A study of the symptom network structure of internet addiction, non-suicidal self-injury and internalization problems in adolescents and its association with physical activity

Abstract

Objective

Internet addiction (IA), internalizing problems (IP), and non-suicidal self-injury (NSSI) are major public health concerns with far-reaching implications. However, few studies have examined their interrelationships with physical activity (PA) from a symptom-level perspective using network analysis. This study aimed to construct an IA-IP-PA symptom network and identify the symptoms most directly associated with NSSI.

Methods

A total of 898 students aged 9 to 16 were recruited from two schools. Data were collected using the Revised Chinese Internet Addiction Scale (CIAS-R), Adolescent Non-Suicidal Self-Injurious Behavior Scale, Physical Activity Rating Scale (PARS-3), and the Brief Problem Monitor. The network structure was estimated using EBICglasso, and central and bridge symptoms were identified.

Results

The prevalence rates of IA and NSSI were 8.4% (n = 76) and 29.6% (n = 266), respectively. The most central symptoms in the network were “interpersonal and health-related problems,” “social anxiety,” and “worthlessness.” Key bridge symptoms included “interpersonal and health-related problems,” “social anxiety,” and “fear.” Among all symptoms, tolerance showed the strongest direct association with NSSI. PA was negatively associated with all symptom nodes, particularly “interpersonal and health-related problems” and tolerance.”

Conclusion

Social anxiety and interpersonal/health-related problems appear to be central drivers of comorbidity across IA, IP, and NSSI. Distress tolerance is closely linked to NSSI, while regular physical activity may buffer against IA, IP, and self-injurious behaviors. Targeted interventions focusing on these key symptoms are essential for promoting adolescent mental and physical well-being.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-25370-9.

Introduction

With the rapid advancement of the digital age, individuals have become increasingly reliant on the internet. According to the 52nd Statistical Report on Internet Development in China, the number of internet users has reached 1.079 billion, among whom 191 million are adolescents [1]. Studies have shown that adolescents are more receptive and adaptable to the internet compared to other age groups [2]. However, due to the immaturity of their prefrontal cortex and ventral striatum, adolescents exhibit heightened sensitivity to cues of high reward or strong stimulation [3, 4], making them more prone to impulsive internet use in pursuit of novelty and excitement. This tendency significantly increases the risk of developing Internet Addiction (IA) [5].

IA, also referred to as “problematic internet use” or “pathological internet use”, is characterized by excessive reliance on the internet, a strong urge to use it, intense focus during use, and an inability to self-regulate or reduce internet engagement [6, 7]. Research has demonstrated that IA is not only closely associated with declining academic performance but also contributes to a range of adverse physical and psychological outcomes [2, 8–10]. Given adolescents’ limited time management skills, they often sacrifice adequate sleep, academic activities, and physical activity (PA) to extend their online time. Such behavior can lead to daytime drowsiness, reduced learning efficiency, and disrupted circadian rhythms, which may in turn cause insomnia, headaches, and mood disorders [11]. The COVID-19 pandemic has further exacerbated this issue, as prolonged home confinement and the widespread shift to online learning during lockdowns have increased the prevalence of IA among adolescents [12]. Existing research indicates that the prevalence of IA is generally higher among students in Asian countries compared to those in other regions [13, 14]. Overall, with the continuous expansion of adolescent internet user populations, IA has emerged as a pressing public health concern that demands urgent scholarly and policy attention.

Adolescence is not merely a period of vulnerability or negative sensitivity but also an important window of developmental potential. During this stage, adolescents, by coping with social and cognitive challenges, have the opportunity to cultivate key skills such as emotion regulation, conflict resolution, time management, and peer interaction [15]. These abilities are considered essential resources for fostering psychological strengths and positive adaptation. However, precisely because adolescence is accompanied by rapid social, cognitive, and physiological changes, individuals are also more prone to internalizing problems (IP), including loneliness, depression, and anxiety [16, 17]. Notably, cross-sectional studies have reported significant associations between IA and IP [18]. According to the Problematic Psychosocial Tendencies Model, IA and psychological distress may form a mutually reinforcing cycle [19]. Specifically, adolescents with psychosocial impairments often lack effective social skills and may prefer online interactions. Over time, they tend to overuse the internet due to its anonymity and reduced demand for nonverbal cues, which mitigates their social anxiety [20, 21]. In turn, uncontrolled internet use may further impair social functioning, contributing to the onset or exacerbation of internalizing symptoms [22, 23]. A meta-analysis encompassing 56 studies revealed that individuals experiencing depression or anxiety are more likely to engage in non-suicidal self-injury (NSSI) [24]. Although IA itself is not considered a direct risk factor for NSSI, Mészáros et al. (2020) suggested that in cases where IA co-occurs with internalizing psychopathologies, it may serve as an indirect contributor to self-injurious behavior [25]. This indicates that comorbid IA and IP may collectively increase the risk of adolescent engagement in NSSI, thereby compounding harm to mental and physical health. Epidemiological surveys show that approximately 17–18% of adolescents globally report engaging in NSSI [26], while the prevalence among Chinese adolescents has reached 24.7% [27]. NSSI has been identified as a significant predictor of future suicide attempts [28], and represents a risk factor for school safety, family harmony, and healthy adolescent development [29]. Given these findings, it is imperative to adopt targeted interventions to reduce the co-occurrence of IA and IP, and to improve the identification of adolescents at high risk for NSSI.

