Abstract
Background
Adolescence represents a critical developmental stage characterized by complex stressors, including academic pressure and interpersonal challenges, which significantly elevate the risk of depressive symptoms. While conventional research predominantly relies on linear methodologies focusing on isolated environmental determinants, the multi-level interactive effects of ecological environments on adolescent depressive symptoms remain under-explored.
Methods
Utilizing longitudinal data from the 2020 and 2022 China Family Panel Studies (CFPS), this study included a sample of 971 adolescents (aged 10–15). Baseline ecological variables and covariates were measured in 2020, while the outcome—depressive symptoms—was assessed in 2022 using the 8-item Center for Epidemiologic Studies Depression Scale (CES-D8). We integrated a machine learning framework (Random Forest, RF) to predict continuous CES-D8 scores and a nomogram model to predict binary depression risk (CES-D8 ≥ 9). The nomogram was internally validated using 1000 bootstrap resamples.
Results
Ecological environments demonstrated significant associations with adolescent depressive symptoms. In the RF model (Out-of-Bag Mean Squared Error = 12.72), the predictor s ranked in descending order of their contribution were social trust, interpersonal relationships, parent-child conflict, teacher satisfaction, and age. The developed visual nomogram demonstrated modest discrimination (AUC = 0.614) but excellent calibration (Mean Absolute Error = 0.011) for quantifying individualized risk probabilities.
Conclusion
These findings highlight specific environmental correlates and underscore the necessity of transitioning from single-factor management to integrated, multi-systemic approaches. While the predictive model requires further external validation before clinical application, it offers a preliminary epidemiological tool to identify potential intervention targets and facilitate coordinated family, school, community, and governmental interventions.
Keywords: ecological environment, machine learning, nomogram, adolescents, depressive symptoms
Introduction
Depression stands as one of the most pervasive mental health disorders affecting children and adolescents globally,1 characterized by persistent low mood, overwhelming feelings of hopelessness, and a profound loss of interest in daily activities.2 As a mental health problem of great concern globally, depression has a significant negative impact on the mental health of adolescents3 and may lead to behaviors such as truancy, delinquency, self-injury, and even suicidality.4,5 Related studies have estimated the prevalence of depression to be 1.4% in the 10–14-year-old adolescent cohort, whereas it rises to 3.5% in 15–19-year-olds.6 In addition, a meta-analysis showed that approximately one in five children and adolescents globally suffer from depression or show symptoms of depression, and this rate tends to increase over time.7 Given that adolescence is a critical stage in the rapid development of an individual’s psychological, physiological, and social roles, mental health problems during this period not only have an impact on adolescents’ academic performance, social competence, and family relationships, but may also have a long-term negative impact on their quality of life in adulthood.8 Therefore, an in-depth investigation of the multidimensional factors influencing adolescent depressive symptoms is of great importance for developing effective preventive measures.
To comprehensively explore these influencing factors, this study is anchored in Bronfenbrenner’s Ecological Systems Theory.9 Rather than merely describing isolated environments, this theory emphasizes that individual development is the result of complex interactions across multiple ecological levels. Applying this framework critically, we can synthesize the four most prominent environments shaping modern adolescents: school, family, society, and the Internet. The family and school act as primary microsystems; factors such as parent-child conflict10,11 and academic pressure12 are prominent proximal risk factors for adolescent depressive symptoms. Concurrently, the broader social environment (exosystem), characterized by social trust and interpersonal relationships, provides a macro-level buffer against psychological distress.13,14 Furthermore, the ubiquitous availability of the Internet has introduced a modern digital dimension to these ecosystems, where online behaviors significantly intersect with adolescent psychosocial well-being.15,16
Although numerous studies have explored the associations between these ecological environments and depressive symptoms, there is no consensus regarding which factors are most critical. Traditional studies have predominantly used classical statistical methods, such as ordinary least squares (OLS) linear regression, to analyze single environmental factors. This linear approach makes it difficult to capture the complex, non-linear interactions and multi-collinear relationships inherent in ecological systems. Driven by advancements in big data analytics, the Random Forest (RF) algorithm stands out for its superior capacity to model non-linear relationships and deliver empirical feature importance rankings.17,18 Nonetheless, the inherent “black-box” architecture of machine learning frequently restricts its direct translation into real-world settings. Nomograms effectively bridge this translational gap by visually representing the outputs of sophisticated models, accurately quantifying individual risk thresholds into intuitive scales.19,20 While widely utilized in clinical medicine, the application of nomograms to visualize adolescent psychosocial risks remains surprisingly sparse. Integrating nomograms with RF thus provides a critical advantage, as RF empirically quantifies the relative contribution of diverse, competing environmental inputs to a given behavioral outcome.21
To capitalize on these methodological innovations, the present study utilizes a two-wave longitudinal sample derived from the 2020 and 2022 China Family Panel Studies (CFPS). Specifically, we aim to investigate whether baseline ecological configurations (measured in 2020) can prospectively predict the severity of subsequent depressive symptoms (measured in 2022). While we leverage this two-wave design to establish temporal predictive associations, baseline CES-D8 scores were not adjusted for in the current analyses. We hypothesize that: (1) specific ecological stressors will significantly predict future depressive symptoms; (2) the RF algorithm will effectively rank these competing environmental inputs to identify the most critical ecological predictors; and (3) a resulting nomogram will offer a valid, visually intuitive epidemiological tool to quantify individualized risk and facilitate coordinated family, school, and community interventions.
