Urbanization and the Network Role of Loneliness in Adolescent Suicide Risk: A Four-Stage Analysis Across 87 Countries

Urbanization and the Network Role of Loneliness in Adolescent Suicide Risk: A Four-Stage Analysis Across 87 Countries | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Urbanization and the Network Role of Loneliness in Adolescent Suicide Risk: A Four-Stage Analysis Across 87 Countries Qingjun Zhu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9083788/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 12 You are reading this latest preprint version Abstract Background The loneliness–suicide association has been confirmed across countries, but what macro-level conditions account for cross-national variation in its strength remains unknown. Methods Four-stage analysis of 439,026 adolescents (13–17 years; 52.1% female) from 87 countries in five non-European WHO regions using GSHS data. Stage 1: psychometric networks (EBICglasso) with regional centrality comparisons. Stage 2: cross-regional XGBoost transfer with SHAP feature importance. Stage 3: ecological correlations between 11 macro-indicators and SHAP profiles ( N = 42 countries), with GLMM and specification curve analysis. Stage 4: urbanization–edge weight associations across 38 country-specific networks. Results Suicidal ideation ranked in the top two by Strength centrality across all WHO regions (Stage 1). Prediction models failed to transfer across regions (mean AUC degradation = 17.3 percentage points), with loneliness as the most variable predictor (Stage 2). Urbanization showed the strongest association with loneliness SHAP importance ( r _s = .655, p _FDR = .0003, N = 42); a GLMM confirmed this at the individual level (OR = 1.096, 95% CI [1.053, 1.140], p = 6.2 x 10^-6), with 94.4% of 432 specification curve variants yielding p < .05 (Stage 3). Urbanization was positively correlated with the loneliness–suicidal ideation edge weight ( r _s = + .639, p _FDR < .001) and negatively with the friendship–loneliness edge weight ( r _s = − .799, p _FDR < .0001; Stage 4). Conclusions The network role of loneliness in adolescent suicide risk varies with national urbanization level: in higher-urbanization contexts, loneliness’s risk connections strengthen while protective connections weaken. If confirmed longitudinally, these patterns would support development-stage-adapted prevention over uniform global templates. adolescent suicide network analysis urbanization cross-cultural comparison loneliness GSHS Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Suicide claimed 727,000 lives globally in 2021 and remains the third leading cause of death among 15–29-year-olds [ 1 ]. Although the age-standardized suicide rate declined from 14.9 to 9.0 per 100,000 between 1990 and 2021 [ 41 ], the trajectory among adolescents is heterogeneous: global suicide mortality among 10–24-year-olds decreased overall, but regional and sociodemographic disparities widened substantially [ 42 ]. Multiple high-income countries have reported rising adolescent suicidal ideation and self-harm [ 2 , 3 ], while LMICs remain underrepresented in the evidence base [ 4 ]. The WHO Global School-based Student Health Survey (GSHS) is one of the largest adolescent health databases and allows large-scale cross-cultural comparison [ 5 , 44 ]. GSHS microdata are publicly released with a lag following data collection, and country participation is staggered across survey cycles [ 44 ]; combined with school closures during the COVID-19 pandemic (2020–2021), this means the 158 datasets spanning 2003–2017 used in the present study represent the most comprehensive publicly available GSHS resource at the time of analysis. The Network Approach Traditional risk factor research adopts a variable-centered perspective, treating each predictor as an independent contributor. However, the comorbidity and multifactorial nature of mental health problems suggest that interactions among risk factors—rather than isolated effects of individual factors—may better capture pathological mechanisms [ 6 ]. Psychometric network analysis models symptoms and risk factors as network nodes and their conditional dependencies as edges. Central nodes—variables with the greatest influence on overall network structure—and cross-domain bridge nodes that connect different symptom clusters can then be identified [ 6 – 9 ]. This approach is now widely applied in clinical psychology [ 7 , 9 ] but remains scarce in large-scale cross-cultural studies of adolescent suicide. Gaps in the Literature Existing cross-cultural analyses using GSHS data leave three questions unanswered. First, although large-scale cross-cultural descriptions exist—Campisi et al. [ 10 ] conducted a pooled analysis of GSHS data from 90 countries and compared suicide behavior prevalence by income group, finding no significant differences—few studies have tested national-level moderating effects. Abio et al. [ 11 ] tested interactions between macro-level indicators (GDP, Gini coefficient, population density) and individual risk factors using multilevel logistic regression in GSHS data from 53 LMICs (with interaction tests conducted in a 33-country analytic subset), but none of the interaction terms reached significance, and urbanization rate was not among the indicators tested. A wider cross-national range may be needed to detect these associations, and alternative analytic approaches may complement the multilevel interaction framework. Second, Hu et al. [ 12 ] applied country-specific logistic regressions followed by random-effects meta-analysis to GSHS data from 93 countries, establishing that the loneliness–suicidal behavior association is present in the vast majority of countries (overall OR = 2.41 for suicidal ideation). This confirmation of cross-cultural universality provides a foundation for a complementary question: given that the association exists nearly everywhere, what macro-level conditions are associated with cross-national variation in its strength? Third, Freije et al. [ 13 ] tracked temporal trends in adolescent loneliness across 38 OECD countries but did not examine urbanization as an explanatory variable. Building on this established universality [ 12 ], the present study took a different analytical approach—psychometric network analysis—for two reasons. First, a network framework models all risk and protective factors as simultaneously interacting nodes rather than separating them into a single predictor and a set of covariates; this makes it possible to examine the structural role of loneliness (e.g., its centrality, its bridge position between protective and risk domains, and the specific edges connecting it to other nodes) within the broader system. Second, once these structural features are quantified at the country level, they can be linked to macro-level indicators to ask not just whether loneliness is associated with suicidal behavior, but under what societal conditions loneliness plays a stronger or weaker role in the network—a question that requires edge-level and node-level variation across countries as input. The Present Research This study adopted a four-stage progressive design. All four studies analyzed the same GSHS dataset from complementary analytical perspectives—node-level (Stage 1), model-level (Stage 2), macro-micro ecological (Stage 3), and edge-level (Stage 4)—and their consistency reflects analytic convergence, not independent replication (see Discussion). The pipeline was developed iteratively; the focus on loneliness and urbanization emerged from data-driven screening in Stages 2–3 rather than from prior hypotheses (see Declarations: Preregistration). Stage 1 estimated global and regional psychometric networks and tested the cross-cultural consistency of central nodes. Stage 2 tested the cultural heterogeneity of prediction pathways through cross-regional model transfer experiments and identified sources of variation. Stage 3 tested associations between macro-indicators such as urbanization and the risk role of loneliness at the country level ( N = 42), with robustness checks via individual-level multilevel modeling and specification curve analysis. Stage 4 tested associations between urbanization rate and specific network edge weights across 38 countries. Stage 1: Global Risk Network and Cross-Cultural Consistency Methods This study is reported following STROBE guidelines for cross-sectional studies [ 14 ]. A completed STROBE checklist is provided in the Supplemental Materials. Participants and Data Data came from the WHO Global School-based Student Health Survey (GSHS), which employs two-stage cluster sampling to select nationally representative schools and classes, covering in-school adolescents aged 13–17. We downloaded all 158 available datasets (100% acquisition rate; survey years spanning 2003–2017) from the WHO public database, yielding a combined sample of 439,026 adolescents from 87 countries across five WHO regions (Americas, Africa, Eastern Mediterranean, South-East Asia, Western Pacific). The GSHS does not cover the European region (European adolescent health data are collected separately by the WHO-Europe Health Behaviour in School-Aged Children [HBSC] survey), and thus no European countries were included. This dataset was shared across all four stages; subsequent stages describe only their stage-specific analytic samples and methods. Of the total sample, 52.1% were female and 46.9% male (0.9% missing); the modal age categories were 14 (22.7%) and 15 (22.0%), with 88.8% in the 13–17 target range. Overall prevalence rates were 17.7% for suicidal ideation, 14.9% for suicide plan, and 20.9% for suicide attempt (at least once). Regional sample sizes ranged from 50,479 (Eastern Mediterranean, 12 countries) to 160,457 (Americas, 29 countries); complete country-level demographic information is provided in Table S1 . Measures Network analysis used 15 node variables: three suicidal behavior variables (suicidal ideation, suicide plan, suicide attempt), three mental health variables (loneliness, worry-induced insomnia, number of close friends), four protective factors (peer support, parental understanding, parental monitoring, parental homework involvement), and five risk factors (hunger, physical attack, physical fight, insufficient physical activity, truancy). Suicidal ideation and suicide plan were binary variables (yes/no); the remaining 13 variables were ordinal (ranging from 2 to 8 categories; e.g., loneliness ranged from “never” to “always” on a 5-point scale; physical attack ranged from 0 to 12 + times on an 8-point scale). All variables were entered in their original ordinal coding into a polychoric correlation matrix, which appropriately handles mixtures of ordinal and binary data. Close friends is both a mental health indicator and a protective factor; the GSHS questionnaire groups it with psychosocial distress items (loneliness, worry-insomnia, close friends) in the Mental Health Core module rather than with family/peer support items, so it was classified accordingly, though its association with loneliness was interpreted as a protective connection in the results. Complete cases (no missing data across all 15 variables) totaled 292,280 individuals. Network Estimation and Comparison The global network was estimated using the EBICglasso method ( N = 292,280 [ 15 ]) based on a polychoric correlation matrix. The EBIC hyperparameter was set to gamma = 0.5 to promote sparse solutions. Because the 15 variables were drawn from distinct GSHS modules (suicidal behavior, psychosocial distress, family environment, violence, and health behavior), the Goldbricker test [ 35 ] confirmed no topologically overlapping node pairs, indicating that all variables contributed unique information to the network. Subnetworks were estimated separately for each of the five WHO regions. The centrality metric was node Strength. Between-region network comparisons used the Network Comparison Test (NCT [ 16 ]). For each pairwise comparison, 5,000 individuals were randomly sampled from each region (the computational complexity of NCT’s permutation test is O( n ^2 x iterations ), making full-sample analysis infeasible; n = 5,000 ensures sufficient convergence of EBICglasso partial correlation estimates). To assess sensitivity to the random subsample, we repeated the procedure across three independent draws; the pattern of significant comparisons was identical across draws. Global strength differences were assessed via 1,000 permutation tests, with p -values corrected using the Benjamini-Hochberg FDR procedure (10 pairwise comparisons [ 17 ]). Bootstrap stability analysis (2,500 resamples) computed the CS-coefficient to assess the stability of centrality rankings. All analyses were conducted in R (version 4.3) using the bootnet (v1.5), qgraph (v1.9), and NetworkComparisonTest (v2.2.2) packages. Results The EBICglasso network (15 nodes, N = 292,280 complete cases; Fig. 1 ) was consistent with the loneliness–suicidal behavior association established by Hu et al. [ 12 ]: descriptive prevalence rates closely matched their weighted estimates (e.g., frequent loneliness: 13.5% vs. 13.2%; regional rank order identical). The network approach complemented this established association by revealing the multivariate conditional dependency structure among all 15 variables simultaneously. Suicidal ideation had the highest Strength centrality (1.257), followed by suicide attempt (1.091) and suicide plan (1.005). Loneliness ranked sixth (Strength = 0.831) but occupied a bridge-like position connecting the mental health and protective factor domains (cf. [ 18 ]). The strongest edge was suicidal ideation–suicide plan (weight = 0.588), followed by parental monitoring–parental homework involvement (0.370) and loneliness–worry-induced insomnia (0.362). The network contained 31 negative edges, primarily between protective factors (e.g., close friends, peer support) and risk/distress nodes, revealing a signed dependency structure across risk and protective domains. Expected Influence (EI) rankings were consistent with Strength rankings ( r _s = .789, p < .001; top three identical), as expected given that both indices derive from the same adjacency matrix; this convergence confirms the dominance of positive edges in the network. Node predictability (mixed graphical models [ 36 ]) yielded mean R² = .19; suicidal ideation was highest (.35) and physical activity lowest (.04). Full indices are in Table S2 . Figure 1 Global adolescent mental health risk network (EBICglasso, N = 292,280, 87 countries). Vermillion = mental health, bluish green = protective factors, sky blue = risk factors. Node size proportional to Strength centrality; edge width proportional to partial correlation weight (blue = positive, vermillion dashed = negative). Abbreviations: SI = suicidal ideation, SP = suicide plan, SA = suicide attempt, Lon = loneliness, Wry = worry-insomnia, Frd = close friends, PeS = peer support, PaU = parental understanding, PaM = parental monitoring, PaH = parental homework, Hng = hunger, Atk = physical attack, Fgt = physical fight, PhA = physical activity, Trc = truancy. Suicidal ideation ranked in the top two by Strength across all five regional subnetworks (range: 1.09–1.32; first in the Americas and Eastern Mediterranean; second elsewhere). NCT indicated partial regional differentiation: 4 of 10 pairwise comparisons reached significance after FDR correction (full results in Table S4). Loneliness Strength varied modestly across regions (Americas: 0.992; Eastern Mediterranean: 0.969; South-East Asia: 0.722; Africa: 0.705; Western Pacific: 0.715), but Stage 2 would show more pronounced variation through SHAP importance. Bootstrap stability analysis confirmed reliable centrality rankings (CS-coefficient = 0.75, above the 0.50 threshold; see Supplemental Fig. S2 ). Stage 2: Cross-Regional Model Transfer Method Using the same GSHS dataset, we trained XGBoost binary classifiers (12 non-suicide features predicting suicidal ideation) separately within each WHO region. Model hyperparameters were set to n _estimators = 300, max_depth = 6, learning_rate = 0.1, subsample = 0.8, and colsample_bytree = 0.8. Class imbalance was addressed via scale_pos_weight = n _negative/ n _positive (overall ratio approximately 4.65, given SI prevalence of 17.7%; computed separately for each region). Model performance was evaluated via stratified 5-fold cross-validation within each region and cross-region transfer, with AUC as the primary performance metric (F1-scores showed a consistent degradation pattern and are reported in Table S5). A 5 x 5 transfer matrix (training region x test region) was constructed. For each country with N > = 2,000 (42 countries), we used SHapley Additive exPlanations (SHAP [ 19 ]) to extract 12-dimensional feature importance vectors and computed cross-regional coefficients of variation to identify the most culturally sensitive predictors. Because SHAP values sum to the model’s prediction for each observation, an increase in one feature’s SHAP importance mechanically reduces the remaining features’ share; the ecological correlations in Stage 3 should be interpreted with this compositional constraint in mind (see Limitations). Analyses were conducted in Python (version 3.10) using the xgboost (v1.7) and shap (v0.42) libraries. Results Models failed to transfer: within-region mean AUC was 0.868, cross-region mean AUC was 0.695, representing an average degradation of 17.3 percentage points (Fig. 2 ). The South-East Asia model performed best within its own region (AUC = 0.921) but dropped to 0.625 when transferred to Africa. Figure 2 Cross-regional model transfer matrix. Diagonal = within-region AUC (mean = 0.868); off-diagonal = cross-region AUC (mean = 0.695). Mean degradation of 17.3 percentage points. Amer = Americas, Afr = Africa, EMed = Eastern Mediterranean, SEA = South-East Asia, WPac = Western Pacific. To localize the sources of