Psychological and contextual risk factors for first‐onset depression among adolescents and young people around the globe: A systematic review and meta‐analysis

Abstract Aim Identifying predictors for future onset of depression is crucial to effectively developing preventive interventions. We conducted a systematic review and meta‐analysis to identify risk factors for first‐onset depression among adolescents and young people. Methods We searched MEDLINE (Ovid), PsycINFO, Cochrane Database, Web of Science, Lilacs, African Journals Online and Global Health (July 2009 to December 2020) for longitudinal studies assessing risk factors for first‐onset depression among adolescents and young people aged 10–25 years. Meta‐analyses generated summary odds ratio (OR) estimates. Registration: PROSPERO CRD42018103973. Results Nineteen studies representing 21 unique populations were included in the meta‐analysis. Among studies reporting race/ethnicity, 79% of participants were of White/European descent. Seventeen studies were from high‐income countries, with only two from an upper‐middle‐income country (China). Odds for first‐onset depression were significantly greater for girls compared to boys (n = 13; OR = 1.78 [1.78, 2.28], p < 0.001) and for youth with other mental health problems at baseline (n = 4; OR = 3.20 [1.95, 5.23], p < 0.001). There were non‐significant associations for negative family environment (n = 8; OR = 1.60 [0.82, 3.10], p = 0.16) and parental depression (n = 3; OR = 2.30 [0.73, 7.24], p = 0.16). Conclusions Most longitudinal studies do not report risk factors specifically for first‐onset depression. Moreover, predictive data are limited to predominantly White populations in high‐income countries. Future research must be more ethnically and geographically representative. Recommendations are provided for consistent and comprehensive reporting of study designs and analyses of risk factors for first‐onset depression.


| INTRODUCTION
Due to its early-onset and chronicity, depression is a leading contributor to disability-adjusted life-years (Whiteford et al., 2013). Neglecting to prevent and treat early-life onset of depression has severe consequences: in 2019, global rates of suicide are highest between ages 20 and 25 years, and most adolescents who died by suicide (88%) were from low-and-middle-income countries (LMIC) (World Health Organization, 2019). This statistic is particularly arresting considering that 90% of the world's adolescents live in LMICs, yet research identifying predictors of depression conducted in these settings is lacking (Kieling et al., 2011). Previous systematic reviews and meta-analyses have been conducted on risk factors for adolescent depression, but the overall generalizability of the evidence is limited.
Identification of a broad scope of psychological and contextual risk factors across settings offers the potential to accurately assess risk for first-onset depression and more reliably inform prevention strategies and mechanisms of action (Cuijpers et al., 2012;Kieling et al., 2019). The World Health Organization and the United Nations agree that a wide age range for adolescence (<25 years) is pivotal, particularly in LMICs, to account for contextual factors that may impact the development of the adolescent brain transitioning into adulthood across cultures (United Nations Department of Economic and Social Affairs, 2010;WHO, 2011). These recommendations corroborate research showing that brain maturation in adolescence is defined between ages 10 and 24 years (Arain et al., 2013), with experts insisting that this more inclusive age range is required to account for major developmental impacts framed by social policies, service systems and laws (Sawyer et al., 2018). Accounting for a range of risk factors also increases the proficiency of existing predictive models that can be evaluated in high-income country (HIC) and LMIC cohorts.
We found nine previous systematic reviews of risk factors for adolescent depression. One review identified 113 studies presenting only modifiable risk and protective factors (e.g., substance use, dieting, negative coping strategies and weight) shown to influence development of adolescent depression (Cairns et al., 2014). Of the 113 studies, only 3% represented populations in LMICs. In another review, modifiable predictors of depression following maltreatment (e.g., physical or sexual abuse) were identified for adolescents <18 years across 22 studies; though authors reported a lack of well-designed prospective studies (Braithwaite et al., 2017). Two reviews identified 76 studies presenting socioeconomic risk factors (community safety and ethnic minority status) (Reiss, 2013;Stirling et al., 2015), though 90% of the studies represented high-and-upper-middle-income populations, and neither review focused on prospective effects for depression onset. Another review identified effects for parental factors increasing risk for anxiety and depression in adolescent offspring aged 12-18 , though results were combined from cross-sectional and prospective studies. All reviews were limited to a 19-year age range. No identified reviews emphasized prospective longitudinal studies evaluating a range of contextual and psychological risk factors among adolescents ≤25 years across universal settings.
Therefore, we conducted a meta-analysis of psychological and contextual risk factors predicting first-onset depression among adolescents ≤25 years in the global literature.

