Classification of Suicide Attempt Risk Using Environmental and Lifestyle Factors in 3 Large Youth Cohorts

Elina Visoki; Tyler M Moore; Xinhe Zhang; Kate T Tran; Christina Ly; Mārtiņš M Gataviņš; Grace E DiDomenico; Leah Brogan; Joel A Fein; Varun Warrier; Sinan Guloksuz; Ran Barzilay

doi:10.1001/jamapsychiatry.2024.1887

. 2024 Jul 17;81(10):1020–1029. doi: 10.1001/jamapsychiatry.2024.1887

Classification of Suicide Attempt Risk Using Environmental and Lifestyle Factors in 3 Large Youth Cohorts

Elina Visoki ^1,², Tyler M Moore ^2,³, Xinhe Zhang ⁴, Kate T Tran ^1,², Christina Ly ², Mārtiņš M Gataviņš ^1,², Grace E DiDomenico ^1,², Leah Brogan ⁵, Joel A Fein ^5,⁶, Varun Warrier ⁴, Sinan Guloksuz ^7,⁸, Ran Barzilay ^1,^2,^3,^✉

¹Department of Child and Adolescent Psychiatry and Behavioral Sciences, Children’s Hospital of Philadelphia (CHOP), Philadelphia, Pennsylvania

²Lifespan Brain Institute of CHOP and Penn Medicine, Philadelphia, Pennsylvania

³Department of Psychiatry, Perelman School of Medicine, University of Pennsylvania, Philadelphia

⁴Cambridge University, Cambridge, United Kingdom

⁵Center for Violence Prevention, Children’s Hospital of Philadelphia (CHOP), Philadelphia, Pennsylvania

⁶Department of Pediatrics, Perelman School of Medicine, University of Pennsylvania, Philadelphia

⁷Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Maastricht University Medical Center, Maastricht, the Netherlands

⁸Department of Psychiatry, Yale University School of Medicine, New Haven, Connecticut

Accepted for Publication: April 30, 2024.

Published Online: July 17, 2024. doi:10.1001/jamapsychiatry.2024.1887

^✉

Corresponding Author: Ran Barzilay, MD, PhD, Department of Psychiatry, 10th Floor, Gates Pavilion, Hospital of the University of Pennsylvania, 34th and Spruce Street, Philadelphia, PA 19104 (ran.barzilay@pennmedicine.upenn.edu).

Author Contributions: Ms Visoki and Dr Barzilay had full access to ABCD and CHOP-ED data, and Ms Zhang had full access to MCS data. Ms Visoki and Dr Barzilay take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Visoki, Moore, Barzilay.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Visoki, DiDomenico, Barzilay.

Critical review of the manuscript for important intellectual content: Visoki, Moore, Zhang, Tran, Ly, Gataviņš, Brogan, Fein, Warrier, Guloksuz, Barzilay.

Statistical analysis: Visoki, Moore, Zhang, Tran, Warrier, Barzilay.

Obtained funding: Barzilay.

Administrative, technical, or material support: Tran, Ly, Gataviņš, DiDomenico, Brogan.

Supervision: Moore, Warrier, Guloksuz, Barzilay.

Conflict of Interest Disclosures: Dr Fein reported having an interest in any licensing of the BHS (2.5% of any licensing fees). Dr Barzilay reported receiving advisory board fees from and having stock ownership in Taliaz Health and receiving consulting fees from Zynerba Pharmaceuticals outside the submitted work. No other disclosures were reported.

Funding/Support: Data used in the preparation of this article were obtained from the Adolescent Brain and Cognitive DevelopmentSM (ABCD) Study, which 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, U24DA041147. A full list of supporters is available at https://abcdstudy.org/federal-partners.html. This study was supported by the National Institute of Mental Health (grant numbers K23MH120437 [Dr Barzilay], P50MH115838 [Dr Barzilay], R01MH117014 [Dr Moore]); the American Foundation for Suicide Prevention (grant number SRG-006-22 [Dr Barzilay]); the Lifespan Brain Institute of Children’s Hospital of Philadelphia and Penn Medicine, University of Pennsylvania; and the YOUTH-GEMs project, funded by the European Union’s Horizon Europe program under grant agreement number 101057182 (Dr Guloksuz).