non-transferability, we computed cross-regional SHAP importance coefficients of variation for each predictor. The three most variable predictors were loneliness (CV = 0.386), hunger (0.352), and parental understanding (0.346). These CV values were closely comparable (spread of 0.040), meaning Stage 2 alone could not distinguish loneliness as the primary source of variation. Stage 3 would systematically test all 12 predictors’ associations with 11 macro-level indicators (132 tests total); the selection of loneliness as the focal variable was based on the results of that systematic test rather than on the CV rankings from Stage 2. The high CV values indicate that these predictors play different predictive roles across regions, constituting the primary sources of cross-regional transfer degradation. Stage 3: Urbanization and the Risk Role of Loneliness Method Macro-Level Indicators Country-level indicators included seven World Bank variables (GDP per capita, Gini coefficient, urbanization rate, internet penetration, health expenditure as percentage of GDP, life expectancy, suicide rate) and four Hofstede cultural dimensions (individualism, power distance, uncertainty avoidance, masculinity). For each country, World Bank indicators were matched to the median survey year of that country’s GSHS data (e.g., if a country’s surveys spanned 2007–2013, the 2010 World Bank values were used); Hofstede dimensions, which are treated as stable cultural parameters, were taken from the most recent available estimates. Ecological Correlations Spearman correlations were computed between each country’s 12-dimensional SHAP vector (from Stage 2) and 11 macro-level indicators across 42 countries (sample size > = 2,000), with Benjamini-Hochberg FDR correction (132 tests). Ninety-five percent confidence intervals for the strongest correlations were estimated via percentile bootstrap (10,000 resamples). Individual-Level GLMM As a complementary analytic approach, we used a generalized linear mixed model (GLMM) to test the same association at the individual level. Because the GLMM and SHAP analyses both used the same GSHS data (at different levels of aggregation), their consistency should be interpreted as analytic convergence rather than independent sample validation [ 20 , 21 ]. We fitted a logistic random-slopes model: suicidal_ideation ~ loneliness + worry_insomnia + … + urbanization_pct + loneliness:urbanization_pct + age + sex + (1 + loneliness | country) where urbanization rate was globally z-standardized ( M = 0, SD = 1), age was grand-mean centered, and 12 individual-level binary predictors were entered in original 0/1 coding. Group-mean centering was not applied because the binary predictors have a natural zero point and the random slope for loneliness already captures between-country variation in the loneliness–SI association; centering would alter the cross-level interaction semantics without improving interpretability [ 37 ]. The loneliness x urbanization_pct interaction was the key term; only this interaction was included because the ecological screening in Stage 3 (132 FDR-corrected tests across all 12 predictors x 11 indicators) had already identified urbanization–loneliness as the strongest macro-micro association, and the GLMM served as a focused individual-level test of that specific result rather than a discovery step. N = 296,863 individuals nested within 42 countries. Model comparison used likelihood ratio tests. The GLMM was fitted in R using lme4 [ 22 ] with the bobyqa optimizer (maxfun = 200,000); the model converged without warnings. Specification Curve Analysis Specification Curve Analysis (SCA [ 23 ]) systematically varied five analytic dimensions: outcome variable (suicidal ideation/plan/attempt), feature subset (all 12/drop weakest 3/top 5/top 3), control variables (none/+age/+age + sex/+age + sex+grade), estimation method (cluster-robust logistic/LMM/GLMM), and country inclusion threshold ( N > = 2,000/1,000/500), generating 432 specifications. The direction, significance, and effect size of the loneliness x urbanization interaction were reported for each specification. Results Ecological Correlations Among 132 macro-micro associations, the urbanization–loneliness SHAP correlation was strongest ( r _s = .655, 95% CI [.41, .81], p _FDR = .0003, N = 42; Fig. 3 ). Two methodological caveats apply to interpreting these ecological correlations. First, because ecological correlations are inflated by aggregation [ 24 ], the individual-level GLMM interaction (OR = 1.096; see below) should be treated as the primary effect size estimate; the ecological r _s = .655 describes the country-level pattern but overstates the individual-level effect. Second, because SHAP values are compositional (they sum to the model’s prediction for each observation), an increase in one feature’s importance mechanically constrains others; the negative correlations reported below (e.g., close friends, hunger) may partly reflect this zero-sum property rather than genuine reductions in those features’ predictive roles. Stage 4’s edge-weight analysis, which does not share this compositional constraint, provides the critical non-compositional check (see Stage 4 Results). Seven associations survived FDR correction (Table 1 ): urbanization and internet penetration were positively correlated with loneliness and worry-insomnia SHAP importance ( r _s = .466–.655), while urbanization was negatively correlated with close friends SHAP importance ( r _s = − .545). Table 1 FDR-Significant Macro-Micro Associations (Stage 3) Macro indicator SHAP feature r _s p _FDR N Urbanization rate Loneliness + .655 .0003 42 Urbanization rate Close friends − .545 .009 42 Internet penetration Loneliness + .542 .009 42 GDP per capita Worry-insomnia + .542 .010 40 GDP per capita Loneliness + .515 .018 40 Internet penetration Worry-insomnia + .466 .036 42 Uncertainty avoidance a Loneliness + .591 .036 25 Note Spearman correlations between country-level macro indicators and SHAP feature importance for suicidal ideation prediction. Only 7 of 132 tests survived Benjamini-Hochberg FDR correction ( p < .05). All World Bank indicators: N = 40–42. a Hofstede cultural dimension; only 25 countries had data (95% CI [.22, .82]). This result should be interpreted with caution given the smaller sample and wider confidence interval. The pattern is consistent with a gradient in which survival-level threats such as hunger and violence carry greater relative weight in suicide risk prediction in low-urbanization countries, while psychosocial factors such as loneliness and worry-induced insomnia gain relative weight in higher-urbanization countries. That only 7 of 132 tests survived FDR correction indicates that the macro-level context is not associated with all risk factors’ roles but selectively linked to the specific loneliness/worry-insomnia pathway. Figure 3 Key macro-micro associations. (a) Urbanization vs. loneliness SHAP ( r _s = .655, p _FDR = .0003). (b) Urbanization vs. close friends SHAP ( r _s = − .545, p _FDR = .009). (c) Internet penetration vs. loneliness SHAP ( r _s = .542, p _FDR = .009). (d) GDP per capita vs. worry-insomnia SHAP ( r _s = .542, p _FDR = .010). Points colored by WHO region. Individual-Level GLMM To test the same association at the individual level (using the same data at a different aggregation level, not an independent sample [ 25 ]), we fitted an individual-level GLMM ( N = 296,863 in 42 countries). The interaction term yielded OR = 1.096 (95% CI [1.053, 1.140], p = 6.2 x 10^-6): in countries one standard deviation higher in urbanization, the loneliness–suicidal ideation association was approximately 9.6% stronger. The random-slopes model was superior to random-intercepts (χ² = 481.51, df = 2, p < 2.2 x 10^-16; null-model ICC = .073). Full random effects are in Table S8. Specification Curve Analysis Among 432 specification variants (3 outcome variables x 4 feature subsets x 4 covariate sets x 3 estimation methods x 3 country thresholds), the loneliness x urbanization interaction yielded p < .05 in 94.4% (408/432) and was positive in 100% (432/432; Fig. S10). The 432 specifications are not independent, so the 94.4% rate is a descriptive consistency indicator. Without the permutation-based p -curve test recommended by Simonsohn et al. [ 23 ], which was computationally prohibitive for 432 multilevel models, SCA results here serve as a sensitivity analysis rather than confirmatory inference (see Supplemental Materials for rationale). Stratified by outcome: 100% of specifications with suicidal ideation and suicide plan were significant, but only 83.3% of suicide attempt specifications reached significance, suggesting that the urbanization–loneliness interaction is more robustly associated with cognitive/ideation-stage outcomes than with behavioral outcomes. This dissociation is consistent with ideation-to-action frameworks [ 26 ], which posit that additional factors (e.g., impulsivity, capability for suicide, means availability) govern the transition from ideation to attempt, and suggests that the urbanization–loneliness pathway primarily operates at the ideation stage. The median interaction coefficient was OR = 1.106 (range: 1.031–1.198; limited to the 288 logistic model specifications; the 144 LMM specifications yielded linear coefficients on a different scale and were not included in the OR summary). Robustness Checks In addition to the GLMM and SCA reported above, we conducted several supplementary robustness checks (Fig. S11; see Supplemental Materials for full details). Partial Correlation Control. After controlling for GDP per capita and internet penetration, the partial Spearman correlation between urbanization rate and loneliness SHAP importance remained significant ( r _s,partial = .466, p = .003), a 29% attenuation from the zero-order r _s = .655. This attenuation indicates that a meaningful portion of the urbanization–loneliness association is shared with economic development and digitalization, but the residual association after partialling out these indicators suggests that urbanization captures additional variance beyond GDP and internet penetration alone. SHAP Stability. Across five random seeds, the mean Spearman correlation among countries’ SHAP vectors was .919, and the urbanization–loneliness correlation ranged from .586 to .676, consistently significant. Leave-One-Out. Sequentially removing any single country from the 42-country sample, the urbanization–loneliness SHAP correlation remained significant in all iterations. No single country drove the results. Missing Data. The suicidal ideation-specific missing rate was not correlated with urbanization rate ( r _s = − .150, p = .343). However, the overall missing rate (proportion with any of 15 variables missing) was significantly negatively correlated with urbanization rate ( r _s = − .454, p = .003), meaning lower-urbanization countries had higher overall missingness. Complete-case analysis may therefore introduce greater selection bias in lower-urbanization countries, a limitation discussed further below. Survey Weight Sensitivity. A weighted GLMM using normalized GSHS sampling weights yielded OR = 1.085 (95% CI [1.041, 1.131], p = 1.10 x 10^-4), with negligible weighted-unweighted difference (delta-OR < .005). Survey Year. Survey year was uncorrelated with urbanization ( r _s = .175, p = .269). Adding survey year as a covariate changed the interaction negligibly (delta-OR = .001). A loneliness x survey year interaction indicated that the loneliness–SI association has strengthened over time (OR_year = 1.017, p < .001), but the urbanization interaction remained unchanged. Multiple Survey Waves. Restricting to each country’s most recent wave only ( N = 216,692, 39 countries) yielded OR = 1.081 (95% CI [1.033, 1.132], p = 8.9 x 10^-4), confirming that wave pooling did not inflate the association. Because SHAP is an aggregate node-level indicator that cannot distinguish which specific edges of loneliness are affected, Stage 4 examined this question at the edge level. Stage 4: Urbanization and Network Edge Weight Association Patterns Method In 38 countries with complete 15-variable data and N > = 500 (36 overlapping with Stage 3), country-specific regularized partial correlation networks were estimated using GraphicalLassoCV (sklearn, Python), which selects the regularization parameter via cross-validation and provides greater stability for batch estimation across many countries than EBICglasso. The lower sample size threshold ( N > = 500 vs. Stage 3’s N > = 2,000) was used because edge-weight estimation is less parameter-intensive than SHAP profiling; sensitivity analysis restricting to N > = 2,000 yielded the same pattern of significant edges. Both are L1-regularized Gaussian graphical models; GraphicalLassoCV uses Pearson rather than polychoric correlations, which may attenuate absolute edge weights but preserves cross-national rankings (the focus of Stage 4) as long as attenuation is approximately uniform. Country-level prevalence rates of loneliness and suicidal ideation were not correlated with urbanization (loneliness: r _s = − .199, p = .073; SI: r _s = .038, p = .741), indicating that differential base-rate attenuation of Pearson correlations was unlikely to produce spurious cross-national associations. The Pearson-polychoric distinction affects absolute edge magnitudes but not the cross-national ranking of edge weights, which is the quantity correlated with urbanization in Stage 4. The convergence between Stage 3 (EBICglasso/polychoric, SHAP) and Stage 4 (GraphicalLassoCV/Pearson, edge weights) should be interpreted with the caveat that both the estimation method and correlation type differ across stages. We extracted 15 x 14 / 2 = 105 unique edge weights and computed Spearman correlations between each edge weight and national urbanization rate, with Benjamini-Hochberg FDR correction (105 tests). Association patterns were analyzed by edge domain classification. Results Seventeen of 105 edges (16.2%) survived FDR correction (Fig. 4 ; Table 2 ), revealing a consistent pattern. Table 2 FDR-Significant Urbanization–Edge Weight Associations (Top 9 of 17; Stage 4) Edge r _s p _FDR Direction Loneliness – Close friends − .799 < .0001 Protection weakened Loneliness – Suicidal ideation + .639 .0008 Risk strengthened Peer support – Physical attack − .611 .0013 Buffer weakened Loneliness – Parental monitoring − .609 .0013 Protection weakened Physical fight – Physical activity + .592 .0019 Strengthened Worry-insomnia – Physical fight − .585 .0020 Weakened Loneliness – Peer support − .543 .0065 Protection weakened Parental homework – Physical fight − .533 .0073 Buffer weakened Peer support – Hunger − .530 .0073 Buffer weakened Note Spearman correlations between country-specific edge weights (GraphicalLassoCV partial correlations, N = 38 countries) and national urbanization rate. Top 9 of 17 FDR-significant edges shown; full results in Table S10. Among the 17 FDR-significant edges, 12 were negative (protective/buffering edges weaker in higher-urbanization countries) and 5 were positive (risk edges stronger). Edges spanning Mental Health and Protective Factor domains showed the strongest urbanization associations (mean | r _s| = .348; 5 of 12 FDR-significant). In higher-urbanization countries, loneliness was subject to compounding structural disadvantages: a stronger direct association with suicidal ideation ( r _s = + .639) and weaker buffering from friendship ( r _s = − .799), parental monitoring ( r _s = − .609), and peer support ( r _s = − .543)—consistent with Stage 3’s SHAP findings. Because edge weights are not subject to the compositional (zero-sum) constraint that affects SHAP values, these results provide a non-compositional confirmation of the Stage 3 pattern. These associations are based on cross-national comparisons and do not support causal inference. Fig. S12 presents a network-level visualization of this rewiring pattern, with FDR-significant edges colored by direction (red = strengthened, blue = weakened in higher-urbanization countries). Figure 4 Discussion Summary of Findings Four-stage network analysis of 439,026 adolescents across 87 countries yielded associational evidence at the node level (centrality), model level (SHAP importance), ecological level (country-level correlations), and edge level (partial correlations). Suicidal ideation consistently occupied a central position in the risk network across all cultural contexts (Stage 1: global Strength rank first, top two in all regions), but specific pathways predicting suicidal ideation failed to transfer across regions (Stage 2: AUC degradation of 17.3 percentage points), with loneliness as the most variable predictor. Among 11 macro-level indicators tested, urbanization rate showed the strongest association with this variation (Stage 3: r _s = .655); an individual-level multilevel model on the same data provided directionally consistent cross-level evidence (OR = 1.096, p = 6.2 x 10^-6), with 94.4% of 432 analytic specifications yielding p < .05. At the edge level, Stage 4 found that in countries with higher urbanization rates, the loneliness–suicidal ideation edge weight was stronger ( r _s = + .639), while protective edge weights from friendship, parental monitoring, and peer support were weaker ( r _s = − .543 to − .799). All four stages analyzed the same GSHS dataset at different levels of aggregation; the consistency across analyses therefore reflects analytic convergence within a single data source, and independent replication using different adolescent health surveys (e.g., HBSC, YRBSS) is an important next step. Theoretical Contributions Hu et al. [ 12 ] confirmed the cross-cultural universality of the loneliness–suicide association (pooled OR range: 2.04–3.13). The present study asks a different question: what macro-level conditions are associated with cross-national variation in this association’s