Identifying internet addiction profiles and bridge connectors among Chinese college students and evaluating CBT vs. CBT+MBI interventions via a randomized controlled trial

Abstract

Background and Aims

The Interaction of Person-Affect-Cognition-Execution (I-PACE) model offers a framework for understanding the interplay between cognitive, affective, and behavioral factors in internet addiction (IA). Our study aims to explore the heterogeneity of IA, identify bridge connectors, and compare the efficacy of cognitive behavioral therapy combined with mindfulness-based intervention (CBT+MBI) versus CBT alone in reducing IA levels among Chinese college students.

Methods

In study 1, 1,030 Chinese college students completed assessments of IA, automatic thoughts, self-control, and anxiety. Latent profile analysis (LPA) was employed to identify distinct symptom profiles of IA across individuals. Network analysis (NA) identified bridge connectors for targeted intervention. In study 2, 36 participants randomly selected from the high IA and low IA groups of study 1 were randomly assigned to CBT+MBI, CBT alone, or a control group. The CBT+MBI group received an 8-week dual-modality intervention and the CBT alone received an 8-week CBT intervention, both designed to target the bridge connectors identified via NA in Study 1, while the control group only completed basic questionnaires.

Results

In study 1, LPA identified four subgroups: regular, at-risk, low IA, and high IA groups. NA pinpointed automatic thoughts and anxiety as bridge connectors. In study 2, targeted interventions significantly reduced college students' levels of IA. CBT+MBI resulted in greater and more sustained improvements compared to CBT alone, with effects maintained for six-month post-intervention.

Conclusions

Our study not only reinforces the I-PACE model but also provides actionable strategies for designing evidence-based, multidimensional interventions to reduce addictive behaviors among college students.

Introduction

Internet addiction (IA) is a complex behavioral pattern characterized by an individual's uncontrollable and compulsion to engage in specific online activities, such as online gaming, social networking, online gambling, or the consumption of pornographic content (Feng et al., 2024; Grant, Potenza, Weinstein, & Gorelick, 2010; Pu et al., 2023; Young & Rogers, 1998). This engagement persists despite the individual's awareness that continued excessive use has caused or exacerbated significant negative psychological, physiological, social, academic, or occupational consequences (Foroughi, Griffiths, Iranmanesh, & Salamzadeh, 2022; Shen et al., 2021). Multiple forms of problematic internet use share a common underlying symptomatology, including tolerance, withdrawal, preoccupation, and compulsive use (Kuss, Kristensen, & Lopez-Fernandez, 2021). These shared features support the use of the broader term ‘Internet Addiction’ to describe a general addictive process involving various online activities. Nonetheless, we acknowledge the ongoing scholarly debate regarding this terminology. A prominent critique advocates replacing the broad ‘IA’ construct with activity-specific terms (e.g., gaming, social networking) to enhance nosological precision (Montag, Wegmann, Sariyska, Demetrovics, & Brand, 2021; Wegmann et al., 2025). However, such refinement is constrained by the absence of standardized diagnostic criteria for many internet-related behaviors. In the present study, we use ‘IA’ as a precise operational term for the construct measured by the Chinese Internet Addiction Scale-Revised (CIAS-R), an instrument validated to capture these core, cross-activity dimensions of addiction. This approach enables us to maintain conceptual clarity while situating our findings within the broader body of existing literature. College students are particularly susceptible to developing IA. This heightened vulnerability is due to their unique developmental stage, which is characterized by active cognitive processes and high receptivity to new technologies. Studies estimate that about 23.7% of Chinese college students experience symptoms of IA (Xu et al., 2020). Due to their increasing reliance on the internet, IA impairs their cognitive abilities (Hou, Xiong, Jiang, Song, & Wang, 2019; Niu et al., 2016; Younes et al., 2016), heightens anxiety and depression (Javaeed, Bint Zafar, Iqbal, & Ghauri, 2019; Longstreet, Brooks, & Gonzalez, 2019; Yen, Chou, Liu, Yang, & Hu, 2014), undermines academic performance through absenteeism and declining grades (Leung & Lee, 2012; Zhang, Lim, Lee, & Ho, 2018), weakens social interaction skills, and fostering isolation (Atroszko et al., 2018; Young, Yue, & Ying, 2007). Collectively, these detrimental effects highlight the urgent need for targeted interventions to address IA among this population.

The Interaction of Person-Affect-Cognition-Execution (I-PACE) model provides a theoretical framework for understanding the mechanisms underlying the development of IA. This model posits that IA arises from a complex interplay of predisposing Person-specific factors (P), affective and cognitive responses to triggers (Affect and Cognition - A & C), and resultant Execution (E) of the addictive behavior, influenced by executive functioning and inhibitory control (Brand, Young, Laier, Wölfling, & Potenza, 2016). Guided by this model, a growing body of research has demonstrated the significant roles of anxiety and deficits in self-control in the context of IA (Zhang & Bian, 2021; Zhang, Lin, Jiang, Zhao, & Dai, 2024). Concurrently, although not always explicitly tested under the I-PACE framework, the crucial role of negative automatic thoughts in IA is also well-established (Coşa, Dobrean, Balazsi, & Poetar, 2025; Park, Lee, & Jun, 2017). However, studies have not comprehensively examined how these three factors dynamically interact, nor have they identified what specific mechanism serves as the most central link connecting IA to broader psychological distress. This gap stems in part from methodological limitations in prior research. Traditional approaches relying on mean or median scores from addiction scales (Spurk, Hirschi, Wang, Valero, & Kauffeld, 2020), often mask significant heterogeneity in IA manifestations. As students with similar scores exhibit diverse symptom patterns, limiting intervention efficacy due to oversimplified analyses. To overcome this, a person-centered approach is paramount. Latent profile analysis (LPA) is ideally suited for this task, enabling the identification of homogeneous subgroups with distinct latent profiles (e.g., high vs. low IA), thereby allowing for targeted interventions with enhanced effectiveness (Mathew & Doorenbos, 2022; Spurk et al., 2020).