A growing body of intervention research has confirmed that PA is an effective strategy for alleviating IA, and it is considered one of the most promising approaches for addressing IA among adolescents [30]. A recent meta-analysis incorporating 19 randomized controlled trials with a total of 760 participants found that exercise interventions significantly reduced IA levels among college students, yielding a moderate to large effect size [31]. From a neurobiological perspective, PA enhances the structure and function of specific regions of the central nervous system and bidirectionally regulates dopamine and its receptors, thereby facilitating recovery from addictive behaviors [32]. Insufficient PA has been associated with an elevated risk of IP [33], whereas regular engagement in PA is widely recognized as a protective factor for adolescent mental health [34]. In a longitudinal study of over 10,000 children aged 9 to 10, Sampasa-Kanyinga et al. (2021) demonstrated that adherence to 24-hour movement guidelines was significantly associated with lower risks of overall behavioral problems and IP during adolescence [35]. Moreover, numerous studies have documented a negative correlation between PA and NSSI [36, 37]. For example, Grasdalsmoen et al. (2020) found a dose-response relationship between PA and NSSI among adolescents: girls who never exercised were approximately 2 to 2.5 times more likely to engage in NSSI compared to those who exercised daily [38]. These findings suggest that IA, PA, IP, and NSSI may be interrelated through complex interactive mechanisms.

However, most existing studies have examined these variables in isolation, primarily exploring pairwise associations or unidirectional pathways, with limited attention to their co-occurrence in adolescent populations. Notably, IP are often represented in the literature by broad indicators such as depression or generalized anxiety, which may oversimplify the multidimensional nature of internalizing symptomatology. In reality, specific symptoms such as social anxiety, feelings of guilt, and worthlessness are common among adolescents and may also be closely linked to IA [39]. Furthermore, the majority of current studies are grounded in latent variable models, which conceptualize comorbidity as a function of shared underlying constructs reflected by surface-level symptoms. While theoretically useful, this approach often neglects the direct relationships and dynamic interactions among individual symptoms. Therefore, investigating the co-occurrence and structural associations among IA, PA, IP, and NSSI at the symptom level may help bridge this gap in the literature and offer novel theoretical insights into the underlying mechanisms of adolescent psychopathology.

As an emerging methodological approach, network analysis has developed at the intersection of multiple scientific disciplines and serves as a modeling tool for complex systems [40]. In the context of psychopathology, this approach challenges the traditional latent variable paradigm by conceptualizing mental symptoms not as manifestations of an underlying disease, but as components of a dynamic system formed through direct causal interactions [41]. Through network analysis, these systems can be visualized and quantitatively examined, offering a framework to explore the intricate interplay between individual symptoms [42]. This method provides a novel lens for understanding symptom interrelations and assessing the mechanisms of psychiatric comorbidity. Moreover, symptom networks can be compared across different populations or time points to explore how their structure varies with developmental, contextual, or demographic factors. In this regard, the Network Comparison Test (NCT) has become a widely used statistical tool to examine whether two independent cross-sectional networks differ significantly. Specifically, NCT assesses three aspects of equivalence: (1) overall network structure, (2) edge strength distribution, and (3) global strength [43].