strength. Abio et al. [ 11 ] tested macro-micro interactions in a 33-country LMIC subset without significant results; the present study’s wider cross-national range (87 countries, urbanization approximately 15%–100% vs. 15%–65%) and two-stage data-driven approach (SHAP screening followed by GLMM confirmation) may account for the different conclusions. Whether the urbanization–loneliness association attenuates when restricted to LMICs remains an empirical question. Urbanization rate should be understood as a composite social-environmental variable that may alter the pathological weight of loneliness through co-occurring processes: weakening of traditional kinship networks and increasing population density without proportionate gains in meaningful social connection [ 27 , 28 , 43 ]. A recent meta-analysis of 80 studies ( N = 539,557) confirmed that urban residence is associated with higher depression prevalence, particularly in developed countries (OR = 1.30) [ 28 ], and city-level analyses have identified loneliness and eroded neighborhood cohesion as mediating pathways between urbanization and poor mental health in adults [ 43 ]. Whether these mechanisms extend to adolescent suicide risk networks has not been tested. Two complementary theoretical frameworks help specify the mechanism. First, Granovetter’s [ 38 ] distinction between strong ties —dense, emotionally intimate bonds characteristic of kinship-based communities—and weak ties —frequent but shallow contacts typical of urban social environments—offers a structural account. Strong ties function as the primary buffer against distress escalation: they sustain the reciprocal emotional support capable of interrupting the loneliness–suicidal ideation pathway. Rapid urbanization is associated with a shift from strong-tie networks toward weak-tie environments [ 27 ], which may reduce this structural buffer and leave distress with fewer sources of relational support. Second, Holt-Lunstad [ 39 ] synthesized evidence that social connection functions as an independent determinant of both mental and physical health outcomes, with perceived isolation carrying risk effects comparable in magnitude to established risk factors such as smoking. High-density urban environments generate frequent but shallow contact, which may paradoxically heighten the perceived gap between actual and desired connection—particularly for adolescents, for whom peer belonging is a central developmental task [ 40 ]. The Stage 4 findings are consistent with both frameworks: in higher-urbanization countries, the loneliness–suicidal ideation edge strengthened ( r _s = + .639) while protective edges from friendship, parental monitoring, and peer support simultaneously weakened ( r _s = − .543 to − .799), a pattern consistent with urbanization being associated with weakened social buffering infrastructure rather than simply elevated loneliness prevalence. Social media penetration is a plausible co-occurring factor [ 29 , 30 ] (though umbrella reviews of the evidence report mixed and often small effect sizes [ 31 ]) that could not be directly measured in the GSHS. After partialing out internet penetration (the closest available proxy), urbanization still predicted loneliness SHAP importance (partial r _s = .452, p = .003), retaining approximately 69% of the original effect. Moreover, 24.3% of the sample was surveyed before 2012, when social media use was minimal in most participating countries, and the GLMM interaction was stable when controlling for survey year (delta-OR < .001). These observations suggest that urbanization captures broader social-environmental restructuring beyond digital connectivity alone [ 27 , 28 ]. Stage 4 showed that a single variable can occupy systematically different structural positions depending on macro-level context: in higher-urbanization countries, loneliness simultaneously has stronger risk edges and weaker protective edges—information that purely variable-level analysis cannot capture. These cross-national differences should not be equated with urbanization “causing” network structure changes. A competing hypothesis is that cross-national variation in loneliness SHAP importance reflects greater reporting tendency in higher-urbanization countries (response bias) rather than genuine changes in network position. Two observations argue against this. First, country-level loneliness prevalence was not positively correlated with urbanization ( r _s = − .199, p = .073, N = 80), contrary to what reporting bias would predict. Second, Stage 4’s loneliness–SI partial correlation (which controls for other variables and is less sensitive to marginal distributions) was itself positively correlated with urbanization ( r _s = + .639). This competing hypothesis is not fully excluded but is inconsistent with the available evidence. The present findings also complement those of Freije et al. [ 13 ] on global trends in adolescent loneliness. Freije et al. documented temporal changes in adolescent loneliness across 38 OECD countries but did not test urbanization rate as an explanatory variable. The present study’s cross-sectional analysis, from the perspective of spatial variation (rather than temporal variation), similarly points to an association between urbanization and adolescent loneliness. Policy Implications The following implications are conditional on longitudinal confirmation of these cross-sectional patterns and should be treated as hypotheses for future testing rather than established guidelines. With that caveat, the relationship between urbanization and mental health has received meta-analytic support in adult samples [ 28 ] and recent attention in adolescent-focused urban health frameworks [ 43 ], but how it relates to adolescent suicide risk network structure remains unexplored. Should future longitudinal work corroborate these patterns, the findings would support tailoring intervention priorities to a country’s development stage instead of applying uniform protocols. In higher-urbanization settings, Stage 4 found that the loneliness–suicidal ideation edge weight is stronger while protective edges from friendship and peer support are weaker. If these patterns are causal, loneliness-focused interventions would be a logical priority—and more specifically, effective programs may need to move beyond generic resilience training toward rebuilding strong-tie opportunities (e.g., sustained cohort-based peer programs rather than one-off social skills workshops). In lower-urbanization settings, survival-level threats such as hunger and school violence appeared to carry greater weight in suicide risk prediction (Stages 2–3). Pending longitudinal evidence, resource allocation in these contexts may need to prioritize food security and school safety rather than transplanting loneliness intervention programs from high-income countries. Limitations First, all GSHS data are cross-sectional, so the four-stage pipeline yields associational evidence only. The GLMM and SHAP analyses used the same data at different levels of aggregation, constituting analytic convergence rather than independent validation [ 20 ]. Social media penetration [ 29 , 30 ] is a plausible unmeasured confounder; after controlling for internet penetration (the closest available proxy), urbanization remained significant (partial r _s = .452, p = .003), but studies with direct social media measurement are needed. Future longitudinal or within-country urban–rural designs [ 21 ] would clarify the causal structure. Second, because the GSHS does not cover the WHO European region, no European countries were included. Replication using HBSC data is a priority. Third, GSHS variables are single-item self-report measures for which cross-cultural measurement invariance cannot be formally tested [ 32 ]. The cultural meaning of “loneliness” may differ across contexts, and cross-national variation in its network role could partly reflect measurement non-equivalence [ 33 ]. The GSHS also samples only in-school adolescents; because school dropout rates are higher in lower-urbanization countries, this could inflate the cross-national gradient. Fourth, survey years span 2003–2017. This window reflects the most comprehensive publicly available GSHS microdata at the time of analysis; post-2017 data collection has been limited by staggered country participation cycles and COVID-19 school closures [ 44 ]. Whether these associations hold in more recent cohorts—particularly given the rapid rise of social media—warrants investigation as newer waves become available. Fifth, urbanization rate is a composite indicator that cannot be decomposed into sub-mechanisms in the present design, and the interaction effect size was small (OR = 1.096; E-value = 1.37, meaning an unmeasured confounder with moderate associations to both urbanization and the loneliness–SI link could explain the observed interaction). Following Rose’s [ 34 ] prevention paradox, small shifts in population-level risk distributions may translate into meaningful changes in absolute case numbers across rapidly urbanizing countries, but this argument applies most directly to within-country temporal trends rather than to the between-country cross-sectional comparisons reported here. The clinical relevance awaits evaluation in longitudinal and intervention designs. Conclusion Four-stage multi-method analysis of nearly 440,000 adolescents across 87 countries in five WHO regions indicates that although suicidal ideation universally occupies a central position in the risk network (top two in Strength across all regions), the relative role of loneliness in suicide risk prediction is systematically associated with national urbanization levels: in countries with higher urbanization rates, loneliness’s risk network connections are stronger while its protective connections are weaker. If these cross-sectional associations are confirmed in longitudinal studies, they would support matching prevention priorities to national development contexts. Declarations Funding The author received no specific funding for this work. Conflicts of Interest The author declares no conflicts of interest. Ethics Approval This study used publicly available, deidentified secondary data from the WHO GSHS public database. No ethical approval was required for secondary analysis of deidentified public data, consistent with institutional review board guidelines for exempt research. Consent to Participate Not applicable. All data were previously collected by WHO-affiliated survey teams with appropriate local ethical approvals and informed consent procedures. Data Availability All primary data are publicly available from the WHO GSHS repository (https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/global-school-based-student-health-survey). Country-level macro indicators were obtained from the World Bank Open Data platform and the Hofstede Insights database. Preregistration This study was not preregistered. The analytic pipeline was developed iteratively, with the focus on loneliness and urbanization emerging from data-driven screening in Stages 2–3. Author Contributions Qingjun Zhu: Conceptualization, Methodology, Formal Analysis, Writing – Original Draft, Visualization. References World Health Organization (2025) Suicide worldwide in 2021: global health estimates. WHO, Geneva. ISBN: 978-92-4-011006-9 Twenge JM, Joiner TE, Rogers ML, Martin GN (2018) Increases in depressive symptoms, suicide-related outcomes, and suicide rates among US adolescents after 2010 and links to increased new media screen time. Clin Psychol Sci 6:3–17. https://doi.org/10.1177/2167702617723376 Plemmons G, Hall M, Doupnik S, Gay J, Brown C, Browning W, Casey R, Freundlich K, Johnson DP, Lind C, Rehm K, Thomas S, Williams D (2018) Hospitalization for suicide ideation or attempt: 2008–2015. Pediatrics 141:e20172426. https://doi.org/10.1542/peds.2017-2426 Uddin R, Burton NW, Maple M, Khan SR, Khan A (2019) Suicidal ideation, suicide planning, and suicide attempts among adolescents in 59 low-income and middle-income countries: a population-based study. Lancet Child Adolesc Health 3:223–233. https://doi.org/10.1016/S2352-4642(18)30403-6 World Health Organization (2024) Global School-based Student Health Survey (GSHS). https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/global-school-based-student-health-survey Borsboom D (2017) A network theory of mental disorders. World Psychiatry 16:5–13. https://doi.org/10.1002/wps.20375 Briganti G, Scutari M, Epskamp S, Borsboom D, Hoekstra RHA, Golino HF, Christensen AP, Morvan Y, Ebrahimi OV, Costantini G, Heeren A, de Ron J, Bringmann LF, Huth K, Haslbeck JMB, Isvoranu AM, Marsman M, Blanken T, Gilbert A, Henry TR, Fried EI, McNally RJ (2024) Network analysis: an overview for mental health research. Int J Methods Psychiatr Res 33:e2034. https://doi.org/10.1002/mpr.2034 Epskamp S, Borsboom D, Fried EI (2018) Estimating psychological networks and their accuracy: a tutorial paper. Behav Res Methods 50:195–212. https://doi.org/10.3758/s13428-017-0862-1 McNally RJ (2021) Network analysis of psychopathology: controversies and challenges. Annu Rev Clin Psychol 17:31–53. https://doi.org/10.1146/annurev-clinpsy-081219-092850 Campisi SC, Carducci B, Akseer N, Zasowski C, Szatmari P, Bhutta ZA (2020) Suicidal behaviours among adolescents from 90 countries: a pooled analysis of the Global School-based Student Health Survey. BMC Public Health 20:1102. https://doi.org/10.1186/s12889-020-09209-z Abio A, Owusu PN, Posti JP, Bärnighausen T, Shaikh MA, Shankar V, Wilson ML (2022) Cross-national examination of adolescent suicidal behavior: a pooled and multi-level analysis of 193,484 students from 53 low- and middle-income countries. Soc Psychiatry Psychiatr Epidemiol 57:1603–1613. https://doi.org/10.1007/s00127-022-02287-x Hu W, Li B, Li X, Luo S, Xu Z, Li J, Chen W, Guo VY (2025) The distribution of frequent perceived loneliness and its association with suicidal behaviors in adolescents: a school-based study across 93 countries/territories. Eur Child Adolesc Psychiatry 34:2779–2789. https://doi.org/10.1007/s00787-025-02698-6 Freije SL, Rhew IC, Evans YN, Chan KCG, Enquobahrie DA (2025) Adolescent loneliness trends and contextual correlates across 38 countries from 2000 to 2022. J Adolesc Health 77:643–650. https://doi.org/10.1016/j.jadohealth.2025.06.017 von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP (2007) The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 370:1453–1457. https://doi.org/10.1016/S0140-6736(07)61602-X Friedman J, Hastie T, Tibshirani R (2008) Sparse inverse covariance estimation with the graphical lasso. Biostatistics 9:432–441. https://doi.org/10.1093/biostatistics/kxm045 van Borkulo CD, van Bork R, Boschloo L, Kossakowski JJ, Tio P, Schoevers RA, Borsboom D, Waldorp LJ (2023) Comparing network structures on three aspects: a permutation test. Psychol Methods 28:1273–1285. https://doi.org/10.1037/met0000476 Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B 57:289–300. https://doi.org/10.1111/j.2517-6161.1995.tb02031.x Jones PJ, Ma R, McNally RJ (2021) Bridge centrality: a network approach to understanding comorbidity. Multivar Behav Res 56:353–367. https://doi.org/10.1080/00273171.2019.1614898 Lundberg SM, Lee SI (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30:4765–4774 Lawlor DA, Tilling K, Davey Smith G (2016) Triangulation in aetiological epidemiology. Int J Epidemiol 45:1866–1886. https://doi.org/10.1093/ije/dyw314 Wakefield J (2009) Multi-level modelling, the ecologic fallacy, and hybrid study designs. Int J Epidemiol 38:330–336. https://doi.org/10.1093/ije/dyp179 Bates D, Machler M, Bolker B, Walker S (2015) Fitting linear mixed-effects models using lme4. J Stat Softw 67:1–48. https://doi.org/10.18637/jss.v067.i01 Simonsohn U, Simmons JP, Nelson LD (2020) Specification curve analysis. Nat Hum Behav 4:1208–1214. https://doi.org/10.1038/s41562-020-0912-z Robinson WS (1950) Ecological correlations and the behavior of individuals. Am Sociol Rev 15:351–357. https://doi.org/10.2307/2087176 Diez-Roux AV (2000) Multilevel analysis in public health research. Annu Rev Public Health 21:171–192. https://doi.org/10.1146/annurev.publhealth.21.1.171 Klonsky ED, May AM (2015) The three-step theory (3ST): a new theory of suicide rooted in the ideation-to-action framework. Int J Cogn Ther 8:114–129. https://doi.org/10.1521/ijct.2015.8.2.114 Patel V, Saxena S, Lund C, Thornicroft G, Baingana F, Bolton P, Chisholm D, Collins PY, Cooper JL, Eaton J, Herrman H, Herzallah MM, Huang Y, Jordans MJD, Kleinman A, Medina-Mora ME, Morgan E, Niaz U, Omigbodun O et al (2018) The Lancet Commission on global mental health and sustainable development. Lancet 392:1553–1598. https://doi.org/10.1016/S0140-6736(18)31612-X Xu C, Miao L, Turner D, DeRubeis R (2023) Urbanicity and depression: a global meta-analysis. J Affect Disord 340:299–311. https://doi.org/10.1016/j.jad.2023.08.030 Haidt J (2024) The anxious generation: how the great rewiring of childhood is causing an epidemic of mental illness. Penguin, New York Twenge JM, Haidt J, Joiner TE, Campbell WK (2020) Underestimating digital media harm. Nat Hum Behav 4:346–348. https://doi.org/10.1038/s41562-020-0839-4 Valkenburg PM, Meier A, Beyens I (2022) Social media use and its impact on adolescent mental health: an umbrella review of the evidence. Curr Opin Psychol 44:58–68. https://doi.org/10.1016/j.copsyc.2021.08.017 Putnick DL, Bornstein MH (2016) Measurement invariance conventions and reporting: the state of the art and future directions for psychological research. Dev Rev 41:71–90. https://doi.org/10.1016/j.dr.2016.06.004 Boer D, Hanke K, He J (2018) On detecting systematic measurement error in cross-cultural research: a review and critical reflection on equivalence and invariance tests. J Cross Cult Psychol 49:713–734. https://doi.org/10.1177/0022022117749042 Rose G (1992) The strategy of preventive medicine. Oxford University Press, Oxford