A second gap arises from the lack of a unified framework for investigating these three factors concurrently. To address this, we argue that the most productive way to understand the interplay among these factors is to conceptualize them through the process-oriented lens of the I-PACE model. Therefore, the core of our study is to posit automatic thoughts, anxiety, and self-control collectively as dynamic, interacting components of the Affect-Cognition-Execution process, rather than as isolated Person-specific components. Specifically, automatic thoughts, often triggered by underlying cognitive schemas or core beliefs (Beck, 1979), are posited to function as the in-the-moment Cognition component. These thoughts are crucial in perpetuating IA through cognitive biases, escapism, and deepened internet immersion (Coşa et al., 2025; Park et al., 2017). This immediate cognitive activity, in turn, can elicit and intensify distressing emotional states, thereby mediating the progression of IA (Lian et al., 2023; Shen, Chen, Ying, Wang, & You, 2024). Within this dynamic, anxiety represents a critical aspect of the Affect component, exhibiting a positive correlation with IA susceptibility by driving internet use as a maladaptive coping mechanism (Swetlitz, Lynch, Propper, Coffman, & Wagner, 2021; Xie, Cheng, & Chen, 2023). Finally, self-control is integral to the Execution component. Enhanced self-control influences inhibitory control to curb impulsive digital engagement, thereby serving as a protective factor against IA (Agbaria, 2021; DeLisi, 2014; Li et al., 2019). To empirically test this novel conceptualization, network analysis (NA) offers a powerful methodology. By mapping the intricate relationships between variables to identify “bridge connectors”, which are defined as the specific psychological constructs that play a central role in linking different factor clusters (Borgatti, Mehra, Brass, & Labianca, 2009; Cramer, Waldorp, Van Der Maas, & Borsboom, 2010; Jones, Ma, & McNally, 2021). In essence, building on the I-PACE model, we selected a priori three theoretically grounded constructs and examined in NA which of them may play a more central role for IA, thereby identifying potential intervention targets.

Given the complexity of IA and the need for targeted approaches informed by a deeper understanding of its mechanisms and heterogeneity, evidence-based psychological interventions are required. Previous intervention studies on IA have predominantly employed cognitive behavioral therapy (CBT) and mindfulness-based interventions (MBI), yielding notable outcomes. CBT, a well-established psychological intervention, has demonstrated efficacy in addressing core IA symptoms among college students, such as improving time management, interpersonal functioning, and preventing relapse through techniques like behavior modification and cognitive restructuring (Kim et al., 2018; Young, 2011). These CBT techniques primarily target maladaptive cognitions and behaviors, which are central components of the I-PACE model. However, the I-PACE model also emphasizes the significant role of affective dysregulation, craving, and cue-reactivity in the maintenance of IA (Brand et al., 2016). MBI can offer a complementary approach by directly targeting these affective and attentional components. MBI fosters non-judgmental awareness and acceptance of internal states such as cravings, negative emotions, which may help decouple these internal experiences from maladaptive coping behaviors like excessive internet use (Brandtner et al., 2022; Fendel, Vogt, Brandtner, & Schmidt, 2024; Kabat-Zinn, 2003; Shonin, Van Gordon, & Griffiths, 2013, 2014). This enhanced awareness and acceptance can be particularly beneficial when complete abstinence from the internet is not feasible, allowing individuals to respond more adaptively to internal triggers (Zhang, Wu, Lu, & Guan, 2025). Recognizing the potential for synergy, combined interventions integrating elements from different approaches have become a growing area of research, with recent meta-analyses suggesting potentially higher efficacy and durability compared to single interventions (López-Navarro & Al-Halabí, 2022; Peng et al., 2021; Zhang et al., 2022; Zhu, Chen, Li, Mei, & Wang, 2023). We hypothesized that our dual-modality intervention combining CBT and MBI might be more effective than CBT alone, particularly by targeting the affective and cognitive aspects and enhancing coping with internal states, which are critical components implicated in the I-PACE model and potentially represented by key bridge connectors.

In this study, we employed a two-part approach to evaluate intervention strategies for IA among Chinese college students. In study 1, guided by the I-PACE model, we a priori identified three specific constructs (automatic thoughts, self-control, and anxiety) as potential transdiagnostic mechanisms of IA. We then used LPA to explore the heterogeneity of IA and applied NA to these constructs to examine which of them may play a more central role for IA, with the aim of identifying the most central bridge connectors that could serve as primary intervention targets. In study 2, we compared the effectiveness of a dual-modality intervention, specifically cognitive behavioral therapy combined with mindfulness-based intervention (CBT+MBI), versus cognitive behavioral therapy alone (CBT alone) in reducing IA among Chinese college students.

Study 1. IA in relation to automatic thoughts, self-control, and anxiety among Chinese college students: based on LPA and NA

Method

Participants

Data were collected using WJX (www.wjx.cn), a widely used online survey platform in China. A total of 1,114 college students from multiple universities in China participated in the survey. Inclusion criteria were: (1) being an enrolled college student; (2) aged between 17 and 25 years. Exclusion criteria were: (1) history of psychiatric or neurological disorders; (2) starting or changing psychotropic medication within the past month. Exclusion criteria were: (1) removing responses with excessively short completion times (below two standard deviations of the mean completion time); (2) invariant response patterns (e.g., straight-lining or uniform response behavior across all items); (3) responses that failed attention check questions were excluded from the final dataset. Following data screening procedures, 1,030 valid responses were included in the final analysis, yielding an effective response rate of 92.45%. Participants were aged between 17 and 25 years (M_age = 20.30, SD = 1.44), with 571 male participants (55.4%) and 459 female participants (44.6%).

Participants received small monetary compensation, approximately $0.7 as an incentive for completing the survey. The data collection period was from March 12, 2024, to March 25, 2024.

Measures

Chinese internet addiction scale-revised (CIAS-R)

The CIAS-R was originally developed by Chen, Weng, Su, Wu, and Yang (2003) and revised by Bai and Fan (2005), measuring the level of IA in individuals. It consists of 19 items, rated on a 4-point Likert scale, ranging from 1 (strongly disagree) to 4 (strongly agree). It is composed of four dimensions: tolerance symptoms of IA, compulsive use of internet & withdrawal symptoms of IA, time management problems, and interpersonal and health-related problems of IA. The first two dimensions constitute the core symptoms of IA, while the latter two represent the related problems associated with it. A higher total score on the scale indicates a greater likelihood and tendency of IA. In this study, the Cronbach's alpha of the CIAS-R was 0.942. Examples include items like “Without the internet, my life lacks enjoyment,” “I find myself spending increasingly more time using the Internet,” and “Internet use has negatively affected my physical health.”

