糖心vlog

Key Points

Questions听 Are hypothesized associations between familial risk of depression and lower cognitive performance evident across the life span for both family history and genetic risk measures?

Findings听 In this cohort study including 57鈥�308 participants from 4 cohorts, among the 3 younger cohorts (age range, 6 to 42 years), family history of depression was primarily associated with lower performance in the memory domain, whereas in the older cohort (age range, 44 to 83 years), the associations were shown for processing speed, attention, and executive function. Associations were similar in the polygenic risk score analyses and were evident even in participants who had never been depressed themselves but had a family history of depression.

Meaning听 In this study, whether assessed by family history or genetic data, depression in prior generations was associated with lower cognitive performance in offspring, which has important implications for understanding and addressing potentially modifiable risk factors.

Importance听 Cognitive impairment in depression is poorly understood. Family history of depression is a potentially useful risk marker for cognitive impairment, facilitating early identification and targeted intervention in those at highest risk, even if they do not themselves have depression. Several research cohorts have emerged recently that enable findings to be compared according to varying depths of family history phenotyping, in some cases also with genetic data, across the life span.

Objective听 To investigate associations between familial risk of depression and cognitive performance in 4 independent cohorts with varied depth of assessment, using both family history and genetic risk measures.

Design, Setting, and Participants听 This study used data from the Three Generations at High and Low Risk of Depression Followed Longitudinally (TGS) family study (data collected from 1982 to 2015) and 3 large population cohorts, including the Adolescent Brain Cognitive Development (ABCD) study (data collected from 2016 to 2021), National Longitudinal Study of Adolescent to Adult Health (Add Health; data collected from 1994 to 2018), and UK Biobank (data collected from 2006 to 2022). Children and adults with or without familial risk of depression were included. Cross-sectional analyses were conducted from March to June 2022.

Exposures听 Family history (across 1 or 2 prior generations) and polygenic risk of depression.

Main Outcomes and Measures听 Neurocognitive tests at follow-up. Regression models were adjusted for confounders and corrected for multiple comparisons.

Results听 A total of 57鈥�308 participants were studied, including 87 from TGS (42 [48%] female; mean [SD] age, 19.7 [6.6] years), 10鈥�258 from ABCD (4899 [48%] female; mean [SD] age, 12.0 [0.7] years), 1064 from Add Health (584 [49%] female; mean [SD] age, 37.8 [1.9] years), and 45鈥�899 from UK Biobank (23鈥�605 [51%] female; mean [SD] age, 64.0 [7.7] years). In the younger cohorts (TGS, ABCD, and Add Health), family history of depression was primarily associated with lower performance in the memory domain, and there were indications that this may be partly associated with educational and socioeconomic factors. In the older UK Biobank cohort, there were associations with processing speed, attention, and executive function, with little evidence of education or socioeconomic influences. These associations were evident even in participants who had never been depressed themselves. Effect sizes between familial risk of depression and neurocognitive test performance were largest in TGS; the largest standardized mean differences in primary analyses were 鈭�0.55 (95% CI, 鈭�1.49 to 0.38) in TGS, 鈭�0.09 (95% CI, 鈭�0.15 to 鈭�0.03) in ABCD, 鈭�0.16 (95% CI, 鈭�0.31 to 鈭�0.01) in Add Health, and 鈭�0.10 (95% CI, 鈭�0.13 to 鈭�0.06) in UK Biobank. Results were generally similar in the polygenic risk score analyses. In UK Biobank, several tasks showed statistically significant associations in the polygenic risk score analysis that were not evident in the family history models.

Conclusions and Relevance听 In this study, whether assessed by family history or genetic data, depression in prior generations was associated with lower cognitive performance in offspring. There are opportunities to generate hypotheses about how this arises through genetic and environmental determinants, moderators of brain development and brain aging, and potentially modifiable social and lifestyle factors across the life span.

