.00, 9.00)Grade 116.00 (3.00, 9.00)BMI categoryUnderweight4.00 (1.00, 8.00)31.750< 0.001Normal5.00 (2.00, 9.00)Overweight5.00 (2.00, 9.00)Obese5.00 (2.00, 9.00)Only childYes4.00 (1.00, 8.00)4.240< 0.001No5.00 (2.00, 9.00)Boarding schoolYes6.00 (3.00, 9.00)−5.133< 0.001No5.00 (2.00, 9.00)Academic performanceLow7.00 (3.00, 13.00)154.312< 0.001Below average6.00 (2.00, 10.00)Average5.00 (2.00, 9.00)Above average4.00 (1.00, 8.00)Excellent4.00 (1.00, 7.00)Father's educational levelJunior high school or below5.00 (2.00, 9.00)24.579< 0.001High school5.00 (2.00, 9.00)Vocational/Technical school5.00 (2.00, 9.00)Associate degree5.00 (2.00, 9.00)line Bachelor's degree4.50 (1.00, 8.00)Graduate degree or above4.00 (1.00, 8.00)Mother's educational levelJunior high school or below5.00 (2.00, 9.00)32.454< 0.001High school5.00 (2.00, 9.00)Vocational/Technical school6.00 (2.00, 9.00)Associate degree5.00 (2.00, 9.00)Bachelor's degree4.00 (1.00, 8.00)Graduate degree or above3.00 (0.00, 9.00)Annual household per capita income< 10,0006.00 (2.00, 10.00)26.250< 0.00110,000–29,9995.00 (2.00, 9.00)30,000–49,9995.00 (2.00, 9.00)50,000–69,9995.00 (1.00, 8.75)70,000–89,9995.00 (2.00, 8.00)90,000–109,9994.00 (1.00, 8.00)≥110,0005.00 (1.00, 8.00)

Differences in depression scores across different demographic characteristics.

4.2 Screen time

The analysis of compositional data in this study revealed significant differences in the geometric mean (in minutes) of screen time (ST) across different demographic groups, as shown in Table 3. Gender analysis indicated that females had significantly higher ST than males (P < 0.001). Significant statistical differences in ST were also observed across grades, BMI levels, and boarding status (P < 0.001). Specifically, ST increased with higher grades and BMI levels, while boarding students had relatively lower ST (P < 0.001). Moreover, variables such as academic performance, parental education level, and family income per capita were significantly correlated with ST (P < 0.001).

General situationST (G)Z/H valuePGenderMale313.476.686< 0.001Female328.12GradeGrade 7270.72347.128< 0.001Grade 8280.99Grade 10355.83Grade 11383.14BMI categoryUnderweight299.3626.186< 0.001Normal321.60Overweight348.22Obese367.58Only childYes316.920.3530.724No322.38Boarding schoolYes316.92−4.873< 0.001No322.38Academic performanceLow345.5923.440< 0.001Below average329.80Average328.51Above average306.42Excellent308.16Father's educational levelJunior high school or below323.5420.244< 0.001High school338.40Vocational/Technical school354.97Associate degree318.83Bachelor's degree283.36Graduate degree or above292.41Mother's educational levelJunior high school or below332.6144.846< 0.001High school319.05Vocational/Technical school352.20Associate degree324.842Bachelor's degree272.26Graduate degree or above266.04Annual household per capita income< 10,000312.3927.706< 0.00110,000–29,999345.9730,000–49,999329.2550,000–69,999301.5170,000–89,999308.9890,000–109,999337.79≥110,000312.39

Screen time across different demographic characteristics.

The usage of different electronic devices by adolescents is shown in Figure 2. Regardless of the severity of depression, the proportion of time spent on TV was consistently the highest, followed by mobile phone usage. As depression severity increased, the proportion of time spent using mobile phones, tablets, video game consoles, and e-readers all showed an upward trend.

Proportion of adolescents using screen devices.

4.3 Time distribution of adolescents' 24-h activity behaviors4.3.1 Central tendency

The 24-h activity behavior time for adolescents is composed of MVPA, LPA, ST, NSST, and SLP. The central tendency was described using geometric means, as shown in Table 4, which presents the geometric and arithmetic means along with their percentage of each activity time. Whether using geometric or arithmetic means, the percentage distribution of 24-h activity time is as follows: SLP > NSST > ST > MVPA > LPA.

Type of activityGeometric mean (Min)Percentage (%)Arithmetic mean (Min)Percentage (%)MVPA44.113.0672.405.03LPA