Automatic thoughts questionnaire (ATQ)

The ATQ was originally developed by Hollon and Kendall (1980). It was used to assess the frequency of negative automatic thoughts occurring in individuals over the past week (Wang, Brennen, & Holte, 2006). The ATQ consists of 30 items, each rated on a 5-point Likert scale, ranging from 1 (at no time) to 5 (all the time). The questionnaire covers four aspects related to depression: poor adaptation and desire for change; negative expectations and self-concept; lack of self-confidence; sense of helplessness. All items reflect negative experiences related to depression, and the scores are positively correlated with the level of depression in the individual. In this study, the ATQ's Cronbach's alpha was 0.977. Examples include statements such as “I'm a failure,” “Things will never get better,” and “It's hopeless.”

Self-control scale (SCS)

This scale was developed by Tangney, Baumeister, and Boone (2004) and revised by Tan and Guo (2008). The revised version consists of 19 items across five dimensions: impulse control, healthy habits, resisting temptation, focus on work, and limiting play, rated on a 5-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree), with higher scores indicating stronger self-control. In this study, the scale's Cronbach's alpha was 0.810. Examples include items like “I am good at resisting temptation,” “I have a hard time breaking bad habits,” and “I can easily stick to my goals.”

Self-rating anxiety scale (SAS)

The SAS was developed by Zung (1971), used to assess the severity of anxiety symptoms. The scale consists of 20 items, rated on a 4-point Likert scale, ranging from 1 (none or a little of the time) to 4 (most or all of the time). Based on the Chinese normative data, a cutoff score of 50 indicated greater severity of anxiety. In this study, the Cronbach's alpha of this scale was 0.923. Examples include items such as “I feel more nervous than usual,” “I get sudden feelings of panic,” and “I have trouble sleeping due to worry.”

Statistical analysis

Descriptive statistics and difference tests were conducted using SPSS (version 26.0). R (version 4.3.2) was employed for latent profile analysis and network analysis.

Latent profile analysis

The R package “tidyLPA” was used to perform the latent profile analysis, evaluating models with 2 to 5 latent profiles. Model fit was evaluated using information criteria, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-size adjusted BIC (aBIC), where lower values indicate superior model fit (Burnham & Anderson, 2004; Nylund, Asparouhov, & Muthén, 2007). An entropy value greater than 0.8 indicates that more than 90% of cases are correctly classified (Jung & Wickrama, 2008). We employed the Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (LMR) and the Bootstrapped Likelihood Ratio Test (BLRT) to compare models. p-values less than 0.05 indicating that the k-class model fits the data better than the k-1 class model (Lubke & Muthén, 2007).

Network analysis

Based on the results from LPA, we constructed the network for the IA group. The high IA and low IA groups identified by LPA were combined to form the ‘IA group’ for network analysis. These groups were selected because they represent individuals clearly exhibiting problematic IA patterns (albeit at different severity levels), making them the most relevant subgroups for understanding the network structure within the IA population and identifying potential bridge connectors for intervention, as opposed to the ‘regular’ or ‘at-risk’ groups. Based on the results from LPA, we constructed the bridge network for the IA group using the R packages “bootnet”, “networktools”, and “qgraph”. While the network contained negative edges, we selected bridge expected influence (BEI) as the metric for identifying bridge connectors (Robinaugh, Millner, & McNally, 2016). The stability of centrality indices was evaluated through correlation stability coefficients (CS-C), the CS-C should not be lower than 0.25, and for optimal stability, it is preferable that this value exceeds 0.5 (Epskamp, Borsboom, & Fried, 2018). Specifically, we used the 1-step bridge expected influence as the bridge centrality metric for the network. We then identified the top 20% of nodes with the highest BEI scores as the bridge connectors (Jones et al., 2021). These nodes were identified as bridge connectors, whose high BEI highlights their crucial role in connecting key factor clusters and potentially maintaining the overall problematic state of the comorbidity network.

Ethics

The study was approved by the Ethics Committee of Inner Mongolia Normal University (No. XL202501), and adhered to the ethical principles outlined in the Declaration of Helsinki. Prior to participation, all participants were provided with detailed information about the study’s aims, voluntary nature, confidentiality protocols, and their right to withdraw at any time. We obtained electronic informed consent from all participants prior to their enrollment in the study.

Results

Demographic data of participants

Table 1 describes the characteristics of each group identified by LPA.

Table 1.

Sociodemographic characteristics of each group

Variables	Total (N = 1,030)	High IA group (n = 118)	Low IA group (n = 364)	At-Risk group (n = 404)	Regular group (n = 144)	F/χ 2
Age	20.30 ± 1.44	20.20 ± 1.68	20.32 ± 1.52	20.09 ± 1.26	20.13 ± 1.23	19.859**
Gender						8.977*
Male	571 (55.44%)	52 (44.07%)	197 (54.12%)	239 (59.16%)	83 (57.64%)
Female	459 (44.56%)	66 (55.93%)	167 (45.88%)	165 (40.84%)	61 (42.36%)
Only child						2.322
Yes	421 (40.87%)	48 (40.68%)	157 (43.13%)	154 (38.12%)	62 (43.06%)
No	609 (59.13%)	70 (59.32%)	207 (56.87%)	250 (61.88%)	82 (56.94%)
Family type						4.013
Nuclear family	854 (82.91%)	96 (81.36%)	310 (85.16%)	330 (81.68%)	118 (81.94%)
Single-parent family	168 (16.31%)	22 (18.64%)	52 (14.29%)	70 (17.33%)	24 (16.67%)
Foster care family	8 (0.78%)	0 (0.00%)	2 (0.55%)	4 (0.99%)	2 (1.39%)
Place of residence						2.081
Town	311 (30.19%)	89 (75.42%)	252 (69.23%)	280 (69.31%)	98 (68.06%)
Village	719 (69.81%)	29 (24.58%)	112 (30.77%)	124 (30.69%)	46 (31.94%)

Notes. **p < 0.01, *p < 0.05 High IA group = High Internet Addiction group; Low IA group = Low Internet Addiction group. Use χ² test for categorical variables and ANOVA for continuous variables.