Methods
Data Sources and Sample
This study analyzed data from the China Family Panel Studies (CFPS), a nationally representative, longitudinal general social survey that provides comprehensive insights into China’s demographic, economic, and public health landscapes.
To explore the prospective associations between ecological environments and adolescent mental health, this study utilized a two-wave longitudinal design. Baseline ecological predictors and demographic covariates were extracted from the 2020 wave, while the outcome variable (depressive symptoms) was assessed in the 2022 follow-up wave. The target population consisted of adolescents aged 10–15 years at the 2020 baseline.
Regarding data cleaning, “invalid responses” (e.g., CFPS codes indicating “refused to answer”, “not applicable”, or “don’t know”) and cases with partial missingness in the multi-item scales were treated as missing data. Given that missingness accounted for a small proportion of the initial eligible pool, listwise deletion was employed to ensure analytical rigor without relying on extensive imputation assumptions. Survey weights and household clustering parameters were not applied in the current machine learning framework, as the primary objective was to evaluate feature importance and build predictive models rather than estimate national prevalence. The final longitudinal sample comprised 971 adolescents, with a mean age of 12.72 ± 1.66 years at baseline. The sample included 522 males, with gender coded as a binary reference variable (0 = Female, 1 = Male). Figure 1 provides a detailed schematic of the screening process.
Figure 1.
Flowchart for screening the study sample.
Notes: T1, Time Point 1 (2020); T2, Time Point 2 (2022).
Measurement
Depressive Symptoms (Outcome)
Adolescent depressive symptoms in 2022 were assessed using the 8-item Center for Epidemiologic Studies Depression Scale (CES-D8).22 The CES-D8 consists of six negative and two positive entries for (1) depressed mood; (2) struggling to get things done; (3) not sleeping well; (4) feeling happy; (5) feeling lonely; (6) happy with life; (7) feeling sad and upset; and (8) feeling that life cannot go on. Positive items were reverse-scored, yielding a continuous total score ranging from 0 to 24, where higher scores indicate more severe depressive symptoms.23 The CES-D8 has demonstrated robust reliability and validity among Chinese populations.24,25 Cronbach’s alpha in this study was 0.768.
We integrated a machine learning framework (Random Forest, RF) and ordinary least squares (OLS) models, utilizing the continuous CES-D8 score as the outcome. For the construction of the nomogram, the outcome was transformed into a binary classification (0 = absence of significant symptoms, 1 = presence of symptoms) using a predefined epidemiological cutoff score. Depressive symptoms were evaluated using the 8-item Center for Epidemiologic Studies Depression Scale (CES-D8). Consistent with established methodological standards, a total score of 9 or higher (≥ 9) was designated as the threshold to define probable binary depression risk.26 It should be noted that baseline CES-D8 scores were not adjusted for in the current models; thus, this study evaluates the prospective predictive value of baseline environments on later symptom severity rather than incident changes.
Influencing Factors
Eleven multidimensional ecological factors across the social, family, school, and Internet domains were measured at the 2020 baseline. While several predictors rely on single-item CFPS standard measures, which inherently limits extensive psychometric evaluation, these items are widely validated in large-scale demographic surveys.
Social
Social Trust
Social trust was measured by trust in different groups in the questionnaire. These were the level of trust in family members, neighbors, strangers, local officials and doctors. Responses were based on a quantitative scale (0–10), with higher scores indicating higher levels of trust in society, and the Cronbach’s alpha coefficient for this study was 0.591. Since the magnitude of the Cronbach’s alpha coefficient increases rapidly with the number of questions, even when all else is equal,27 some researchers have argued that with a very small number of questions, coefficients that appear to be on the low side of the scale in the study have actually been very difficult to obtain.28
Interpersonal Relationships
Interpersonal relationships were measured by the question “How would you rate your popularity among your peers?”. The responses were based on quantitative criteria, with higher scores indicating better interpersonal relationships. The quantitative range is 0–10 points.
Family
Parent-Child Communication
Parent-child communication was measured by “The number of times you talked to your parents in the last month”. Responses were based on quantitative criteria, with higher scores indicating more frequent or closer parent-child communication. The quantitative range is 0–50 times.
Parent-Child Conflict
Measure parent-child conflict by “The number of arguments with parents in the last month”. Responses were based on quantitative criteria, with higher scores indicating more frequent or more severe parent-child conflicts. The quantitative range is 0–50 times.
School
School Satisfaction
School satisfaction was measured by the question “Based on your situation, how satisfied are you with your current school?”. Responses were on a 5-point Likert scale ranging from 1 (dissatisfied) to 5 (very satisfied). The higher the score, the more satisfied you are with your current school.