Previous studies have confirmed significant gender differences among adolescents in IA, IP, NSSI and PA. Specifically, the prevalence of IA is generally lower in females than in males [44], yet the impact of IA on internalizing symptoms appears to be more pronounced among females [39]. In addition, females are approximately 2.85 times more likely to engage in NSSI compared to males [45]. Further evidence suggests that gender may also moderate the relationship between PA and IA. However, other studies have reported no significant gender differences in the network structure linking insomnia, PA, and suicidal ideation [13]. Taken together, the findings regarding gender remain inconsistent. Moreover, adolescence is a developmental stage marked by rapid psychological and behavioral changes [15], and differences between early and middle adolescence may further shape the associations among IA, IP, NSSI, and PA. Thus, the present study incorporates age as an additional factor to capture potential developmental variations. In addition to gender, little is known about whether the structural associations among these variables differ according to adolescents’ history of physical illness. Existing research indicates that chronic physical conditions may lead to social withdrawal, reduced participation in daily activities, and lower quality of life, thereby increasing psychological distress and the risk of NSSI [46]. Adolescents characterized by internalizing symptoms or problematic internet use are at elevated risk for somatization compared to regular internet users [47]. Taken together, these findings suggest that the associations among these variables may vary depending on whether adolescents have a history of physical illness. Therefore, the present study also aims to examine the role of physical illness in these associations.

Accordingly, the present study aims to: (a) construct a symptom network model of IA, IP, and PA, and identify central and bridge symptoms within the network; (b) detect symptoms directly associated with NSSI; and (c) compare the network structure by gender, age group (early vs. middle adolescence), and history of physical illness. This study aims to uncover the potential connections among IA, IP, and PA in adolescents through network analysis, and to identify core symptoms associated with NSSI, thereby providing a reference for interventions targeting adolescent psychological symptoms and alleviating the burden on public health systems and family education.

Materials and methods

Participants and procedure

In this study, participants were recruited from two middle schools in Shandong and Henan through a cluster sampling method between May and June 2025; data were collected using an online platform (name, if applicable) (Wen Juan Xing). Inclusion criteria included: (1) aged 9–16 years and (2) able to complete all questionnaires independently. Exclusion criteria were: (1) individuals who self-reported a diagnosis of severe mental illness in the initial screening item, and (2) those who declined participation, identified on the platform when consent was not provided or the survey was exited prematurely. A total of 1,031 questionnaires were distributed, of which 898 were valid, yielding an effective response rate of 87.1%.

Informed consent from both parents and adolescent participants was obtained online through the survey platform. Only those who provided dual consent were granted access to the questionnaire, while cases without consent were automatically excluded. The study was approved by the Biomedical Ethics Committee of Henan University.

Measures

Demographics

Demographic information was collected through a self-designed questionnaire that included variables such as gender, age, student origin, sibling status, mental health history and somatic disease history. (e.g., “Have you ever had a physical illness or been hospitalized for it?”).

Internet addiction

IA was assessed using the Revised Chen Internet Addiction Scale (CIAS-R) [48], which consists of 19 items rated on a 4-point Likert scale ranging from 1 (“very much in my situation”) to 4 (“very much not in my situation”). The scale comprises five distinct dimensions: compulsive Internet use, withdrawal reactions, tolerance, interpersonal and health problems, and time management problems. CIAS-R has been shown to have good reliability and validity among Chinese adolescents [49]. A higher total score indicates a more severe level of Internet addiction. Participants scoring below 46 on the CIAS-R scale were classified within the normal group, whereas those who scored above 53 were classified in the Internet addiction group. The scale exhibited high reliability with a Cronbach’s alpha coefficient of 0.87 in this research.

Non-suicidal self-injury

NSSI was measured using the self-report Adolescent Non-Suicidal Self-Injurious Behavior Scale [50], a 12-item measure developed for use in the Chinese cultural context, sample items include “scratch yourself on purpose”, and “bite yourself on purpose”. Each item is rated on a 5-point Likert scale from 0 (none) to 4 (always). A total score of 0 indicates no NSSI; higher scores indicate more frequent NSSI. Total scores range from 0 to 48, with higher scores indicating more frequent non-suicidal self-injurious behavior in adolescents. The Cronbach’s α for the scale in this study was 0.94.

Physical activity

In this study, the Physical Activity Rating Scale (PARS-3), revised by Liang [51], was used to assess the physical activity levels of adolescents. This scale has been widely applied in related research [52] and evaluates three dimensions of physical activity: intensity, duration, and frequency. Each item is rated on a five-point Likert scale. A sample item is: “How often do you engage in the above physical activities?” The total physical activity score is calculated using the formula: Exercise amount = Intensity × Duration × Frequency. Both intensity and frequency are scored from 1 to 5, while duration is graded from 1 to 5 and scored from 0 to 4. The total score ranges from 0 to 100, with higher scores indicating greater levels of physical activity. According to the scale, physical activity levels were categorized as follows: ≤ 19 = low activity; 20–42 = moderate activity; ≥ 43 = high activity. This resulted in coding low, moderate and high activity levels as 1, 2 and 3 in that order. In the present study, the Cronbach’s α coefficient of the scale was 0.74, indicating acceptable internal consistency reliability.