Latent profile analysis of IA

We conducted LPA using mean scores from the four dimensions of the CIAS-R to IA among 1,030 Chinese college students. As the number of model classes increased, the values of AIC, BIC, and aBIC gradually decreased, stabilizing between the 4- and 5-class models. Despite the 3-class model's higher Entropy, the 4-class model better differentiated IA types, identifying regular, at-risk, low IA, and high IA subgroups. Table 2 presents the fit indices of the latent profiles. Figure 1 presents latent classes based on the dimension of the CIAS-R.

Table 2.

Fit indices of the latent profile models of IA

Classes	Log-likelihood	AIC	BIC	aBIC	Entropy	BLRT_p	LMRT_p	Probability of classes
1	−4600.745	9217.490	9256.989	9231.580
2	−3573.909	7173.818	7238.003	7196.714	0.877	<0.001	<0.001	49.3%/50.7%
3	−3049.339	6134.678	6223.550	6166.380	0.926	<0.001	<0.001	47.4%/14.8%/37.8%
4	−2747.572	5541.143	5654.701	5581.651	0.919	<0.001	<0.001	35.3%/39.2%/14.0%/11.5%
5	−2687.474	5430.949	5569.194	5480.262	0.918	<0.001	<0.001	34.7%/38.2%/8.6%/7.1%/11.4%

Notes. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; aBIC = Adjusted Bayesian Information Criterion; BLRT = bootstrap likelihood ratio test; LMRT = Lo-Mendell-Rubin Adjusted Likelihood Ratio Test.

Fig. 1.

Latent classes based on the dimension of the Chinese Internet Addiction Scale-Revised (CIAS-R)

Notes. High IA group = High Internet Addiction group; Low IA group = Low Internet Addiction group; IA1 = Compulsive Use of Internet and Withdrawal Symptoms of Internet Addiction; IA2 = Tolerance Symptoms of Internet Addiction; IA3 = Interpersonal and Health-Related Problems of Internet Addiction; IA4 = Time Management Problems.

Network analysis of IA

We combined the high IA group and low IA group into the IA group for network analysis. Figure 2a shows the bridge network structure of IA, self-control, automatic thoughts, and anxiety in the IA group. Figure 2b shows the bridge nodes in the network of the IA group are Automatic Thoughts (AT) and Anxiety (ANX), indicating that changes in these two connectors have the greatest impact on the overall network.

Fig. 2.

a. The undirected network of internet addiction, self-control, automatic thoughts, and anxiety in the IA group. The blue solid lines represent positive correlations, and the red dashed lines represent negative correlations. Thicker lines indicate stronger correlations. b. Bridge expected influence (1-step) of each node

Narrow bootstrapped 95% confidence intervals for edge weights indicate good reliability and accuracy of the network model (Fig. S1). The CS-Cs of expected influence and bridge expected influence generated by the case-dropping bootstrap method all >0.25 (Fig. S2), indicating that the network structure did not change significantly after most samples were discarded. Figures S3 and S4 present the outcomes of non-parametric bootstrapped difference tests concerning edge weights and node bridge expected influence. The analyses revealed significant differences for the bridge centrality of AT and ANX and their linking edges compared to most other edges and nodes in the network, supporting the robustness of our primary finding identifying AT and ANX as key bridge connectors.

In summary, study 1 identified four distinct profiles of IA among college students using LPA and revealed through NA that automatic thoughts (the ‘C’ component) and anxiety (the ‘A’ component) function as key bridge connectors linking function as key bridge connectors connecting different factor clusters within the comorbidity network. Building on these findings, Study 2 was designed to evaluate the efficacy of interventions, specifically Cognitive Behavioral Therapy and a combined Mindfulness-Based Intervention with CBT, in a randomized controlled trial targeting these identified mechanisms among Chinese college students with internet addiction.

Study 2. CBT and MBI intervention for internet addiction among Chinese college students: a randomized controlled trial

Method

Participants

We conducted an a priori power analysis using G*Power to determine the required sample size for this investigation. For a 3 (CBT group, CBT+MBI group, control group) × 4 (pre-intervention, post-intervention, 1 month follow-up, 6 months follow-up) mixed design, the analysis indicated that a minimum of 30 participants would be needed to achieve 80% power to detect a moderate effect size (f = 0.25) at a 0.05 significance level (Reangsing, Wongchan, Trakooltorwong, Thaibandit, & Oerther, 2025). To account for an anticipated 20% dropout rate, informed by a meta-analysis on Internet addiction treatment (Winkler, Dörsing, Rief, Shen, & Glombiewski, 2013), the recruitment target was increased to 38 participants.

Recruitment for study 2 was conducted from April 2 to April 12, 2024, using an IRB-approved, two-step process to maintain participant anonymity. This process began at the end of the study 1 survey, where consenting participants voluntarily provided telephone numbers via a separate, unlinked web form. Based on this contact list, we initiated telephone calls to individuals from study 1 who were identified with high IA or low IA profiles and were geographically accessible for a face-to-face intervention at our university counseling center. This initial call served as both an invitation to the current study and a preliminary screening to assess their willingness to participate and motivation for change, as well as to check for key exclusion criteria. To be included in the study, participants had to meet the following criteria: (1) belonging to the high IA or low IA profile identified in study 1; (2) agreeing to participate in the 8-week intervention and follow-up assessments. Exclusion criteria were: (1) currently receiving other psychological therapy; (2) a history of severe mental disorders such as schizophrenia or bipolar disorder; (3) the presence of current severe suicidal ideation, substance use disorder, or psychotic symptoms identified during screening. Interventions were delivered face-to-face at the university counseling center. Finally, 39 eligible participants were randomly assigned in a 1:1:1 ratio to the CBT+MBI, CBT, and control groups. The randomization sequence was generated by an independent researcher using a computer-based random number generator. Due to the nature of the intervention, participants and intervention providers were not blinded to group assignment. However, we blinded outcome assessors at post-intervention and follow-up to group assignment to minimize bias. A sample of 36 participants completed the full intervention and all requirements, with 12 participants in each group. The 3 participants who did not complete the study included 2 from the CBT+MBI group and 1 from the CBT group. The reasons for dropout included: (1) conflict with course schedule (n = 2); (2) withdrawal due to personal interest (n = 1). The flow of participants through the study phases is summarized in the CONSORT participant flow diagram (Fig. 3). Paper informed consent was obtained from all participants prior to their enrollment in the study. Participants who completed the entire intervention and follow-up assessments were reimbursed ¥150 (approximately $ 20) as compensation for their time and participation (Hopewell et al., 2025). To minimize potential bias, data analysis was conducted by personnel who were blinded to the group assignments.