Teacher Satisfaction
Teacher satisfaction was assessed on the basis of satisfaction with different teachers such as mathematics teachers, language teachers and foreign language teachers. For example, satisfaction with mathematics teachers was assessed on an individual basis. A 5-point Likert scale ranging from 1 (dissatisfied) to 5 (very satisfied) was used. The total score ranged from 3 to 15, with higher scores indicating greater satisfaction with the school’s teachers. The Cronbach’s alpha coefficient for this study was 0.707.
Academic Pressure
Academic pressure was measured by the question “How much academic stress do you consider yourself under?”. Responses were on a 5-point Likert scale ranging from 1 (no stress) to 5 (a lot of stress). Higher scores indicated greater perceived academic stress.
Internet
Online Shopping
Online shopping was indicated by “Whether or not you shop on the Internet”. Answers were given in terms of whether or not (0 = No, 1 = Yes).
Online Gaming
Online gaming was indicated by “Playing games online or not”. Answers were expressed in terms of whether or not (0 = No, 1 = Yes).
Watch Short Videos
Watch short videos was indicated by “Whether to watch short videos online”. Answers were given in terms of whether or not (0 = No, 1 = Yes).
Online Education
Online education was indicated by “Whether participants participated in online education”. Answers were expressed as yes or no (0 = No, 1 = Yes).
Analytical Process
Statistical analyses were executed sequentially through a multi-stage workflow. Data mapping and descriptive statistics were performed using SPSS 24.0, while advanced modeling and validations were executed in R version 4.1.2.
Parametric Regression
First, an ordinary least squares (OLS) linear regression was utilized to evaluate the baseline parametric associations between the 2020 ecological predictors, covariates (age and sex), and the 2022 continuous CES-D8 score. Prior to interpretation, OLS statistical assumptions were rigorously assessed: normality was checked via Q-Q plots, homoscedasticity via residuals versus fitted plots, and multicollinearity was quantified using the Variance Inflation Factor (VIF), with a VIF < 5 confirming the absence of collinearity.
Machine Learning Architecture
Second, to transcend the limitations of OLS in capturing non-linear relationships and complex interactions, a Random Forest (RF) algorithm was deployed utilizing the randomForest package. All predictors, alongside age and sex, were included simultaneously. Hyperparameters were tuned for optimal performance, establishing the forest architecture with 500 decision trees (ntree = 500) and 4 variables tried at each split (mtry = 4). A random seed was set for reproducibility. Model performance was evaluated using the Out-of-Bag (OOB) Mean Squared Error (MSE). Feature importance was quantified using the Increase in Node Purity (IncNodePurity) metric.
Nomogram Construction and Validation
Finally, to translate the high-dimensional findings into an intuitive epidemiological tool, a prognostic nomogram was established based on a multivariable logistic regression model utilizing the binary depression outcome. The nomogram’s performance was subjected to rigorous internal validation using 1000 bootstrap resamples. Discrimination was measured using the bias-corrected concordance index (C-index, equivalent to the AUC), while calibration was evaluated visually via a calibration curve and quantitatively through the Mean Absolute Error (MAE).
Results
Correlation Analysis of Research Variables
The correlation results shown in Figure 2 indicated that school satisfaction (r = −0.09, p < 0.01), teacher satisfaction (r = −0.12, p < 0.01), and social trust (r = −0.13, p < 0.01) were negatively correlated with depressive symptoms. Conversely, parent-child conflict (r = 0.12, p < 0.01), academic pressure (r = 0.14, p < 0.01), and online shopping (r = 0.08, p < 0.01) were positively associated with depressive symptoms. Specifically, depressive symptoms showed the strongest associations with school-related environmental factors and weaker associations with online-related factors. In addition, age (r = 0.11, p < 0.01) and sex (r = −0.12, p < 0.01) were also significantly associated with adolescent depressive symptoms. All correlation coefficients between the independent variables were below 0.80, suggesting no extreme bivariate collinearity issues. To investigate whether multicollinearity existed among the independent variables, tolerance and variance inflation factor (VIF) were examined, with tolerance values below 0.1 or VIF values above 10 considered as evidence of multicollinearity. The results showed that all variables had tolerance values greater than 0.1 and VIF values less than 10, indicating that no severe multicollinearity was present among the predictors. As shown in Table 1.
Figure 2.
Pearson correlation matrix of ecological factors and adolescent depressive symptoms.
Abbreviations: The lower triangle displays the Pearson correlation coefficients. The upper triangle indicates statistical significance, with color intensity reflecting correlation strength (color bar on the right). Significance levels are denoted as follows: *p < 0.05, **p < 0.01, ***p < 0.001. Diagonal values represent self-correlations (r = 1.00). CESD8, Center for Epidemiologic Studies Depression Scale 8-item score; PCCF, Parent-child conflict; PCCM, Parent-child communication; AP, Academic pressure; SS, School satisfaction; TS, Teacher satisfaction; IR, Interpersonal relationships; ST, Social trust; OG, Online gaming; OS, Online shopping; WSV, Watching short videos; OE, Online education.
Table 1.