| METHODS
This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) 2015 checklist (PRISMA-P Group et al., 2015). A systematic review protocol for the evaluation of risk factors associated with depression among adolescents and young people was registered in PROSPERO (https://www.crd.york.ac.uk/prospero/#myprospero CRD42018103973) and published (Pedersen et al., 2019

| Study selection
The database results were exported to EndNote X8, a reference management system that was also used to remove duplicates, resulting in 12 753 articles titles and abstracts for review. Using a charting form in Google sheets, four reviewers screened the same title and abstracts in batches of 10 articles from the EndNote export results to assess inter-rater reliability (IRR) on inclusion/exclusion criteria for title and abstract screening, with any discrepancies resolved by a fifth reviewer. The reviewers repeated this process with subsequent batches of 10 articles, recording all final agreements per article. Upon the fifth batch of 10 articles, a blinded 95% IRR was reached. After achieving this IRR, the remaining titles and abstracts (n = 12, 703) were then divided among the reviewers for independent screening with discussions and stages of IRR reassessment (see Figure S1 in supplementary file for details on the title and abstract screening and IRR processes). After the 12 753 titles and abstracts were screened, 6394 titles and abstracts were left for potential inclusion, which then underwent a re-screening. IRR was reassessed at this time following the same procedure as described above (e.g., all reviewers screening batches of 10 until an IRR of 95% was achieved).  . Table S1).
Where analysis of first-onset depression was unclear (e.g., baseline or history of MDD was not reported or first-onset was combined with previous onset at follow-up), but all other criteria were met, we contacted authors by email for clarification. Contextual and psychological risk factors for first-onset depression were included. After full-text screening was completed, but prior to conducting the meta-analysis, we created risk factor subgroups a priori by thematically grouping similar risk factors into categories following guidance by the IDEA consortium experts and existing predictive models for adolescent depression (Wahid et al., 2021). For performing meta-analysis of randomized controlled trials (RCT), Cochrane recommends 'Two studies is a sufficient number to perform a meta-analysis, provided that those two studies can be meaningfully pooled and provided their results are sufficiently "similar",' (Ryan, 2016) with the basis of being 'sufficiently similar' depending on whether the studies are fairly homogeneous and lower risk of bias. Given observational cohort study designs introduce more bias overall than RCT designs, and because this study did not limit inclusion criteria to a single tool for measuring depression, nor a single context, the authors of this review agreed that a more robust analysis would require at least three or more studies per thematic risk factor category in order to be included in the meta-analysis.

| Outcomes
The primary outcome was first-onset depression among young people ages 10-25 years in any population, for example, high-, middle-and low-income countries. At baseline, the age of young people should have been <25 years, with first-onset depression reported as a dichotomous outcome and measured ≥6 months from baseline. Depression could include self-and adult-informant reports with a cut-off score, structured observations in clinical settings, and/or clinical records.
Additionally, we included mixed population studies if outcomes were reported separately for the subgroup who did not have current or prior depression at baseline or if risk factors were reported controlling for baseline depression status.

| Data extraction and analysis
For the extraction stage, a data extraction sheet for descriptive and quantitative data was developed in Google sheets with predetermined fields (Pedersen et al., 2019) and was approved by the IDEA consortium. Three reviewers conducted an IRR process where each reviewer extracted information for all categories for the same article. Each author was blind to each-other's extracted content to ensure rich discussion and assess unique discrepancies across reviewers. The extraction reviewers reached a 95% IRR on the third full-text article. Two reviewers extracted data for the remaining included studies. A third reviewer validated extraction for all studies, using weekly meetings to discuss discrepancies, update extractions and refer to a fourth reviewer if any discrepancy could not be resolved.
We assessed the quality of evidence for each study using the Systematic Assessment of Quality in Observational Research (SAQOR) (Ross et al., 2011) with converted modified Grading of Recommendations, Assessment, Development and Evaluations (GRADE) rankings (Guyatt et al., 2011;Ross et al., 2011). Reviewers discussed and piloted application of SAQOR criteria with three articles. A Google doc was created to specify discussion points and charting information for reviewers' reference during assessment of studies. After agreement was reached, two reviewers assessed the quality of the included articles, and the third reviewer validated the SAQOR criteria per article to ensure accuracy.
To assess the effect of risk factors predicting first-onset depres- Lack of analysis of first-onset depression included studies that met any of the following criteria: Baseline assessment of current or prior major depressive disorder (MDD) not reported; Baseline assessment of current or prior MDD combined with first-onset at follow-up outcome reporting; Analysis of any disorder first-onset (not specific to MDD) combined with first-onset depression.

