Association between trajectories of the neighborhood social exposome and mental health in late adolescence: A FinnTwin12 cohort study

Journal of Affective Disorders 358 (2024) 70–78 Available online 30 April 2024 0165-0327/© 2024 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Research paper Association between trajectories of the neighborhood social exposome and mental health in late adolescence: A FinnTwin12 cohort study Zhiyang Wang a , Gabin Drouard a , Alyce M. Whipp a , Marja Heinonen-Guzejev b , Gabriele Bolte c , Jaakko Kaprio a , b , * a Institute for Molecular Medicine Finland, Helsinki Institute of Life Science, University of Helsinki, Helsinki, Finland b Department of Public Health, Clinicum, University of Helsinki, Helsinki, Finland c Department of Social Epidemiology, Institute of Public Health and Nursing Research, University of Bremen, Bremen, Germany ARTICLE INFO Keywords: Adolescence Exposome Mental health Trajectory Social inequality ABSTRACT Background: Adolescent mental health problems impose a significant burden. Exploring evolving social environments could enhance comprehension of their impact on mental health. We aimed to depict the trajectories of the neighborhood social exposome from middle to late adolescence and assess the intricate relationship between them and late adolescent mental health. Methods: Participants (n =3965) from the FinnTwin12 cohort with completed questionnaires at age 17 were used. Nine mental health measures were assessed. The social exposome comprised 28 neighborhood social indicators. Trajectories of these indicators from ages 12 to 17 were summarized via latent growth curve modeling into growth factors, including baseline intercept. Mixture effects of all growth factors were assessed through quantile-based g-computation. Repeated generalized linear regressions identified significant growth factors. Sex stratification was performed. Results: The linear-quadratic model was the most optimal trajectory model. No mixture effect was detected. Regression models showed some growth factors saliently linked to the p-factor, internalizing problems, anxiety, hyperactivity, and aggression. The majority of them were baseline intercepts. Quadratic growth factors about mother tongues correlated with anxiety among sex-combined participants and males. The linear growth factor in the proportion of households of couples without children was associated with internalizing problems in females. Limitations: We were limited to including only neighborhood-level social exposures, and the multilevel contextual exposome situation interfered with our assessment. Conclusions: Trajectories of the social neighborhood exposome modestly influenced late adolescent mental health. Tackling root causes of social inequalities through targeted programs for living conditions could improve adolescent mental health. 1. Introduction Mental health problems are escalating as a severe public health burden, with a striking increasing trend among younger individuals. Empirical studies across different countries have demonstrated that nowadays adolescents and young adults experience more mental health problems than older people (Goodwin et al., 2022; Krokstad et al., 2022), and the pandemic has exacerbated this dramatic situation (COVID-19 Mental Disorders Collaborators, 2021). As a critical developmental stage, adolescents mental health is susceptible to the influence of the external environment, of which the social environment is an important component. Multiple social determinants have been raised such as poverty and employment status (Alegría et al., 2018). In social epidemiology, lifecourse research typically considers the relationship between exposure and health in three types of risk models: critical period, accumulation, and trajectory (Glymour et al., 2014). Distinguishing the risk models helps to investigate the nature of the relationship and guide the intervention. Some studies have detected significant relationships between cross-sectional social exposures and adolescent mental health (Xue et al., 2005; Keles et al., 2020; Huang et al., 2023). Nevertheless, our previous exposome-wide association study included various social indicators at the postal code or * Corresponding author at: Institute for Molecular Medicine, University of Helsinki, PL 20 (Tukholmankatu 8), FI-00014 Helsinki, Finland. E-mail address: [email protected] (J. Kaprio). Contents lists available at ScienceDirect Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad https://doi.org/10.1016/j.jad.2024.04.096 Received 18 October 2023; Received in revised form 14 April 2024; Accepted 21 April 2024 Journal of Affective Disorders 358 (2024) 70–78 71 municipality level at three time points and did not find any significant association of these neighborhood-level exposures with depressive symptoms in late adolescence and young adulthood (Wang et al., 2023). Another study also rejected the hypothesis of the critical period between individual financial hardship in adolescence and mental health conditions in the British population (Morrissey and Kinderman, 2020). Furthermore, in the US, changes in neighborhood economic disadvantage from early childhood to adolescence significantly influenced major depressive disorder symptoms in late adolescence (Buu et al., 2009). During severe economic recession in Finland in the early 1990s, the reduction in disposable family income deteriorated children’s mental health, and there was a long-term reciprocal influence between compromised parenting and child mental health, contributing to children’s internalizing and externalizing symptoms (Solantaus et al., 2004). Therefore, we deduced that multiple risk models of social exposures’ effect on mental health may coexist. Longitudinal design with growth modeling could be suitable for assessing time-varying mechanisms (both baseline and change) underlying the effects of the social environment on mental health, such as adaptive response early in development (Glymour et al., 2014). However, previous investigations have mostly focused on single or limited numbers of social exposures. The social exposome depicts a holistic and comprehensive