Cognitive impairment is a key cause of disability in adults with depression. It is evident at the first depressive episode¹ and persists even after remission,² leading to worse functioning³ and lower quality of life.⁴ Cognitive impairment in depression is poorly understood but likely involves a complex interplay between background risk factors for both depression and cognitive dysfunction and other factors that operate further downstream after depression onset.

Background risk can be elucidated by studying biological relatives of people with depression. A meta-analysis of studies of never-depressed first-degree relatives of people with major depressive disorder⁵ showed consistent effect sizes across all cognitive domains (standardized mean difference, 鈭�0.2), which were statistically significant for intelligence, memory, and language but not for attention, speed, or executive function. Therefore, family history of depression has potential to be a clinically useful risk marker, opening the possibility of early identification and targeted prevention or intervention for cognitive dysfunction in those at highest risk.

There are challenges with studying familial risk of depression. Retrospective reporting of family history is liable to missingness and recall bias, but direct prospective assessment is resource intensive and difficult to implement at scale. The family study Three Generations at High and Low Risk of Depression Followed Longitudinally (TGS)⁶ offers a unique opportunity to investigate family history and cognitive function using criterion-standard methods. Prospective clinical assessment of depression by trained clinical interviewers has been undertaken on multiple occasions across more than 30 years, together with high-quality cognitive testing and neuroimaging. The inclusion of multiple generations enables family risk to be characterized in greater detail than most studies to date, which have included only first-degree relatives. We have shown that there is a dose effect in this cohort whereby offspring with both a parent and grandparent with major depression were at highest risk of developing depression themselves.⁶ To our knowledge, no study to date has investigated whether a dose effect is also present for offspring cognitive outcomes. If that were found to be the case, it would enable better targeting of early intervention on the basis of number of prior generations affected.

Although the TGS cohort is uniquely well placed to enable this research, it is essential that findings are replicable and generalizable to the wider population, especially where direct assessment of relatives is not feasible. We have demonstrated that a dose effect on offspring depression outcomes is also evident in the general population-based Adolescent Brain Cognitive Development (ABCD) study,^7,8 which relied on family history reported retrospectively by a single informant, and similar research is needed on cognitive outcomes.

It is also important to include cohorts with different age ranges; there are indications that processing speed deficits are less prominent (compared with deficits in other domains) in unaffected relatives⁵ and emerge later in life in those with depression,² implicating downstream effects of depressive illness or differential aspects of brain aging. A further advantage of studying large population cohorts, such as ABCD, is that many include genotyping data, enabling the derivation of polygenic risk scores (PRS). Polygenic risk for depression represents genetic aspects of familial depression risk based on common genotypic variants and has been shown to be associated with a wide range of phenotypes relating to mental and physical health and brain structure in independent cohorts.⁹ Socially diverse population cohorts can also shed light on nongenetic aspects of familial risk, for example, lower socioeconomic resources in families affected by depression may reduce opportunities for cognitive development in offspring,^5,10 as well as modifying genetic risk in an interactive manner.⁹

In this study, we quantified the association of familial risk of depression with cognitive outcomes in TGS and in 3 general population cohorts spanning childhood to old age, including the ABCD,⁷ the National Longitudinal Study of Adolescent to Adult Health (Add Health),¹¹ and UK Biobank.¹² Our aims were to ascertain whether the hypothesized associations with lower cognitive performance were evident in all cohorts and for both family history and genetic data and to elucidate the patterns of association across cognitive domains and across the life span.

This study used a cohort design within TGS, ABCD, and UK Biobank, with family history data collected at one assessment wave and cognitive outcomes measured at a later wave. In Add Health, the family history data and cognitive data were only available at the same wave, and so these analyses were cross-sectional. Each cohort鈥檚 study procedures were approved by the relevant institutional review board or ethics committee, and participants gave written informed consent. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology () reporting guideline.¹³

Full details regarding the design and composition of each cohort are provided in the eMethods in Supplement 1. Race and ethnicity data were self-reported for all 4 studies. In TGS, race was coded as non-Hispanic White or other; in ABCD, race and ethnicity was coded as Asian, Black, Hispanic, White, or other; in Add Health, race and ethnicity were coded as American Indian or Alaska Native, Asian, Black/African American, Hispanic, multiple (selected more than 1 category), Native Hawaiian or Other Pacific Islander, White, or other race; and in UK Biobank, race and ethnicity was coded as Asian/Asian British, Black/Black British, Chinese, mixed or other ethnic group, and White.