| RESULTS
Of the 27 803 articles from the research output, 12 753 titles and abstracts were screened of which 6394 titles and abstracts were rescreened, resulting in 496 full-text articles screened, with a final 19 papers included in the meta-analysis. During full-text screening, 325 (66%) were excluded because of lack of reporting for first-onset depression. Typical reasons for not meeting our depression outcome criteria included not controlling for depression at baseline, combining results for multiple age ranges including ages above 25, combining depression outcomes at follow-up for those with and without a history of depression, or presenting outcome data on age ranges outside of 10-25 years. All exclusion criteria, including for the other 152 articles at full-text screening, can be found in the PRISMA diagram in Figure 1.
Each of the 19 articles included in the meta-analysis selected its population nationally, with one article reporting a national sample stratified by three ethnic groups (Smith et al., 2015). Therefore, we  Table 2.
The studies featured a wide variation in baseline sample sizes, ranging from 62 to 60 066 young people. Studies that met the inclusion criteria did not include young people above age 19 at baseline; young people' baseline ages ranged from 4.5 to 19 years across the studies. Table 2 includes the mean age at baseline for studies that included enough information for calculation; however, we cannot report a mean age across studies due to lack of reporting. The estimated length of follow-up for first-onset depression across the 19 studies ranged from 9 months to 14.5 years, with a mean of 4.6 years and a median of 3 years. Reporting the exact length of follow-up varied across studies, where some studies were less clear, and therefore the authors calculated an estimation and included this in Table 2.
Depression was the primary outcome in each study (see Suppl. Table S2). The most common depression assessment instrument was the Kiddie-Schedule for Affective Disorders and Schizophrenia Prevalence of first-onset depression, as a percentage of nondepressed baseline young people with complete data, ranged from 2.90% for a sample size of 999 to 53.23% for a sample size of 62.
Details of the study risk factors, risk factor categories, measurement type, and analysis type in relation to first-onset depression can be found in Table 3. Final risk factor categories for this study included   (Higgins et al., 2003).
We used SAQOR quality assessment criteria with modified GRADE rankings (Ross et al., 2011) (see Table 4). Ten studies were marked 'Adequate' across four or more of the six SAQOR categories and remained at the initial GRADE rating for observational studies of 'low' quality. The other nine studies had three or more categories marked as 'Inadequate' and were downgraded to a GRADE rating of "very low" (Guyatt et al., 2011). Four out of the six SAQOR domains were typically not fully met. Papers typically did not meet one or more criteria in four of the six SAQOR domains. In quality in measurement, 37% of studies did not meet full criteria, such as authors not clearly defining the tools or methods used to measure the study outcome. In reporting on follow-up, 58% of studies did not meet full criteria, in some cases due to authors' not providing an explanation for how or why study participants were not followed-up on. In reporting of distorting influences, 53% of studies did not meet one or more criteria, for reasons such as not accounting for other psychiatric comorbidities  T A B L E 3 Study (n = 19) risk factor(s), corresponding risk factor category, risk factor measurement type, analysis type for risk factor relation to first-onset depression