portrayal of the social environment with three core principles: multidimensionality, reciprocity, and timing and continuity (Gudi-Mindermann et al., 2023). However, conventional investigations cannot simultaneously estimate the complexity, integrity, or spatiotemporality of the social exposome. Several previous exposome studies targeting mental health with a range of social exposures neither employed the trajectory model nor emphasized interior complexity (Choi et al., 2020; Lin et al., 2022; Ni et al., 2020; Pries et al., 2022; van de Weijer et al., 2022). To fulfill this research gap, our study aimed to investigate the intricate relationship between trajectories of the neighborhood social exposome and mental health in late adolescence with the following objectives: 1) to characterize the trajectory of each social indicator from mid to late adolescence, 2) to assess the mixture effect of growth factors of the social exposome on mental health in late adolescence, and 3) to identify growth factors of the social exposome on mental health in late adolescence. 2. Material and methods 2.1. Participants This study was based on the longitudinal FinnTwin12 cohort, which is a population-based prospective cohort of Finnish twins born between 1983 and 1987. At baseline, 5522 twins were invited to participate, and 87 % of their families agreed to take part. Three follow-ups were conducted at age 14 (wave 2), age 17 (wave 3), and young adulthood (wave 4). At baseline and at waves 2 and 3, data collection was conducted over a 5-year period, one birth year annually, with questionnaires sent out close to the twin’s birthdays. For the purposes of the current study, only data from twins who participated in wave 3 were used. Of the 4594 twins who were invited in this wave, 4239 completed the questionnaires at a mean age of 17.6 (standard deviation (SD): 0.3) and 2191 (51.7 %) were female. An updated review of this cohort is available elsewhere (Rose et al., 2019). 2.2. Measures 2.2.1. Indicators of the neighborhood social exposome We derived social indicators at the postal code (2022’s version) level between 1987 and 2021 from Statistics Finland and geocodes of twins’ residences from birth to 2021 from the Digital and Population Data Services Agency, Finland. Then, the map provided by Esri Finland, containing 2022’s postal code information, was utilized to link the twins to social indicators for each year from ages 12 (1995–1999) to 17 (2000–2004). Age 12 was selected as the starting point to avoid the impact of the early 1990s severe economic depression in Finland (Solantaus et al., 2004). We combined several education and age-related indicators, resulting in a total of 28 social indicators. They were categorized into six domains: 1) age structure, 2) educational level, 3) unemployment, 4) religion, language, and migration, 5) household structure, and 6) household income. These social indicators represented the population structure at the postal code, which is a good proxy for describing a neighborhood (Weckroth et al., 2022). There were over 1100 postal codes used in each age in the study, and the average number of residents inside the postal codes was 1802 (Kemppainen et al., 2023). The variable name, description, and statistics of indicators are presented in Supplementary Table 1. 2.2.2. Mental health measure The self-reported Multidimensional Peer Nomination Inventory (MPNI) was used to assess twins’ mental health at age 17 (Pulkkinen et al., 1999). It is a 37-item inventory to represent a model of emotional and behavioral regulation with a 4-point scale from 0 to 3. There were three classes of mental health measures, at different levels of aggregation of items. In the first class, we used the five subscales: depression (two items), social anxiety (anxiety hereafter, three items), hyperactivity-impulsivity (hyperactivity hereafter, six items), aggressive behaviors (aggression hereafter, six items), and inattention (four items). Mean scores, which were the average across all the response items in each subscale, were calculated as outcomes, and no missing item was allowed in any subscale. In the second class, there were threedimensional measures: internalizing problems, externalizing problems, and prosocial behaviors. Internalizing problems consisted of depression and anxiety subscale items as well as an item on victimization. Externalizing problems consisted of hyperactivity, aggression, and inattention subscale items. Twelve items composed the measure of prosocial behaviors. Two missing items were allowed for these measures in the second class. The psychopathology factor (p-factor) was the third class’s measure, which was composed of two-dimensional components in the second class (internalizing problems and externalizing problems). Four missing items were allowed for the p-factor. For both second and third classes, we used mean scores of measures. Thus, there are a total of nine MPNI measures. A higher score implied more occurrences of the problem/behavior. The MPNI scale has been validated and showed predictive power for psychiatric disorder diagnoses in early adulthood (Pulkkinen, 2017; Whipp et al., 2019). 2.2.3. Covariates We included several covariates a priori in analyses: sex (male, female), zygosity (monozygotic, dizygotic, unknown), parental education, smoking (never, former, occasional, current), and study and work status (neither study nor work, only study, only work). Sex was assigned at birth. If both parents had low education (less than 12 years) or high education (12 or more years), parental education was categorized as “limited” or “high,” respectively. In cases of differing parental education levels, it was classified as “intermediate” (Huppertz et al., 2017). Then, there was a covariate used to reflect the physical environment: Normalized Difference Vegetation Index (NDVI). It describes the difference between visible and near-infrared reflectance of vegetation cover to quantify vegetation greenness (Schinasi et al., 2018). The NDVI was measured at the geocode of twins’ residences at age 17, within a 100 m buffer during the whole year. Because we emphasized contextual aspects of the social exposome at the neighborhood level in this study, we considered sex and parental education as covariates, although they are also important social determinants. 