Familial risk of depression was measured using 2 sources of data: reported/assessed biological family history and PRS.

TGS was the only cohort in which depression was directly assessed in all generations by direct interview with the participant. In the other cohorts, family history was ascertained from retrospective reporting by the participant or their parent. The cohorts varied in how depression was defined (details in the eMethods in Supplement 1): TGS used a best-estimate major depressive disorder diagnosis with an additional requirement of impaired functioning; ABCD asked about 鈥渄epression, that is, have they felt so low for a period of at least two weeks that they hardly ate or slept or couldn't work or do whatever they usually do?鈥�; Add Health asked about depression (not further defined); and UK Biobank asked about severe depression (not further defined). The primary family history measure used in the main analyses was a binary variable based on lifetime parental history (at least 1 biological parent with depression vs no parent with depression); this is in keeping with the previous meta-analysis in which parental history was the exposure in most studies.⁵ Three of the cohorts (TGS, ABCD, and Add Health) also collected data on biological grandparent history, enabling the creation of secondary exposure measures: (1) binary variable for at least 1 parent or grandparent with depression vs no parent or grandparent with depression and (2) 4-category dose variable⁸ representing the number of prior generations with depression (both generations; parent only; grandparent only; neither generation).

This was available in 3 cohorts (ABCD, Add Health, and UK Biobank). In ABCD, we created LDpred PRS¹⁴ based on a 2019 genome-wide association study (GWAS) meta-analysis of various depression phenotypes (self-reported or clinically confirmed).¹⁵ Details are provided in the eMethods in Supplement 1. The Add Health PRS was created centrally by the Add Health team¹⁶ based on the same 2019 meta-analysis. The UK Biobank PRS was not created from the 2019 meta-analysis because UK Biobank was a discovery cohort in that GWAS. We instead created the UK Biobank LDpred PRS from a 2018 GWAS of various depression phenotypes,¹⁷ using summary statistics that excluded UK Biobank participants. Details are provided in the eMethods in Supplement 1. All PRS were standardized as z scores (mean [SD] of 0 [1]) within each analysis sample.

The primary PRS analyses were restricted to White or European participants because of potential biases with applying PRS derived from single-ancestry GWAS in multiancestry samples. Secondary analyses in ABCD and Add Health included all participants (multiancestry sample), with ethnic group included as a covariate; it was not possible to do this in UK Biobank, as the genotyping imputation that was done in this cohort is not appropriate for multiancestry analysis.

In each cohort, all available tests of neurocognition were analyzed (details in the eMethods in Supplement 1). TGS wave 6 follow-up included a detailed battery of assessor-administered criterion-standard tests of speed, reasoning/intelligence, attention, executive function, and memory. This was administered only to participants who were assessed in person. ABCD year 2 follow-up included assessor-administered brief computerized tests of vocabulary, speed, attention/executive function, and memory using a mix of in-person and videoconferencing assessment. Add Health wave V follow-up included 3 assessor-administered brief bespoke measures of attention/executive function and memory, administered only to a representative subsample who were assessed in person. The UK Biobank imaging visit follow-up (in person) included self-administered brief computerized touchscreen tests of speed, reasoning, attention, executive function, and memory. Composite scores (representing the mean performance across tests within a cognitive domain) were also analyzed.