Internalizing problems

IP were assessed using a self-reported Brief Problem Monitor [53], which includes six items assessing internalizing problems. Response options ranged from 0 (not true) to 2 (certainly true), with higher scores representing higher levels of internalizing problems. The answers were dichotomized to indicate the presence (somewhat true and certainly true = 1) or absence (not true = 0) of symptoms. The self-reported BPM scale demonstrated good reliability and validity with Chinese Youth [54]. In this study, Cronbach’s α for internalizing problems was 0.74.

Statistical analyses

Descriptive analyses

Descriptive analyses were conducted using SPSS version 26.0. Categorical variables were presented as frequencies and percentages, while continuous variables were summarized using means and standard deviations. Group differences were assessed using the chi-square test or the Mann–Whitney U test, as appropriate. All statistical analyses were two-tailed, and a p-value of less than 0.05 was considered statistically significant.

Network Estimation

In this study, we constructed five distinct symptom-level networks to examine the interrelations among IA, NSSI, IP, and PA. The networks were estimated separately for the full sample, male and female subgroups, and participants with and without a history of physical illness. Each network incorporated five dimensions of the CIAS-R (e.g., compulsive internet use), six items of internalizing behaviors (e.g., depressive mood), and physical activity frequency, aiming to capture the complex interplay among these variables.

Network estimation was performed using the estimateNetwork function from the bootnet package in R Studio [55]. Following model selection recommendations proposed by Isvoranu and Epskamp [56], and considering the moderately high sample size (N = 898), we employed the default EBICglasso method. This approach combines graphical LASSO regularization with the EBIC, enabling the construction of sparse and interpretable networks by optimizing edge selection while penalizing model complexity. Prior literature has demonstrated that EBICglasso performs robustly in this sample size range, offering accurate edge weight estimation, stable network structures, and high replicability. Moreover, it is computationally efficient and particularly sensitive to identifying the strongest edges, making it well suited for detecting the network backbone [56].Given that the raw data did not fully meet the assumption of multivariate normality, we applied Spearman rank correlations prior to estimating the EBICglasso-based Gaussian Graphical Models (GGMs) [57]. In the final networks, each node represents a symptom variable, and edges reflect conditional associations between node pairs after controlling for all other nodes. Blue edges indicate positive associations, red edges indicate negative associations, and thicker edges represent stronger relationships.

Centrality and stability

To evaluate the importance of each node within the network, we estimated centrality indices using the qgraph package, and reported standardized values (z-scores) [58]. The centrality measures examined in this study included strength, closeness, expected influence (EI), and bridge expected influence (BEI). Strength and closeness are conventional centrality metrics that reflect, respectively, the overall connectivity of a node and its proximity to all other nodes in the network [55, 59].

However, given that the networks constructed in this study included both positive and negative edge weights, traditional centrality measures may fail to capture the nuanced effects of nodes with mixed influences. Therefore, we additionally employed expected influence [55], which retains the sign of edge weights and thus provides a more comprehensive estimation of a node’s net influence on others. Furthermore, to identify nodes that serve as connectors between distinct network communities, we calculated bridge expected influence (BEI) using the “bridge” function in the networktools package [60]. Bridge metrics are particularly useful for detecting “key transmission nodes” in psychological networks, especially in contexts where multiple psychological domains (e.g., internalizing symptoms and NSSI) are intertwined, helping to identify potential intervention targets.

To ensure the robustness of our findings, we assessed the stability of strength and expected influence using case-dropping bootstrap procedures with 1,000 subsamples, implemented via the bootnet package (version 1.6). We calculated the correlation stability coefficient (CS-coefficient) to determine how consistently centrality estimates remained across subsets of the data. Following recommendations by Epskamp et al. (2018), CS-coefficients of at least 0.25 are considered acceptable, and values above 0.5 indicate high stability.

Network comparison

In this study, the “NetworkComparisonTest” package in R [61] was used to examine differences in psychological networks between different groups (e.g., by gender and presence or absence of physical illness). A total of 1,000 permutation tests were conducted to evaluate differences in network structure invariance, edge strength invariance, and global strength invariance. Structure invariance tests whether at least one edge differs significantly between networks; global strength reflects the overall level of connectivity, with higher values indicating denser networks. Additionally, edge weight distribution was analyzed to further assess structural differences.