Jones PJ (2018) Networktools: tools for identifying important nodes in networks. R package version 1.2.3. https://CRAN.R-project.org/package=networktools Haslbeck JMB, Waldorp LJ (2020) mgm: estimating time-varying mixed graphical models in high-dimensional data. J Stat Softw 93:1–46. https://doi.org/10.18637/jss.v093.i08 Enders CK, Tofighi D (2007) Centering predictor variables in cross-sectional multilevel models: a new look at an old issue. Psychol Methods 12:121–138. https://doi.org/10.1037/1082-989X.12.2.121 Granovetter MS (1973) The strength of weak ties. Am J Sociol 78:1360–1380. https://doi.org/10.1086/225469 Holt-Lunstad J (2024) Social connection as a critical factor for mental and physical health: evidence, trends, challenges, and future implications. World Psychiatry 23:312–332. https://doi.org/10.1002/wps.21224 Baumeister RF, Leary MR (1995) The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychol Bull 117:497–529. https://doi.org/10.1037/0033-2909.117.3.497 GBD 2021 Suicide Collaborators (2025) Global, regional, and national burden of suicide, 1990–2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Public Health 10:e189–e202. https://doi.org/10.1016/S2468-2667(25)00006-4 Yan N, Luo Y, Mackay LE, Wang Y, Wang Y, Wang Y, Shiferaw BD, Wang J, Tang J, Yan W, Wang Q, Gao X, Wang W (2024) Global patterns and trends of suicide mortality and years of life lost among adolescents and young adults from 1990 to 2021: a systematic analysis for the Global Burden of Disease Study 2021. Epidemiol Psychiatr Sci 33:e52. https://doi.org/10.1017/S2045796024000532 Collins PY, Sinha M, Concepcion T, Patton G, Way T, McCay L et al (2024) Making cities mental health friendly for adolescents and young adults. Nature 627:137–148. https://doi.org/10.1038/s41586-023-07005-4 Bischops AC, Radev ST, Köthe U, Chen S, Geldsetzer P, Sarker M et al (2023) Data resource profile: the Global School-based Student Health Survey—behavioural risk and protective factors among adolescents. Int J Epidemiol 52:e102–e109. https://doi.org/10.1093/ije/dyac208 Supplemental, Materials STROBE, Checklist Additional Declarations No competing interests reported. Supplementary Files ESM1.pdf ESM2STROBE.docx Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 17 May, 2026 Reviews received at journal 04 May, 2026 Reviews received at journal 30 Mar, 2026 Reviewers agreed at journal 26 Mar, 2026 Reviews received at journal 25 Mar, 2026 Reviewers agreed at journal 25 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers invited by journal 18 Mar, 2026 Editor assigned by journal 11 Mar, 2026 Submission checks completed at journal 11 Mar, 2026 First submitted to journal 10 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9083788","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":608816717,"identity":"204e1b4d-1094-47be-abd7-046a1055e437","order_by":0,"name":"Qingjun Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYDACZiCWYGCQY29gMADxGRuI1WLMc4BoLVCQ2EO0Fr7jvIdfWLbdSe+RSN74uYDBRnbDAeZnD/BpkTzMl2Yh2fYst0cirVh6BkOa8YYDbOYG+LQYHOYxM5BsO5y7XyLHQJqH4XDihgM8bBLEaEnnkcgx/s3D8J8oLcYPgFoSgFrMgLYcIKxFEmgLg8S5w4Y9PM/KrHkMko1nHmYzw6uF7/wZ488SZYflediTN9/mqbCT7Tve/AyvFoYDDGzSCBWgoGLGqx6shfnjB0KKRsEoGAWjYGQDAPzPRWIdRx+zAAAAAElFTkSuQmCC","orcid":"","institution":"Fudan University","correspondingAuthor":true,"prefix":"","firstName":"Qingjun","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2026-03-10 12:11:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9083788/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9083788/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105564461,"identity":"230ba298-ccff-43d6-b529-5d6eda21b009","added_by":"auto","created_at":"2026-03-27 12:49:40","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":140143,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal adolescent mental health risk network (EBICglasso, \u003cem\u003eN\u003c/em\u003e = 292,280, 87 countries). Vermillion = mental health, bluish green = protective factors, sky blue = risk factors. Node size proportional to Strength centrality; edge width proportional to partial correlation weight (blue = positive, vermillion dashed = negative). Abbreviations: SI = suicidal ideation, SP = suicide plan, SA = suicide attempt, Lon = loneliness, Wry = worry-insomnia, Frd = close friends, PeS = peer support, PaU = parental understanding, PaM = parental monitoring, PaH = parental homework, Hng = hunger, Atk = physical attack, Fgt = physical fight, PhA = physical activity, Trc = truancy.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/8050e66cacff9e238b3baf11.jpg"},{"id":105235841,"identity":"ed2d1cde-7ed5-476f-a4aa-376b21fc9d07","added_by":"auto","created_at":"2026-03-23 20:00:47","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":126784,"visible":true,"origin":"","legend":"\u003cp\u003eCross-regional model transfer matrix. Diagonal = within-region AUC (mean = 0.868); off-diagonal = cross-region AUC (mean = 0.695). Mean degradation of 17.3 percentage points. Amer = Americas, Afr = Africa, EMed = Eastern Mediterranean, SEA = South-East Asia, WPac = Western Pacific.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/9449f1bd1484e2867a162879.jpg"},{"id":105235843,"identity":"68066009-8b8d-45dd-accb-1d0a2f96cc46","added_by":"auto","created_at":"2026-03-23 20:00:47","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":129155,"visible":true,"origin":"","legend":"\u003cp\u003eKey macro-micro associations. (a) Urbanization vs. loneliness SHAP (\u003cem\u003er\u003c/em\u003e_s = .655, \u003cem\u003ep\u003c/em\u003e_FDR = .0003). (b) Urbanization vs. close friends SHAP (\u003cem\u003er\u003c/em\u003e_s = -.545, \u003cem\u003ep\u003c/em\u003e_FDR = .009). (c) Internet penetration vs. loneliness SHAP (\u003cem\u003er\u003c/em\u003e_s = .542, \u003cem\u003ep\u003c/em\u003e_FDR = .009). (d) GDP per capita vs. worry-insomnia SHAP (\u003cem\u003er\u003c/em\u003e_s = .542, \u003cem\u003ep\u003c/em\u003e_FDR = .010). Points colored by WHO region.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/0ca0a146c78dc965ca9e864b.jpg"},{"id":105235844,"identity":"f7e3cc88-cd8f-494f-a54b-018f1d6df7f3","added_by":"auto","created_at":"2026-03-23 20:00:47","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":182522,"visible":true,"origin":"","legend":"\u003cp\u003eUrbanization and network edge weight associations. (a) Volcano plot of 105 edge–urbanization correlations. (b) FDR-significant edges ranked by \u003cem\u003er\u003c/em\u003e_s with 95% CI. (c) Loneliness–SI edge weight vs. urbanization (\u003cem\u003er\u003c/em\u003e_s = .639). (d) Mean sensitivity by domain pair (error bars = SD). Blue = negative (protective weakened), vermillion = positive (risk strengthened).\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/b224e7c3f4feb6a99dcae373.jpg"},{"id":105569687,"identity":"492e176a-9f62-4cd6-81f2-d16d3e225293","added_by":"auto","created_at":"2026-03-27 13:13:04","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1375836,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/75150c25-4f99-4a9b-8414-5d46c7d59938.pdf"},{"id":105235845,"identity":"3807c485-41fc-48f4-a984-a73bb20ad721","added_by":"auto","created_at":"2026-03-23 20:00:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":5736319,"visible":true,"origin":"","legend":"","description":"","filename":"ESM1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/8567a5364ca834ef3f2bb920.pdf"},{"id":105563957,"identity":"2c0c0f9c-2e3f-4988-a2a1-6859c1df6566","added_by":"auto","created_at":"2026-03-27 12:48:18","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":25784,"visible":true,"origin":"","legend":"","description":"","filename":"ESM2STROBE.docx","url":"https://assets-eu.researchsquare.com/files/rs-9083788/v1/8ba8e6f570bd9722f34de1f9.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Urbanization and the Network Role of Loneliness in Adolescent Suicide Risk: A Four-Stage Analysis Across 87 Countries","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSuicide claimed 727,000 lives globally in 2021 and remains the third leading cause of death among 15\u0026ndash;29-year-olds [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Although the age-standardized suicide rate declined from 14.9 to 9.0 per 100,000 between 1990 and 2021 [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e], the trajectory among adolescents is heterogeneous: global suicide mortality among 10\u0026ndash;24-year-olds decreased overall, but regional and sociodemographic disparities widened substantially [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Multiple high-income countries have reported rising adolescent suicidal ideation and self-harm [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], while LMICs remain underrepresented in the evidence base [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The WHO Global School-based Student Health Survey (GSHS) is one of the largest adolescent health databases and allows large-scale cross-cultural comparison [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. GSHS microdata are publicly released with a lag following data collection, and country participation is staggered across survey cycles [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]; combined with school closures during the COVID-19 pandemic (2020\u0026ndash;2021), this means the 158 datasets spanning 2003\u0026ndash;2017 used in the present study represent the most comprehensive publicly available GSHS resource at the time of analysis.\u003c/p\u003e \u003cp\u003eThe Network Approach\u003c/p\u003e \u003cp\u003eTraditional risk factor research adopts a variable-centered perspective, treating each predictor as an independent contributor. However, the comorbidity and multifactorial nature of mental health problems suggest that interactions among risk factors\u0026mdash;rather than isolated effects of individual factors\u0026mdash;may better capture pathological mechanisms [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Psychometric network analysis models symptoms and risk factors as network nodes and their conditional dependencies as edges. Central nodes\u0026mdash;variables with the greatest influence on overall network structure\u0026mdash;and cross-domain bridge nodes that connect different symptom clusters can then be identified [\u003cspan additionalcitationids=\"CR7 CR8\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. This approach is now widely applied in clinical psychology [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] but remains scarce in large-scale cross-cultural studies of adolescent suicide.\u003c/p\u003e \u003cp\u003eGaps in the Literature\u003c/p\u003e \u003cp\u003eExisting cross-cultural analyses using GSHS data leave three questions unanswered. First, although large-scale cross-cultural descriptions exist\u0026mdash;Campisi et al. [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] conducted a pooled analysis of GSHS data from 90 countries and compared suicide behavior prevalence by income group, finding no significant differences\u0026mdash;few studies have tested national-level moderating effects. Abio et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] tested interactions between macro-level indicators (GDP, Gini coefficient, population density) and individual risk factors using multilevel logistic regression in GSHS data from 53 LMICs (with interaction tests conducted in a 33-country analytic subset), but none of the interaction terms reached significance, and urbanization rate was not among the indicators tested. A wider cross-national range may be needed to detect these associations, and alternative analytic approaches may complement the multilevel interaction framework.\u003c/p\u003e \u003cp\u003eSecond, Hu et al. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] applied country-specific logistic regressions followed by random-effects meta-analysis to GSHS data from 93 countries, establishing that the loneliness\u0026ndash;suicidal behavior association is present in the vast majority of countries (overall OR\u0026thinsp;=\u0026thinsp;2.41 for suicidal ideation). This confirmation of cross-cultural universality provides a foundation for a complementary question: given that the association exists nearly everywhere, what macro-level conditions are associated with cross-national variation in its strength? Third, Freije et al. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] tracked temporal trends in adolescent loneliness across 38 OECD countries but did not examine urbanization as an explanatory variable.\u003c/p\u003e \u003cp\u003eBuilding on this established universality [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], the present study took a different analytical approach\u0026mdash;psychometric network analysis\u0026mdash;for two reasons. First, a network framework models all risk and protective factors as simultaneously interacting nodes rather than separating them into a single predictor and a set of covariates; this makes it possible to examine the \u003cem\u003estructural role\u003c/em\u003e of loneliness (e.g., its centrality, its bridge position between protective and risk domains, and the specific edges connecting it to other nodes) within the broader system. Second, once these structural features are quantified at the country level, they can be linked to macro-level indicators to ask not just \u003cem\u003ewhether\u003c/em\u003e loneliness is associated with suicidal behavior, but \u003cem\u003eunder what societal conditions\u003c/em\u003e loneliness plays a stronger or weaker role in the network\u0026mdash;a question that requires edge-level and node-level variation across countries as input.\u003c/p\u003e \u003cp\u003eThe Present Research\u003c/p\u003e \u003cp\u003eThis study adopted a four-stage progressive design. All four studies analyzed the same GSHS dataset from complementary analytical perspectives\u0026mdash;node-level (Stage 1), model-level (Stage 2), macro-micro ecological (Stage 3), and edge-level (Stage 4)\u0026mdash;and their consistency reflects analytic convergence, not independent replication (see Discussion). The pipeline was developed iteratively; the focus on loneliness and urbanization emerged from data-driven screening in Stages 2\u0026ndash;3 rather than from prior hypotheses (see Declarations: Preregistration).\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eStage 1 estimated global and regional psychometric networks and tested the cross-cultural consistency of central nodes.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eStage 2 tested the cultural heterogeneity of prediction pathways through cross-regional model transfer experiments and identified sources of variation.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eStage 3 tested associations between macro-indicators such as urbanization and the risk role of loneliness at the country level (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;42), with robustness checks via individual-level multilevel modeling and specification curve analysis.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eStage 4 tested associations between urbanization rate and specific network edge weights across 38 countries.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eStage 1: Global Risk Network and Cross-Cultural Consistency\u003c/p\u003e "},{"header":"Methods","content":" \u003cp\u003eThis study is reported following STROBE guidelines for cross-sectional studies [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. A completed STROBE checklist is provided in the Supplemental Materials.\u003c/p\u003e\n\u003ch3\u003eParticipants and Data\u003c/h3\u003e\n\u003cp\u003eData came from the WHO Global School-based Student Health Survey (GSHS), which employs two-stage cluster sampling to select nationally representative schools and classes, covering in-school adolescents aged 13\u0026ndash;17. We downloaded all 158 available datasets (100% acquisition rate; survey years spanning 2003\u0026ndash;2017) from the WHO public database, yielding a combined sample of 439,026 adolescents from 87 countries across five WHO regions (Americas, Africa, Eastern Mediterranean, South-East Asia, Western Pacific). The GSHS does not cover the European region (European adolescent health data are collected separately by the WHO-Europe Health Behaviour in School-Aged Children [HBSC] survey), and thus no European countries were included. This dataset was shared across all four stages; subsequent stages describe only their stage-specific analytic samples and methods. Of the total sample, 52.1% were female and 46.9% male (0.9% missing); the modal age categories were 14 (22.7%) and 15 (22.0%), with 88.8% in the 13\u0026ndash;17 target range. Overall prevalence rates were 17.7% for suicidal ideation, 14.9% for suicide plan, and 20.9% for suicide attempt (at least once). Regional sample sizes ranged from 50,479 (Eastern Mediterranean, 12 countries) to 160,457 (Americas, 29 countries); complete country-level demographic information is provided in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eMeasures\u003c/h2\u003e \u003cp\u003eNetwork analysis used 15 node variables: three suicidal behavior variables (suicidal ideation, suicide plan, suicide attempt), three mental health variables (loneliness, worry-induced insomnia, number of close friends), four protective factors (peer support, parental understanding, parental monitoring, parental homework involvement), and five risk factors (hunger, physical attack, physical fight, insufficient physical activity, truancy). Suicidal ideation and suicide plan were binary variables (yes/no); the remaining 13 variables were ordinal (ranging from 2 to 8 categories; e.g., loneliness ranged from \u0026ldquo;never\u0026rdquo; to \u0026ldquo;always\u0026rdquo; on a 5-point scale; physical attack ranged from 0 to 12\u0026thinsp;+\u0026thinsp;times on an 8-point scale). All variables were entered in their original ordinal coding into a polychoric correlation matrix, which appropriately handles mixtures of ordinal and binary data. Close friends is both a mental health indicator and a protective factor; the GSHS questionnaire groups it with psychosocial distress items (loneliness, worry-insomnia, close friends) in the Mental Health Core module rather than with family/peer support items, so it was classified accordingly, though its association with loneliness was interpreted as a protective connection in the results. Complete cases (no missing data across all 15 variables) totaled 292,280 individuals.