Fig. 3.

Intervention flowchart

Notes. CBT+MBI = cognitive behavioral therapy integrated mindfulness-based intervention; CBT = cognitive behavior therapy; T₀ = pre-intervention; T₁ = post-intervention; T₂ = 1 month follow-up; T₃ = 6 months follow-up.

Measures

This study utilized most of the measures employed in study 1, including the CIAS-R (IA), SCS (Self-Control), and SAS (Anxiety). All participants completed CIAS-R, SCS, and SAS at pre-intervention (T₀), post-intervention (T₁), 1 month follow-up (T₂), and 6 months follow-up (T₃). Due to study design limitations, automatic thoughts (AT) were measured only at baseline (T₀) and not repeated at subsequent time points.

Additionally, study 2 included the following measures (assessed at T₀, T₁, T₂, T₃ unless specified).

Self-rating depression scale (SDS)

The SDS developed by Zung (1965), was used to measure the severity of depressive symptoms, encompassing affective, psychological, and somatic symptoms. The SDS consists of 20 items, rated on a 4-point Likert scale from 1 (not at all) to 4 (most of the time). In this study, the Cronbach's alpha of this scale was 0.863. Examples include items like “I feel downhearted and blue,” “I don't enjoy things I used to,” and “I feel that I am no use to anybody.”

Mindful attention awareness scale (MAAS)

The MAAS was developed by Brown and Ryan (2003). This scale was designed to measure the frequency with which individuals experience mindful presence in their daily lives. The scale consists of 15 items, rated on a 6-point Likert scale. Items are reverse-scored such that higher total scores reflect greater dispositional mindfulness. In this study, the Cronbach's alpha of this scale was 0.924. Examples include items such as “I could be experiencing some emotion and not be conscious of it until some time later,” “I rush through activities without paying much attention to them,” and “I find myself preoccupied with the future or the past.”

Intervention satisfaction and usability questionnaire

We also collected a custom questionnaire to assess post-intervention satisfaction and usability. Post-intervention (T₁), participants rated their overall satisfaction on a 4-point Likert scale (1 = very dissatisfied, 4 = very satisfied) and reported on the intervention's effectiveness for managing daily routines, study habits, interpersonal relationships, and emotional regulation based on a binary (yes/no) response. Furthermore, intent to continue applying intervention-acquired skills in future life was assessed via a binary response. Sample items from this questionnaire included “What help do you think this activity provided you?” and “How would you rate your overall satisfaction with this activity?”.

Procedures

Both CBT and MBI protocols were tailored to target automatic thoughts and anxiety, bridge connectors identified via network analysis in Study 1. Our approach was designed to integrate CBT's “top-down” cognitive restructuring with MBI's “bottom-up” affective and metacognitive regulation. As detailed in Table S2, this dual-mechanism model aimed to provide a more comprehensive intervention than either modality alone. The CBT protocol was adapted from The Internet Addiction Assessment and Treatment Manual (Young & De Abreu, 2010), encompassing eight weekly 1–1.5-h sessions focused on building connections, unveiling IA, identifying automatic thoughts, challenging negative thoughts, crafting coping strategies, enhancing interpersonal trust, mastering self-control, and preventing relapse. The MBI protocol, derived from The Mindfulness Way Workbook (Teasdale, Williams, & Segal, 2013), consisted of eight weekly 0.5–1-h sessions emphasizing embracing relaxation, focusing on the present, body scanning, cultivating acceptance, walking with focus, observing thoughts, managing emotions mindfully, and fostering grateful awareness. Two experienced CBT and MBI psychologists, proficient in both CBT and MBI, delivered all sessions individually, each fully dedicated to one intervention condition. To ensure intervention fidelity and professional standards, all intervention sessions were supervised by a qualified clinical supervisor. The CBT+MBI group received an 8-week dual-modality intervention (1–1.5 h CBT and 0.5–1 h MBI sessions per week), the CBT alone group only received the CBT protocol, and the control group received no intervention beyond completing baseline questionnaires at longitudinal tracking time points. The design and analysis plan for this study were registered retrospectively on the Open Science Framework (https://doi.org/10.17605/OSF.IO/RKC43) after data collection had begun. A brief description of the dual-modality intervention is shown in Table S2.

Statistical analysis

Descriptive statistics and repeated measures analysis of covariance (RM-ANCOVA) were conducted using SPSS (version 26.0).

Repeated measures analysis of covariance (RM-ANCOVA)

We conducted an RM-ANCOVA to examine group (MBI+CBT group, CBT alone group, Control group) × time (pre-intervention [T₀], post-intervention [T₁], 1 month follow-up [T₂], 6 months follow-up [T₃]) effects on IA, with age and gender included as covariate. The group × time interaction assessed intervention effects, with LSD post-hoc tests comparing pairwise group differences (MBI+CBT vs. Control, CBT vs. Control, MBI+CBT vs. CBT) at each time point.

Additionally, we also conducted RM-ANCOVA analysis to examine the effects of group (MBI+CBT, CBT, Control) and time (pre-intervention [T₀], post-intervention [T₁], 1 month follow-up [T₂]) on mindfulness, anxiety, and depression. Assessed by mindful attention awareness scale (MAAS), self-rating anxiety scale (SAS), self-rating depression scale (SDS) separately, with age and gender included as covariates.

Ethics

The study was approved by the Ethics Committee of Inner Mongolia Normal University (No. XL202501) and adhered to the ethical principles outlined in the Declaration of Helsinki.

Results

Participant characteristics

There were no statistical differences at T₀ between each group in terms of demographic characteristics or primary study variables (see Table 3).

Table 3.