Age, sex, race and ethnicity, country of birth (as an indicator of linguistic/cultural variation, which may affect performance on US-designed or UK-designed cognitive tests), and duration between exposure and outcome waves were analyzed as potential confounders. We also extracted data on highest level of educational qualifications (except in ABCD, where all participants were still in education) and socioeconomic status (SES); these may act as mediators rather than confounders (ie, if they are influenced by parental/grandparental depression and in turn affect opportunities for cognitive development in offspring), and their potential role was evaluated by adding them as additional covariates in sensitivity analyses. For the purpose of secondary analyses, we classified participants according to whether they had a lifetime history of depression or of neurological disorders that may affect cognitive performance (eMethods in Supplement 1).

Analyses were conducted in Stata version 15 or version 17 (StataCorp) and took account of complex survey structure and relatedness in the data sets using weighting and cluster standard errors. Descriptive statistics are reported for the whole sample and split by family history status. The validity of the familial risk exposure measures was checked by examining their association with lifetime history of depression in the analysis sample. Analyses of the association between familial risk of depression and cognitive outcome were conducted using unadjusted and adjusted regression models. All but one of the cognitive outcome measures were z scores, so these were analyzed in linear models and the coefficients can be interpreted as standardized mean differences in cognitive score per unit of the exposure. The Prospective Memory score in UK Biobank was binary, so this was analyzed in a logistic model with results expressed as the odds ratio for a correct response per unit of the exposure. We report 95% CIs, and 2-tailed P values are reported with and without correction for multiple comparisons (false discovery rate maintained at .05). Full details of all models are provided in the eMethods in Supplement 1.

Demographic, health, and family history characteristics in each cohort are summarized in the Table. A total of 57鈥�308 participants were studied, including 87 from TGS (42 [48%] female; mean [SD] age, 19.7 [6.6] years), 10鈥�258 from ABCD (4899 [48%] female; mean [SD] age, 12.0 [0.7] years), 1064 from Add Health (584 [49%] female; mean [SD] age, 37.8 [1.9] years), and 45鈥�899 from UK Biobank (23鈥�605 [51%] female; mean [SD] age, 64.0 [7.7] years). Further descriptive statistics for all measures, stratified by family history status, are provided in eTables 1, 2, 3, and 4 in Supplement 1. The validity of the family history and PRS exposures was demonstrated by their clear associations with lifetime depression history in each analysis sample (eResults in Supplement 1).

Sample sizes were small and so estimates have relatively wide confidence intervals and should be interpreted with caution. In the primary adjusted models (parental history of depression), dual-task decrement had an effect size of medium magnitude (Figure 1). Additional adjustment for SES showed similar results on most tasks but shifted the results for IQ in a positive direction (eFigure 1C in Supplement 1). Using the dose exposure measure, the specific contrast analysis between the subgroups with both vs neither prior generations affected was strongest for dual-task decrement (eFigure 2B in Supplement 1). The exclusion of individuals with depression attenuated some estimates toward the null, with the exception of the visual delayed memory task (eFigure 3A in Supplement 1). After taking account of missing data, results showed lower performance on some attention/executive tasks (eFigure 4 in Supplement 1). It was not possible to conduct sensitivity analyses in an unrelated subgroup due to very small sample sizes.

The primary adjusted models (parental history) showed that performance on the picture memory task was lower in the group with a family history of depression, with verbal memory, the memory composite score, and processing speed also showing slightly lower performance (Figure 2A). Effect sizes were very small. These differences attenuated toward the null after additional adjustment for SES (eFigure 5B in Supplement 1). Participants with a family history of depression showed higher performance on vocabulary tasks in the unadjusted model and in the adjusted model including SES (eFigure 5 in Supplement 1). Compared with the primary models, the pattern of results across cognitive domains was similar in models that took into account grandparental as well as parental history, that excluded participants with depression or neurological disorders, that were restricted to unrelated participants, and that took account of missing data (eFigures 6, 7, and 8 in Supplement 1).