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eNetwork Estimation and Comparison\u003c/h3\u003e\n\u003cp\u003eThe global network was estimated using the EBICglasso method (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;292,280 [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]) based on a polychoric correlation matrix. The EBIC hyperparameter was set to gamma\u0026thinsp;=\u0026thinsp;0.5 to promote sparse solutions. Because the 15 variables were drawn from distinct GSHS modules (suicidal behavior, psychosocial distress, family environment, violence, and health behavior), the Goldbricker test [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] confirmed no topologically overlapping node pairs, indicating that all variables contributed unique information to the network. Subnetworks were estimated separately for each of the five WHO regions. The centrality metric was node Strength.\u003c/p\u003e \u003cp\u003eBetween-region network comparisons used the Network Comparison Test (NCT [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]). For each pairwise comparison, 5,000 individuals were randomly sampled from each region (the computational complexity of NCT\u0026rsquo;s permutation test is O(\u003cem\u003en\u003c/em\u003e^2 x \u003cem\u003eiterations\u003c/em\u003e), making full-sample analysis infeasible; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5,000 ensures sufficient convergence of EBICglasso partial correlation estimates). To assess sensitivity to the random subsample, we repeated the procedure across three independent draws; the pattern of significant comparisons was identical across draws. Global strength differences were assessed via 1,000 permutation tests, with \u003cem\u003ep\u003c/em\u003e-values corrected using the Benjamini-Hochberg FDR procedure (10 pairwise comparisons [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]). Bootstrap stability analysis (2,500 resamples) computed the CS-coefficient to assess the stability of centrality rankings. All analyses were conducted in R (version 4.3) using the bootnet (v1.5), qgraph (v1.9), and NetworkComparisonTest (v2.2.2) packages.\u003c/p\u003e\n\u003ch3\u003eResults \u003c/h3\u003e\n\u003cp\u003eThe EBICglasso network (15 nodes, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;292,280 complete cases; Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) was consistent with the loneliness\u0026ndash;suicidal behavior association established by Hu et al. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]: descriptive prevalence rates closely matched their weighted estimates (e.g., frequent loneliness: 13.5% vs. 13.2%; regional rank order identical). The network approach complemented this established association by revealing the multivariate conditional dependency structure among all 15 variables simultaneously.\u003c/p\u003e \u003cp\u003eSuicidal ideation had the highest Strength centrality (1.257), followed by suicide attempt (1.091) and suicide plan (1.005). Loneliness ranked sixth (Strength\u0026thinsp;=\u0026thinsp;0.831) but occupied a bridge-like position connecting the mental health and protective factor domains (cf. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]). The strongest edge was suicidal ideation\u0026ndash;suicide plan (weight\u0026thinsp;=\u0026thinsp;0.588), followed by parental monitoring\u0026ndash;parental homework involvement (0.370) and loneliness\u0026ndash;worry-induced insomnia (0.362). The network contained 31 negative edges, primarily between protective factors (e.g., close friends, peer support) and risk/distress nodes, revealing a signed dependency structure across risk and protective domains.\u003c/p\u003e \u003cp\u003eExpected Influence (EI) rankings were consistent with Strength rankings (\u003cem\u003er\u003c/em\u003e_s = .789, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001; top three identical), as expected given that both indices derive from the same adjacency matrix; this convergence confirms the dominance of positive edges in the network. Node predictability (mixed graphical models [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]) yielded mean R\u0026sup2; = .19; suicidal ideation was highest (.35) and physical activity lowest (.04). Full indices are in Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e Global adolescent mental health risk network (EBICglasso, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;292,280, 87 countries). Vermillion\u0026thinsp;=\u0026thinsp;mental health, bluish green\u0026thinsp;=\u0026thinsp;protective factors, sky blue\u0026thinsp;=\u0026thinsp;risk factors. Node size proportional to Strength centrality; edge width proportional to partial correlation weight (blue\u0026thinsp;=\u0026thinsp;positive, vermillion dashed\u0026thinsp;=\u0026thinsp;negative). Abbreviations: SI\u0026thinsp;=\u0026thinsp;suicidal ideation, SP\u0026thinsp;=\u0026thinsp;suicide plan, SA\u0026thinsp;=\u0026thinsp;suicide attempt, Lon\u0026thinsp;=\u0026thinsp;loneliness, Wry\u0026thinsp;=\u0026thinsp;worry-insomnia, Frd\u0026thinsp;=\u0026thinsp;close friends, PeS\u0026thinsp;=\u0026thinsp;peer support, PaU\u0026thinsp;=\u0026thinsp;parental understanding, PaM\u0026thinsp;=\u0026thinsp;parental monitoring, PaH\u0026thinsp;=\u0026thinsp;parental homework, Hng\u0026thinsp;=\u0026thinsp;hunger, Atk\u0026thinsp;=\u0026thinsp;physical attack, Fgt\u0026thinsp;=\u0026thinsp;physical fight, PhA\u0026thinsp;=\u0026thinsp;physical activity, Trc\u0026thinsp;=\u0026thinsp;truancy.\u003c/p\u003e \u003cp\u003eSuicidal ideation ranked in the top two by Strength across all five regional subnetworks (range: 1.09\u0026ndash;1.32; first in the Americas and Eastern Mediterranean; second elsewhere). NCT indicated partial regional differentiation: 4 of 10 pairwise comparisons reached significance after FDR correction (full results in Table S4). Loneliness Strength varied modestly across regions (Americas: 0.992; Eastern Mediterranean: 0.969; South-East Asia: 0.722; Africa: 0.705; Western Pacific: 0.715), but Stage 2 would show more pronounced variation through SHAP importance. Bootstrap stability analysis confirmed reliable centrality rankings (CS-coefficient\u0026thinsp;=\u0026thinsp;0.75, above the 0.50 threshold; see Supplemental Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eStage 2: Cross-Regional Model Transfer\u003c/p\u003e \u003cp\u003eMethod\u003c/p\u003e \u003cp\u003eUsing the same GSHS dataset, we trained XGBoost binary classifiers (12 non-suicide features predicting suicidal ideation) separately within each WHO region. Model hyperparameters were set to \u003cem\u003en\u003c/em\u003e_estimators\u0026thinsp;=\u0026thinsp;300, max_depth\u0026thinsp;=\u0026thinsp;6, learning_rate\u0026thinsp;=\u0026thinsp;0.1, subsample\u0026thinsp;=\u0026thinsp;0.8, and colsample_bytree\u0026thinsp;=\u0026thinsp;0.8. Class imbalance was addressed via scale_pos_weight\u0026thinsp;=\u0026thinsp;\u003cem\u003en\u003c/em\u003e_negative/\u003cem\u003en\u003c/em\u003e_positive (overall ratio approximately 4.65, given SI prevalence of 17.7%; computed separately for each region). Model performance was evaluated via stratified 5-fold cross-validation within each region and cross-region transfer, with AUC as the primary performance metric (F1-scores showed a consistent degradation pattern and are reported in Table S5). A 5 x 5 transfer matrix (training region x test region) was constructed.\u003c/p\u003e \u003cp\u003eFor each country with \u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;2,000 (42 countries), we used SHapley Additive exPlanations (SHAP [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]) to extract 12-dimensional feature importance vectors and computed cross-regional coefficients of variation to identify the most culturally sensitive predictors. Because SHAP values sum to the model\u0026rsquo;s prediction for each observation, an increase in one feature\u0026rsquo;s SHAP importance mechanically reduces the remaining features\u0026rsquo; share; the ecological correlations in Stage 3 should be interpreted with this compositional constraint in mind (see Limitations). Analyses were conducted in Python (version 3.10) using the xgboost (v1.7) and shap (v0.42) libraries.\u003c/p\u003e\n\u003ch3\u003eResults \u003c/h3\u003e\n\u003cp\u003eModels failed to transfer: within-region mean AUC was 0.868, cross-region mean AUC was 0.695, representing an average degradation of 17.3 percentage points (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The South-East Asia model performed best within its own region (AUC\u0026thinsp;=\u0026thinsp;0.921) but dropped to 0.625 when transferred to Africa.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Cross-regional model transfer matrix. Diagonal\u0026thinsp;=\u0026thinsp;within-region AUC (mean\u0026thinsp;=\u0026thinsp;0.868); off-diagonal\u0026thinsp;=\u0026thinsp;cross-region AUC (mean\u0026thinsp;=\u0026thinsp;0.695). Mean degradation of 17.3 percentage points. Amer\u0026thinsp;=\u0026thinsp;Americas, Afr\u0026thinsp;=\u0026thinsp;Africa, EMed\u0026thinsp;=\u0026thinsp;Eastern Mediterranean, SEA\u0026thinsp;=\u0026thinsp;South-East Asia, WPac\u0026thinsp;=\u0026thinsp;Western Pacific.\u003c/p\u003e \u003cp\u003eTo localize the sources of non-transferability, we computed cross-regional SHAP importance coefficients of variation for each predictor. The three most variable predictors were loneliness (CV\u0026thinsp;=\u0026thinsp;0.386), hunger (0.352), and parental understanding (0.346). These CV values were closely comparable (spread of 0.040), meaning Stage 2 alone could not distinguish loneliness as the primary source of variation. Stage 3 would systematically test all 12 predictors\u0026rsquo; associations with 11 macro-level indicators (132 tests total); the selection of loneliness as the focal variable was based on the results of that systematic test rather than on the CV rankings from Stage 2. The high CV values indicate that these predictors play different predictive roles across regions, constituting the primary sources of cross-regional transfer degradation.\u003c/p\u003e \u003cp\u003eStage 3: Urbanization and the Risk Role of Loneliness\u003c/p\u003e \u003cp\u003eMethod\u003c/p\u003e\n\u003ch3\u003eMacro-Level Indicators\u003c/h3\u003e\n\u003cp\u003eCountry-level indicators included seven World Bank variables (GDP per capita, Gini coefficient, urbanization rate, internet penetration, health expenditure as percentage of GDP, life expectancy, suicide rate) and four Hofstede cultural dimensions (individualism, power distance, uncertainty avoidance, masculinity). For each country, World Bank indicators were matched to the median survey year of that country\u0026rsquo;s GSHS data (e.g., if a country\u0026rsquo;s surveys spanned 2007\u0026ndash;2013, the 2010 World Bank values were used); Hofstede dimensions, which are treated as stable cultural parameters, were taken from the most recent available estimates.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eEcological Correlations\u003c/h2\u003e \u003cp\u003eSpearman correlations were computed between each country\u0026rsquo;s 12-dimensional SHAP vector (from Stage 2) and 11 macro-level indicators across 42 countries (sample size\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;2,000), with Benjamini-Hochberg FDR correction (132 tests). Ninety-five percent confidence intervals for the strongest correlations were estimated via percentile bootstrap (10,000 resamples).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eIndividual-Level GLMM\u003c/h3\u003e\n\u003cp\u003eAs a complementary analytic approach, we used a generalized linear mixed model (GLMM) to test the same association at the individual level. Because the GLMM and SHAP analyses both used the same GSHS data (at different levels of aggregation), their consistency should be interpreted as analytic convergence rather than independent sample validation [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. We fitted a logistic random-slopes model:\u003c/p\u003e \u003cp\u003esuicidal_ideation\u0026thinsp;~\u0026thinsp;loneliness\u0026thinsp;+\u0026thinsp;worry_insomnia + \u0026hellip; + urbanization_pct\u0026thinsp;+\u0026thinsp;loneliness:urbanization_pct\u0026thinsp;+\u0026thinsp;age\u0026thinsp;+\u0026thinsp;sex + (1\u0026thinsp;+\u0026thinsp;loneliness | country)\u003c/p\u003e \u003cp\u003ewhere urbanization rate was globally z-standardized (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1), age was grand-mean centered, and 12 individual-level binary predictors were entered in original 0/1 coding. Group-mean centering was not applied because the binary predictors have a natural zero point and the random slope for loneliness already captures between-country variation in the loneliness\u0026ndash;SI association; centering would alter the cross-level interaction semantics without improving interpretability [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. The loneliness x urbanization_pct interaction was the key term; only this interaction was included because the ecological screening in Stage 3 (132 FDR-corrected tests across all 12 predictors x 11 indicators) had already identified urbanization\u0026ndash;loneliness as the strongest macro-micro association, and the GLMM served as a focused individual-level test of that specific result rather than a discovery step. \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;296,863 individuals nested within 42 countries. Model comparison used likelihood ratio tests. The GLMM was fitted in R using lme4 [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] with the bobyqa optimizer (maxfun\u0026thinsp;=\u0026thinsp;200,000); the model converged without warnings.\u003c/p\u003e\n\u003ch3\u003eSpecification Curve Analysis\u003c/h3\u003e\n\u003cp\u003eSpecification Curve Analysis (SCA [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]) systematically varied five analytic dimensions: outcome variable (suicidal ideation/plan/attempt), feature subset (all 12/drop weakest 3/top 5/top 3), control variables (none/+age/+age\u0026thinsp;+\u0026thinsp;sex/+age\u0026thinsp;+\u0026thinsp;sex+grade), estimation method (cluster-robust logistic/LMM/GLMM), and country inclusion threshold (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;2,000/1,000/500), generating 432 specifications. The direction, significance, and effect size of the loneliness x urbanization interaction were reported for each specification.\u003c/p\u003e\n\u003ch3\u003eResults\u003c/h3\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eEcological Correlations\u003c/h2\u003e \u003cp\u003eAmong 132 macro-micro associations, the urbanization\u0026ndash;loneliness SHAP correlation was strongest (\u003cem\u003er\u003c/em\u003e_s = .655, 95% CI [.41, .81], \u003cem\u003ep\u003c/em\u003e_FDR = .0003, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;42; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Two methodological caveats apply to interpreting these ecological correlations. First, because ecological correlations are inflated by aggregation [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], the individual-level GLMM interaction (OR\u0026thinsp;=\u0026thinsp;1.096; see below) should be treated as the primary effect size estimate; the ecological \u003cem\u003er\u003c/em\u003e_s = .655 describes the country-level pattern but overstates the individual-level effect. Second, because SHAP values are compositional (they sum to the model\u0026rsquo;s prediction for each observation), an increase in one feature\u0026rsquo;s importance mechanically constrains others; the negative correlations reported below (e.g., close friends, hunger) may partly reflect this zero-sum property rather than genuine reductions in those features\u0026rsquo; predictive roles. Stage 4\u0026rsquo;s edge-weight analysis, which does not share this compositional constraint, provides the critical non-compositional check (see Stage 4 Results).\u003c/p\u003e \u003cp\u003eSeven associations survived FDR correction (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e): urbanization and internet penetration were positively correlated with loneliness and worry-insomnia SHAP importance (\u003cem\u003er\u003c/em\u003e_s = .466\u0026ndash;.655), while urbanization was negatively correlated with close friends SHAP importance (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.545).