Variance Inflation Factors (VIF) for the Independent Variables
| Independent Variables | Tolerance | VIF |
|---|---|---|
| Self | ||
| Sex (Male) | 0.834 | 1.200 |
| Age | 0.849 | 1.178 |
| Social | ||
| Social trust | 0.894 | 1.119 |
| Interpersonal relationship | 0.945 | 1.058 |
| Family | ||
| Parent-child communication | 0.971 | 1.030 |
| Parent-child conflict | 0.972 | 1.029 |
| School | ||
| School satisfaction | 0.627 | 1.594 |
| Teacher satisfaction | 0.629 | 1.590 |
| Academic pressure | 0.969 | 1.032 |
| Internet | ||
| Online Shopping (Yes) | 0.860 | 1.162 |
| Online Game (Yes) | 0.863 | 1.159 |
| Online Education (Yes) | 0.979 | 1.021 |
| Watch short videos (Yes) | 0.974 | 1.027 |
Factors Affecting the Development of Depressive Symptoms in Adolescents
Linear regression was used in this study to examine the associations of various factors with adolescent depressive symptoms (Model fit: adjusted R2 = 0.063). The results showed that among the social dimensions, social trust (B =-0.060, p = 0.002) was significantly and inversely associated with depressive symptoms. In the family dimension, parent-child conflict (B = 0.086, p = 0.005) was positively associated with higher depressive symptom scores. In the school dimension, academic presssure (B = 0.355, p < 0.001)was significantly and positively associated with depressive symptoms, whereas teacher satisfaction (B = -0.133, p = 0.041) showed significant inverse associations. On the Internet use dimension, none of the factors exhibited significant associations. In addition, for individual demographic characteristics, sex (B = -0.798, p = 0.001) was significantly associated with depressive symptoms (with females exhibiting higher scores given the male binary coding reference), and age (B = 0.175, p = 0.016) exhibited a significant positive association, indicating that older adolescents reported more severe depressive symptoms. As shown in Table 2.
Table 2.
Linear Regression Analysis of Ecological Factors Associated with Depressive Symptoms in Adolescents
| Factors | B | 95% CI | p |
|---|---|---|---|
| Self | |||
| Sex (Male) | −0.798 | [−1.278, −0.318] | 0.001 |
| Age | 0.175 | [0.032, 0.317] | 0.016 |
| Social | |||
| Social trust | −0.060 | [−0.097, −0.023] | 0.002 |
| Interpersonal relationship | 0.004 | [−0.110, 0.118] | 0.949 |
| Family | |||
| Parent-child communication | −0.034 | [−0.084, 0.015] | 0.170 |
| Parent-child conflict | 0.086 | [0.027, 0.145] | 0.005 |
| School | |||
| School satisfaction | 0.038 | [−0.258, 0.334] | 0.800 |
| Teacher satisfaction | −0.133 | [−0.260, −0.005] | 0.041 |
| Academic pressure | 0.355 | [0.152, 0.559] | <0.001 |
| Internet | |||
| Online Shopping (Yes) | 0.201 | [−0.359, 0.761] | 0.481 |
| Online Game (Yes) | 0.179 | [−0.303, 0.661] | 0.466 |
| Online Education (Yes) | 0.133 | [−0.311, 0.578] | 0.555 |
| Watch short videos (Yes) | −0.159 | [−0.742, 0.423] | 0.592 |
Random Forest Results and Variable Importance
To capture potential non-linear relationships and complex interactions, a Random Forest (RF) algorithm was deployed. Unlike traditional stepwise selection, all 13 ecological predictors and demographic covariates (including age and sex) were entered simultaneously into the model as independent predictors. The model was constructed with 500 decision trees (ntree = 500) and 4 variables tried at each split (mtry = 4). The RF regression model yielded an Out-of-Bag (OOB) Mean Squared Error (MSE) of 12.72.
Feature importance was quantified using both the percentage increase in Mean Squared Error (%IncMSE) and the Increase in Node Purity (IncNodePurity) metrics. As visualized in the dual-metric importance plot (Figure 3) and the standardized importance bar chart (Figure 4), the analysis revealed that social trust (ST) was the most critical predictor of adolescent depressive symptoms. The top predictors in descending order of their contribution were social trust, interpersonal relationships, parent-child conflict, teacher satisfaction, and age. This non-linear approach highlights the substantial predictive importance of demographic factors (age) and social buffering mechanisms (social trust) alongside proximal family and school stressors. The specific importance values (Increase in Node Purity, IN) of each independent predictor derived from the Random Forest model are detailed in Table 3.
Figure 3.
Variable importance in the Random Forest model (%IncMSE and IncNodePurity).
Abbreviations: PCCF, Parent-child conflict; PCCM, Parent-child communication; AP, Academic pressure; SS, School satisfaction; TS, Teacher satisfaction; IR, Interpersonal relationships; ST, Social trust; OG, Online gaming; OS, Online shopping; WSV, Watching short videos; OE: Online education.
Figure 4.
Variable importance bar chart of the Random Forest model measured by Increase in Node Purity (IncNodePurity).
Abbreviations: PCCF, Parent-child conflict; PCCM, Parent-child communication; AP, Academic pressure; SS, School satisfaction; TS, Teacher satisfaction; IR, Interpersonal relationships; ST, Social trust; OG, Online gaming; OS, Online shopping; WSV, Watching short videos; OE, Online education.
