This longitudinal study used data from the China Family Panel Study (CFPS), which covers populations from 25 provinces, municipalities, and autonomous regions in China. The CFPS uses a multistage sampling approach with implicit stratification based on urban-rural distinctions and regional socioeconomic status. Using a probability proportional to size sampling strategy, 32 streets and townships from Shanghai and 144 districts and counties from the other 24 provinces were selected as primary sampling units. Subsequently, 640 communities were randomly chosen from these units, and 25 households were systematically sampled from each community. The baseline survey in 2010 collected data from 33,600 adults and 8990 children across 14,960 families, who were invited to participate in follow-up surveys every 2 years. These surveys gathered comprehensive information on participants’ economic conditions, education, lifestyle, family relationships, and physical and mental health.

For this study, we used 3 waves of data from the 2014-2018 cycle of the CFPS to ensure consistent measurement of internet use duration. In 2014, the sample included 2567 adolescents aged 10-15 years, of whom 1491 completed surveys in both the 2016 and 2018 cycles. By 2018, the participants’ age ranged from 14 to 19 years. To capture longitudinal developmental patterns, adolescents with self-reported data from at least 2 of the 3 waves were included, resulting in a final sample of 2237 participants. The flow of participants in this study is illustrated in Figure 1.

The statistical analysis involved a 2-step procedure. In the first step, we conducted latent class growth modeling (LCGM) to determine trajectories of internet use across the 3 waves using Mplus (version 8.3; Muthén & Muthén). LCGM is a technique that identifies subgroups with distinct developmental patterns within a population over time [35]. With the 3-wave data, both linear and quadratic models were fitted to determine the optimal number of latent trajectory classes. The best-fitting model was determined based on the Akaike information criterion (AIC), Bayesian information criterion (BIC), sample size–adjusted BIC (aBIC), entropy, the Lo-Mendell-Rubin likelihood ratio test (LMR), and the bootstrapped likelihood ratio test (BLRT). Lower values of AIC, BIC, and aBIC indicate a better model fit. Higher entropy represents better classification accuracy. A significant P value (P<.05) for LMR and BLRT in a k-class model suggests that the model is better than the k–1 class model. Further, the sample size was taken into account, with at least 5% of the total population required in each class to define a meaningful latent entity [36].

In the second step, multivariable logistic regressions were conducted within the best-fitting model to examine the associations between internet use trajectories and distal outcomes. Each individual was assigned to their most likely class based on the highest posterior probability estimate, which was then used as the independent variable. The outcome variable was depressive symptoms in 2018. Three regression models were constructed. In model 1, adjustments were made for baseline K6 scores. Model 2 included additional demographic covariates, and model 3 further adjusted for lifestyle and health-related covariates. Stratified analysis was performed based on urban and rural classifications.

The Little missing completely at random (MCAR) test indicated that missing data for internet use followed an MCAR pattern (χ²₉=6.616; P=.68). Therefore, missing data on internet use were handled using the robust full information maximum-likelihood estimation approach [37]. Missing data on covariates and the distal outcome were imputed using the missForest function in R (R Project for Statistical Computing), a method capable of handling both continuous and categorical variables [38]. We imputed missing values for the father’s educational level (n=32, 1.4%), mother’s educational level (n=41, 1.8%), family income (n=118, 5.4%), grandparental caregiving (n=13, 0.6%), residential area (n=15, 0.6%), physical activity (n=11, 0.5%), sleep duration (n=28, 1.2%), BMI (n=154, 7.1%), perceived health condition in 2014 (n=5, 0.2%), and depressive symptoms in 2018 (n=467, 21.5%).

The Biomedical Ethics Review Committee of Peking University approved the study (IRB00001052-14,010). All eligible participants provided informed consent. The questionnaire was anonymous and did not collect any identifying information, such as name or address. Eligible participants were given CN ¥20-CN ¥60 (US $3.26-US $9.77) in cash as compensation.

Table 1 presents the characteristics of the 2237 participants. The sample consisted of 1067 (47.7%) girls and 1170 (52.3%) boys, with a mean age of 12.46 (SD 1.73) years in 2014. Approximately half of the participants’ parents had a high school education or higher (father’s educational level: n=1187, 53.1%; mother’s educational level: n=938, 41.9%). The median annual household income was CN ¥39,000 (IQR CN ¥20,100-CN ¥61,200; US $3272-US $9962.60). A total of 60.4% (n=1351) of the participants were from rural areas, and 17.3% (n=386) were cared for by their grandparents. The median weekly hours of internet use were 0.0 (IQR 0.0-3.0) in 2014, 3.0 (IQR 0.0-8.0) in 2016, and 7.0 (IQR 2.0-20.0) in 2018. In 2018, the prevalence of depressive symptoms was 19.8% (n=444). CESD-20 scores in 2018 were positively correlated with K6 scores in 2014 (Spearman ρ=0.166; P<.001).