Delayed memory and the memory composite score showed lower performance in those with a family history (primary adjusted analysis for parental history), with small effect sizes (Figure 3A). This attenuated slightly after additional adjustment for education and SES (eFigure 9 in Supplement 1). Results were similar in secondary models taking into account grandparental history (eFigure 10 in Supplement 1), in models that excluded people with depression or neurological conditions (eFigure 11 in Supplement 1), and after accounting for missing data (eFigure 12B in Supplement 1). In models restricted to unrelated participants, all estimates shifted toward the null or positive direction (eFigure 12A in Supplement 1).

Figure 4A shows associations in the primary adjusted analyses between family (parental) history and lower performance on tests of processing speed, attention, and executive function. Effect sizes were very small. Results were essentially the same after additional adjustment for education and SES (eFigure 13 in Supplement 1). Results attenuated after excluding people with depression (though still showed lower performance), but there was little or no evidence of attenuation after excluding those with neurological conditions (eFigure 14 in Supplement 1) or restricting to unrelated participants (eFigure 15A in Supplement 1). Results were the same after accounting for missing data (eFigure 15B in Supplement 1).

Primary adjusted models in the White subgroup (Figure 2B) comparing those with higher vs lower polygenic risk for depression showed lower performance on picture memory, similar to the family history models, but also showed lower performance on picture vocabulary and lower performance on other tasks except verbal memory. Effect sizes were very small. After additional adjustment for SES (eFigure 16B in Supplement 1), the picture memory result was essentially unchanged, but the vocabulary estimates attenuated toward the null. Results were virtually the same in the larger multiancestry sample (eFigure 17 in Supplement 1). Compared with the primary models, results were almost the same in models that excluded participants with depression or neurological disorders, that were restricted to unrelated participants, and that took account of missing data (eFigures 18 and 19 in Supplement 1).

Primary adjusted models in the European subgroup (Figure 3B) comparing those with higher vs lower polygenic risk for depression showed no association with memory performance, but there was a positive association on the attention task (digit span) that had not been evident in the family history analyses. This remained evident after additional adjustment for education and SES (eFigure 20 in Supplement 1) and was also seen in the larger multiancestry sample (eFigure 21 in Supplement 1). Excluding participants with depression or neurological disorders, restricting to unrelated participants, and taking account of missing data did not make any appreciable difference to the results (eFigures 22 and 23 in Supplement 1).

Lower performance was seen on all but 2 of the cognitive tests in the primary adjusted models comparing those with higher vs lower polygenic risk for depression in the White British subgroup (Figure 4B). The general pattern of performance across domains was quite similar compared with the family history results with similarly small effect sizes, but several tasks showed statistically significant associations in the PRS analysis only (reasoning, digit span, memory). Additional adjustment for education and SES did not change the results (eFigure 24 in Supplement 1), nor did excluding participants with depression or neurological disorders (eFigure 25 in Supplement 1), restricting to unrelated participants (eFigure 26A in Supplement 1), or accounting for missing data (eFigure 26B in Supplement 1).

This study provides evidence for lower cognitive performance in people with familial risk of depression, which appears to manifest differently across the life span. In the younger cohorts (primarily ABCD and Add Health), family history of depression was associated with lower performance in the memory domain, albeit inconsistently, and there were indications that this may be partly because of educational and socioeconomic factors. In contrast, family history in the older UK Biobank cohort was associated with lower performance in the domains of processing speed, attention, and executive function but not memory, and there was little evidence of an influence of education or SES. Although there was a dose effect for depression itself, with participants with 2 prior generations affected showing greater odds of depression, this effect was not clearly evident with regard to the strength of association with cognitive performance.

The largest effect sizes between familial risk of depression and neurocognitive test performance were found in TGS, albeit with wider confidence intervals due to the small sample size. Effect sizes in the other cohorts were smaller than in TGS and the previous systematic review.⁵ Larger effect sizes in TGS may reflect the criterion-standard assessments used for both family history and cognitive testing, which increases measurement reliability, as well as the strict eligibility criteria in the first generation at cohort inception. The other cohorts had broader inclusion criteria and relied on responses from the participant or their parent to retrospective questions about family history; similarly, the PRS were created from GWAS of a broad depression phenotype. These factors may have biased associations toward the null, although the large sample sizes nevertheless enabled weaker associations to be detected from less reliable measures. This demonstrates the value of using population cohorts for this type of research, where criterion-standard phenotyping is not feasible at such a large scale. A major strength of our study is that we have used small-scale, carefully phenotyped data alongside big data sets with less detailed phenotyping. Using only the former may mean that results might not be replicable, while using only the latter risks generating large numbers of statistically significant yet trivial results that are not clinically meaningful.