Sociodemographic characteristics of each group

Variables	MBI+CBT (n = 12) N (%)/M±SD	CBT (n = 12) N (%)/M±SD	Control (n = 12) N (%)/M±SD	F/χ2
Age	21.58 ± 2.11	20.67 ± 0.89	20.83 ± 1.11	1.325
Gender				1.200
Male	1 (8.33)	2 (16.67)	3 (25.00)
Female	11 (91.67)	10 (83.33)	9 (75.00)
Only child				0.234
Yes	4 (33.33)	5 (41.67)	5 (41.67)
No	8 (66.67)	7 (58.33)	7 (58.33)
Family type				2.245
Nuclear family	8 (66.67)	7 (58.33)	7 (58.33)
Single-parent family	4 (33.33)	4 (33.33)	5 (41.67)
Foster care family	0 (0.00)	1 (8.33)	0 (0.00)
Place of residence				0.241
Town	5 (41.67)	4 (33.33)	4 (33.33)
Village	7 (58.33)	8 (66.67)	8 (66.67)
IA	62.50 ± 2.11	63.67 ± 2.87	63.92 ± 4.60	0.608
SA	57.58 ± 8.35	59.42 ± 8.13	54.75 ± 6.80	1.093
SD	51.33 ± 6.07	54.25 ± 7.03	52.75 ± 6.41	0.601
MAA	40.92 ± 3.45	41.08 ± 3.92	42.00 ± 3.64	0.302

Note. IA = internet addiction; SA = Anxiety; SD = Depression; MAA = Mindful Attention Awareness.

Effectiveness of the intervention

RM-ANCOVA revealed a significant main effect of group on IA (F = 84.190, p < 0.001, η_p² = 0.845), with the CBT+MBI group exhibiting significantly lower IA scores than the CBT group (p < 0.001, 95% CI [−8.478, −4.180]), and the CBT group outperforming the control group (p < 0.001, 95% CI [−9.393, −5.231]). No significant main effect of time was observed (F = 0.066, p = 0.978, η_p² = 0.02). The group × time interaction was significant (F = 17.817, p < 0.001, η_p² = 0.535). As shown in Tables 4 and 5, Fig. 4. Simple effects analysis indicated no baseline differences in IA scores across groups at T₀ (all p > 0.05, MBI+CBT vs. Control: 95% CI [−4.239, 1.739], CBT vs. Control: 95% CI [−3.095, 2.689], MBI+CBT vs. CBT: 95% CI [−4.034, 1.939]). Post-intervention (T₁), both intervention groups had significantly lower IA scores than the control group (all p < 0.001, MBI+CBT vs. Control: 95% CI [−23.516, −17.932], CBT vs. Control: 95% CI [−14.763, −9.361]), with CBT+MBI scores significantly below CBT scores (p < 0.001, 95% CI [−11.451, −5.872]). At 1 month follow-up (T₂), intervention groups maintained significantly lower IA scores than the control group (all p < 0.001, MBI+CBT vs. Control: 95% CI [−24.847, −15.548], CBT vs. Control: 95% CI [−17.761, −8.763]), with CBT+MBI consistently outperforming CBT (p = 0.005, 95% CI [−11.581, −2.290]). At 6 months follow-up (T₃), CBT group also maintained significantly lower IA scores than the control group (p = 0.034, 95% CI [−7.152, −0.292]), CBT+MBI group maintained significantly lower IA scores than the control group (p < 0.001, 95% CI [−15.936, −8.847]), with CBT+MBI consistently outperforming CBT (p < 0.001, 95% CI [−12.211, −5.128]).

Table 4.

The IA scores for each group at different time point

Group	Time
	T0 (M±SD)	T1 (M±SD)	T2 (M±SD)	T3 (M±SD)
CBT+MBI	62.50 ± 2.11	40.33 ± 3.92	43.58 ± 4.01	47.92 ± 3.37
CBT	63.67 ± 2.87	49.42 ± 3.39	51.00 ± 6.94	57.00 ± 3.86
Control	63.92 ± 4.60	61.33 ± 2.10	63.92 ± 4.87	60.92 ± 4.96

Notes. IA = internet addiction; T₀ = pre-intervention; T₁ = post-intervention; T₂ = 1 month follow-up; T₃ = 6 months follow-up; CBT+MBI = cognitive behavioral therapy integrated mindfulness-based intervention; CBT = cognitive behavioral therapy.

Table 5.

Results of RM-ANCOVA for IA

Effects	F	p	η 2	Post-hoc		p	95%CI
Main effects
Age	1.742	0.197	0.053
Gender	0.001	0.978	<0.001
Group	84.190	<0.001	0.845	a v.s. b		<0.001	−8.478, −4.180
				a v.s. c		<0.001	−15.791, −11.490
				b v.s. c		<0.001	−9.393, −5.231
Time	0.066	0.978	0.002
Interactions				Simple effects
Time*Age	0.488	0.691	0.016
Time*Gender	1.713	0.170	0.052
Time*Group	17.817	<0.001	0.535	T1	a v.s. b	0.480	−4.034, 1.939
					a v.s. c	0.400	−4.239, 1.739
					b v.s. c	0.887	−3.095, 2.689
				T2	a v.s. b	<0.001	−11.451, −5.872
					a v.s. c	<0.001	−23.516, −17.932
					b v.s. c	<0.001	−14.763, −9.361
				T3	a v.s. b	0.005	−11.581, −2.290
					a v.s. c	<0.001	−24.847, −15.548
					b v.s. c	<0.001	−17.761, −8.763
				T4	a v.s. b	<0.001	−12.211, −5.128
					a v.s. c	<0.001	−15.936, −8.847
					b v.s. c	0.034	−7.152, −0.292

Note. a: CBT+MBI group; b: CBT group; c: Control group.

Fig. 4.

The interaction of group and time

Notes. CBT+MBI = cognitive behavioral therapy and mindfulness-based intervention, CBT = cognitive behavior therapy. T₀ = pre-intervention, T₁ = post-intervention, T₂ = 1 month follow-up, T₃ = 6 months follow-up.

Besides, the result of RM-ANCOVA analysis about mindfulness, anxiety and depression indicated significant group × time interaction effect. CBT+MBI group and CBT alone exhibited the highest level of mindfulness in T₁, higher level in T₂ compared to T₀ point. In terms of anxiety and depression, CBT+MBI group and CBT alone exhibited the lowest score in T₂, lower score in T₁ compared to T₀ point, see Tables S4–S6 and Figs S5–S7.

