Conference Agenda

Introduction

Addictive screen use in adolescence represents an emerging suicide risk factor, with high and increasing social media and mobile phone trajectories associating with 2-3 times of suicide behaviors in our previous JAMA paper (Xiao, et al, 2025). Yet, little is known about which youth follow high-risk trajectories. Understanding early predictors of high-risk addictive screen use trajectories can inform targeted suicide prevention strategies. This study applied explainable machine learning to identify key baseline factors distinguishing developmental trajectories of addictive screen use across social media, mobile phones, and video games.

Methods

Using data from the Adolescent Brain Cognitive Development (ABCD) Study (Baseline to Year 5), we reproduced three distinct trajectories (low-stable, increasing, high-persistent) for each screen modality through latent growth curve modeling. We utilized XGBoost classifiers to predict trajectory membership. Feature importance was interpreted using SHAP (SHapley Additive exPlanations) values across five domains: sociodemographic factors, psychological (including suicidal ideation, suicidal behaviors, family suicide history), neurocognitive functioning, family/environmental dynamics, and physical health behaviors.

Results

For high-risk social media addiction trajectories, the strongest predictors included age (SHAP=0.21), male sex (0.15), picture vocabulary task (0.16), peer number of friends (0.16), family substance use (0.14), and CBCL externalizing behaviors (0.12). In contrast, mobile phone addiction was uniquely predicted by physical health markers, specifically Waist Circumference (0.13) and Puberty status (0.12), alongside cognitive markers like Working Memory (0.10). Video game addiction displayed the most distinct demographic risk, overwhelmingly driven by Male Sex (0.62) and low Family Parental Monitoring (0.20). Neurocognitive markers and environmental factors often outperformed traditional clinical risk factors.

Conclusions

These findings enable early identification of at-risk youth and inform precision interventions targeting modifiable factors to reduce screen addiction and suicide risks. Findings suggest rethinking screen addiction not merely as a symptom of distress, but as a behavior rooted in neurocognitive development and environmental context.

Introduction

Social media addiction trajectories predict 2-3 fold increased suicide risk in adolescence, yet the neural mechanisms distinguishing established addiction from predisposition to addiction remain unclear. Understanding reward circuitry and cognitive control abnormalities underlying different addiction trajectories could inform circuit-specific interventions for suicide prevention.

Methods

Using the Adolescent Brain Cognitive Development (ABCD) Study sample (N=11,000) with extended 5-year follow-up, we identified three social media addiction trajectories through latent growth curve modeling: T1 (established addiction at age 10), T2 (addiction predisposition, non-addicted at age 11 but addicted by age 16), and T3 (resilient, non-addicted throughout). We analyzed baseline task-based fMRI data examining reward/loss anticipation, outcome feedback, and cognitive control (Stop Signal Task) to compare neural correlates across trajectory groups. We hypothesized that T1 would show blunted reward/loss sensitivity and impaired cognitive control, T2 would show impaired cognitive control with normal reward/loss processing, and T3 would show intact functioning across domains.

Results

The Addicted group (T1) exhibited a significantly blunted reward anticipation response at baseline, which varied depending on the size of the anticipated reward. Contrary to our hypothesis regarding cognitive control, group differences were less apparent in the Stop Signal Task, loss anticipation, or outcome feedback. This indicates that neural deficits in reward circuitry are detectable in established addiction at age 10, distinguishing it from future addiction or resilience.

Conclusions

This study identifies specific reward circuitry abnormalities at age 10 in established addiction that differentiate it from future addiction or resilience, despite shared suicide risk. These neural signatures may enable early identification of highest-risk youth and inform circuit-specific biofeedback interventions targeting reward processing to prevent addiction development, potentially reducing concurrent suicide risk in adolescents.

Objectives: To determine the causal relationship of time spent on social media, texting/messaging, and video chat—distinct from addiction-like mobile phone usage—to suicidal thoughts and behaviors (STB) and mental health in early adolescence.

Methods: We analyzed longitudinal data from 6,991 adolescents (ages 10–14) in the US-based Adolescent Brain Cognitive Development (ABCD) Study (2019–2021). To address parameter bias from confounding, random measurement error, and bidirectional association, we employed instrumental variable (IV) probit models using peer behavior as instruments. We examined the distinct effects of daily duration of use versus addiction-like mobile phone attachment (measured by the Mobile Phone Involvement Questionnaire) on suicidal ideation (SI), suicide attempts (SB), and internalizing/externalizing behaviors.

Results: Assessing effects by addiction status revealed distinct patterns. For adolescents with addiction-like attachment, moderate use proved protective against suicidality: two hours of daily social media use reduced the probability of SI (marginal effect: -0.016; 95% CI: -0.032, -0.001) and SB (-0.017; 95% CI: -0.033, -0.001). Similarly, one hour of texting or video chat reduced SB probability (-0.034 and -0.021, respectively). However, a trade-off emerged: this same level of use in the addiction-like group increased the probability of borderline externalizing behaviors. Conversely, for adolescents without addiction-like attachment, moderate use showed no significant impacts on STBs.

Conclusions: These findings underscore the importance of distinguishing addiction-like mobile phone attachment from simple time-use metrics. For adolescents with addiction-like attachment, moderate digital social connectivity appears to mitigate suicide risk, potentially due to peer connection, despite a concurrent increase in mild externalizing symptoms. Prevention strategies should prioritize treating addiction-like attachment behaviors rather than broadly restricting time, which may offer protective benefits for at-risk youth.

Background: As social media has become embedded in adolescents’ daily lives, understanding how routine digital behaviors relate to short-term fluctuations in suicide risk has become increasingly urgent. Yet prior findings remain mixed, in part because most studies do not distinguish between-person vulnerabilities from within-person changes in social media behavior and context.

Methods: Ninety-nine adolescents aged 11–18 at high risk for suicide completed questionnaires and used an app that gathered daily smartphone data over six months. The dataset included sociodemographic details, total smartphone activity, social media usage time, and sleep indicators derived from nighttime screen inactivity. Two-level linear mixed-effects models analyzed longitudinal change and group differences by sexual and gender minority (SGM) and immigrant status, as well as associations to suicidal thoughts and behaviors (STB).

Results: Participants contributed 1,500 participant-weeks of data (median completion = 74%). Average daily smartphone use was 14,249 seconds (~4 hours), social media use 3,431 seconds (~1 hour), and nightly sleep 8.7 hours. Feasibility was high, with moderate-to-strong within-person stability across outcomes. No longitudinal differences emerged by SGM status, whereas immigrant-origin adolescents exhibited shorter but more stable sleep. Social media use was not directly associated with STB. Higher than usual social media use in the preceding week was associated with higher suicidal ideation among adolescents with higher social functioning, but with lower suicidal ideation among adolescents with lower social functioning.

Conclusions: Findings highlight the nuanced interplay between stable vulnerabilities and weekly digital behaviors, underscoring the value of ecologically informed, within-person approaches for understanding social media’s role in adolescent suicide risk.