| DISCUSSION
In this meta-analysis, we identified 19 prospective studies representing 21 unique populations that identified risk factors predicting first-onset depression among young people aged 10-25. Heterogeneity was high across all analyses conducted, for example, I 2 > 70%.
Given the small number of studies for each risk factor, it was not possible to do subgroup analyses to determine sources of heterogeneity.
Other previously reported risk factors were not found to be statistically associated with future first-onset depression in this analysis.    (Kohrt et al., 2012;Marahatta et al., 2017;Upadhaya et al., 2019). Also, prevalence of depression and incidence of suicide are highest among females under age 25 in this region (Hagaman et al., 2017;Luitel et al., 2013).
Of the existing cohorts in HIC and LMICs which have collected data on risk factors for first-onset depression, only a few have published results that represent participants without depression at F I G U R E 2 Forest plot of risk factor categories predicting first-onset depression outcome among study populations (n = 19) †Forest plot shows four risk factor categories (gender, family environment, other mental health problems, and parental depression) in which three or more studies were compiled per category. We conducted random-effects meta-analyses in Cochrane Review Manager version 5.3 and calculated odds ratios (OR) for binary outcomes with 95% confidence intervals and two-sided p-values for each risk factor category. Heterogeneity was assessed using the I-squared statistic within Cochrane Review Manager 5.3. * Forest Plot Abbreviations: Chi-square (Chi 2 ); Confidence interval (CI); Degrees of freedom (df); Heterogeneity (I 2 ); Interval variable (IV); Standard error (SE) baseline. Among the full-text articles we excluded for lack of analysis for first-onset depression in this study, at least 34% (112 out of 325) resulted from not reporting or controlling for baseline or history of MDD. Those that have access to these dataset types are encouraged to publish these findings. Zurich Cohort Study (Angst et al., 1984;Angst et al., 2005), Early Developmental Stages of Psychopathology (EDSP) (Wittchen et al., 1998), and Christchurch Health and Development Study (Fergusson & Horwood, 2001). The majority of these studies were excluded in the full-text stage due to lack of clear reporting for firstonset of depression, with the exception of one EDSP study (Beesdo et al., 2007), which was excluded due to being published outside of our included date range. Common reasons for exclusion were not controlling for baseline depression, combining depression outcomes for those with and without a history of depression, or age range issues such as combining results for age ranges below and above 25. For instance, the EDSP cohort data was analysed for negative emotionality as a risk factor associated with depression at age 18 but did not report a measurement of depression at baseline or any point before age 18 (Bould et al., 2014). As a result, depression diagnoses at follow-up did not meet our first-onset depression outcome criteria.
Another analysis of EDSP cohort data examined incidence, comorbidity, and risk patterns for anxiety and depressive disorders (Beesdo et al., 2010). However, the data presented for onset of depressive disorder was a cumulative age of 34, and we could not extract onset for depression within our included age range (10-25 years). An analysis from the Oregon cohort data was evaluated for the relationship between childhood respiratory symptoms and later depression but did not distinguish between those with and without a history of depression (Goodwin et al., 2013). In this case, combining depression outcomes for those with and without a history of depression prevented the determination of a clear link between the risk factor and firstonset depression. Analyses using data from the Christchurch cohort reported pooled results for onset of MDD for age groups 18, 21, 25 and 30, and therefore we could not include these in our study (Fergusson et al., 2014;Fergusson, Boden, & Horwood, 2013;Fergusson, McLeod, & Horwood, 2013;Fergusson, McLeod, Horwood, Swain, et al., 2015;Newton-Howes et al., 2015).
Overall, due to the heterogeneity in design, analyses, and reporting of data in major longitudinal studies, it was not possible to draw upon many of the large cohort studies for the current meta-analysis. Therefore, to inform prevention programs and increase accurate detection of risk and protective factors for first-onset depression among young people globally, we offer recommendations for future research and analyses in Box 1. Youth living in LMICs experience the highest risk for firstonset depression and suicide due to their environmental and social circumstances, such as increased rates of poverty, violence, natural disasters and lack of available psychological treatments (Belfer, 2008;Erskine et al., 2015;Kieling et al., 2011;Leckman & Leventhal, 2008;WHO, 2014 (Brathwaite, Rocha, Kieling, Gautam, et al., 2021;Brathwaite, Rocha, Kieling, Kohrt, et al., 2020;Rocha et al., 2021) 2. Sample selection: Longitudinal study designs with young people should clearly recruit, confirm, and report on subsamples without current or prior depression at baseline. Harmonization of how risk and protective factors are operationalized and measured will enhance the ability to conduct cross-cultural data analysis, including testing prediction models and measurement of at-risk young people from multiple data sets worldwide (Brathwaite, Rocha, Kieling, Gautam, et al., 2021). For example, for the risk factor, 'family environment', included studies assessed maternal aggressive parenting behaviours (Callaghan et al., 2017;Little et al., 2015), mother-infant interaction (Schmid et al., 2011), report of parent involvement with school (Smith et al., 2015), report of living with only one parent (Li et al., 2018), and satisfaction of perceived family functioning (Wu et al., 2017). Studies in the gender category lacked clarity in measurement: most included a self-report (male/ female) but did not decipher gender as a biological variable (defined by DNA encoded, physiological characteristics). Risk factors should not be studied in isolation. The epidemiology and psychiatry fields have established the increasing importance of using multicausal models for illnesses; yet most risk factors for psychiatric illnesses continue to be studied in isolation (Hill, 1965;Kendler, 2019). Rather than attempting a reductionist approach to determine a single cause that is 'necessary and sufficient', researchers should