Role of the Funder/Sponsor: The ABCD Consortium investigators designed and implemented the study and/or provided data but did not participate in analysis or writing of this report. The funding organizations 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.

Disclaimer: 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. Neither the Centre for Longitudinal Studies nor the UK Data Service bear any responsibility for the analysis or interpretation of study data.

Data Sharing Statement: See Supplement 3.

Additional Contributions: We thank the Centre for Longitudinal Studies, UCL Social Research Institute, for the use of the study data and to the UK Data Service for making them available.

^✉

Corresponding author.

PMCID: PMC11255979 PMID: 39018056

This cohort study investigates the cumulative contribution of environmental and lifestyle (exposome) factors to risk of suicide attempt among youth.

Key Points

Question

What is the cumulative contribution of environmental and lifestyle (exposome) factors to risk of suicide attempt among youth?

Findings

In this cohort study including 40 364 youth, aggregate exposomic scores of risk and protective factors consistently explained substantial variance in suicide attempt over and above clinical risk benchmarks in diverse clinical and population youth cohorts from the US and the UK.

Meaning

Results of this cohort study suggest that exposomic scores of suicide attempt provide a generalizable method for risk classification that can be applied in clinical or population settings and can advance reproducibility in suicide research.

Abstract

Importance

Suicide is the third-leading cause of death among US adolescents. Environmental and lifestyle factors influence suicidal behavior and can inform risk classification, yet quantifying and incorporating them in risk assessment presents a significant challenge for reproducibility and clinical translation.

Objective

To quantify the aggregate contribution of environmental and lifestyle factors to youth suicide attempt risk classification.

Design, Setting, and Participants

This was a cohort study in 3 youth samples: 2 national longitudinal cohorts from the US and the UK and 1 clinical cohort from a tertiary pediatric US hospital. An exposome-wide association study (ExWAS) approach was used to identify risk and protective factors and compute aggregate exposomic scores. Logistic regression models were applied to test associations and model fit of exposomic scores with suicide attempts in independent data. Youth from the Adolescent Brain Cognitive Development (ABCD) study, the UK Millennium Cohort Study (MCS), and the Children’s Hospital of Philadelphia emergency department (CHOP-ED) were included in the study.

Exposures

A single-weighted exposomic score that sums significant risk and protective environmental/lifestyle factors.

Main Outcome and Measure

Self-reported suicide attempt.

Results

A total of 40 364 youth were included in this analysis: 11 564 from the ABCD study (3 waves of assessment; mean [SD] age, 12.0 [0.7] years; 6034 male [52.2%]; 344 attempted suicide [3.0%]; 1154 environmental/lifestyle factors were included in the ABCD study), 9000 from the MCS cohort (mean [SD] age, 17.2 [0.3] years; 4593 female [51.0%]; 661 attempted suicide [7.3%]; 2864 environmental/lifestyle factors were included in the MCS cohort), and 19 800 from the CHOP-ED cohort (mean [SD] age, 15.3 [1.5] years; 12 937 female [65.3%]; 2051 attempted suicide [10.4%]; 36 environmental/lifestyle factors were included in the CHOP-ED cohort). In the ABCD discovery subsample, ExWAS identified 99 risk and protective exposures significantly associated with suicide attempt. A single weighted exposomic score that sums significant risk and protective exposures was associated with suicide attempt in an independent ABCD testing subsample (odds ratio [OR], 2.2; 95% CI, 2.0-2.6; P < .001) and explained 17.6% of the variance (based on regression pseudo-R²) in suicide attempt over and above that explained by age, sex, race, and ethnicity (2.8%) and by family history of suicide (6.3%). Findings were consistent in the MCS and CHOP-ED cohorts (explaining 22.6% and 19.3% of the variance in suicide attempt, respectively) despite clinical, demographic, and exposure differences. In all cohorts, compared with youth at the median quintile of the exposomic score, youth at the top fifth quintile were substantially more likely to have made a suicide attempt (OR, 4.3; 95% CI, 2.6-7.2 in the ABCD study; OR, 3.8; 95% CI, 2.7-5.3 in the MCS cohort; OR, 5.8; 95% CI, 4.7-7.1 in the CHOP-ED cohort).

Conclusions and Relevance

Results suggest that exposomic scores of suicide attempt provided a generalizable method for risk classification that can be applied in diverse samples from clinical or population settings.