To our knowledge, this study is the first to examine both polygenic risk and family history of depression in multiple cohorts; we found that both exposures showed similar results, although the PRS models showed associations with lower performance on a greater number of cognitive tests. An exception was the digit span test in Add Health, on which higher PRS was associated with better performance. We did not directly compare the contribution of family history and polygenic risk in the same models and so we cannot infer the relative strength of their distinct associations with cognitive outcome. This would require detailed multivariate modeling to take account of the mediating paths between genetic and nongenetic aspects of family history and their interactions.

The memory domain findings in the younger cohorts are congruent with neuroimaging markers in depression that also underpin memory function; hippocampal volumes are lower on average¹⁸ and cortical gray matter is thinner on average in various regions, including the temporal lobes,¹⁹ in people with depression, and we have previously shown in TGS that family history of depression is associated with hippocampal microstructure differences,²⁰ cortical thinning,²¹ and default mode network hyperconnectivity.²² The speed, attention, and executive function findings in the older UK Biobank cohort may point to differences in brain aging (eg, white matter disease), even in never-depressed participants, although evidence is currently lacking on neuroimaging in older people with high familial risk of depression, and this should be investigated in future UK Biobank analyses. It should also be borne in mind that the different pattern of results in UK Biobank may not be fully attributable to older age but rather to the other differences in the methods used in this cohort, including the use of a bespoke test battery with an emphasis on timed and executive function tasks.

There was little impact on the results after excluding participants who themselves had depression. Only UK Biobank showed clear evidence of attenuation in those models but not enough to negate the findings. This suggests that lifetime experience of depression may have some influence on cognitive outcomes, especially in older participants, but other factors must be at play.

Education and SES may explain some of the association. This was evident in the 3 younger cohorts and may reflect a mediating role of household/neighborhood environment, resource access, and opportunities in influencing cognitive development and reserve in families affected by parental or multigenerational depression. This warrants further research within a mediation framework, with important implications for early intervention on potentially modifiable intermediate risk factors.

This study has limitations. The 4 cohorts we analyzed have various strengths and limitations with regard to sample size, representativeness, and depth and completeness of measures, which means that it is difficult to disentangle age-related and generational effects from methodological differences when interpreting the patterns of findings. TGS was the only cohort with clinically confirmed depression diagnoses in all generations, but PRS data are not available at present in this cohort. We focused on biological family history and so have not captured the influence of nonbiological relatives, such as step-parents, in the household. It would also be of interest to analyze the number of affected biological relatives in detail (eg, whether 1 or both parents had a depression history), but this was not feasible owing to the amount of missing data. We aimed to analyze exposures and outcomes from different assessment waves (to reduce the possibility of reverse causality and allow for future mediation analyses to examine intermediate measures, such as brain imaging), but data from different waves were not available in Add Health.

In this study, whether assessed by family history or genetic data, depression in prior generations was associated with lower cognitive performance in offspring. The next challenge is to elucidate the pathways by which this arises, which may include genetic and environmental determinants and moderators of brain development and brain aging, and potentially modifiable social and lifestyle factors at play across the life span. These and other cohorts enable such research at a scale and depth never before possible, opening new research directions for prevention and early intervention in at-risk individuals.

Accepted for Publication: February 16, 2023.

Published Online: April 19, 2023. doi:10.1001/jamapsychiatry.2023.0716

Corresponding Author: Breda Cullen, PhD, School of Health and Wellbeing, University of Glasgow, Clarice Pears Building, 90 Byres Rd, Glasgow G12 8TB, United Kingdom (breda.cullen@glasgow.ac.uk).