2.3. Statistical analysis 2.3.1. Latent growth curve modeling for trajectories We employed latent growth curve models to summarize trajectories Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 72 of social indicators into latent variables, which allows good correspondence between the statistical model and the theory (Preacher et al., 2008). These latent variables, commonly referred to as growth factors, were calculated with pathways influencing the social indicators at age 12, 13, 14, 15, 16, and 17. The pathway coefficients were set according to the characteristics of each growth factor. We specified four models for each social indicator at first: 1) linear model, 2) quadratic model, 3) linear-quadratic model, and 4) logarithmic model. A diagram of the linear-quadratic model is shown in Fig. 1. Bayesian information criterion (BIC), Comparative Fit Index (CFI), and standardized root mean square residual (SRMR) were used to evaluate model fitness and select the most-optimistic model (Xia and Yang, 2019). Growth factors from the most-optimistic model were then generated and standardized (i.e., mean zero and 1-unit variance) for following analyses. CFI over 0.9 and SRMR below 0.09 were considered good model-data fit (Xia and Yang, 2019). We included all twins with valid geocodes enrolled at baseline (individual twin n =5089) for larger gains in statistical power. The maximum likelihood and robust standard errors were used to allow missing values during model fitting and to control for clustering in the data due to familial relatedness, respectively. A small number of the unemployment indicators had not been collected for some postal codes for a period, and we replaced them with their counterparts at the municipality level. For example, 5.5 % of individual twins were missing information for the variable “unemp55” in all six years. The R package “lavaan” (version 0.6-15) was used (Rosseel, 2012). 2.3.2. Exposome pluralistic analysis Participants without information on mental health measures and covariates were excluded, thus 3965 individual twins were included in the final analysis. The patterns of the included and excluded participants on growth factors were assessed by univariate linear regression. A Pearson correlation matrix accounting for the clustering in the data due to familial relatedness was created. The normality of mental health measures was visually assessed through histograms. Most mental health measures differed between sexes, so sex stratification was performed in the following pluralistic analyses (Pulkkinen et al., 1999). Initially, we set an a priori hypothesis: “Every factor matters”. Quantile g-computation, leveraging the correlation between growth factors, was used to generate unbiased estimates of the overall effect of the mixture of growth factors on each mental health measure with individual factor weights in both directions (Keil et al., 2020). It is based on the conventional weighted quantile sum regression and is an efficient approach to assess mixture effects (Keil et al., 2020). Cluster effects induced by family relatedness was controlled for. The R package “qgcomp” (version 2.10.1) was used (Keil et al., 2020). When the hypothesis of the mixture effect was null, we fitted generalized linear regression models repeatedly for each growth factor with each mental health measures to select the most important growth factors. Models were adjusted for covariates, and robust standard errors were calculated to control for family relatedness. We a priori specified a P-value of 0.001 as a significant threshold for multiple testing. The R package “rexposome” (version 1.25.2) was used (Hernandez-Ferrer et al., 2019). 3. Results 3.1. Characteristics of included participants and growth factors Of the 3965 included individual twins (Table 1), the majority were female (52 %) and dizygotic twins (62 %), and 58 % of parents had a limited level of education. At age 17, 25 % of individual twins reported being current smokers and 83 % were full-time students and not working. The characteristic of mental health measures by sex are shown in Supplementary Table 2. Through univariate linear regression, females had significantly higher mean scores for the p-factor, internalizing problems, depression, anxiety, and hyperactivity, while the mean score Fig. 1. Path diagram of linear-quadratic latent growth curve modeling. Table 1 Characteristics of twins included in the pluralistic analysis. Characteristics N (%) / Mean (SD) Sex-combined (individual twin n =3965) Male (individual twin n =1898) Female (individual twin n =2067) Zygosity Monozygotic 1317 (33.2) 593 (31.2) 724 (35.0) Dizygotic 2464 (62.1) 1226 (64.6) 1238 (59.9) Unknown 184 (4.6) 79 (4.2) 105 (5.1) Parental education Limited 2290 (57.8) 1082 (57.0) 1208 (58.4) Intermediate 914 (23.1) 453 (23.9) 461 (22.3) High 761 (19.2) 363 (19.1) 398 (19.3) Smoking Never 1178 (29.7) 589 (31.0) 589 (28.5) Former 1363 (34.4) 655 (34.5) 708 (34.3) Occasional 430 (10.8) 173 (9.1) 257 (12.4) Current 994 (25.1) 481 (25.3) 513 (24.8) Study and work status Neither study nor work 144 (3.6) 54 (2.9) 90 (4.4) Only study 3275 (82.6) 1631 (85.9) 1644 (79.5) Both study and work 546 (13.8) 213 (11.2) 333 (16.1) Normalized difference vegetation index 0.5 (0.1) 0.5 (0.1) 0.5 (0.1) Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 73 for inattention was significantly higher in males. The largest correlation was between the p-factor and externalizing problems in sex-combined and sex-stratified twins (Supplementary Fig. 1). Prosocial behavior was negatively correlated with all other measures. For all 28 social indicators, the linear-quadratic model was the most optimistic and best fitted the data (Supplementary Table 3). A total of 84 growth factors were derived from the linear-quadratic model, and all showed significant variations between individuals. Most growth factors did not differ between participants included or excluded in the final analysis, suggesting a low risk of selection bias (Supplementary Table 4). We observed strong correlations within the domains of social indicators and between domains: “household structure” and “age structure”, “household income” and “unemployment”, and “household income” and “religion, language, and migration” in the analyzed neighborhood (Supplementary Fig. 2). 