Flow network construction

In addition to the primary network, we incorporated NSSI as an additional node and constructed a flow network model based on the full sample to identify factors directly associated with NSSI. The visualization was performed using the “flow” function from the qgraph package [58]. In the flow network, the NSSI node was positioned on the left, while all other nodes were placed on the right. The proximity of each node to the SI node indicates the strength of their associations.

Results

Demographics of study samples

Table 1 presents the demographic characteristics of the study sample. A total of 898 participants were included in the final analysis, with a median age of 13 years; among them, 349 were female (38.9%) and 549 were male (61.1%). Overall, the prevalence rates of IA and NSSI were 8.4% (n = 76) and 29.6% (n = 266), respectively. Males were more likely to be classified as having IA. Furthermore, individuals with IA were more likely to exhibit NSSI behaviors, have a history of physical illness, spend less time on PA, and report more severe IP. However, no statistically significant differences were found between the groups in terms of age (p = 0.508) or history of psychiatric disorders (p = 0.089). The correlation analysis between the different variables is shown in Fig. 1. The underlined portion shall be amended as follows: " Overall, the prevalence rates of IA and NSSI were 8.4% (n = 76) and 29.6% (n = 266), respectively. Males were more likely to be classified as having IA.

Table 1.

Descriptive statistics of samples

	IA (N = 76)	Without IA (N = 822)	Total (N = 898)	P
Age	13(10,14)	13(10,14)	13(10,14)	0.508
Gender				< 0.005
Female	17(22.4%)	332(40.4%)	349(38.9%)
Male	59(77.6%)	490(59.6%)	549(61.1%)
Residence				< 0.001
Urban	61(80.3%)	344(41.8%)	405(45.1%)
Rural	15(19.7%)	478(58.2%)	493(54.9%)
Mental health history				0.082
Yes	5(6.6%)	20(2.4%)	25(2.8%)
No	71(93.4%)	802(97.6%)	873(97.2%)
Somatic disease history				< 0.005
Yes	26 (34.8%)	164(20.0%)	190 (21.1%)
No	50 (65.2)	658(80.0%)	708 (78.9%)
NSSI ideation				< 0.001
Yes	36(47.4%)	230(28.0%)	266(29.6%)
No	40(52.6%)	592(72.0%)	632(70.4%)
Physical activity	1(1,2)	2(1,2)		< 0.001
Internalization issues	5(3,7)	4(1,5)		< 0.001

^an (%); Median (IQR)

Fig. 1.

Spearman's correlation heatmap of variables. CLAS1: Compulsive internet use; CLAS2: Withdrawal reaction; ClAS3: Tolerance; CLAS4: Interpersonal and health-related problems; CLAS5: Time management; NSSI: non-suicidal self-injurious; N1: Worthlessness; N2: Fear; N3: Guilt; N4: Social anxiety; N5: Depressed mood; N6: Worry

Network stability and accuracy

The case-dropping bootstrap procedure revealed that the CS-cofficients for Strength, EI, and BEI were 0.62, 0.70, and 0.64, respectively. All values exceeded the recommended threshold of 0.50, indicating a high level of stability and reliability in the centrality estimates (see Fig. 2).

Fig. 2.

Network stability factor

Network structure

The IA-PA-IP symptom network (see Fig. 3) consists of 12 nodes and up to 66 possible edges, of which 49 are nonzero (density: 74.2%, average edge weight: 0.06). Among the Internet addiction symptom cluster, the strongest associations were observed between health-related problems (CIAS4) and time management (CIAS5), as well as between withdrawal reaction (CIAS2) and compulsive Internet use (CIAS1). Within the cluster of internalizing symptoms, the strongest association was between Guilt (N3), depressed mood (N5), and worry (N6). Internet addiction and internalizing problems were linked through several peripheral edges, the most prominent of which was the connection between Interpersonal and health-related problems (CIAS4) and social anxiety (N4). Most node associations were positive. In contrast, PA was negatively associated with all nodes, particularly with Tolerance (CIAS3) and Interpersonal and health-related problems (CIAS4). The specific edge weight matrix in this network is shown in Supplementary Table 1.

Fig. 3.

Network Structure of Internet addiction, Internalisation problems and Physical activity in adolescents

Central and Bridge symptoms

Table 2; Fig. 4 display the centrality of each node in the network. The core symptoms identified were interpersonal and health-related problems (CLAS4), social anxiety (N4), worthlessness (N1) and depressed mood (N5), with their respective EI values being 0.84, 0.79, 0.54, and 0.61. Figure 5 illustrates the bridge expected influence between network communities, highlighting CLAS4, N4, and N2 as the key bridge nodes connecting IA, PE, and IP. These nodes may serve as critical links in the co-occurrence of IA, PA, and IP.