Author Contributions: Dr Cullen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Cullen, van Dijk, Whalley, Cavanagh, Weissman.

Acquisition, analysis, or interpretation of data: Cullen, Gameroff, Ward, Bailey, D. Lyall, L. Lyall, MacSweeney, Murphy, Sangha, Shen, Strawbridge, van Dijk, Zhu, Smith, Talati, Weissman.

Drafting of the manuscript: Cullen, Ward, D. Lyall, Talati, Whalley, Cavanagh.

Critical revision of the manuscript for important intellectual content: Cullen, Gameroff, Ward, Bailey, L. Lyall, MacSweeney, Murphy, Sangha, Shen, Strawbridge, van Dijk, Zhu, Smith, Talati, Whalley, Weissman.

Statistical analysis: Cullen.

Obtained funding: Cullen, Weissman.

Administrative, technical, or material support: Gameroff, Ward, D. Lyall, Shen, Smith.

Study supervision: Talati, Whalley, Cavanagh.

Conflict of Interest Disclosures: Dr Cullen has received grants from the Scottish Executive Chief Scientist Office during the conduct of the study. Dr Weissman has received grants from the National Institute of Mental Health during the conduct of the study; grants from John Templeton Foundation, National Institute of Mental Health, and Brain and Behavior Foundation; and royalties from Oxford Press, APA Publishing, Perseus Press, and Multihealth Systems outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported in part by grant DTF/14/03 from the Scottish Executive Chief Scientist Office (Dr Cullen) and grant R01MH036197 from the National Institute of Mental Health (Dr Weissman). Dr L. Lyall is supported by a JMAS Sim Fellowship from the Royal College of Physicians of Edinburgh and a Lord Kelvin Adam Smith Fellowship from the University of Glasgow. Ms MacSweeney is supported by a Mental Health Research UK PhD Studentship. Dr Strawbridge is supported by a UK Research and Innovation Health Data Research-UK Fellowship MR/S003061/1 and a Lord Kelvin Adam Smith Fellowship from the University of Glasgow. Dr van Dijk is funded by National Institute of Mental Health grant K99MH129611 and American Foundation for Suicide Prevention Young Investigator Award YIG-R-001-19.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 2.

Additional Information: Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development (ABCD) study, held in the National Institute of Mental Health data archive. This is a multisite, longitudinal study designed to recruit more than 10鈥�000 children aged 9 to 10 years and observe them over 10 years into early adulthood. The ABCD study is supported by the National Institutes of Health and additional federal partners under award numbers U01DA041048, U01DA050989, U01DA051016, U01DA041022, U01DA051018, U01DA051037, U01DA050987, U01DA041174, U01DA041106, U01DA041117, U01DA041028, U01DA041134, U01DA050988, U01DA051039, U01DA041156, U01DA041025, U01DA041120, U01DA051038, U01DA041148, U01DA041093, U01DA041089, U24DA041123, and U24DA041147. A full list of supporters is available at . A listing of participating sites and a complete listing of the study investigators can be found at . ABCD Consortium Investigators designed and implemented the study and/or provided data but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the National Institutes of Health or ABCD Consortium Investigators. The ABCD data repository grows and changes over time. The ABCD data used in this report came from Annual Release 4.0 (study number 1299). The National Longitudinal Study of Adolescent to Adult Health (Add Health) is directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01 AG071448 (Hummer) and U01 AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Waves I to V data are from the Add Health Program Project, grant P01 HD31921 (Harris) from Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Add Health was designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill. This research has been conducted using the UK Biobank Resource under Application Number 11332 (Dr Cullen). UK Biobank was established by the Wellcome Trust medical charity, Medical Research Council, Department of Health, Scottish Government, and the Northwest Regional Development Agency. It has also had funding from the Welsh Government, British Heart Foundation, Cancer Research UK, and Diabetes UK.