Table 3.
Scale of Importance of IN Index for Independent Variables
| Factors | IN |
|---|---|
| Social trust | 1981.0259 |
| Interpersonal relationships | 1205.1218 |
| Parent-child conflict | 1095.6452 |
| Teacher satisfaction | 1026.6001 |
| Age | 1016.8986 |
| Parent-child communication | 1000.1393 |
| Academic pressure | 902.0598 |
| School satisfaction | 577.2840 |
| Online Game | 362.8828 |
| Online Education | 359.4229 |
| Sex | 357.8919 |
| Online Shopping | 302.4129 |
| Watch short videos | 233.2183 |
Note: IN index: Increase in Node Purity, a measure of variable importance derived from the random forest algorithm.
Nomogram Construction and Validation
Based on the identified significant predictors, a prognostic nomogram was constructed to translate high-dimensional statistical arrays into an intuitive visual tool for estimating individualized risk probabilities of clinically significant depressive symptoms (Figure 5). Categorical variables were appropriately mapped as discrete points, while continuous variables were represented on scales.
Figure 5.
Nomogram of factors influencing depressive symptoms.
Abbreviations: PCCF, Parent-child conflict; AP, Academic pressure; TS, Teacher satisfaction; ST, Social trust.
Crucially, the nomogram’s performance was subjected to rigorous internal validation using 1000 bootstrap resamples. The model demonstrated a modest discriminative ability, yielding a bootstrap-corrected C-index (AUC) of 0.614, which was further corroborated by the ROC curve analysis (Figure 6, AUC = 0.642, 95% CI: 0.590-0.694). Furthermore, the model exhibited excellent calibration. As depicted in the calibration curve (Figure 7), the bias-corrected line aligned closely with the ideal reference line, supported by a remarkably low Mean Absolute Error (MAE) of 0.011. These metrics indicate that while the nomogram is not a definitive clinical diagnostic test, it serves as a highly reliable and calibrated epidemiological tool for visualizing risk trajectories based on ecological profiles.
Figure 6.
ROC curve for predicting adolescent depressive symptoms.
Notes: bootstrap-corrected AUC = 0.614, 95% CI: 0.590-0.694 . The solid line represents the ROC curve of the model, and the dashed diagonal line represents the reference line (chance level, AUC = 0.5).
Figure 7.
Calibration curve of the Nomogram model for predicting adolescent depressive symptoms.
Notes: Apparent (dotted line), Bias-corrected (solid line), Ideal (dashed red line), and C.L. (confidence limit). Evaluated via 1000 bootstrap resamples; Mean Absolute Error = 0.011.
Discussion
Adolescence is a critical stage of life development, during which individuals are challenged by complex issues such as escalating academic pressure and evolving interpersonal relationships. Consequently, adolescents often exhibit an elevated risk for psychological distress, making depressive symptoms a major mental health concern within this demographic. While previous studies have explored individual correlates of adolescent depression, they have frequently overlooked the combined predictive value and relative importance of these multi-domain ecological factors. Therefore, the present study aimed to explore the relative associations between various ecological domains and depressive symptoms in adolescents.
The Associations of Ecological Factors with Adolescent Depression
Research consistently indicates that female adolescents report higher levels of depressive symptoms than males, a phenomenon that can be interpreted through several intersecting psychosocial frameworks. First, adolescent girls often navigate more complex interpersonal networks and may be more susceptible to peer-related relational stress.29 Second, females are typically observed to possess higher affective sensitivity to environmental stressors, increasing their likelihood of experiencing internalizing symptoms such as grief and anxiety.30 In addition, traditional sex role socialization impacts adolescents’ mental health; males are often encouraged to exhibit autonomy and assertiveness, whereas females are frequently socialized toward empathy and hyper-sensitivity to the needs of others.31 Furthermore, age emerged as an important demographic correlate. In our models, age exhibited a positive association with depression, indicating that older adolescents reported more severe depressive symptoms. Rather than acting as a protective factor, this trajectory may reflect the cumulative psychological burden of increasingly rigorous academic demands and intensifying social stressors as adolescents progress toward late high school.
Our study further explored environmental correlates using Random Forest algorithms and nomograms to rank and visualize their predictive importance. Notably, social trust emerged as the strongest predictor of adolescent depressive symptoms. Interpreted through the lens of Ecological Systems Theory, social trust acts as a vital macrosystemic resource that shapes how adolescents perceive their immediate microsystems. High social trust provides individuals with a cognitive schema of a safe, predictable, and supportive broader environment, which is inversely associated with psychological distress.32 Conversely, low social trust reflects a profound sense of insecurity and alienation, leaving adolescents more vulnerable to depressive symptoms.13 Rather than acting in isolation, social trust serves as a foundational psychological safety net that modulates an adolescent’s capacity to process proximal stressors.
In the school setting, academic pressureand teacher satisfaction were significantly associated with adolescent depression. Adolescents reporting higher levels of academic pressure were more likely to exhibit depressive symptoms, particularly in the context of high-stakes testing where performance pressure correlates with severe anxiety.12 Conversely, positive evaluations of school lifeare key components of overall adolescent well-being.33 Higher teacher and school climate were negatively correlated with depressive symptoms, suggesting that supportive teacher-student interactions and a positive school climate serve as crucial ecological buffers.34 Students who are satisfied with their educational environment may be more likely to exhibit positive engagement and adaptive coping mechanisms.
