Social-Media-Based Mental Health Interventions: Meta-Analysis of Randomized Controlled Trials

The study addresses the global burden of mental health disorders and barriers to access, proposing social-media-based interventions as scalable, low-cost, and acceptable options. It aims to evaluate whether rigorously designed social-media-based RCTs reduce negative mental health outcomes among adolescents and adults and to examine moderators influencing effectiveness. Research questions: (RQ1) What are the overall impacts of social-media-based RCTs on negative mental health outcomes versus care as usual (CAU) or waitlist? Hypothesis H1: social-media-based RCTs will alleviate negative outcomes. (RQ2) To what extent do outcomes differ by methodological and intervention criteria (age, recruitment type, control group type, program duration, delivery guidance, sex, and program focus)? Hypothesis H2: larger effects are expected with clinical populations, younger age, passive controls, human guidance, social orientation, majority female samples, and longer duration.

Prior meta-analyses have broadly covered online, digital, eHealth, computer therapy, and internet-based mental health interventions, but few examined social-media-based interventions specifically. Existing reviews focused on cancer patients, children, and young people using social networking sites. This meta-analysis fills a gap by focusing on rigorously designed social-media-based mental health RCTs for general populations and exploring moderators informed by prior digital intervention literature.

Registration: Preregistered on OSF, with deviations to include all age ranges (analyzed via age moderators), restrict to social-media-based interventions only, and focus on negative mental health outcomes to reduce heterogeneity. Search Strategy: Comprehensive searches across seven databases (ERIC, PsycINFO, Scopus, PsycArticles, Communication and Mass Media Complete, PubMed, ProQuest), targeted hand searching via Paperfetcher, and forward/backward citation tracking using CitationChaser. Searches completed by April 2025; 11,658 records imported into Covidence. Eligibility Criteria: RCTs; ≥30 participants per arm at baseline; interventions delivered largely via social media platforms (eg, Facebook, Instagram, WhatsApp, WeChat); exclude social media abstinence; baseline differences <0.25 SDs; differential attrition <15%; delivery by nonresearchers; quantitative measures of negative mental health outcomes (depression, anxiety, stress, negative affect, psychological distress) with sufficient data for Hedges g; full-text English; published ≥2005; primary studies; exclude one-item measures and single-session interventions. Screening and Coding: Double-blind screening of titles/abstracts and full texts in Covidence; double coding using Google Spreadsheets; conflicts resolved in group discussions; open data and code shared on GitHub. Analytical Plan: Random-effects meta-regression using R (metafor). Effect sizes computed as Hedges g (escalc), with inverse-variance weights adjusted per Hedges. Moderators included recruitment type (clinical vs nonclinical), age group (adolescents, early, middle, late adulthood), control type (waitlist vs CAU/active), delivery (self-guided vs human-guided), duration (weeks), sex composition (>70% female), and program focus (social-oriented vs task-oriented). Publication bias assessed via selection modeling (weightr). Risk of bias assessed with JBI Critical Appraisal Checklist for RCTs, coding allocation concealment, blinding of participants, deliverers, assessors, and follow-up completeness; double-blind quality assessment with a third reviewer for disagreements.

• Included studies: 17 RCTs, 22 programs, total n=5624; 73 effect sizes (depression=31, anxiety=27, stress=12, negative affect=2, psychological distress=1).
• Overall effectiveness: ES=0.32 (95% CI 0.24-0.45; P<.001), random-effects; I²=88.1% (28.87% between-study; 59.23% within-cluster), τ²≈0.13; prediction interval ~(-0.38 to 1.08).
• Outcome subgroups: Depression ES=0.31 (P<.001, n=31); Anxiety ES=0.33 (P=.04, n=27); Stress ES=0.69 (P=.02, n=12); overall suggests reductions in negative mental health outcomes.
• Moderator results: • Sex: >70% female associated with larger effects (β=1.40, P=.01); marginal means ES=1.81 (>70% female) vs ES=0.41 (<70% female). • Delivery: Human-guided more effective than self-guided (β=-0.72, P=.02); marginal means ES=1.35 (human-guided) vs ES=0.63 (self-guided). • Program focus: Social-oriented more effective than task-oriented (β=-0.76, P=.03); marginal means ES=1.20 (social) vs ES=0.44 (task). • Control type: CAU/active controls associated with larger effects than waitlist (β=-0.49, P=.02); marginal means ES=1.37 (CAU/active) vs ES=0.88 (waitlist). • Age, duration, clinical vs nonclinical: No significant moderation (age comparisons mostly nonsignificant; duration P=.26; clinical P=.17). Late adulthood showed largest marginal mean ES but based on limited data.
• Publication bias: Weight-function model suggested selective reporting (nonsignificant results less likely); estimates indicated potential upward adjustment.
• Risk of bias: Overall low risk with mean JBI appraisal score 9.29/13; several domains frequently unclear; stringent inclusion criteria ensured baseline rigor.

Findings support H1: social-media-based interventions yield small-to-moderate benefits in reducing depression, anxiety, and stress across general populations, addressing access and scalability challenges. Moderator patterns suggest mechanisms: social-oriented designs may enhance therapeutic alliance and peer support; human guidance likely improves adherence and engagement; majority-female samples may benefit more due to higher help-seeking and social support dynamics. The unexpected superiority of CAU/active controls over waitlist may reflect waitlisted participants seeking alternative treatments in accessible postpandemic environments, complicating comparisons. No significant age or duration effects indicate broad applicability across age groups and that program length alone may not drive effectiveness. Despite heterogeneity, risk of bias was low and sensitivity analyses robust, though selection modeling indicates possible publication bias.

This meta-analysis provides high-quality evidence that social-media-based mental health interventions can reduce negative mental health symptoms and are particularly effective when human-guided, socially oriented, and implemented with CAU/active controls, especially among majority-female samples. Given their scalability and cost-effectiveness, integrating social-media-based interventions into routine mental health care and public health strategies is recommended. Future research should design rigorous multi-arm RCTs, explore interactions between program orientation and user characteristics (sex, personality, tech comfort, specific disorders), improve reporting (eg, race/ethnicity), and examine strategies to optimize effectiveness for male-majority samples and various age groups.

• Limited number of included studies (n=17) constraining statistical power and moderator analyses (low degrees of freedom).
• Small subgroup sample sizes for certain outcomes and categories (eg, psychological distress n=1, negative affect n=2, late adulthood n=1), warranting cautious interpretation.
• High heterogeneity (I²=88.1%), with substantial within-cluster variability.
• Publication bias indicated by selection modeling, suggesting upwardly adjusted mean effects due to underreporting of nonsignificant findings.
• Insufficient reporting on race/ethnicity across studies, limiting equity-related analyses.