Introduction

Suicide is among the 3 leading causes of death among US youth.¹ Suicide risk is influenced by a combination of distal and proximal environmental factors and by biological liability.² Several individual early-life environmental adversities are established clinical risk indicators for suicide,² and specific adversities were linked to adolescent suicide attempt, including physical and sexual abuse,³ bullying,⁴ and cyberbullying.⁵ However, studying individual exposures in isolation does not capture the complexity of environment, which has a multilevel structure. Indeed, theoretical frameworks that address the contribution of the environment to human development, such as the Bronfenbrenner ecological system theory,⁶ highlight the need to consider multiple levels of exposures in order to understand human development.⁷ To capture the contribution of environment to the development of suicidal behavior, it is, therefore, critical to adopt a social-ecological framework that incorporates information on multiple environmental risk and protective factors.⁸ However, such efforts of aggregating environmental data have been limited due to challenges of measuring and quantifying environment.⁹ In contrast, advances in statistical genetics methods have made substantial progress in quantifying aggregate genetic risk, moving from the concept of individual candidate genes to polygenic models of suicide attempt risk.¹⁰ Novel reproducible methods are needed to consistently quantify polyenvironmental contributions to suicide attempts to improve risk classification.

The exposome was conceptualized as the environmental analogue of the genome. Exposome encompasses all nongenetic exposures throughout the lifespan and can explain health-related outcomes.^11,12 Unlike the genome, however, measurement of environment varies across studies, whereby the term environment is defined differently, and its measurement is limited to certain facets of exposures (eg, trauma, maltreatment). Novel data-driven approaches such as exposome-wide association studies (ExWAS) allow identification of multiple exposures for disease risk and enable calculation of exposomic (polyenvironmental) risk.¹³ ExWAS share conceptual similarity with genome-wide association studies¹⁰ that identify genetic risk markers that can be aggregated to quantify polygenic disease risk.¹⁴ Exposomic risk models have shown promise in explaining disease outcomes in various medical fields^15,16,17 and are gaining interest in psychiatry.^18,19 One advantage of exposomic risk models is that they may reveal how disparities in environmental burden are linked to heightened suicide risk among marginalized youth (eg, Black youth²⁰ and sexual or gender minoritized youth²¹).

Here, we applied an ExWAS approach to model environmental contribution to suicide attempt in 3 large youth cohorts that differ significantly in terms of age (from early to postadolescence), settings (research observational studies and electronic health records), and geography (US and UK). We used ExWAS discovery findings on risk and protective exposures to quantify aggregate risk and applied exposomic scores in independent data. We tested the (1) potential of exposomic scores to explain variance in youth suicide attempt and (2) generalizability of this approach in different diverse cohorts.

Methods

Study Design

The Adolescent Brain Cognitive Development Study²² (ABCD) study protocol was approved by the University of California, San Diego, institutional review board (IRB), and the current analysis was exempted from a full review by the University of Pennsylvania IRB owing to the use of deidentified patient data. Data collection for the Millennium Cohort Study²³ (MCS) cohort was approved by the UK National Health Service Research Ethics Committee. Written consent was obtained from all parents in the ABCD and MCS cohorts. Analyses of deidentified Children’s Hospital of Philadelphia emergency department (CHOP-ED) data were approved by the CHOP IRB.

We applied a similar analytic design in 3 cohorts. After compiling exposures and curating the data for ExWAS, we divided each sample into discovery and testing subsamples, balanced by demographics and suicide attempt rates. Then, we conducted an ExWAS in each discovery subsample and identified risk and protective exposures significantly associated with suicide attempt. We then calculated aggregate exposomic risk scores (ERS) for each participant in the testing subsamples, using the weights (coefficients) obtained from the ExWAS. Next, we tested associations of ERS with suicide attempt in the testing subsamples. Lastly, we tested differential exposure to exposomic risk and tested differential association of exposomic risk with suicide attempt across marginalized subpopulations. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Study Participants

We included participants from 3 cohorts that systematically assessed suicide attempts: the ABCD study,²² the MCS cohort,^23,24 and a youth cohort from CHOP-ED.²⁵ Race and ethnicity were reported by caregivers. In the ABCD and CHOP-ED cohorts, the participant’s caregiver identified the child’s race with the following races and ethnicities (multiselection): American Indian or Alaska Native, Asian, Black, Hispanic, multiracial, Native Hawaiian or Other Pacific Islander, White, or other (not specified). In the MCS cohort, participants identified with the following: Asian (British Indian, British Pakistani, etc), Black, multiracial, White or other ethnic group.