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e\u003c/div\u003e \u003c/caption\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFDR-Significant Macro-Micro Associations (Stage 3)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMacro indicator\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSHAP feature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003er\u003c/em\u003e_s\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e_FDR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eN\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrbanization rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLoneliness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.0003\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrbanization rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClose friends\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.545\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInternet penetration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLoneliness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.542\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGDP per capita\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWorry-insomnia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.542\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.010\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGDP per capita\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLoneliness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.515\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInternet penetration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWorry-insomnia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.466\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.036\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUncertainty avoidance\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLoneliness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;.591\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.036\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eNote\u003c/strong\u003e \u003cp\u003eSpearman correlations between country-level macro indicators and SHAP feature importance for suicidal ideation prediction. Only 7 of 132 tests survived Benjamini-Hochberg FDR correction (\u003cem\u003ep\u003c/em\u003e \u0026lt; .05). All World Bank indicators: \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;40\u0026ndash;42. \u003csup\u003ea\u003c/sup\u003e Hofstede cultural dimension; only 25 countries had data (95% CI [.22, .82]). This result should be interpreted with caution given the smaller sample and wider confidence interval.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eThe pattern is consistent with a gradient in which survival-level threats such as hunger and violence carry greater relative weight in suicide risk prediction in low-urbanization countries, while psychosocial factors such as loneliness and worry-induced insomnia gain relative weight in higher-urbanization countries. That only 7 of 132 tests survived FDR correction indicates that the macro-level context is not associated with all risk factors\u0026rsquo; roles but selectively linked to the specific loneliness/worry-insomnia pathway.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e Key macro-micro associations. (a) Urbanization vs. loneliness SHAP (\u003cem\u003er\u003c/em\u003e_s = .655, \u003cem\u003ep\u003c/em\u003e_FDR = .0003). (b) Urbanization vs. close friends SHAP (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.545, \u003cem\u003ep\u003c/em\u003e_FDR = .009). (c) Internet penetration vs. loneliness SHAP (\u003cem\u003er\u003c/em\u003e_s = .542, \u003cem\u003ep\u003c/em\u003e_FDR = .009). (d) GDP per capita vs. worry-insomnia SHAP (\u003cem\u003er\u003c/em\u003e_s = .542, \u003cem\u003ep\u003c/em\u003e_FDR = .010). Points colored by WHO region.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eIndividual-Level GLMM\u003c/h2\u003e \u003cp\u003eTo test the same association at the individual level (using the same data at a different aggregation level, not an independent sample [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]), we fitted an individual-level GLMM (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;296,863 in 42 countries). The interaction term yielded OR\u0026thinsp;=\u0026thinsp;1.096 (95% CI [1.053, 1.140], \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.2 x 10^-6): in countries one standard deviation higher in urbanization, the loneliness\u0026ndash;suicidal ideation association was approximately 9.6% stronger. The random-slopes model was superior to random-intercepts (χ\u0026sup2; = 481.51, \u003cem\u003edf\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;2.2 x 10^-16; null-model ICC = .073). Full random effects are in Table S8.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eSpecification Curve Analysis\u003c/h2\u003e \u003cp\u003eAmong 432 specification variants (3 outcome variables x 4 feature subsets x 4 covariate sets x 3 estimation methods x 3 country thresholds), the loneliness x urbanization interaction yielded \u003cem\u003ep\u003c/em\u003e \u0026lt; .05 in 94.4% (408/432) and was positive in 100% (432/432; Fig. S10).\u003c/p\u003e \u003cp\u003eThe 432 specifications are not independent, so the 94.4% rate is a descriptive consistency indicator. Without the permutation-based \u003cem\u003ep\u003c/em\u003e-curve test recommended by Simonsohn et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], which was computationally prohibitive for 432 multilevel models, SCA results here serve as a sensitivity analysis rather than confirmatory inference (see Supplemental Materials for rationale). Stratified by outcome: 100% of specifications with suicidal ideation and suicide plan were significant, but only 83.3% of suicide attempt specifications reached significance, suggesting that the urbanization\u0026ndash;loneliness interaction is more robustly associated with cognitive/ideation-stage outcomes than with behavioral outcomes. This dissociation is consistent with ideation-to-action frameworks [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], which posit that additional factors (e.g., impulsivity, capability for suicide, means availability) govern the transition from ideation to attempt, and suggests that the urbanization\u0026ndash;loneliness pathway primarily operates at the ideation stage. The median interaction coefficient was OR\u0026thinsp;=\u0026thinsp;1.106 (range: 1.031\u0026ndash;1.198; limited to the 288 logistic model specifications; the 144 LMM specifications yielded linear coefficients on a different scale and were not included in the OR summary).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eRobustness Checks\u003c/h2\u003e \u003cp\u003eIn addition to the GLMM and SCA reported above, we conducted several supplementary robustness checks (Fig. S11; see Supplemental Materials for full details).\u003c/p\u003e \u003cp\u003e \u003cem\u003ePartial Correlation Control.\u003c/em\u003e After controlling for GDP per capita and internet penetration, the partial Spearman correlation between urbanization rate and loneliness SHAP importance remained significant (\u003cem\u003er\u003c/em\u003e_s,partial = .466, \u003cem\u003ep\u003c/em\u003e = .003), a 29% attenuation from the zero-order \u003cem\u003er\u003c/em\u003e_s = .655. This attenuation indicates that a meaningful portion of the urbanization\u0026ndash;loneliness association is shared with economic development and digitalization, but the residual association after partialling out these indicators suggests that urbanization captures additional variance beyond GDP and internet penetration alone.\u003c/p\u003e \u003cp\u003e \u003cem\u003eSHAP Stability.\u003c/em\u003e Across five random seeds, the mean Spearman correlation among countries\u0026rsquo; SHAP vectors was .919, and the urbanization\u0026ndash;loneliness correlation ranged from .586 to .676, consistently significant.\u003c/p\u003e \u003cp\u003e \u003cem\u003eLeave-One-Out.\u003c/em\u003e Sequentially removing any single country from the 42-country sample, the urbanization\u0026ndash;loneliness SHAP correlation remained significant in all iterations. No single country drove the results.\u003c/p\u003e \u003cp\u003e \u003cem\u003eMissing Data.\u003c/em\u003e The suicidal ideation-specific missing rate was not correlated with urbanization rate (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.150, \u003cem\u003ep\u003c/em\u003e = .343). However, the overall missing rate (proportion with any of 15 variables missing) was significantly negatively correlated with urbanization rate (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.454, \u003cem\u003ep\u003c/em\u003e = .003), meaning lower-urbanization countries had higher overall missingness. Complete-case analysis may therefore introduce greater selection bias in lower-urbanization countries, a limitation discussed further below.\u003c/p\u003e \u003cp\u003e \u003cem\u003eSurvey Weight Sensitivity.\u003c/em\u003e A weighted GLMM using normalized GSHS sampling weights yielded OR\u0026thinsp;=\u0026thinsp;1.085 (95% CI [1.041, 1.131], \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.10 x 10^-4), with negligible weighted-unweighted difference (delta-OR \u0026lt; .005).\u003c/p\u003e \u003cp\u003e \u003cem\u003eSurvey Year.\u003c/em\u003e Survey year was uncorrelated with urbanization (\u003cem\u003er\u003c/em\u003e_s = .175, \u003cem\u003ep\u003c/em\u003e = .269). Adding survey year as a covariate changed the interaction negligibly (delta-OR = .001). A loneliness x survey year interaction indicated that the loneliness\u0026ndash;SI association has strengthened over time (OR_year\u0026thinsp;=\u0026thinsp;1.017, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001), but the urbanization interaction remained unchanged.\u003c/p\u003e \u003cp\u003e \u003cem\u003eMultiple Survey Waves.\u003c/em\u003e Restricting to each country\u0026rsquo;s most recent wave only (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;216,692, 39 countries) yielded OR\u0026thinsp;=\u0026thinsp;1.081 (95% CI [1.033, 1.132], \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.9 x 10^-4), confirming that wave pooling did not inflate the association.\u003c/p\u003e \u003cp\u003eBecause SHAP is an aggregate node-level indicator that cannot distinguish which specific edges of loneliness are affected, Stage 4 examined this question at the edge level.\u003c/p\u003e \u003cp\u003eStage 4: Urbanization and Network Edge Weight Association Patterns\u003c/p\u003e \u003cp\u003eMethod\u003c/p\u003e \u003cp\u003eIn 38 countries with complete 15-variable data and \u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;500 (36 overlapping with Stage 3), country-specific regularized partial correlation networks were estimated using GraphicalLassoCV (sklearn, Python), which selects the regularization parameter via cross-validation and provides greater stability for batch estimation across many countries than EBICglasso. The lower sample size threshold (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;500 vs. Stage 3\u0026rsquo;s \u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;2,000) was used because edge-weight estimation is less parameter-intensive than SHAP profiling; sensitivity analysis restricting to \u003cem\u003eN\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;2,000 yielded the same pattern of significant edges. Both are L1-regularized Gaussian graphical models; GraphicalLassoCV uses Pearson rather than polychoric correlations, which may attenuate absolute edge weights but preserves cross-national rankings (the focus of Stage 4) as long as attenuation is approximately uniform. Country-level prevalence rates of loneliness and suicidal ideation were not correlated with urbanization (loneliness: \u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.199, \u003cem\u003ep\u003c/em\u003e = .073; SI: \u003cem\u003er\u003c/em\u003e_s = .038, \u003cem\u003ep\u003c/em\u003e = .741), indicating that differential base-rate attenuation of Pearson correlations was unlikely to produce spurious cross-national associations. The Pearson-polychoric distinction affects absolute edge magnitudes but not the cross-national ranking of edge weights, which is the quantity correlated with urbanization in Stage 4. The convergence between Stage 3 (EBICglasso/polychoric, SHAP) and Stage 4 (GraphicalLassoCV/Pearson, edge weights) should be interpreted with the caveat that both the estimation method and correlation type differ across stages.\u003c/p\u003e \u003cp\u003eWe extracted 15 x 14 / 2\u0026thinsp;=\u0026thinsp;105 unique edge weights and computed Spearman correlations between each edge weight and national urbanization rate, with Benjamini-Hochberg FDR correction (105 tests). Association patterns were analyzed by edge domain classification.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eSeventeen of 105 edges (16.2%) survived FDR correction (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e; Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), revealing a consistent pattern.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e\u003c/div\u003e \u003c/caption\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eFDR-Significant Urbanization\u0026ndash;Edge Weight Associations (Top 9 of 17; Stage 4)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEdge\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003er\u003c/em\u003e_s\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e_FDR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDirection\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoneliness \u0026ndash; Close friends\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.799\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.0001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProtection weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoneliness \u0026ndash; Suicidal ideation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u0026thinsp;.639\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRisk strengthened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeer support \u0026ndash; Physical attack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.611\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0013\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBuffer weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoneliness \u0026ndash; Parental monitoring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.609\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0013\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProtection weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhysical fight \u0026ndash; Physical activity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u0026thinsp;.592\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStrengthened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWorry-insomnia \u0026ndash; Physical fight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.585\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWeakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoneliness \u0026ndash; Peer support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.543\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0065\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProtection weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParental homework \u0026ndash; Physical fight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.533\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0073\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBuffer weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeer support \u0026ndash; Hunger\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.530\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.0073\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBuffer weakened\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eNote\u003c/strong\u003e \u003cp\u003eSpearman correlations between country-specific edge weights (GraphicalLassoCV partial correlations, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;38 countries) and national urbanization rate. Top 9 of 17 FDR-significant edges shown; full results in Table S10.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eAmong the 17 FDR-significant edges, 12 were negative (protective/buffering edges weaker in higher-urbanization countries) and 5 were positive (risk edges stronger). Edges spanning Mental Health and Protective Factor domains showed the strongest urbanization associations (mean |\u003cem\u003er\u003c/em\u003e_s| = .348; 5 of 12 FDR-significant).\u003c/p\u003e \u003cp\u003eIn higher-urbanization countries, loneliness was subject to compounding structural disadvantages: a stronger direct association with suicidal ideation (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;+\u0026thinsp;.639) and weaker buffering from friendship (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.799), parental monitoring (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.609), and peer support (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.543)\u0026mdash;consistent with Stage 3\u0026rsquo;s SHAP findings. Because edge weights are not subject to the compositional (zero-sum) constraint that affects SHAP values, these results provide a non-compositional confirmation of the Stage 3 pattern. These associations are based on cross-national comparisons and do not support causal inference. Fig. S12 presents a network-level visualization of this rewiring pattern, with FDR-significant edges colored by direction (red\u0026thinsp;=\u0026thinsp;strengthened, blue\u0026thinsp;=\u0026thinsp;weakened in higher-urbanization countries).\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eSummary of Findings\u003c/p\u003e \u003cp\u003eFour-stage network analysis of 439,026 adolescents across 87 countries yielded associational evidence at the node level (centrality), model level (SHAP importance), ecological level (country-level correlations), and edge level (partial correlations). Suicidal ideation consistently occupied a central position in the risk network across all cultural contexts (Stage 1: global Strength rank first, top two in all regions), but specific pathways predicting suicidal ideation failed to transfer across regions (Stage 2: AUC degradation of 17.3 percentage points), with loneliness as the most variable predictor. Among 11 macro-level indicators tested, urbanization rate showed the strongest association with this variation (Stage 3: \u003cem\u003er\u003c/em\u003e_s = .655); an individual-level multilevel model on the same data provided directionally consistent cross-level evidence (OR\u0026thinsp;=\u0026thinsp;1.096, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.2 x 10^-6), with 94.4% of 432 analytic specifications yielding \u003cem\u003ep\u003c/em\u003e \u0026lt; .05. At the edge level, Stage 4 found that in countries with higher urbanization rates, the loneliness\u0026ndash;suicidal ideation edge weight was stronger (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;+\u0026thinsp;.639), while protective edge weights from friendship, parental monitoring, and peer support were weaker (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.543 to \u0026minus;\u0026thinsp;.799). All four stages analyzed the same GSHS dataset at different levels of aggregation; the consistency across analyses therefore reflects analytic convergence within a single data source, and independent replication using different adolescent health surveys (e.g., HBSC, YRBSS) is an important next step.