3.2. Mixture effect by quantile g-computation After adjusting for all covariates, we did not find any significant associations between the mixture of growth factors and any of the nine mental health measures in sex-combined twins (Table 2). After sex stratification, still, no significant association was identified (Table 2). 3.3. Associations between growth factors and mental health measures using generalized linear regression The results among sex-combined twins are presented in Fig. 2 After adjusting for covariates, the higher intercept of edu1 (proportion of unknown or lower secondary education level or lower in the neighborhood) was significantly associated with higher mean scores of internalizing problems (beta: 0.03, 95 % confidence interval (CI): 0.01, 0.04) (Fig. 2B) and anxiety (beta: 0.04, 95 % CI: 0.02, 0.06) (Fig. 2D). The quadratic growth factors of finsa (proportion with Finnish as mother tongue) (beta: -0.03, 95 % CI: −0.05, −0.01) and swe (proportion with Swedish as mother tongue) (beta: 0.03, 95 % CI: 0.01, 0.05) were significantly associated with anxiety (Fig. 2D). Furthermore, the intercept of household1 (proportion of single households) (beta: -0.02, 95 % CI: −0.04, −0.01) was significantly associated with hyperactivity (Fig. 2F). In males only, the higher intercept of household1 was associated with lower mean scores of the p-factor (beta: -0.02, 95 % CI: −0.04, −0.01) (Fig. 3A). The intercepts of edu3 (proportion of education level of bachelor or above) (beta: -0.05, 95 % CI: −0.08, −0.02) and evluth (belonging to Lutheran Church community) (beta: 0.05, 95 % CI: 0.03, 0.08) were significantly associated with anxiety (Fig. 3D). In females only, the higher linear growth factors of household3 (proportion of households of couples without children) and swe were significantly associated with smaller mean scores of internalizing problems (beta: -0.04, 95 % CI: −0.06, −0.02) (Fig. 4B) and anxiety (beta: -0.05, 95 % CI: −0.09, −0.02) (Fig. 4D), respectively. The intercepts of unemp1824 (unemployment rate between age 18 and 24) (beta: 0.04, 95 % CI: 0.02, 0.06) and umep55 (unemployment rate over age 55) (beta: 0.04, 95 % CI: 0.02, 0.06) were positively associated with aggression, but the intercept of swe was negatively associated with aggression (beta: -0.03, 95 % CI: −0.05, −0.01) (Fig. 4G). 4. Discussion In the FinnTwin12 cohort, we explored the intricate effects of trajectories of indicators of the neighborhood social exposome on a range of mental health measures in late adolescence. We did not detect any mixture effect in sex-combined or -stratified analyses. Assuming some growth factors are potentially more important, we performed a factor selection via repeated regression models. Growth factors from the domains of “educational level,” “unemployment,” “religion, language, and migration,” and “household structure” were associated with mental health in sex-combined or sex-stratified analyses, while their effect sizes were relatively small. Nonetheless, some associations, for example between the trajectory of the unemployment status of the residential area and aggression, were seen even after accounting for the effects of parental education and own study and work status. The neighborhood social exposome and its trajectory modestly affected adolescent mental health. Although most growth factors identified were intercept (baseline) factors, some factors indicating changes were also found to be related to language and household. Findings with baseline factors were consistent with previous studies that mental health and well-being were frequently linked to neighborhood deprivation including aspects of poverty, income, unemployment, and education among young people (Jivraj et al., 2020; Visser et al., 2021; Weckroth et al., 2022). Although few studies have directly bridged the trajectories of neighborhood household and mother tongue structures to mental health, these trajectories were usually correlated with the change of socioeconomic status (Anders´ en et al., 2021; Hyypp¨ a and M¨ aki, 2001; Koops et al., 2021). From a more holistic perspective, trajectories represented the changes undergone in the social environment, driving social segregation. Segregation was shown to induce regional and socioeconomic gradients and disparities in healthcare resources in Finland (Kajantie et al., 2006). A UK natural experiment showed that neighborhood-level regeneration projects, ameliorating social inequality, improved residents’ mental health, which partially supports the trajectory effect (White et al., 2017). Additionally, sex-specific effects of the trajectories of the social exposome were observed in our findings. Brazil and Clark found that the relationship between changes in mental health and neighborhood poverty was stronger in females than males (Brazil and Clark, 2017), however, there is evidence to the contrary (Barr, 2018; Humphrey and Root, 2017). This may be due to sex differences in susceptibility and the capacity to respond to a social context (Person Waye et al., 2023; Stinson, 1985). Overall, the lifecourse Table 2 Adjusted estimates for associations between the mixture of standardized growth factors of the neighborhood social indicators and mental health measures using quantile g-computation. Mental health measure Adjusted coefficient (95 % CI) a Sex-combined Male Female P-factor −0.01 (−0.21, 0.18) −0.11 (−0.39, 0.17) 0.12 (−0.16, 0.39) Internalizing problems 0.15 (−0.15, 0.46) −0.11 (−0.54, 0.32) 0.33 (−0.11, 0.77) Depression 0.13 (−0.28, 0.55) −0.23 (−0.81, 0.34) 0.37 (−0.23, 0.97) Anxiety 0.18 (−0.20, 0.57) −0.05 (−0.60, 0.49) 0.39 (−0.16, 0.94) Externalizing problems −0.07 (−0.30, 0.16) −0.11 (−0.44, 0.22) 0.04 (−0.29, 0.37) Hyperactivity −0.24 (−0.55, 0.07) −0.41 (−0.85, 0.04) 0.01 (−0.43, 0.45) Aggression −0.03 (−0.32, 0.26) 0.10 (−0.32, 0.51) 0.03 (−0.38, 0.44) Inattention 0.11 (−0.18, 0.41) 0.03 (−0.39, 0.44) 0.11 (−0.31, 0.53) Prosocial behavior 0.02 (−0.19, 0.24) −0.02 (−0.34, 0.31) 0.01 (−0.28, 0.30) a Adjusted for sex, zygosity, parental education, smoking at age 17, study and work status at age 17, and normalized difference vegetation index with in a 100 m buffer at age 17. Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 74 approach, focusing on the history of social contexts, assists us to illuminate the temporal impact of the exposome on adolescent mental health (Wheaton and Clarke, 2003). As an implication, we suggest that policymakers strive to reduce differing conditions and promote better neighborhoods from an integrated and dynamic perspective. Of note, since social indicators reflect the external social environment rather than the individual level, our interpretation primarily focuses on the neighborhood effects. Only a small number of growth factors were selected, for which effect sizes were small, and some effects’ directions (for example, a higher proportion of single households at baseline was associated with lower hyperactivity symptoms) were not straightforward to understand. Therefore, we inferred that the effects of trajectories of the social exposome on mental health were either not substantial or were “confounded” by something else. A multilevel contextual exposome situation was proposed, but only one level (neighborhood) was considered in our analyses (Hu et al., 2023). Imagining a Swedish-speaking family residing in the predominantly Finnish-speaking eastern part of Finland, notable differences between familial and neighborhood environments potentially exist in the domains of religion, language, and migration, which may reciprocally mitigate or exacerbate their effects on mental health. In Australia, second-generation immigrants from a non-English-speaking background had a higher prevalence of mental disorders than those from an Englishspeaking background (Liddell et al., 2016). Flouri et al. (2016) also pointed out an interaction between the neighborhood single-parent rate and family marital status on child emotional symptoms. Weckroth et al. (2022) found that mental health and loneliness, as covariates, attenuated the impact of living in the inner urban area and unemployment rate on the quality of life in Finland. In the traditional assessment, neighborhood-, familialor individual-level exposures mutually act like confounders, but we rather refer to this situation as a multilevel structure within the exposome. In addition, exposures used in this study were based on individual’s geocodes rather than self-reports, which again requires consideration of information granularity. A study in Switzerland suggested that perceived environmental stressors predict mental health better than objective measures among residents over 17 years (Gomm and Bernauer, 2023). In the future, it is imperative to incorporate multilevel exposures from different sources to capture all dimensions of the social exposome (Gudi-Mindermann et al., 2023) and consider their possible cascading effects on mental health to enhance our understanding of this intricate relationship. Under the social exposome framework, we need sophisticated and advanced methodologies to deepen our understanding of the relationship between the exposome and mental health. An ABCD cohort study found that exposome scores calculated with factor analysis, including neighborhood exposures, were significantly associated with adolescent p-factor scores (Pries et al., 2022). Many machine learning methods have been employed in other types of exposome analyses and outcomes (Ohanyan et al., 2022), recommending the application of pluralistic analysis. Neufcourt et al. (2023) raised the consideration of agnostics in social exposome research, towards testing all possible hypotheses without prejudice. In the context of the exposome framework that integrates a pluralistic analysis, we propose a “semi-agnostic” paradigm wherein hypotheses are formulated at the exposome level rather than on individual exposures. Thus, we made a hypothesis on the mixture effect in this study initially, which describes the co-occurrence of multiple exposures into one latent mixture, particularly from similar sources (Traini et al., 2022). The mixture effect is usually hypothesized in the Fig. 2. Volcano plots of associations, via generalized linear regression, of growth factors with p-factor (A), internalizing problems (B), depression (C), anxiety (D), externalizing problems (E), hyperactivity (F), aggression (G), inattention (H), and prosocial behavior (I) among sex-combined participants (individual twins n = 3965). The x-axis represents regression coefficients and the Y-axis represents –log 10 (P-value) of regressions. The prefix i means the intercept, the prefix s means the linear growth factor, and the prefix q means quadratic the growth factor. Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 75 field of toxicoepidemiology, and we employed it to social indicators since they are from the same source and relatively at the same level of structure (Hu et al., 2023). Although we did not identify any significant findings related to the mixture effect, we extended our investigation to identify crucial factors and explore alternative assumptions. The “semiagnostic” paradigm offers the advantage of providing both theoretical guidance for further mechanism explanation and a broader perspective to uncover unknowns. Practical application should also consider the dimensionality, exposure structure, research question, and future implications to cautiously choose paradigms. This study was strengthened by its longitudinal design with reliable residential history, enabling the accurate assessment of trajectories of indicators of the social exposome at the neighborhood level. The incorporation of the MPNI and a range of social indicators allowed us to comprehensively appraise the influence of social environmental trends on mental health in late adolescence. However, there are also some limitations. First, the lack of other dimensions of the social exposome including individual exposures may have affected our assessment. Therefore, one could consider novel models such as neural networks to disentangle the multilevel contextual exposome situation in the future. Second, we only included cross-sectional mental health measures, which were unable to assess the effect on the development of mental health. Third, as a cohort study, we did not have the counterfactual condition, so we did not imply any causal inference and suggested being more cautious on interpretation. At last, we used an arbitrary significance threshold, which may limit the power. Further studies are welcomed to verify our findings. 