Table 2.

The z-scores of the node centrality indices in the network

Node	Strength	Closeness	ExpectedInfluence	Betweenness
PA	−1.50	−1.20	−2.91	−0.61
N1	0.67	0.02	0.54	−0.61
N2	−0.16	−0.19	0.06	−0.61
N3	−0.38	−0.39	0.23	−0.61
N4	0.70	1.96	0.79	2.24
N5	0.47	0.02	0.61	0.10
N6	−0.61	−0.59	−0.16	−0.07
CLAS1	−0.03	−0.08	0.28	−0.07
CLAS2	−0.51	−0.66	−0.12	−0.61
CLAS3	−0.93	−1.12	−0.50	−0.43
CLAS4	2.43	1.83	0.84	1.88
CLAS5	−0.15	0.40	0.35	−0.61

Fig. 4.

Centrality metrics for each node of the network (ranked by z-scores)

Fig. 5.

Bridge expected influence for each node of the network (ranked by z-scores)

Flow network diagram of symptoms associated with NSSI

The flow chart suicidal ideantion is shown in Fig. 6. Except for Compulsive Internet Use (CLAS1), Guilt (N3), and Fear (N2), most nodes are directly connected to NSSI. Tolerance (CLAS3) exhibits the strongest direct positive association with NSSI, followed by Social Anxiety (N4), Withdrawal Reaction (CLAS2), and Depressed Mood (N5). In contrast, Interpersonal and health-related problems (CLAS) and PA were directly negatively associated with NSSI. Supplementary Table 2 shows the correlation weight matrix of NSSI with IA, IP and PA.

Fig. 6.

Flow network of Non-suicidal self-injury, Internet addiction, Internalizing problems and physical activity in adolescents

Network comparison of gender, history of somatic disease and age groups

The network comparison test revealed no statistically significant differences in overall network strength between males and females (Males: 4.578 vs. Females: 4.288; S = 0.289, p = 0.954) (Supplementary Fig. 1), and no significant differences in network structure between the groups (M = 0.283, p = 0.144). Similarly, there were no statistically significant differences in overall network strength between the groups with somatic disease and those without (Yes: 4.360844 vs. No: 3.649411; S = 0.711, p = 0.937) (Supplementary Fig. 2), and no significant differences in network structure between the groups (M = 0.281, p = 0.212). In addition, the comparison between the early adolescence group (10–13 years) and the middle adolescence group (14–16 years) also revealed no statistically significant differences in overall network strength (Early: 2.880126 vs. Middle: 0.086537; S = 2.794, p = 0.327) (Supplementary Fig. 3), nor in network structure (M = 0.206, p = 0.287).

Discussion

According to existing literature, studies involving large-scale adolescent samples, particularly those exploring the relationship between IA, IP, NSSI, and PA from a network analysis perspective, are relatively scarce. Our study indicates that interpersonal and health-related problems (CIAS4), social anxiety (N4), and Worthlessness (N1) are key factors that sustain the entire network structure and lead to symptom co-occurrence. Additionally, social anxiety (N4), interpersonal and health-related issues (CLAS4), and compulsive internet use (CLAS1) were identified as bridging factors between symptoms. Notably, PA exhibits negative correlations with all nodes, particularly between interpersonal and health-related issues (CLAS4) and tolerance (CIAS3). Social anxiety (N4) is most strongly associated with suicidal ideation. This study provides new insights into the co-occurrence of NSSI, internet addiction, and IP in adolescents, and offers potential directions for future prevention and treatment strategies.

In this study, the prevalence of NSSI among adolescents was 29.6%, which is consistent with earlier research conducted in China [27], but slightly lower than the 34% reported in a U.S. observational study [62]. However, our study found that 8.4% of adolescents were suffering from IA, a proportion lower than the findings of a longitudinal study conducted during the COVID period in mainland China [63], which may be attributed to differences in the study population and measurement timing. Additionally, consistent with other research [18, 24, 25], our study suggests that individuals with IA are more likely to experience NSSI and IP, which not only impact their quality of life but also place a heavy burden on their families and society.