Intervention satisfaction and usability

Post-intervention (T₁), 81.1% of participants in the CBT+MBI and CBT groups rated their overall satisfaction as “very satisfied”. Additionally, 75.7% reported the intervention as effective for managing daily routines, study habits, interpersonal relationships, and emotional regulation. Furthermore, 83.8% indicated intent to continue applying intervention-acquired skills in future life.

In conclusion, study 2 demonstrated that both CBT and CBT+MBI interventions were effective in reducing IA levels among college students, with CBT+MBI showing superior and more sustained effects, alongside positive impacts on mindfulness, anxiety, and depression. No adverse events or harms related to the interventions were reported during the study period.

Discussion

Subgroup identification and bridge connectors via LPA and NA

The results of LPA identified four IA profiles (regular, at-risk, low IA, high IA) among college students. The regular group had the lower scores in all items of CIAS-R and showed a lower risk of internet addiction. Specifically, they had the lowest score in compulsive use of the internet, tolerance symptoms of internet addiction, interpersonal and health-related problems, and time management problems. Conversely, the high IA group exhibited the most severe symptoms across these dimensions. Additionally, we also found an at-risk group, which exhibited moderate levels of IA symptoms, and highly likely to develop into high IA individuals (Brand et al., 2016). However, it is worth noting that the class trajectories presented in Fig. 1 reveal highly parallel lines across all IA subdimensions. This pattern might suggest that the profiles differ primarily in the severity of symptoms rather than in qualitative symptom configurations. In other words, the LPA solution might appear to reflect a dimensional gradient rather than categorically distinct subtypes of IA. This implies that the observed “heterogeneity” might be more quantitative (degree) than qualitative (kind). Despite this, LPA remains useful for identifying clinically relevant groups like the at-risk group that could benefit from targeted prevention or early intervention efforts. Further investigation into the dimensionality versus typology of IA is warranted.

We merged the high and low IA subgroups into an IA group for network analysis, which elucidated complex relationships among IA, self-control, automatic thoughts, and anxiety. Notably, our undirected network analysis results, specifically the high bridge centrality of AT and ANX within the comorbidity network comprising IA and other relevant variables, provide support for interpreting these connectors as playing potential “dual roles” (vulnerability factors and consequences), indicating their significant interplay within the network structure, potentially consistent with the dynamic feedback loops proposed by the ACE process in the I-PACE model. Specifically, the interventions appear to have successfully disrupted this maladaptive cycle: CBT primarily targeted the Cognitive (C) component by restructuring automatic thoughts, while the addition of MBI offered a powerful tool to de-escalate the Affective (A) component by fostering non-judgmental awareness of anxiety. The superior, sustained effects of the combined intervention suggest that simultaneously decoupling both the cognitive and affective drivers is a particularly potent strategy for dismantling the addictive process. These findings align with the assumptions of the I-PACE model, suggesting that automatic thoughts and anxiety are not only potential consequences of IA, but also key connectors that drive and sustain this behavioral pattern, forming a negative feedback loop (Brand et al., 2016). While existing research presents conflicting findings regarding the relationship between IA, automatic thought, and anxiety, with some studies indicating bidirectional influences among these variables, complicating the design of subsequent intervention research (Dong il & Yeoju, 2005; Odaci, Degerli, & Cikrikci, 2021; Wang et al., 2024; Xie et al., 2023). NA effectively addresses this issue by clarifying the complex bidirectional relationships between these variables, providing a clearer framework for intervention design.

Targeted intervention outcomes: CBT+MBI vs CBT

Regarding the intervention effects, our results demonstrate that both CBT+MBI and CBT alone effectively alleviate symptoms of IA, mindfulness, anxiety, and depression among college students and sustained effects for 6 months. A key objective of study 2 was to assess whether interventions targeting the bridge connectors of automatic thoughts (AT) and anxiety, identified in study 1, would yield superior outcomes. The superior efficacy of the dual-modality CBT+MBI intervention compared to CBT alone, particularly in producing more significant and lasting reductions in IA and anxiety, provides evidence supporting this targeted approach. Specifically, while both interventions likely addressed AT through CBT components (Nicolai, 2014; Thakral, Von Korff, McCurry, Morin, & Vitiello, 2020) and anxiety to some extent, the addition of MBI in the combined group may have offered a more comprehensive strategy for managing anxiety. MBI emphasis on non-judgmental observation and emotional regulation (López-Navarro & Al-Halabí, 2022; Thakral et al., 2020)= could have provided participants with enhanced skills to disengage from anxious rumination and maladaptive coping mechanisms often linked to excessive internet use (Dawson et al., 2020; Van Gordon et al., 2017). This enhanced capacity for affect regulation, when combined with CBT's cognitive restructuring, likely contributed to the synergistic effect observed, enabling students to more effectively address the interplay between cognitive (AT) and affective (anxiety) drivers of their IA. This interpretation aligns with the notion that multifaceted interventions targeting multiple identified mechanisms can yield more robust outcomes.

From a theoretical perspective, our findings further support the I-PACE model, which posits that IA is a result of the interplay among affect, cognition, and executive functions (Brand et al., 2019). We observed reductions in IA and anxiety, alongside improvements in mindfulness, are consistent with the I-PACE model's emphasis on the interconnectedness of these connectors. The sustained nature of these improvements, especially in the CBT+MBI group, suggests that addressing these core components can indeed disrupt the maladaptive positive feedback loops (e.g., anxiety leading to internet use for escape, which in turn exacerbates negative consequences and anxiety) described within the I-PACE framework (Brand et al., 2019; Tokunaga, 2017).

The contribution of this research extends beyond simply demonstrating intervention efficacy. By first identifying specific psychological mechanisms (bridge connectors of AT and anxiety) through a data-driven approach (LPA and NA in study 1) and then evaluating interventions tailored to these mechanisms, our study offers a more nuanced understanding of how and why certain interventions might be effective for IA. While both interventions showed promise, the enhanced and more durable effects of the CBT+MBI combination suggest that a dual-focus on both cognitive restructuring (primarily CBT for AT) and affect regulation/mindfulness (enhanced by MBI for anxiety and general awareness) is particularly beneficial. This aligns with the multidimensional treatment approach advocated by the I-PACE model, aiming to improve mental health across cognitive, emotional, and behavioral domains for long-term outcomes (Brand et al., 2016, 2019). This two-study approach, therefore, provides a replicable framework for future research aiming to develop and test personalized or mechanism-based interventions for behavioral addictions.