5. Conclusion This study explored the intricate relationship between trajectories of the neighborhood social exposome and adolescent mental health. Although we did not identify mixture effects between trajectories and mental health, a small number of neighborhood social exposures’ trajectories showed a noticeable effect on mental health in late adolescence after growth factor selection. However, the multilevel contextual exposome situation may obstruct our assessment, and interpretation should be cautious. Other studies are to be encouraged to replicate and extend our findings. Additionally, we advocate a “semi-agnostic” paradigm in social exposome research to better navigate and comprehend its inherent complexity. Funding This research was partly funded by the European Union’s Horizon 2020 research and innovation program under grant agreement No 874724 (Equal-Life). Equal-Life is part of the European Human Exposome Network. Data collection in FinnTwin12 has been supported by the National Institute on Alcohol Abuse and Alcoholism (grants AA-12502, AA-00145, and AA-09203 to Richard J. Rose) and the Academy of Finland (grants 100499, 205585, 118555, 141054, 264146, 308248, 312073, 336823, and 352792 to Jaakko Kaprio). Jaakko Kaprio acknowledges support by the Academy of Finland (grants 265240, 263278). Fig. 3. Volcano plots of associations, via generalized linear regression, of growth factors with p-factor (A), internalizing problems (B), depression (C), anxiety (D), externalizing problems (E), hyperactivity (F), aggression (G), inattention (H), and prosocial behavior (I) among males (individual twins n =1898). The x-axis represents regression coefficients and the Y-axis represents –log 10 (P-value) of regressions. The prefix i means the intercept, the prefix s means the linear growth factor, and the prefix q means quadratic the growth factor. Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 76 Ethical statement The ethics committee of the Department of Public Health of the University of Helsinki (Helsinki, Finland), the ethics committee of the Helsinki University Central Hospital District (Helsinki, Finland), and the Institutional Review Board of Indiana University (Bloomington, Indiana, USA) approved the FinnTwin12 study protocol. All participants and their parents/legal guardians gave informed written consent to participate in the study. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. CRediT authorship contribution statement Zhiyang Wang: Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft. Gabin Drouard: Conceptualization, Methodology, Writing – review & editing. Alyce M. Whipp: Conceptualization, Data curation, Methodology, Writing – review & editing. Marja Heinonen-Guzejev: Conceptualization, Methodology, Writing – review & editing. Gabriele Bolte: Conceptualization, Writing – review & editing. Jaakko Kaprio: Conceptualization, Data curation, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability The FinnTwin12 data is not publicly available due to the restrictions of informed consent. However, the FinnTwin12 data is available through the Institute for Molecular Medicine Finland (FIMM) Data Access Committee (DAC) ([email protected]) for authorized researchers who have IRB/ethics approval and an institutionally approved study plan. To ensure the protection of privacy and compliance with national data protection legislation, a data use/transfer agreement is needed, the content and specific clauses of which will depend on the nature of the requested data. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.jad.2024.04.096. References Alegría, M., NeMoyer, A., Falg` as Bagu´ e, I., Wang, Y., Alvarez, K., 2018. Social determinants of mental health: where we are and where we need to go. Curr. Psychiatry Rep. 20, 95. https://doi.org/10.1007/s11920-018-0969-9. Anders´ en, H., Kankaanranta, H., Tuomisto, L.E., Piiril¨ a, P., Sovij¨ arvi, A., Langhammer, A., Backman, H., Lundb¨ ack, B., R¨ onmark, E., Lehtim¨ aki, L., Ilmarinen, P., 2021. Multimorbidity in Finnish and Swedish speaking Finns; association with daily habits and socioeconomic status – Nordic EpiLung crosssectional study. Prev. Med. Rep. 22, 101338 https://doi.org/10.1016/j. pmedr.2021.101338. Fig. 4. Volcano plots of associations, via generalized linear regression, of growth factors with p-factor (A), internalizing problems (B), depression (C), anxiety (D), externalizing problems (E), hyperactivity (F), aggression (G), inattention (H), and prosocial behavior (I) among females (individual twins n =2067). The x-axis represents regression coefficients and the Y-axis represents –log 10 (P-value) of regressions. The prefix i means the intercept, the prefix s means the linear growth factor, and the prefix q means quadratic the growth factor. Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 77 Barr, P.B., 2018. Early neighborhood conditions and trajectories of depressive symptoms across adolescence and into adulthood. Adv. Life Course Res. 35, 57–68. https://doi. org/10.1016/j.alcr.2018.01.005. Brazil, N., Clark, W.A.V., 2017. Individual mental health, life course events and dynamic neighbourhood change during the transition to adulthood. Health Place 45, 99–109. https://doi.org/10.1016/j.healthplace.2017.03.007. Buu, A., DiPiazza, C., Wang, J., Puttler, L.I., Fitzgerald, H.E., Zucker, R.A., 2009. Parent, family, and neighborhood effects on the development of child substance use and other psychopathology from preschool to the start of adulthood. J. Stud. Alcohol Drugs 70, 489–498. https://doi.org/10.15288/jsad.2009.70.489. Choi, K.W., Stein, M.B., Nishimi, K.M., Ge, T., Coleman, J.R.I., Chen, C.