We found that the most significant association in the symptoms of IA is between compulsive internet use (CIAS1) and interpersonal and health-related problems (CIAS4). Compulsive internet use is characterized by loss of control, excessive focus, conflicts, withdrawal symptoms, and using the internet as a coping strategy, with negative physiological reactions and emotions arising when use is stopped [64]. According to Kraut et al. (2002), heavy internet use does indeed increase social connections, but this is limited to individuals with strong social support [65]. Their study indicates that heavy internet users with weaker social support actually have poorer social connections, which may suggest that those with weaker social support are more likely to use the internet for social purposes, but this usage does not necessarily yield psychosocial benefits. In terms of internalizing problems, the strongest correlation is between worthlessness (N1) and social anxiety (N4), a finding consistent with Heeren et al. (2018) network analysis of 174 individuals with social disorders [66]. Worthlessness refers to the devaluation of one’s self-worth, especially in social interactions. When individuals feel low self-worth, they often doubt their abilities and status in interpersonal relationships [67]. This finding further clarifies the core symptoms of IP in adolescents and provides valuable insights for implementing targeted mental health interventions.

Research has pointed out that various factors, such as loneliness, shyness, anxiety, and stress, play a crucial role in the association between IA and IP, which aligns with the network structure we present [68]. Based on EI and BEI values, we further identified social anxiety and interpersonal and health-related problems as the two most critical symptoms. A previous study used network analysis to examine the correlations between symptoms of internet addiction. Unlike our findings, they identified tolerance and time management as core symptoms [13]. This difference may stem from different research focuses, as our model also includes IP, PA, and NSSI. Furthermore, in terms of methodology, our study focuses on investigating these associations at the dimensional level, while the previous study analyzed them at the item level [69]. In earlier research, the association between social anxiety in IP and IA had already been identified. IA is viewed as a way of seeking relief from social anxiety and other distressing emotions in real life, which may lead to behavioral difficulties [70]. From this perspective, adolescents with fragile personalities face a higher risk of internet abuse due to the increased risk of IP. Additionally, the term “internalization” is associated with the fact that these IP make individuals more likely to close off, not revealing or expressing symptoms to others. In turn, these adolescents’ internalizing behavioral issues may evolve into internet abuse, communicating through cold and detached symbolic interactions in the virtual world, thereby exacerbating their real-world interpersonal difficulties [68, 71]. Consequently, excessive reliance on the internet will consume much of their time, leading to reduced social participation and increasingly strained interpersonal relationships.

In the network model, we observed a negative correlation between PA and IA as well as IP, particularly in interpersonal and health-related issues (CLAS4), worthlessness (N1), and tolerance (CLAS3). Firstly, engaging in PA helps adolescents with internet addiction to quickly extract relevant information, reduce external distractions, and improve cognitive processing abilities, which decreases the psychological need for internet use and lays a foundation for overcoming internet addiction [72]. Secondly, PA significantly predicts self-efficacy, psychological resilience, and self-control, thus contributing to the intervention of internet addiction behavior [73]. Additionally, PA helps shift attention biases, reduce dependence on internet-related cues, and enhance the individual’s ability to positively cope with the impacts of the internet [74]. Thirdly, the pleasurable experiences, peak experiences, and smooth processes associated with PA play a significant role in improving and preventing internet addiction [75]. Furthermore, exercise strengthens the body, improves physical function, and active participation in outdoor and group activities helps to enhance social connections, thereby assisting adolescents in building healthy peer relationships [76]. Empirical evidence suggests that engagement in PA is significantly linked to various dimensions of adolescents’ self-worth—such as physical self-concept, perceived competence, body image consciousness, and self-efficacy—and may further contribute to the development of motor skills and a heightened sense of personal control [76, 77]. Therefore, Regular physical activity maintenance may serve as an effective measure to prevent IA and IP.

In this study, no significant differences were found in the network comparisons across gender, age groups, and physical illness history, which is consistent with some previous findings [13]. This result suggests that the associations among PA, IA, and IP in adolescents may have a certain degree of universality and robustness, and are not easily moderated by demographic or health-related factors. A possible explanation is that adolescents in early to middle adolescence generally face similar developmental tasks and psychological challenges [78], resulting in consistent mechanisms through which PA mitigates the associations between IA and IP across groups. However, it should also be noted that the relatively small sample sizes of the IA group and the subgroup with a history of physical illness may have limited the statistical power to detect subtle differences. Future research with larger samples and longitudinal designs is needed to further explore potential group differences and to validate the present findings.