Table 2 presents the fit indices from the LCGM analysis for models with 1-4 trajectory classes. First, linear models were fitted. Reductions in AIC, BIC, and aBIC were observed across 1- to 4-class models. While both the LMR and BLRT statistics favored the 3-class model over the 2-class model, the latter exhibited higher entropy (0.909) with its smallest group >5%. Consequently, the 2-class model was selected as the best-fitting linear model. Subsequently, quadratic models were fitted, showing a similar pattern in fit indices to the linear models. The AIC, BIC, and aBIC values indicated an improvement in model fit upon adding the quadratic term. Therefore, the 2-class quadratic model was identified as the final optimal model of internet use trajectories.

Figure 2 shows the 2 latent trajectory classes of internet use. Class 1 (89.8%) is labelled the “low-growth group,” representing the majority of adolescents who, on average, had relatively short internet use duration in 2014 (intercept=2.723; P<.001) and experienced gradual growth over 4 years (slope=1.080; P=.01; quadratic term=0.998; P<.001). Class 2 (10.2%) is labeled the “high-growth group,” characterized by initially longer average internet use duration (intercept=8.421; P<.001) and rapid growth until 2018 (slope=20.183; P=.008; quadratic term=–1.293; P=.74).

There were statistically significant differences across the trajectory classes in the distributions of age, parental educational level, annual family income, and residential area (Multimedia Appendix 1, Table S1). Specifically, urban participants were more likely to belong to the low-growth group compared to their rural counterparts.

The percentage of adolescents with depressive symptoms in 2018 was 19.2% (385/2008) in the low-growth group and 25.8% (59/229) in the high-growth group. The difference in the percentages of depressive symptoms between the 2 groups was 6.6% (95% CI 1.1%-12%). Table 3 presents the results from multivariable logistic regressions examining the association between internet use trajectories and depressive symptoms. After adjusting for baseline psychological distress (model 1), the high-growth group was associated with higher odds of depressive symptoms compared to the low-growth group (OR 1.460, 95% CI 1.062-2.008; P=.02). The association remained significant after further adjustment for demographic, lifestyle, and health-related variables (model 3), with the high-growth group showing higher odds of depressive symptoms (OR 1.520, 95% CI 1.090-2.120; P=.01).

When stratified by residential area, the prevalence rates of depressive symptoms were 18.7% (232/1241) in the low-growth group and 28.2% (31/110) in the high-growth group among rural participants. The difference in the percentages of depressive symptoms between the 2 groups was 9.5% (95% CI 1.8%-17.2%). Results from the multivariable logistic regression indicated that the high-growth group was associated with higher odds of depressive symptoms among rural participants (OR 1.853, 95% CI 1.162-2.955; P=.01). Among urban participants, the prevalence rates were 19.9% (153/767) for the low-growth group and 23.5% (28/119) for the high-growth group. The difference in the percentages of depressive symptoms between the 2 groups was 3.6% (95% CI –4.2% to 11.4%). In the multivariable logistic regression, the association between internet use trajectories and depressive symptoms was not significant among urban participants (OR 1.124, 95% CI 0.686-1.840; P=.64).

We performed a sensitivity analysis to assess the robustness of our findings (Multimedia Appendix 1, Table S2). This analysis included additional adjustments for the CESD-20 scores in 2016. The results were consistent with the main analysis, with the association between internet use trajectories and depressive symptoms remaining significant among rural participants.

In this nationally representative study using data from the CFPS, 2 latent trajectory classes of internet use were identified among Chinese adolescents aged 10 to 19 years: a low-growth group (89.8%) and a high-growth group (10.2%). After delineating these distinct trajectories, we examined their associations with depressive symptoms and found that the high-growth group was associated with higher odds of depressive symptoms. Furthermore, our stratified analysis revealed variations in this association based on residential area. Specifically, while the association between trajectories of internet use and depressive symptoms remained significant among rural adolescents, it was not observed among their urban counterparts.

In this study, we observed a notable increase in the duration of internet use among adolescents over 4 years. This trend aligns with previous research indicating that adolescent internet use intensifies with age [18]. Zhu et al [39] found that adolescents tend to allocate more time to chatting or browsing the internet as they age. This increase in internet use among adolescents can be linked to the popularity of social media platforms, as well as their inclination toward social interaction and the development of virtual identities [39]. Furthermore, our analysis revealed heterogeneous developmental patterns within this trend, with 89.8% of adolescents exhibiting gradual growth in their internet use duration. In comparison, the remaining 10.2% increased their internet use to an average level approximately 5 times the baseline by the end of the follow-up. The diverse developmental patterns of these trajectories underscore the importance of early identification of excessive internet use. Understanding the contextual drivers of internet use behaviors is essential for altering the upward trajectory.