\u003c/p\u003e \u003cp\u003eTheoretical Contributions\u003c/p\u003e \u003cp\u003eHu et al. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] confirmed the cross-cultural universality of the loneliness\u0026ndash;suicide association (pooled OR range: 2.04\u0026ndash;3.13). The present study asks a different question: what macro-level conditions are associated with cross-national variation in this association\u0026rsquo;s strength. Abio et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] tested macro-micro interactions in a 33-country LMIC subset without significant results; the present study\u0026rsquo;s wider cross-national range (87 countries, urbanization approximately 15%\u0026ndash;100% vs. 15%\u0026ndash;65%) and two-stage data-driven approach (SHAP screening followed by GLMM confirmation) may account for the different conclusions. Whether the urbanization\u0026ndash;loneliness association attenuates when restricted to LMICs remains an empirical question.\u003c/p\u003e \u003cp\u003eUrbanization rate should be understood as a composite social-environmental variable that may alter the pathological weight of loneliness through co-occurring processes: weakening of traditional kinship networks and increasing population density without proportionate gains in meaningful social connection [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. A recent meta-analysis of 80 studies (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;539,557) confirmed that urban residence is associated with higher depression prevalence, particularly in developed countries (OR\u0026thinsp;=\u0026thinsp;1.30) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], and city-level analyses have identified loneliness and eroded neighborhood cohesion as mediating pathways between urbanization and poor mental health in adults [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Whether these mechanisms extend to adolescent suicide risk networks has not been tested. Two complementary theoretical frameworks help specify the mechanism. First, Granovetter\u0026rsquo;s [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e] distinction between \u003cem\u003estrong ties\u003c/em\u003e\u0026mdash;dense, emotionally intimate bonds characteristic of kinship-based communities\u0026mdash;and \u003cem\u003eweak ties\u003c/em\u003e\u0026mdash;frequent but shallow contacts typical of urban social environments\u0026mdash;offers a structural account. Strong ties function as the primary buffer against distress escalation: they sustain the reciprocal emotional support capable of interrupting the loneliness\u0026ndash;suicidal ideation pathway. Rapid urbanization is associated with a shift from strong-tie networks toward weak-tie environments [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], which may reduce this structural buffer and leave distress with fewer sources of relational support. Second, Holt-Lunstad [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e] synthesized evidence that social connection functions as an independent determinant of both mental and physical health outcomes, with perceived isolation carrying risk effects comparable in magnitude to established risk factors such as smoking. High-density urban environments generate frequent but shallow contact, which may paradoxically heighten the perceived gap between actual and desired connection\u0026mdash;particularly for adolescents, for whom peer belonging is a central developmental task [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. The Stage 4 findings are consistent with both frameworks: in higher-urbanization countries, the loneliness\u0026ndash;suicidal ideation edge strengthened (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;+\u0026thinsp;.639) while protective edges from friendship, parental monitoring, and peer support simultaneously weakened (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.543 to \u0026minus;\u0026thinsp;.799), a pattern consistent with urbanization being associated with weakened social buffering infrastructure rather than simply elevated loneliness prevalence.\u003c/p\u003e \u003cp\u003eSocial media penetration is a plausible co-occurring factor [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] (though umbrella reviews of the evidence report mixed and often small effect sizes [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]) that could not be directly measured in the GSHS. After partialing out internet penetration (the closest available proxy), urbanization still predicted loneliness SHAP importance (partial \u003cem\u003er\u003c/em\u003e_s = .452, \u003cem\u003ep\u003c/em\u003e = .003), retaining approximately 69% of the original effect. Moreover, 24.3% of the sample was surveyed before 2012, when social media use was minimal in most participating countries, and the GLMM interaction was stable when controlling for survey year (delta-OR \u0026lt; .001). These observations suggest that urbanization captures broader social-environmental restructuring beyond digital connectivity alone [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eStage 4 showed that a single variable can occupy systematically different structural positions depending on macro-level context: in higher-urbanization countries, loneliness simultaneously has stronger risk edges and weaker protective edges\u0026mdash;information that purely variable-level analysis cannot capture. These cross-national differences should not be equated with urbanization \u0026ldquo;causing\u0026rdquo; network structure changes.\u003c/p\u003e \u003cp\u003eA competing hypothesis is that cross-national variation in loneliness SHAP importance reflects greater reporting tendency in higher-urbanization countries (response bias) rather than genuine changes in network position. Two observations argue against this. First, country-level loneliness prevalence was not positively correlated with urbanization (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.199, \u003cem\u003ep\u003c/em\u003e = .073, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;80), contrary to what reporting bias would predict. Second, Stage 4\u0026rsquo;s loneliness\u0026ndash;SI partial correlation (which controls for other variables and is less sensitive to marginal distributions) was itself positively correlated with urbanization (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;+\u0026thinsp;.639). This competing hypothesis is not fully excluded but is inconsistent with the available evidence.\u003c/p\u003e \u003cp\u003eThe present findings also complement those of Freije et al. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] on global trends in adolescent loneliness. Freije et al. documented temporal changes in adolescent loneliness across 38 OECD countries but did not test urbanization rate as an explanatory variable. The present study\u0026rsquo;s cross-sectional analysis, from the perspective of spatial variation (rather than temporal variation), similarly points to an association between urbanization and adolescent loneliness.\u003c/p\u003e \u003cp\u003ePolicy Implications\u003c/p\u003e \u003cp\u003eThe following implications are conditional on longitudinal confirmation of these cross-sectional patterns and should be treated as hypotheses for future testing rather than established guidelines.\u003c/p\u003e \u003cp\u003eWith that caveat, the relationship between urbanization and mental health has received meta-analytic support in adult samples [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and recent attention in adolescent-focused urban health frameworks [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e], but how it relates to adolescent suicide risk network structure remains unexplored. Should future longitudinal work corroborate these patterns, the findings would support tailoring intervention priorities to a country\u0026rsquo;s development stage instead of applying uniform protocols.\u003c/p\u003e \u003cp\u003eIn higher-urbanization settings, Stage 4 found that the loneliness\u0026ndash;suicidal ideation edge weight is stronger while protective edges from friendship and peer support are weaker. If these patterns are causal, loneliness-focused interventions would be a logical priority\u0026mdash;and more specifically, effective programs may need to move beyond generic resilience training toward rebuilding strong-tie opportunities (e.g., sustained cohort-based peer programs rather than one-off social skills workshops).\u003c/p\u003e \u003cp\u003eIn lower-urbanization settings, survival-level threats such as hunger and school violence appeared to carry greater weight in suicide risk prediction (Stages 2\u0026ndash;3). Pending longitudinal evidence, resource allocation in these contexts may need to prioritize food security and school safety rather than transplanting loneliness intervention programs from high-income countries.\u003c/p\u003e \u003cp\u003eLimitations\u003c/p\u003e \u003cp\u003eFirst, all GSHS data are cross-sectional, so the four-stage pipeline yields associational evidence only. The GLMM and SHAP analyses used the same data at different levels of aggregation, constituting analytic convergence rather than independent validation [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Social media penetration [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] is a plausible unmeasured confounder; after controlling for internet penetration (the closest available proxy), urbanization remained significant (partial \u003cem\u003er\u003c/em\u003e_s = .452, \u003cem\u003ep\u003c/em\u003e = .003), but studies with direct social media measurement are needed. Future longitudinal or within-country urban\u0026ndash;rural designs [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] would clarify the causal structure.\u003c/p\u003e \u003cp\u003eSecond, because the GSHS does not cover the WHO European region, no European countries were included. Replication using HBSC data is a priority.\u003c/p\u003e \u003cp\u003eThird, GSHS variables are single-item self-report measures for which cross-cultural measurement invariance cannot be formally tested [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. The cultural meaning of \u0026ldquo;loneliness\u0026rdquo; may differ across contexts, and cross-national variation in its network role could partly reflect measurement non-equivalence [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. The GSHS also samples only in-school adolescents; because school dropout rates are higher in lower-urbanization countries, this could inflate the cross-national gradient.\u003c/p\u003e \u003cp\u003eFourth, survey years span 2003\u0026ndash;2017. This window reflects the most comprehensive publicly available GSHS microdata at the time of analysis; post-2017 data collection has been limited by staggered country participation cycles and COVID-19 school closures [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Whether these associations hold in more recent cohorts\u0026mdash;particularly given the rapid rise of social media\u0026mdash;warrants investigation as newer waves become available.\u003c/p\u003e \u003cp\u003eFifth, urbanization rate is a composite indicator that cannot be decomposed into sub-mechanisms in the present design, and the interaction effect size was small (OR\u0026thinsp;=\u0026thinsp;1.096; E-value\u0026thinsp;=\u0026thinsp;1.37, meaning an unmeasured confounder with moderate associations to both urbanization and the loneliness\u0026ndash;SI link could explain the observed interaction). Following Rose\u0026rsquo;s [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] prevention paradox, small shifts in population-level risk distributions may translate into meaningful changes in absolute case numbers across rapidly urbanizing countries, but this argument applies most directly to within-country temporal trends rather than to the between-country cross-sectional comparisons reported here. The clinical relevance awaits evaluation in longitudinal and intervention designs.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eFour-stage multi-method analysis of nearly 440,000 adolescents across 87 countries in five WHO regions indicates that although suicidal ideation universally occupies a central position in the risk network (top two in Strength across all regions), the relative role of loneliness in suicide risk prediction is systematically associated with national urbanization levels: in countries with higher urbanization rates, loneliness\u0026rsquo;s risk network connections are stronger while its protective connections are weaker. If these cross-sectional associations are confirmed in longitudinal studies, they would support matching prevention priorities to national development contexts.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch3\u003eFunding\u003c/h3\u003e\n\u003cp\u003eThe author received no specific funding for this work.\u003c/p\u003e\n\u003ch3\u003eConflicts of Interest\u003c/h3\u003e\n\u003cp\u003eThe author declares no conflicts of interest.\u003c/p\u003e\n\u003ch3\u003eEthics Approval\u003c/h3\u003e\n\u003cp\u003eThis study used publicly available, deidentified secondary data from the WHO GSHS public database. No ethical approval was required for secondary analysis of deidentified public data, consistent with institutional review board guidelines for exempt research.\u003c/p\u003e\n\u003ch3\u003eConsent to Participate\u003c/h3\u003e\n\u003cp\u003eNot applicable. All data were previously collected by WHO-affiliated survey teams with appropriate local ethical approvals and informed consent procedures.\u003c/p\u003e\n\u003ch3\u003eData Availability\u003c/h3\u003e\n\u003cp\u003eAll primary data are publicly available from the WHO GSHS repository (https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/global-school-based-student-health-survey). Country-level macro indicators were obtained from the World Bank Open Data platform and the Hofstede Insights database.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003ePreregistration\u003c/h3\u003e\n\u003cp\u003eThis study was not preregistered. The analytic pipeline was developed iteratively, with the focus on loneliness and urbanization emerging from data-driven screening in Stages 2–3.\u003c/p\u003e\n\u003ch3\u003eAuthor Contributions\u003c/h3\u003e\n\u003cp\u003eQingjun Zhu: Conceptualization, Methodology, Formal Analysis, Writing – Original Draft, Visualization.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWorld Health Organization (2025) Suicide worldwide in 2021: global health estimates. WHO, Geneva. ISBN: 978-92-4-011006-9\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTwenge JM, Joiner TE, Rogers ML, Martin GN (2018) Increases in depressive symptoms, suicide-related outcomes, and suicide rates among US adolescents after 2010 and links to increased new media screen time. Clin Psychol Sci 6:3\u0026ndash;17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2167702617723376\u003c/span\u003e\u003cspan address=\"10.1177/2167702617723376\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlemmons G, Hall M, Doupnik S, Gay J, Brown C, Browning W, Casey R, Freundlich K, Johnson DP, Lind C, Rehm K, Thomas S, Williams D (2018) Hospitalization for suicide ideation or attempt: 2008\u0026ndash;2015. Pediatrics 141:e20172426. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1542/peds.2017-2426\u003c/span\u003e\u003cspan address=\"10.1542/peds.2017-2426\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUddin R, Burton NW, Maple M, Khan SR, Khan A (2019) Suicidal ideation, suicide planning, and suicide attempts among adolescents in 59 low-income and middle-income countries: a population-based study. Lancet Child Adolesc Health 3:223\u0026ndash;233. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S2352-4642(18)30403-6\u003c/span\u003e\u003cspan address=\"10.1016/S2352-4642(18)30403-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization (2024) Global School-based Student Health Survey (GSHS). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/global-school-based-student-health-survey\u003c/span\u003e\u003cspan address=\"https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/global-school-based-student-health-survey\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBorsboom D (2017) A network theory of mental disorders. World Psychiatry 16:5\u0026ndash;13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/wps.20375\u003c/span\u003e\u003cspan address=\"10.1002/wps.20375\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBriganti G, Scutari M, Epskamp S, Borsboom D, Hoekstra RHA, Golino HF, Christensen AP, Morvan Y, Ebrahimi OV, Costantini G, Heeren A, de Ron J, Bringmann LF, Huth K, Haslbeck JMB, Isvoranu AM, Marsman M, Blanken T, Gilbert A, Henry TR, Fried EI, McNally RJ (2024) Network analysis: an overview for mental health research. Int J Methods Psychiatr Res 33:e2034. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mpr.2034\u003c/span\u003e\u003cspan address=\"10.1002/mpr.2034\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEpskamp S, Borsboom D, Fried EI (2018) Estimating psychological networks and their accuracy: a tutorial paper. Behav Res Methods 50:195\u0026ndash;212. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/s13428-017-0862-1\u003c/span\u003e\u003cspan address=\"10.3758/s13428-017-0862-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcNally RJ (2021) Network analysis of psychopathology: controversies and challenges. Annu Rev Clin Psychol 17:31\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev-clinpsy-081219-092850\u003c/span\u003e\u003cspan address=\"10.1146/annurev-clinpsy-081219-092850\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCampisi SC, Carducci B, Akseer N, Zasowski C, Szatmari P, Bhutta ZA (2020) Suicidal behaviours among adolescents from 90 countries: a pooled analysis of the Global School-based Student Health Survey. BMC Public Health 20:1102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12889-020-09209-z\u003c/span\u003e\u003cspan address=\"10.1186/s12889-020-09209-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbio A, Owusu PN, Posti JP, B\u0026auml;rnighausen T, Shaikh MA, Shankar V, Wilson ML (2022) Cross-national examination of adolescent suicidal behavior: a pooled and multi-level analysis of 193,484 students from 53 low- and middle-income countries. Soc Psychiatry Psychiatr Epidemiol 57:1603\u0026ndash;1613. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00127-022-02287-x\u003c/span\u003e\u003cspan address=\"10.1007/s00127-022-02287-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu W, Li B, Li X, Luo S, Xu Z, Li J, Chen W, Guo VY (2025) The distribution of frequent perceived loneliness and its association with suicidal behaviors in adolescents: a school-based study across 93 countries/territories. Eur Child Adolesc Psychiatry 34:2779\u0026ndash;2789. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00787-025-02698-6\u003c/span\u003e\u003cspan address=\"10.1007/s00787-025-02698-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreije SL, Rhew IC, Evans YN, Chan KCG, Enquobahrie DA (2025) Adolescent loneliness trends and contextual correlates across 38 countries from 2000 to 2022. J Adolesc Health 77:643\u0026ndash;650. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jadohealth.2025.06.017\u003c/span\u003e\u003cspan address=\"10.1016/j.jadohealth.2025.06.017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP (2007) The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 370:1453\u0026ndash;1457. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0140-6736(07)61602-X\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(07)61602-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFriedman J, Hastie T, Tibshirani R (2008) Sparse inverse covariance estimation with the graphical lasso. Biostatistics 9:432\u0026ndash;441. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/biostatistics/kxm045\u003c/span\u003e\u003cspan address=\"10.1093/biostatistics/kxm045\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Borkulo CD, van Bork R, Boschloo L, Kossakowski JJ, Tio P, Schoevers RA, Borsboom D, Waldorp LJ (2023) Comparing network structures on three aspects: a permutation test. Psychol Methods 28:1273\u0026ndash;1285. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/met0000476\u003c/span\u003e\u003cspan address=\"10.1037/met0000476\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B 57:289\u0026ndash;300. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.2517-6161.1995.tb02031.x\u003c/span\u003e\u003cspan address=\"10.1111/j.2517-6161.1995.tb02031.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJones PJ, Ma R, McNally RJ (2021) Bridge centrality: a network approach to understanding comorbidity. Multivar Behav Res 56:353\u0026ndash;367. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00273171.2019.1614898\u003c/span\u003e\u003cspan address=\"10.1080/00273171.2019.1614898\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Lee SI (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30:4765\u0026ndash;4774\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLawlor DA, Tilling K, Davey Smith G (2016) Triangulation in aetiological epidemiology. Int J Epidemiol 45:1866\u0026ndash;1886. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ije/dyw314\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyw314\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWakefield J (2009) Multi-level modelling, the ecologic fallacy, and hybrid study designs. Int J Epidemiol 38:330\u0026ndash;336. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ije/dyp179\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyp179\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBates D, Machler M, Bolker B, Walker S (2015) Fitting linear mixed-effects models using lme4. J Stat Softw 67:1\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v067.i01\u003c/span\u003e\u003cspan address=\"10.18637/jss.v067.i01\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSimonsohn U, Simmons JP, Nelson LD (2020) Specification curve analysis. Nat Hum Behav 4:1208\u0026ndash;1214. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41562-020-0912-z\u003c/span\u003e\u003cspan address=\"10.1038/s41562-020-0912-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRobinson WS (1950) Ecological correlations and the behavior of individuals. Am Sociol Rev 15:351\u0026ndash;357. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/2087176\u003c/span\u003e\u003cspan address=\"10.2307/2087176\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDiez-Roux AV (2000) Multilevel analysis in public health research. Annu Rev Public Health 21:171\u0026ndash;192. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev.publhealth.21.1.171\u003c/span\u003e\u003cspan address=\"10.1146/annurev.publhealth.21.1.171\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKlonsky ED, May AM (2015) The three-step theory (3ST): a new theory of suicide rooted in the ideation-to-action framework. Int J Cogn Ther 8:114\u0026ndash;129. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1521/ijct.2015.8.2.114\u003c/span\u003e\u003cspan address=\"10.1521/ijct.2015.8.2.114\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel V, Saxena S, Lund C, Thornicroft G, Baingana F, Bolton P, Chisholm D, Collins PY, Cooper JL, Eaton J, Herrman H, Herzallah MM, Huang Y, Jordans MJD, Kleinman A, Medina-Mora ME, Morgan E, Niaz U, Omigbodun O et al (2018) The Lancet Commission on global mental health and sustainable development. Lancet 392:1553\u0026ndash;1598. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0140-6736(18)31612-X\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(18)31612-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu C, Miao L, Turner D, DeRubeis R (2023) Urbanicity and depression: a global meta-analysis. J Affect Disord 340:299\u0026ndash;311. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jad.2023.08.030\u003c/span\u003e\u003cspan address=\"10.1016/j.jad.2023.08.030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaidt J (2024) The anxious generation: how the great rewiring of childhood is causing an epidemic of mental illness. Penguin, New York\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTwenge JM, Haidt J, Joiner TE, Campbell WK (2020) Underestimating digital media harm. Nat Hum Behav 4:346\u0026ndash;348. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41562-020-0839-4\u003c/span\u003e\u003cspan address=\"10.1038/s41562-020-0839-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eValkenburg PM, Meier A, Beyens I (2022) Social media use and its impact on adolescent mental health: an umbrella review of the evidence. Curr Opin Psychol 44:58\u0026ndash;68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.copsyc.2021.08.017\u003c/span\u003e\u003cspan address=\"10.1016/j.copsyc.2021.08.017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePutnick DL, Bornstein MH (2016) Measurement invariance conventions and reporting: the state of the art and future directions for psychological research. Dev Rev 41:71\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.dr.2016.06.004\u003c/span\u003e\u003cspan address=\"10.1016/j.dr.2016.06.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoer D, Hanke K, He J (2018) On detecting systematic measurement error in cross-cultural research: a review and critical reflection on equivalence and invariance tests. J Cross Cult Psychol 49:713\u0026ndash;734. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0022022117749042\u003c/span\u003e\u003cspan address=\"10.1177/0022022117749042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRose G (1992) The strategy of preventive medicine. Oxford University Press, Oxford\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJones PJ (2018) Networktools: tools for identifying important nodes in networks. R package version 1.2.3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://CRAN.R-project.org/package=networktools\u003c/span\u003e\u003cspan address=\"https://CRAN.R-project.org/package=networktools\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaslbeck JMB, Waldorp LJ (2020) mgm: estimating time-varying mixed graphical models in high-dimensional data. J Stat Softw 93:1\u0026ndash;46. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v093.i08\u003c/span\u003e\u003cspan address=\"10.18637/jss.v093.i08\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEnders CK, Tofighi D (2007) Centering predictor variables in cross-sectional multilevel models: a new look at an old issue. Psychol Methods 12:121\u0026ndash;138. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/1082-989X.12.2.121\u003c/span\u003e\u003cspan address=\"10.1037/1082-989X.12.2.121\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGranovetter MS (1973) The strength of weak ties. Am J Sociol 78:1360\u0026ndash;1380. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1086/225469\u003c/span\u003e\u003cspan address=\"10.1086/225469\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolt-Lunstad J (2024) Social connection as a critical factor for mental and physical health: evidence, trends, challenges, and future implications. World Psychiatry 23:312\u0026ndash;332. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/wps.21224\u003c/span\u003e\u003cspan address=\"10.1002/wps.21224\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaumeister RF, Leary MR (1995) The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychol Bull 117:497\u0026ndash;529. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/0033-2909.117.3.497\u003c/span\u003e\u003cspan address=\"10.1037/0033-2909.117.3.497\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGBD 2021 Suicide Collaborators (2025) Global, regional, and national burden of suicide, 1990\u0026ndash;2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Public Health 10:e189\u0026ndash;e202. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S2468-2667(25)00006-4\u003c/span\u003e\u003cspan address=\"10.1016/S2468-2667(25)00006-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan N, Luo Y, Mackay LE, Wang Y, Wang Y, Wang Y, Shiferaw BD, Wang J, Tang J, Yan W, Wang Q, Gao X, Wang W (2024) Global patterns and trends of suicide mortality and years of life lost among adolescents and young adults from 1990 to 2021: a systematic analysis for the Global Burden of Disease Study 2021. Epidemiol Psychiatr Sci 33:e52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S2045796024000532\u003c/span\u003e\u003cspan address=\"10.1017/S2045796024000532\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins PY, Sinha M, Concepcion T, Patton G, Way T, McCay L et al (2024) Making cities mental health friendly for adolescents and young adults. Nature 627:137\u0026ndash;148. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-023-07005-4\u003c/span\u003e\u003cspan address=\"10.1038/s41586-023-07005-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBischops AC, Radev ST, K\u0026ouml;the U, Chen S, Geldsetzer P, Sarker M et al (2023) Data resource profile: the Global School-based Student Health Survey\u0026mdash;behavioural risk and protective factors among adolescents. Int J Epidemiol 52:e102\u0026ndash;e109. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ije/dyac208\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyac208\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSupplemental, Materials\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSTROBE, Checklist\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"european-child-and-adolescent-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ecap","sideBox":"Learn more about [European Child \u0026 Adolescent Psychiatry](http://link.springer.com/journal/787)","snPcode":"787","submissionUrl":"https://submission.nature.com/new-submission/787/3","title":"European Child \u0026 Adolescent Psychiatry","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"adolescent suicide, network analysis, urbanization, cross-cultural comparison, loneliness, GSHS","lastPublishedDoi":"10.21203/rs.3.rs-9083788/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9083788/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eThe loneliness\u0026ndash;suicide association has been confirmed across countries, but what macro-level conditions account for cross-national variation in its strength remains unknown.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eFour-stage analysis of 439,026 adolescents (13\u0026ndash;17 years; 52.1% female) from 87 countries in five non-European WHO regions using GSHS data. Stage 1: psychometric networks (EBICglasso) with regional centrality comparisons. Stage 2: cross-regional XGBoost transfer with SHAP feature importance. Stage 3: ecological correlations between 11 macro-indicators and SHAP profiles (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;42 countries), with GLMM and specification curve analysis. Stage 4: urbanization\u0026ndash;edge weight associations across 38 country-specific networks.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eSuicidal ideation ranked in the top two by Strength centrality across all WHO regions (Stage 1). Prediction models failed to transfer across regions (mean AUC degradation\u0026thinsp;=\u0026thinsp;17.3 percentage points), with loneliness as the most variable predictor (Stage 2). Urbanization showed the strongest association with loneliness SHAP importance (\u003cem\u003er\u003c/em\u003e_s = .655, \u003cem\u003ep\u003c/em\u003e_FDR = .0003, \u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;42); a GLMM confirmed this at the individual level (OR\u0026thinsp;=\u0026thinsp;1.096, 95% CI [1.053, 1.140], \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.2 x 10^-6), with 94.4% of 432 specification curve variants yielding \u003cem\u003ep\u003c/em\u003e \u0026lt; .05 (Stage 3). Urbanization was positively correlated with the loneliness\u0026ndash;suicidal ideation edge weight (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;+\u0026thinsp;.639, \u003cem\u003ep\u003c/em\u003e_FDR \u0026lt; .001) and negatively with the friendship\u0026ndash;loneliness edge weight (\u003cem\u003er\u003c/em\u003e_s\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;.799, \u003cem\u003ep\u003c/em\u003e_FDR \u0026lt; .0001; Stage 4).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe network role of loneliness in adolescent suicide risk varies with national urbanization level: in higher-urbanization contexts, loneliness\u0026rsquo;s risk connections strengthen while protective connections weaken. If confirmed longitudinally, these patterns would support development-stage-adapted prevention over uniform global templates.\u003c/p\u003e","manuscriptTitle":"Urbanization and the Network Role of Loneliness in Adolescent Suicide Risk: A Four-Stage Analysis Across 87 Countries","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-23 20:00:42","doi":"10.21203/rs.3.rs-9083788/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-18T03:15:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-04T07:01:58+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-30T10:20:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"198067206908108534520251084823827396317","date":"2026-03-26T13:17:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-25T21:45:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"273111492283125856707966463496101097088","date":"2026-03-25T10:25:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"120173010872429249056929092430216179518","date":"2026-03-18T20:51:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"183238763350889359634938639976567705943","date":"2026-03-18T16:28:14+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-18T16:25:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-11T05:20:01+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-11T05:19:12+00:00","index":"","fulltext":""},{"type":"submitted","content":"European Child \u0026 Adolescent Psychiatry","date":"2026-03-10T11:56:08+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"european-child-and-adolescent-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ecap","sideBox":"Learn more about [European Child \u0026 Adolescent Psychiatry](http://link.springer.com/journal/787)","snPcode":"787","submissionUrl":"https://submission.nature.com/new-submission/787/3","title":"European Child \u0026 Adolescent Psychiatry","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"19e5c3a9-8887-4ef6-8cad-04842b905d8a","owner":[],"postedDate":"March 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-18T03:15:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-04T07:01:58+00:00","index":39,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-05-18T03:24:57+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-23 20:00:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9083788","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9083788","identity":"rs-9083788","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}