NSSI is a serious public health and safety issue that impacts the physical and mental health of adolescents [79]. Previous empirical studies have found significant associations between NSSI, IA [25], IP [80, 81], and PA [37]. The flowchart of the network model reveals that IA and IP are directly associated with an increased frequency of NSSI, particularly in terms of tolerance (CLAS3). This finding aligns with a study by Anestis et al., which suggests that individuals with low tolerance and difficulty persevering in the face of negative influences are more likely to engage in NSSI [82]. Van Orden et al. further emphasized that low pain tolerance may lead to NSSI, as the pain and/or provocation may increase suicidal capability, ultimately elevating the risk of suicidal behavior [83]. However, we found a negative correlation between interpersonal and health-related problems and the increase in NSSI. This phenomenon may be attributed to the social support provided by the internet [84], which fosters the development of intimate relationships and a sense of belonging, thus effectively alleviating feelings of loneliness, depression, and anxiety, and reducing the risk of NSSI [85]. Additionally, we observed that higher levels of PA are closely related to a decrease in NSSI. On one hand, PA helps improve mood by increasing serotonin, endorphin, and dopamine levels in the brain [86]. On the other hand, exercise enhances individuals’ sense of self-efficacy and self-esteem [87], which in turn improves their ability to cope with negative emotions and stress. These factors may explain why PA can reduce NSSI.

Strength

This study reveals the complex relationships between IA, IP, PA, and NSSI. Our findings suggest that social anxiety and interpersonal and health-related problems are key symptoms in the adolescent IA-IP-PA symptom network. Targeted and personalized interventions addressing these two symptoms may help dismantle the entire network, providing a foundation for intervention strategies in schools and families. For example, schools can assist emotionally vulnerable adolescents by promoting outdoor activities or conducting internet addiction education to help improve interpersonal relationships and reduce IA symptoms. Parents can also establish family exercise plans, gradually increasing their children’s physical activity time, thereby reducing screen time. Moreover, particular attention should be given to adolescents’ emotional problems, which may be closely related to the reduction in NSSI. Additionally, encouraging adolescents to engage in regular physical activity, especially group sports, could help them gain emotional support and reduce their reliance on the internet. PA not only promotes improvements in negative emotions but also enhances neural plasticity and encourages social interaction, thus improving adolescents’ overall health through multiple mechanisms. These intervention strategies could provide more comprehensive support for adolescents’ physical and mental well-being, particularly with regard to reducing IA and IP.

Limitations

The current study has several limitations. Firstly, the network structure is based on cross-sectional data, which does not allow for the determination of directionality or causal relationships between symptoms. Future research will need to utilize large-scale longitudinal data to validate the current findings. Additionally, it has been observed that symptom networks may change over time. Our results may only represent the relative stability of a single point in time, and therefore, should be interpreted with caution. Secondly, the measurements of IP and NSSI in this study rely on self-report measures, which may introduce recall and reporting biases. To reduce these biases, future investigations could incorporate objective or clinical assessments. Thirdly, we used physical illness as a meta-variable (presence or absence of physical illness) in the analysis, but did not further differentiate between specific types of physical illnesses. However, the impact of short-term acute illness or mild injury history on IA, IP, and NSSI may be relatively small or insignificant. Future studies should consider treating different types of physical illnesses as separate variables to explore their varying effects on the psychological symptom network. Another limitation is that gender was measured only with binary options (male/female), which may not capture the diversity of gender identities. Future research should include more inclusive categories to better reflect adolescent experiences. Finally, regarding the measurement and categorization of PA, future research could classify different types of PA (e.g., group sports, outdoor activities, or family activities) to clarify which type of activity has the greatest benefit in alleviating IA, IP, and NSSI. Additionally, more precise measurement tools, such as accelerometers, should be used to provide a more accurate assessment of PA.

Conclusion

This study employed network analysis to explore the relationships between IA, IP, PA, and NSSI. Social anxiety and interpersonal and health-related problems were identified as the most critical symptoms, significantly influencing the maintenance of the entire network and the transition between different symptom clusters. Tolerance was found to be the symptom most closely associated with NSSI. Encouraging adolescents to develop regular exercise habits may help reduce the severity of IA and IP, as well as decrease the frequency of NSSI. It is also essential to address these key symptoms to protect the physical and mental health of adolescents.

Authors’ contributions

YB-Z was responsible for writing the original draft of the manuscript. ZJ-D contributed to the review and editing of the manuscript. YB-Z and ZY conceptualized the study. XL and JM managed the data curation. ZJ-D conducted the formal analysis. YB-Z and ZY contributed to the investigation, methodology, and project administration. YB-Z and XL were responsible for software development and data visualization. ZJ-D and MJ supervised the entire research process. All authors read and approved the final version of the manuscript.

Funding

There is no funding or incentive to carry out this research.

Data availability

The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Henan University Biomedical Research Ethics Sub-Committee (approval number HUSOM2025-680). All procedures were carried out in accordance with the Declaration of Helsinki. Written informed consent was obtained from each subject prior to participation in the study. Participants retained the right to withdraw from the study without any consequences.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Clinical trial number

Not applicable.