The high-growth group was associated with higher odds of depressive symptoms compared to the low-growth group. Our findings align with prior longitudinal research indicating that adolescent internet use can act as a risk factor for developing depressive symptoms [40]. However, previous studies have largely focused on static measures of internet use levels without examining how internet use evolves during adolescence. This study contributes to the existing evidence by demonstrating how trajectories of internet use can predict subsequent depressive symptoms. Various explanations may account for the observed association between internet use and depressive symptoms. For example, internet use may affect adolescents’ mental well-being by reducing time spent on physical activity and sleep [13,41]. However, even after adjusting for lifestyle factors, the association remained significant in this study, suggesting that time reallocation alone may not fully explain the relationship. Another possibility is that excessive internet use leads to reduced face-to-face interaction [42] and worsened academic performance [43], thereby increasing the risk of depressive symptoms among adolescents.

This study identified an urban-rural difference in how internet use trajectories were associated with depressive symptoms, specifically affecting adolescents residing in rural areas. Previous research has consistently highlighted the adverse effects of internet use on the mental well-being of rural adolescents in China [44,45]. For example, Zhou and Ding [46] examined the impacts of internet use across diverse adolescent groups and found stronger associations with mental health outcomes among those from less-developed regions. However, to our knowledge, this study is the first to reveal disparities between urban and rural adolescents in the impacts of long-term changes in internet use on depressive symptoms. Although rural adolescents showed a lower proportion of rapidly increasing internet use than their urban counterparts, this trajectory had a more pronounced effect on their mental health.

The primary factor driving urban-rural differences may be the insufficient parental supervision commonly experienced by rural adolescents. Only 38.3% of rural underage internet users report having their internet use regularly restricted by parents [7]. While grandparents in rural areas often provide caregiving in the absence of parents, their limited educational attainment and reliance on traditional caregiving practices may weaken their ability to effectively monitor adolescents’ internet use [47,48]. This lack of supervision can increase the likelihood of internet addiction, cyberbullying victimization, and engagement in high-risk online behaviors [49-51], potentially contributing to depressive symptoms. Additionally, individuals in rural areas with lower socioeconomic status often face limited social resources and reduced social support, which may amplify the association between internet use and depressive symptoms [52]. Rural schools, for example, tend to place less emphasis on internet literacy and education, resulting in lower awareness of healthy internet use practices [53,54]. The insufficient guidance may further heighten the risks to rural adolescents’ mental health.

This study provides valuable insights into the distinct trajectories of internet use and their associations with adolescent depressive symptoms. The varying initial levels displayed by the trajectories in early adolescence highlight the critical need for early identification of excessive internet use. Given the demographic differences observed in this study, more attention is warranted to adolescents from lower socioeconomic backgrounds and rural areas.

The urban-rural divide in the association between internet use trajectories and depressive symptoms further emphasizes the importance of prioritizing interventions for rural adolescents. Previous research has shown that rural adolescents in China report higher levels of depressive symptoms compared to their urban peers [55]. Our findings suggest that rural adolescents are particularly vulnerable to the adverse effects of internet use, potentially widening existing mental health disparities between urban and rural areas. Effective interventions for rural adolescents should equip them with essential internet use skills, which can be fostered through enhanced parental supervision or integrated into school education programs. Additionally, parents should serve as positive role models by dedicating more time to engaging in outdoor activities with their children, providing alternatives to excessive screen time and encouraging a balanced lifestyle.

This study has several limitations. First, this study relied on self-reported measures of internet use and depressive symptoms. Although the CESD has demonstrated robust psychometric properties among adolescent populations [30,31], self-reported scales are subject to social desirability bias. Moreover, participants may inaccurately report their internet use due to recall bias, introducing potential measurement errors that could attenuate or exaggerate the observed associations between internet use and depressive symptoms. Future research could benefit from incorporating objective measures of internet use and clinically validated tools for assessing depressive symptoms to enhance the reliability of findings. Second, the CFPS did not assess depressive symptoms in 2014. However, baseline psychological distress was controlled using the K6 scale, which was positively correlated with CESD-20 scores in 2018. Furthermore, we found that additional adjustments for CESD-20 scores in 2016 did not alter the association between internet use trajectories and depressive symptoms, indicating the robustness of our findings. Third, the assessment of internet use duration was based on a single question, limiting insights into the specific online activities adolescents engaged in. Future research could consider incorporating comprehensive measures of diverse online activities. Fourth, our sample was limited to adolescents in China. Disparities in health issues between rural and urban areas may exhibit different patterns among adolescents from other countries. Therefore, our findings may not be generalizable to a global population. Finally, in this study, the sample size was relatively small, particularly when considering the urban-rural stratification. Future research should aim to use larger, nationally representative datasets to further explore variations in the effects of adolescent internet use between urban and rural regions.

This study elucidates urban-rural differences in the associations between internet use trajectories and adolescent depressive symptoms. Over 4 years, we observed diverse trajectories of internet use among adolescents, with one subgroup showing a rapid increase in internet use duration that was associated with higher odds of depressive symptoms. Further stratified analysis revealed that this association was significant only among rural adolescents. These findings offer valuable insights into regional disparities in the impacts of adolescent internet use. Tailored intervention strategies are needed to address the challenges faced by rural adolescents, thereby mitigating the negative effects of excessive internet use on their mental health.