Within the family setting, parent-child conflict demonstrated a robust positive association with adolescent depressive symptom levels. Frequent friction in the household is strongly correlated with emotional dysregulation and psychological distress.11 Strained parent-child dynamics are frequently conceptualized as markers of insecure attachment, which can foster feelings of neglect and are linked to internalizing problems.35
Regarding the online environment, engagement in online shopping was positively associated with depressive symptoms in initial exploratory and univariate evaluations, though its relative effect size was small. While some regional studies suggest that internet-driven consumerism may correlate with lower subjective well-being among youth,36,37 our binary measurement (“yes/no”) strictly precludes any evaluation of frequency, overindulgence, or dose-response effects. Therefore, the observed association must be interpreted cautiously; it may reflect specific behavioral phenotypes, unmeasured cultural specificities, or residual confounding, rather than a direct psychological consequence of online shopping itself.
Methodological Implications
By utilizing a Random Forest approach, this study mapped the relative predictive importance of various ecological domains, allowing for the visualization of complex associations. The nomogram quantifies these variables into an epidemiological visual tool. However, the application and interpretation of these machine learning tools warrant strict caution. Without external validation, and given the modest predictive accuracy and calibration of the models, these findings cannot currently support clinical diagnostic screening or the definitive prioritization of localized intervention resources. The findings are best utilized to conceptualize potential contextual targets for supportive initiatives—such as addressing family communication—rather than confirming direct causal pathways for clinical intervention.
Strengths and Limitations
The primary strength of this study lies in its innovative methodological approach, which integrates traditional regression analysis with advanced machine learning algorithms (Random Forest) to evaluate complex, multi-dimensional ecological data. This dual approach not only identifies the relative predictive importance of diverse environmental and demographic factors but also constructs a comprehensive ecological framework for understanding adolescent depressive symptoms. Furthermore, by using a nomogram, the study successfully translates high-dimensional statistical arrays into an intuitive visual tool, offering a novel epidemiological perspective on how combined ecological factors are associated with mental health risks.
Despite these methodological strengths, several limitations must be acknowledged. First, although this study used a two-wave prospective design, the lack of adjustment for baseline depressive symptoms and its observational nature fundamentally preclude definitive causal inferences. The findings remain susceptible to reverse causality, as adolescents with unmeasured prior depressive symptoms may inherently perceive and report their environments more negatively. Second, the reliance on self-reported questionnaires introduces potential social desirability and recall biases, while the use of binary measurements for certain behaviors (such as online shopping) restricts a more nuanced evaluation of exposure. Additionally, the study remains vulnerable to residual confounding from unmeasured factors, such as baseline psychiatric history, precise socioeconomic status, and genetic predispositions, alongside potential selection bias from excluded participants.
Finally, while machine learning models provide robust analytical capabilities, they are inherently susceptible to overfitting and the potential for complex feature interactions was not exhaustively examined. Because the predictive models generated in this study lack external validation in independent cohorts, their current applicability for clinical diagnostic screening or resource allocation is severely limited. To address these constraints, future longitudinal studies incorporating multi-informant designs, continuous behavioral measurements, and externally validated algorithms are essential to establish temporal precedence and further substantiate these ecological associations.
Conclusion
This observational study used linear regression, Random Forest algorithms, and nomogram visualization to explore the ecological and demographic factors associated with depressive symptoms in adolescents. The findings indicated that adolescent depressive symptoms are associated with a complex interplay of multiple variables. Based on the variable importance ranking, social trust emerged as the most critical predictor, followed by interpersonal relationships, parent-child conflict, teacher satisfaction, and age.
Rather than confirming causal effects, these results highlight potential contextual targets for future mental health initiatives. The identified associations suggest that macrosystemic factors (such as social trust) and proximal micro-environmental elements (such as family dynamics) are vital components of adolescent well-being. Furthermore, the positive association between age and depressive symptoms, along with the higher vulnerability in females, suggests that older adolescents and adolescent girls might warrant closer psychological attention.
While the developed nomogram provides an intuitive visualization of these multi-factor associations, its modest discriminative performance, good calibration, and lack of external validation mean it is not currently suitable for clinical diagnostic screening or definitive resource allocation. Finally, the interpretation of these findings requires strict caution. The observational nature and design preclude any causal inferences, and the study’s generalizability is constrained by the exclusion of participants with missing data, reliance on self-reported measures, potential unmeasured residual confounding, and the relatively small effect sizes of certain variables. Future longitudinal research incorporating externally validated predictive models is needed to substantiate these associations and further explore their temporal dynamics.
Funding Statement
This research has been supported by the National Social Science Foundation of China (Grant No. 24BSH122), awarded to Prof. Songli Mei.
Ethical Approval
This study is a secondary analysis of de-identified, publicly available data from the China Family Panel Studies (CFPS). The CFPS project received ethical approval from the Peking University Biomedical Institutional Review Board (approval number: IRB00001052-14010).