-Y., Ratanatharathorn, A., Zheutlin, A.B., Dunn, E.C., Breen, G., Koenen, K.C., Smoller, J. W., 2020. An exposure-wide and Mendelian randomization approach to identifying modifiable factors for the prevention of depression. Am. J. Psychiatry 177, 944–954. https://doi.org/10.1176/appi.ajp.2020.19111158. COVID-19 Mental Disorders Collaborators, 2021. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic. Lancet 398, 1700–1712. https://doi.org/10.1016/S0140-6736 (21)02143-7. Flouri, E., Midouhas, E., Ruddy, A., 2016. Socio–economic status and family structure differences in early trajectories of child adjustment: individual and neighbourhood effects. Health Place 37, 8–15. https://doi.org/10.1016/j.healthplace.2015.11.005. Glymour, M.M., Avendano, M., Kawachi, I., 2014. Social Epidemiology, 2nd ed. Oxford University Press. https://doi.org/10.1093/med/9780195377903.003.0002. Gomm, S., Bernauer, T., 2023. Are actual and perceived environmental conditions associated with variation in mental health? Environ. Res. 223, 115398 https://doi. org/10.1016/j.envres.2023.115398. Goodwin, R.D., Dierker, L.C., Wu, M., Galea, S., Hoven, C.W., Weinberger, A.H., 2022. Trends in U.S. depression prevalence from 2015 to 2020: the widening treatment gap. Am. J. Prev. Med. 63, 726–733. https://doi.org/10.1016/j. amepre.2022.05.014. Gudi-Mindermann, H., White, M., Roczen, J., Riedel, N., Dreger, S., Bolte, G., 2023. Integrating the social environment with an equity perspective into the exposome paradigm: a new conceptual framework of the social exposome. Environ. Res. 233, 116485 https://doi.org/10.1016/j.envres.2023.116485. Hernandez-Ferrer, C., Wellenius, G.A., Tamayo, I., Basaga˜ na, X., Sunyer, J., Vrijheid, M., Gonzalez, J.R., 2019. Comprehensive study of the exposome and omic data using rexposome Bioconductor Packages. Bioinformatics 35. https://doi.org/10.1093/ bioinformatics/btz526. Hu, H., Liu, X., Zheng, Y., He, X., Hart, J., James, P., Laden, F., Chen, Y., Bian, J., 2023. Methodological challenges in spatial and contextual exposome-health studies. Crit. Rev. Environ. Sci. Technol. 53, 827–846. https://doi.org/10.1080/ 10643389.2022.2093595. Huang, Y., Lu, J., ˇ Sirůˇ cek, J., 2023. The associations between social environment and adolescents’ psychosomatic health: an ecological perspective. Front. Psychol. 14, 1141206 https://doi.org/10.3389/fpsyg.2023.1141206. Humphrey, J.L., Root, E.D., 2017. Spatio-temporal neighborhood impacts on internalizing and externalizing behaviors in U.S. elementary school children: effect modification by child and family socio-demographics. Soc. Sci. Med. 180, 52–61. https://doi.org/10.1016/j.socscimed.2017.03.014. Huppertz, C., Bartels, M., de Geus, E.J.C., van Beijsterveldt, C.E.M., Rose, R.J., Kaprio, J., Silventoinen, K., 2017. The effects of parental education on exercise behavior in childhood and youth: a study in Dutch and Finnish twins. Scand. J. Med. Sci. Sports 27, 1143–1156. https://doi.org/10.1111/sms.12727. Hyypp¨ a, M.T., M¨ aki, J., 2001. Why do Swedish-speaking Finns have longer active life? An area for social capital research. Health Promot. Int. 16, 55–64. https://doi.org/ 10.1093/heapro/16.1.55. Jivraj, S., Murray, E.T., Norman, P., Nicholas, O., 2020. The impact of life course exposures to neighbourhood deprivation on health and well-being: a review of the long-term neighbourhood effects literature. Eur. J. Pub. Health 30, 922–928. https://doi.org/10.1093/eurpub/ckz153. Kajantie, M., Manderbacka, K., McCallum, A., Notkola, I.-L., Arffman, M., Forssas, E., Karvonen, S., Kortteinen, M., Leyland, A., Keskim¨ aki, I., 2006. How to Carry out Register-Based Health Services Research in Finland?: Compiling Complex Study Data in the REDD Project. Stakes, Helsinki. Keil, A.P., Buckley, J.P., O’Brien, K.M., Ferguson, K.K., Zhao, S., White, A.J., 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ. Health Perspect. 128, 47004. Keles, B., McCrae, N., Grealish, A., 2020. A systematic review: the influence of social media on depression, anxiety and psychological distress in adolescents. Int. J. Adolesc. Youth 25, 79–93. https://doi.org/10.1080/02673843.2019.1590851. Kemppainen, L., Kemppainen, T., Fokkema, T., Wrede, S., Kouvonen, A., 2023. Neighbourhood ethnic density, local language skills, and loneliness among older migrants—a population-based study on Russian speakers in Finland. Int. J. Environ. Res. Public Health. https://doi.org/10.3390/ijerph20021117. Koops, J.C., Liefbroer, A.C., Gauthier, A.H., 2021. Socio-economic differences in the prevalence of single motherhood in North America and Europe. Eur. J. Popul. 37, 825–849. https://doi.org/10.1007/s10680-021-09591-3. Krokstad, S., Weiss, D.A., Krokstad, M.A., Rangul, V., Kvaløy, K., Ingul, J.M., Bjerkeset, O., Twenge, J., Sund, E.R., 2022. Divergent decennial trends in mental health according to age reveal poorer mental health for young people: repeated cross-sectional population-based surveys from the HUNT Study, Norway. BMJ Open 12, e057654. https://doi.org/10.1136/bmjopen-2021-057654. Liddell, B.J., Nickerson, A., Sartor, L., Ivancic, L., Bryant, R.A., 2016. The generational gap: mental disorder prevalence and disability amongst first and second generation immigrants in Australia. J. Psychiatr. Res. 83, 103–111. https://doi.org/10.1016/j. jpsychires.2016.08.011. Lin, B.D., Pries, L.-K., Sarac, H.S., van Os, J., Rutten, B.P.F., Luykx, J., Guloksuz, S., 2022. Nongenetic factors associated with psychotic experiences among UK biobank participants: exposome-wide analysis and Mendelian randomization analysis. JAMA Psychiatry 79, 857–868. https://doi.org/10.1001/jamapsychiatry.2022.1655. Morrissey, K., Kinderman, P., 2020. The impact of childhood socioeconomic status on depression and anxiety in adult life: testing the accumulation, critical period and social mobility hypotheses. SSM - Popul. Heal. 11, 100576 https://doi.org/10.1016/ j.ssmph.2020.100576. Neufcourt, L., Castagn´ e, R., Mabile, L., Khalatbari-Soltani, S., Delpierre, C., KellyIrving, M., 2023. Assessing how social exposures are integrated in exposome research: a scoping review. Environ. Health Perspect. 