| Limitations
Our study has several limitations. Due to the small number of studies included across most of the risk factor categories, we cannot uncover sources of heterogeneity across studies or run tests for asymmetry.
Also, we are unable to address additional risk factors that have been previously shown to increase risk of depression. For example, we attempted to include risk categories that covered substance use (e.g., smoking and alcohol use) and physical health problems (e.g., asthma, body mass index and blood pressure). We identified only two studies that were initially eligible but then excluded for not meeting criteria to substantiate a risk factor category in our analyses (Chen et al., 2014;Hammerton et al., 2013). Due to the lack of evidence, this review was unable to address or control for protective factors. Finally, this review focused on identifying exposures to psychological and contextual factors that could lead to a depression outcome, rather than assessing the effects of an intervention (e.g., preventative interventions) on a depression outcome. Given the absence of preventive interventions in this review, it is not possible to make any GRADE classification of recommendations for intervention strategies. A recent review has summarized the prevention literature for young people (Fusar-Poli et al., 2021).

| CONCLUSION
This systematic review and meta-analysis strengthen evidence on risk factors predicting first-onset depression in adolescence and young people, revealing longitudinal prospective effects for gender and other mental health problems in high-income and one uppermiddle-income countries. We were unable to identify any longitudinal studies that further support evidence for these risk factors in LMIC. Based on common shortcomings of the current literature, we provided five recommendations for future research on risk factors for first-onset depression among young people: first, include more ethnically, racially, and diverse populations; second, in recruitment and analysis, clearly identify the subsample of participants who had no current or prior depression at baseline assessment; third, validate the depression assessment within the given study population for both the age ranges and the specific linguistic and cultural groups participating; fourth, during protective and risk factor measurement, use common definitions, measurement strategies, and data collection time points to promote conduct of cross-cultural data analysis; fifth, assess multiple risk and protective factors, and consider development of risk factor scores, to identify pathways for predicting the onset of depression. Reducing the incidence of depression is a global health necessity, and preventive strategies have shown promising results (Cuijpers et al., 2012). To support the prevention of firstonset depression in young people, future research could develop strategies that give greater attention to the strongest risk factors consistently evidenced in the literature (Ormel et al., 2019), such as focusing on the mental health needs of children that identify as female, or children with psychological comorbidities. In addition, prevention strategies could target a constellation of risk factors and go beyond the individual or relationship level. Prevention strategies taking place in local institutions (e.g., schools) have been found to be acceptable, increase reach, allow for modular approaches that can be tailored to the individual or group, and increase the chance for sustainability (Fazel & Kohrt, 2019;Ormel et al., 2019;Weist et al., 2019). A larger scope of risk and protective factors for firstonset depression must be identified globally to inform best prevention strategies across settings.

CONFLICT OF INTEREST
VM has received research funding from Johnson & Johnson, a pharmaceutical company interested in the development of antiinflammatory strategies for depression, but the research described in this paper is unrelated to this funding. All other authors declare that they have no competing interests.

AUTHORS CONTRIBUTIONS
BAK and CK conceived the idea for the review. GAP, BAK, CK, HLF, VM, and ZZ developed the review protocol. GAP, KG, CL, and MH conducted the search with supervision by BAK. CL, MH and GAP performed extraction. GAP and BAK did the statistical analysis. GAP and BAK wrote the first draft of the manuscript with input from CL and MH. All authors critically appraised and edited the manuscript. All authors contributed to revision and finalization of the manuscript.

DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no new data were created or analyzed in this study.
identify a 'web of causation' (Kendler, 2019;Krieger, 1994) and assess multiple interactions of multiple causes to identify pathways that predict the onset of depression, including both risk and protective factors. Regardless of whether it is adequate or essential, blocking a component factor or cause of first-onset depression offers the ability to prevent onset of depression (Kendler, 2019;Rothman & Greenland, 2005). The IDEA consortium studies have used a constellation of 11 risk factors to calculate future depression with greater than chance accuracy (Brathwaite, Rocha, Kieling, Gautam, et al., 2021;Brathwaite, Rocha, Kieling, Kohrt, et al., 2020;Rocha et al., 2021).