As our research involves the use of (1) lawfully obtained, publicly available data and (2) anonymized information data, it qualifies for an exemption from further ethical review according to Article 32, Items 1 and 2 of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects (jointly issued by the National Health Commission and other competent authorities of China, effective February 18, 2023).
To comply fully with the journal’s editorial policies, we additionally submitted this study to our local Institutional Review Board. The research was reviewed and granted an exemption by the Ethics Committee of the School of Public Health, Jilin University (Ethics Approval Exemption Number: 2025-10-05).
Disclosure
The authors report no conflicts of interest in this work.
References
- 1.Zhang H, Yu PL, Liu XM, et al. Predictive factors for the development of depression in children and adolescents: a clinical study. Front Psychiatry. 2024;15:1460801. doi: 10.3389/fpsyt.2024.1460801 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Havaei F, Ma A, Staempfli S, et al. Nurses’ workplace conditions impacting their mental health during COVID-19: a cross-sectional survey study. Healthcare. 2021;9(1):84. doi: 10.3390/healthcare9010084 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Li ZY, Li J, Liang JH, et al. Depressive symptoms among children and adolescents in China: a systematic review and meta-analysis. Med Sci Monit. 2019;25:7459–15. doi: 10.12659/Msm.916774 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Johnson D, Dupuis G, Piche J, et al. Adult mental health outcomes of adolescent depression: a systematic review. Depression Anxiety. 2018;35(8):700–716. doi: 10.1002/da.22777 [DOI] [PubMed] [Google Scholar]
- 5.Rowe R, Maughan B, Eley TC. Links between antisocial behavior and depressed mood: the role of life events and attributional style. J Abnormal Child Psychol. 2006;34(3):293–302. doi: 10.1007/s10802-006-9032-0 [DOI] [PubMed] [Google Scholar]
- 6.World Health Organization. Adolescent mental health. World Health Organization. Available from: https://www.who.int/zh/news-room/fact-sheets/detail/adolescent-mental-health. Accessed August 1, 2026. [Google Scholar]
- 7.Lu BQ, Lin LX, Su XJ. Global burden of depression or depressive symptoms in children and adolescents: a systematic review and meta-analysis. J Affect Disord. 2024;354:553–562. doi: 10.1016/j.jad.2024.03.074 [DOI] [PubMed] [Google Scholar]
- 8.Kessler RC, Berglund P, Demler O, et al. Lifetime prevalence and ge-of-onset distributions of DSM-IV disorders in the national comorbidity survey replication.(vol 62, pg 593, 2005). Arch Gen Psychiatry. 2005;62(6):593–602. doi: 10.1001/archpsyc.62.6.593 [DOI] [PubMed] [Google Scholar]
- 9.Bronfenbrenner U. Ecological Systems Theory. American Psychological Association; 2000. [Google Scholar]
- 10.Jin Y, Liu JY, Li P, et al. Longitudinal associations between family conflict, intergenerational transmission, and adolescents’ depressive symptoms: evidence from China family panel studies (2016-2020). Child Adolescent Psychiatry Mental Health. 2025;19(1):10. doi: 10.1186/s13034-025-00866-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Nowakowski-Sims E, Rowe A. The relationship between childhood adversity, attachment, and internalizing behaviors in a diversion program for child-to-mother violence. Child Abuse Neglect. 2017;72:266–275. doi: 10.1016/j.chiabu.2017.08.015 [DOI] [PubMed] [Google Scholar]
- 12.Fröjd SA, Nissinen ES, Pelkonen MUI, et al. Depression and school performance in middle adolescent boys and girls. J Adolescence. 2008;31(4):485–498. doi: 10.1016/j.adolescence.2007.08.006 [DOI] [PubMed] [Google Scholar]
- 13.Fahmi M, Panjaitan NA, Habibie I, et al. Does your neighborhood protect you from being depressed? A study on social trust and depression in Indonesia. Bmc Public Health. 2019;19(1):1371. doi: 10.1186/s12889-019-7657-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Kehusmaa J, Ruotsalainen H, Männikkö N, et al. The association between the social environment of childhood and adolescence and depression in young adulthood - A prospective cohort study. J Affect Disord. 2022;305:37–46. doi: 10.1016/j.jad.2022.02.067 [DOI] [PubMed] [Google Scholar]
- 15.Khan A, Uddin R, Lee EY. Excessive recreational Internet use was associated with poor mental health in adolescents. Acta Paediatrica. 2021;110(2):571–573. doi: 10.1111/apa.15528 [DOI] [PubMed] [Google Scholar]
- 16.Twenge JM, Joiner TE, Rogers ML, et al. Increases in depressive symptoms, suicide-related outcomes, and suicide rates among US adolescents after 2010 and links to increased new media screen time. Clin Psychol Sci. 2018;6(1):3–17. doi: 10.1177/2167702617723376 [DOI] [Google Scholar]