130, 116001 https://doi.org/ 10.1289/EHP11015. Ni, M.Y., Yao, X.I., Cheung, F., Wu, J.T., Schooling, C.M., Pang, H., Leung, G.M., 2020. Determinants of physical, mental and social well-being: a longitudinal environmentwide association study. Int. J. Epidemiol. 49, 380–389. https://doi.org/10.1093/ije/ dyz238. Ohanyan, H., Portengen, L., Huss, A., Traini, E., Beulens, J.W.J., Hoek, G., Lakerveld, J., Vermeulen, R., 2022. Machine learning approaches to characterize the obesogenic urban exposome. Environ. Int. 158, 107015 https://doi.org/10.1016/j. envint.2021.107015. Person Waye, K., L¨ ove, J., Lercher, P., Dzhambov, A.M., Klatte, M., Schreckenberg, D., Belke, C., Leist, L., Ristovska, G., Jeram, S., Kanninen, K.M., Selander, J., Arat, A., Lachmann, T., Clark, C., Botteldooren, D., White, K., Julvez, J., Foraster, M., Kaprio, J., Bolte, G., Psyllidis, A., Gulliver, J., Boshuizen, H., Bozzon, A., Fels, J., Hornikx, M., van den Hazel, P., Weber, M., Brambilla, M., Braat-Eggen, E., Van Kamp, I., Vincens, N., 2023. Adopting a child perspective for exposome research on mental health and cognitive development - conceptualisation and opportunities. Environ. Res. 117279 https://doi.org/10.1016/j.envres.2023.117279. Preacher, K.J., Wichman, A.L., MacCallum, R.C., Briggs, N.E., 2008. Latent Growth Curve Modeling. SAGE Publications, Inc., Thousand Oaks https://doi.org/10.4135/ 9781412984737. Pries, L.-K., Moore, T.M., Visoki, E., Sotelo, I., Barzilay, R., Guloksuz, S., 2022. Estimating the association between exposome and psychosis as well as general psychopathology: results from the ABCD study. Biol. Psychiatry Glob. Open Sci. 2, 283–291. https://doi.org/10.1016/j.bpsgos.2022.05.005. Pulkkinen, L., 2017. Human development from middle childhood to middle adulthood: growing up to be middle-aged. In: Human Development from Middle Childhood to Middle Adulthood: Growing Up to be Middle-aged, 1st ed. Routledge, London. https://doi.org/10.4324/9781315732947. Pulkkinen, L., Kaprio, J., Rose, R.J., 1999. Peers, teachers and parents as assessors of the behavioural and emotional problems of twins and their adjustment: the Multidimensional Peer Nomination Inventory. Twin Res. Hum. Genet. 2, 274–285. https://doi.org/10.1375/twin.2.4.274. Rose, R.J., Salvatore, J.E., Aaltonen, S., Barr, P.B., Bogl, L.H., Byers, H.A., Heikkil¨ a, K., Korhonen, T., Latvala, A., Palviainen, T., Ranjit, A., Whipp, A.M., Pulkkinen, L., Dick, D.M., Kaprio, J., 2019. FinnTwin12 cohort: an updated review. Twin Res. Hum. Genet. 22, 302–311. https://doi.org/10.1017/thg.2019.83. Rosseel, Y., 2012. Lavaan: an R package for structural equation modeling. J. Stat. Softw. 48, 1–36. https://doi.org/10.18637/jss.v048.i02. Schinasi, L.H., Benmarhnia, T., De Roos, A.J., 2018. Modification of the association between high ambient temperature and health by urban microclimate indicators: a systematic review and meta-analysis. Environ. Res. 161, 168–180. https://doi.org/ 10.1016/j.envres.2017.11.004. Solantaus, T., Leinonen, J., Punam¨ aki, R.-L., 2004. Children’s mental health in times of economic recession: replication and extension of the family economic stress model in Finland. Dev. Psychol. https://doi.org/10.1037/0012-1649.40.3.412. Stinson, S., 1985. Sex differences in environmental sensitivity during growth and development. Am. J. Phys. Anthropol. 28, 123–147. https://doi.org/10.1002/ ajpa.1330280507. Traini, E., Huss, A., Portengen, L., Rookus, M., Verschuren, W.M.M., Vermeulen, R.C.H., Bellavia, A., 2022. A multipollutant approach to estimating causal effects of air pollution mixtures on overall mortality in a large, prospective cohort. Epidemiology 33. van de Weijer, M.P., Baselmans, B.M.L., Hottenga, J.-J., Dolan, C.V., Willemsen, G., Bartels, M., 2022. Expanding the environmental scope: an environment-wide association study for mental well-being. J. Expo. Sci. Environ. Epidemiol. 32, 195–204. https://doi.org/10.1038/s41370-021-00346-0. Visser, K., Bolt, G., Finkenauer, C., Jonker, M., Weinberg, D., Stevens, G.W.J.M., 2021. Neighbourhood deprivation effects on young people’s mental health and well-being: a systematic review of the literature. Soc. Sci. Med. 270, 113542 https://doi.org/ 10.1016/j.socscimed.2020.113542. Wang, Z., Zellers, S., Whipp, A.M., Heinonen-Guzejev, M., Foraster, M., Júlvez, J., van Kamp, I., Kaprio, J., 2023. The effect of environment on depressive symptoms in late adolescence and early adulthood: an exposome-wide association study and twin modeling. Nat. Ment. Heal. https://doi.org/10.1038/s44220-023-00124-x. Weckroth, M., Ala-Mantila, S., Ballas, D., Ziogas, T., Ikonen, J., 2022. Urbanity, neighbourhood characteristics and perceived quality of life (QoL): analysis of individual and contextual determinants for perceived QoL in 3300 postal code areas in Finland. Soc. Indic. Res. 164, 139–164. https://doi.org/10.1007/s11205-02102835-z. Wheaton, B., Clarke, P., 2003. Space meets time: integrating temporal and contextual influences on mental health in early adulthood. Am. Sociol. Rev. 68, 680–706. https://doi.org/10.2307/1519758. Z. Wang et al. Journal of Affective Disorders 358 (2024) 70–78 78 Whipp, A.M., Korhonen, T., Raevuori, A., Heikkil¨ a, K., Pulkkinen, L., Rose, R.J., Kaprio, J., Vuoksimaa, E., 2019. Early adolescent aggression predicts antisocial personality disorder in young adults: a population-based study. Eur. Child Adolesc. Psychiatry 28, 341–350. https://doi.org/10.1007/s00787-018-1198-9. White, J., Greene, G., Farewell, D., Dunstan, F., Rodgers, S., Lyons, R.A., Humphreys, I., John, A., Webster, C., Phillips, C.J., Fone, D., 2017. Improving mental health through the regeneration of deprived neighborhoods: a natural experiment. Am. J. Epidemiol. 186, 473–480. https://doi.org/10.1093/aje/kwx086. Xia, Y., Yang, Y., 2019. RMSEA, CFI, and TLI in structural equation modeling with ordered categorical data: the story they tell depends on the estimation methods. Behav. Res. Methods 51, 409–428. https://doi.org/10.3758/s13428-018-1055-2. Xue, Y., Leventhal, T., Brooks-Gunn, J., Earls, F.J., 2005. Neighborhood residence and mental health problems of 5to 11-year-olds. Arch. Gen. Psychiatry 62, 554–563. https://doi.org/10.1001/archpsyc.62.5.554. Z. Wang et al.