During the intervention process, we also discovered that most of participants are satisfied and willing to use this strategy to following daily activities. Through the intervention, enhancing students' awareness of IA and its negative consequences is crucial for prevention and intervention. Existing studies have shown that the severe symptoms of individuals with high IA are not only derived from their high sensitivity to stimuli but also from their unconscious responses to online stimuli (Ryan, 2002). Therefore, it is crucial for college students to recognize which cognitive and behavioral patterns are associated with high sensitivity. This increased awareness can serve as a catalyst for self-reflection, encouraging students to evaluate and modify their internet usage habits, ultimately reducing the prevalence and severity of IA (Silvia & Phillips, 2011). As students become more aware of the psychological and behavioral aspects of their internet use, they are better equipped to manage potential risks, thereby enhancing their long-term well-being and reducing the likelihood of developing more severe forms of IA.

By providing targeted interventions, this study not only enriches the theoretical foundation but also offers practical guidelines. Mental health practitioners in higher education institutions can utilize the findings of this research to promptly intervene in populations with IA. Future research should focus on deepening existing findings while exploring new perspectives and methods, such as in-depth investigation of the impact of different individual characteristics on intervention effects to more precisely customize personalized intervention plans; considering the introduction of digital tools and technological support to enable students to access help and support anytime and anywhere, enhancing the convenience and coverage of interventions; and analyzing the influence of family, school, and social environments on students' internet use habits to design more combined intervention strategies.

Limitations and future directions

When interpreting the results of this study, some limitations should be considered. First, our use of the broad term ‘Internet Addiction’, while operationally defined and justified in the introduction, represents a necessary simplification. The field is rightly moving toward more specific, activity-based terminologies (e.g., gaming disorder, social media addiction) to enhance nosological precision (Montag et al., 2021; Wegmann et al., 2025). While our focus on shared underlying mechanisms made a general term appropriate for this study's scope, we acknowledge that this approach may obscure important differences between specific online behaviors. In our subsequent work, we are committed to employing more granular terminology where feasible, aiming to contribute to the field's ongoing efforts to refine its conceptual and diagnostic frameworks. Secondly, this study relied on participants' self-reports for both subgroup classification in study 1 and outcome assessment in study 2, which may be susceptible to common method variance. Crucially, participant selection and eligibility for study 2 (based on study 1 LPA profiles) were not confirmed using structured clinical interviews, limiting diagnostic accuracy and the generalizability of findings. The non-use of more objective diagnostic methods was primarily due to feasibility and resource constraints. Furthermore, the study experienced participant dropouts (3 of 39 enrolled did not complete). The analysis was conducted using only the 36 participants with complete data across all time points. This approach, known as complete case analysis, can introduce selection bias if participants who dropped out differ systematically from those who completed the study (e.g., being less motivated or experiencing less favorable outcomes), thus potentially limiting the generalizability of the findings to the entire initially recruited sample or broader population. Future researchers could mitigate this bias by incorporating measures beyond self-report for IA and other behavioral dimensions, such as objective screen time tracking or behavioral tasks assessing impulse control related to internet use. Third, this study identified an at-risk group for IA with significant clinical implications through LPA but did not conduct longitudinal tracking and intervention validation for this group. According to the I-PACE model's framework, this group has not yet met the cut-off criteria. They are at a higher risk of transitioning into the internet addiction group (Brand et al., 2019). Therefore, it is recommended that future practice should prioritize the development of preventive interventions targeting this group, focusing on strengthening cognitive reappraisal training and the cultivation of impulse control abilities. Implementing early interventions can effectively prevent their pathological conversion into the internet addiction group. Fourth, due to resource limitations and the preliminary nature of the trial, an MBI-only group was not included; the focus was on testing the added value of MBI to the established CBT. Future studies should incorporate an MBI only condition to clarify MBI's independent efficacy for IA and to better understand its specific contributions versus synergistic effects when combined with CBT. This would help delineate optimal and potentially more resource-efficient intervention pathways for individuals with IA. Finally, another limitation pertains to the assessment of intervention mechanisms. Although study 1 highlighted automatic thoughts (AT) as a crucial bridge connector, and the interventions in study 2 were designed to target them, the lack of longitudinal AT measurement in study 2 precludes a direct evaluation of the intervention's impact on this specific cognitive target. While reductions in IA and anxiety suggest that targeting I-PACE-related bridge connectors is a viable strategy, the mediating role of changes in AT could not be empirically tested in this trial.

Conclusion

In conclusion, IA is a complex, multifactorial disorder influenced by complex physiological, psychological, and social factors. This two-study investigation provided insights into both the heterogeneity of IA among college students and the potential efficacy of targeted interventions. Study 1 identified four distinct IA profiles (regular, at-risk, low IA, and high IA) using LPA, highlighting the varied manifestations of IA within this population. Furthermore, NA pinpointed automatic thoughts and anxiety as key bridge connectors connecting different factor clusters within the comorbidity network, offering specific targets for intervention. Building on these exploratory findings, study 2 offered preliminary evidence for the effectiveness of a dual-modality CBT+MBI intervention and CBT alone in reducing IA symptoms and improving related psychological outcomes (mindfulness, anxiety, depression) among Chinese college students, with effects sustained for up to six months. Notably, the dual-modality CBT+MBI intervention appeared to demonstrate enhanced and more durable improvements compared to CBT alone, suggesting potential benefits of this dual-modality intervention that addresses both cognitive and affective mechanisms identified as bridge connectors.

Supplementary material

jba-14-1590-s001.pdf ^{(663.4KB, pdf)}

Funding Statement

Funding sources: This research was supported by the Inner Mongolia Natural Science Foundation (2022MS03016), Inner Mongolia Social Science Foundation (2025DY44), Inner Mongolia Higher Education Science Project (NJZZ22566), Research Funds for the Inner Mongolia Normal University (2022JBQN118), High-level Talent Research Start-up Fund for the Inner Mongolia Normal University (2021YJRC010).