- 17.Futoma J, Morris J, Lucas J. A comparison of models for predicting early hospital readmissions. J Biomed Informat. 2015;56:229–238. doi: 10.1016/j.jbi.2015.05.016 [DOI] [PubMed] [Google Scholar]
- 18.Zhou Y, Zhang XL, Gong J, et al. Identifying the risk of depression in a large sample of adolescents: an artificial neural network based on random forest. J Adolescence. 2024;96(7):1485–1497. doi: 10.1002/jad.12357 [DOI] [PubMed] [Google Scholar]
- 19.Lv JP, Ren H, Guo XM, et al. Nomogram predicting bullying victimization in adolescents. J Affect Disord. 2022;303:264–272. doi: 10.1016/j.jad.2022.02.037 [DOI] [PubMed] [Google Scholar]
- 20.Chung SM, Park JC, Moon JS, et al. Novel nomogram for screening the risk of developing diabetes in a Korean population. Diabetes Res Clin Pract. 2018;142:286–293. doi: 10.1016/j.diabres.2018.05.036 [DOI] [PubMed] [Google Scholar]
- 21.Yarkoni T, Westfall J. Choosing prediction over explanation in psychology: lessons from machine learning. Perspect Psychol Sci. 2017;12(6):1100–1122. doi: 10.1177/1745691617693393 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psycholog Measurement. 1977;1(3):385–401. doi: 10.1177/014662167700100306 [DOI] [Google Scholar]
- 23.Eikemo TA, Bambra C, Huijts T, et al. The first pan-european sociological health inequalities survey of the general population: the european social survey rotating module on the social determinants of health. Euro Sociol Rev. 2017;33(1):137–153. doi: 10.1093/esr/jcw019 [DOI] [Google Scholar]
- 24.Zhao HF, Shi H, Ren Z, et al. The mediating role of extra-family social relationship between personality and depressive symptoms among Chinese adults. Int J Public Health. 2022;67:1604797. doi: 10.3389/ijph.2022.1604797 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Zhou T, Cheng G, Wu XH, et al. The associations between sleep duration, academic pressure, and depressive symptoms among Chinese adolescents: results from China family panel studies. Int J Environ Res Public Health. 2021;18(11):6134. doi: 10.3390/ijerph18116134 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Pan B, Ge P, Liu R, et al. Association between educational attainment and depressive symptom severity among young-old adults aged 60–74 years: the mediating role of self-reported memory function. Behav Sci. 2026;16(8):1264. doi: 10.3390/bs16081264 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Guan S. A review of Cronbach’s alpha coefficient studies. Psychol Sci. 2009;32(03):685–687. doi: 10.16719/j.cnki.1671-6981.2009.03.019 [DOI] [Google Scholar]
- 28.Webster GD, Bryan A. Sociosexual attitudes and behaviors: why two factors are better than one. J Res Personality. 2007;41(4):917–922. doi: 10.1016/j.jrp.2006.08.007 [DOI] [Google Scholar]
- 29.Kuehner C. Why is depression more common among women than among men? Lancet Psychiatry. 2017;4(2):146–158. doi: 10.1016/S2215-0366(16)30263-2 [DOI] [PubMed] [Google Scholar]
- 30.Chaplin TM, Hong K, Bergquist K, et al. Gender differences in response to emotional stress: an assessment across subjective, behavioral, and physiological domains and relations to alcohol craving. Alcohol Clin Exp Res. 2008;32(7):1242–1250. doi: 10.1111/j.1530-0277.2008.00679.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Gibson PA, Baker EH, Milner AN. The role of sex, gender, and education on depressive symptoms among young adults in the United States. J Affect Disord. 2016;189:306–313. doi: 10.1016/j.jad.2015.08.067 [DOI] [PubMed] [Google Scholar]
- 32.Rothstein B, Uslaner EM. All for all - Equality, corruption, and social trust. World Politics. 2005;58(1):41. doi: 10.1353/wp.2006.0022 [DOI] [Google Scholar]
- 33.Holopainen L, Waltzer K, Hoang N, et al. The relationship between students’ self-esteem, schoolwork difficulties and subjective school well-being in finnish puper-secondary education. Int J Educ Res. 2020;104:101688. doi: 10.1016/j.ijer.2020.101688 [DOI] [Google Scholar]
- 34.Wang MT, Eccles JS. Social support matters: longitudinal effects of social support on three dimensions of school engagement from middle to high school. Child Development. 2012;83(3):877–895. doi: 10.1111/j.1467-8624.2012.01745.x [DOI] [PubMed] [Google Scholar]
- 35.Bradford AB, Burningham KL, Sandberg JG, et al. The association between the parent-child relationship and symptoms of anxiety and depression: the roles of attachment and perceived spouse attachment behaviors. J Marital Family Ther. 2017;43(2):291–307. doi: 10.1111/jmft.12190 [DOI] [PubMed] [Google Scholar]
- 36.Ma YA, Gu JF. Internet and depression in adolescents: evidence from China. Front Psychol. 2023;14:1026920. doi: 10.3389/fpsyg.2023.1026920 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Seki T, Hamazaki K, Natori T, et al. Relationship between internet addiction and depression among Japanese university students. J Affect Disord. 2019;256:668–672. doi: 10.1016/j.jad.2019.06.055 [DOI] [PubMed] [Google Scholar]