Measures

Exposures

We included individual-level environmental and lifestyle factors with less than 10% missing data (based on prior works²⁶). We considered all environmental or lifestyle experiences that were endorsed by youth (in all 3 cohorts) or their caregivers (in the ABCD and MCS studies) as exposures. Exposures encompassed different levels of exposure including interpersonal experiences, family, school, and neighborhood environment. Continuous exposures were standardized, whereas most ordinal exposures were converted into binary variables. We removed features with less than 1% nonzero values and filtered out exposures that exhibited a correlation exceeding 0.9 to address collinearity, as in prior works.²⁷

Primary Outcome

We chose self-reported lifetime history of suicide attempt as the primary outcome because it is a clinical indicator of adolescent suicide risk.²⁸ In the ABCD study, history of suicide attempt was captured using the Kiddie Schedule for Affective Disorders and Schizophrenia structured interview.²² In the MCS cohort,²⁴ we used the question “Have you ever hurt yourself on purpose in an attempt to end your life?” and in the CHOP-ED cohort, we used the question “Have you ever tried to kill yourself?”²⁵

Missing Data

In discovery subsamples, we used listwise deletion of missing data of exposures. In the ABCD study, this was done by time point, such that a missing value affected only the time point in which data were missing, avoiding participant-level listwise deletion. In the testing subsamples, to prevent the computation of missing data as 0 in the ERS, we imputed data on exposures using the missForest library in R (R Project for Statistical Computing).

Statistical Analyses

ExWAS and Calculation of ERS

In each cohort, we conducted an ExWAS and calculated ERS as described previously.²⁹ Briefly, we first ran an ExWAS using half of the cohort (discovery subsample). Each exposure (independent variable) was tested individually for its association with suicide attempt (dependent variable) in a logistic regression model, adjusting for age and sex. This allowed derivation of a regression coefficient and P value for the association of each exposure included in the ExWAS with suicide attempt.

After the ExWAS, in each cohort, we selected all risk and protective exposures that remained associated with suicide attempt following Benjamini-Hochberg false discovery rate (FDR) correction (P threshold = .05) and calculated ERS for each participant in the testing subsamples by aggregating their exposures multiplied by their corresponding regression coefficient obtained from the ExWAS (positive or negative coefficient for risk or protective factors, respectively).

All P values were 2-sided, and a P value <.05 was considered significant. All analyses were conducted using R statistical software, version 4.2 (R Project for Statistical Computing).

Testing Associations of ERS With Suicide Attempt

We used logistic regression in the testing subsample of each cohort with ERS (independent variable) and suicide attempt (dependent variable), adjusting for demographics (age, sex, race, Hispanic ethnicity).

In the ABCD study, we estimated 4 mixed models. The base model tested association of demographics with suicide attempt (model 1). We then added ERS to the model (model 2). We also tested the association between parental suicide/attempted suicide with the youth suicide attempt (model 3). This model was used as a benchmark to evaluate the additive association of ERS over and above family history (model 4). In the MCS and CHOP-ED cohorts, we ran 2 models (models 1 and 2) as family history was not available.

In all 3 cohorts, we used adjusted pseudo-R² (referred to as R² hereafter; Nakagawa R² for mixed-effects logistic regression models³⁰; Nagelkerke R² for logistic regression models³¹) to test performance of ERS in explaining variance in suicide attempt. To test additive contributions of ERS to explain suicide attempt, we compared goodness of fit using the χ² likelihood ratio test.

Additionally, we compared the area under the receiver operating characteristic curve (AUROC) between models with ERS vs models that only included demographics in each of the 3 cohorts. Note that in the ABCD and MCS cohorts, we calculated a marginal AUROC using only the fixed effects of the models, excluding the random effects of site/region, family, and participant.

Differential Exposures and Associations of Exposomic Risk With Suicide Attempt Across Subpopulations

Because minoritized youth are at increased risk for suicide attempts,³² we examined differences in ERS across race and ethnicity groups (non-Hispanic Black, Hispanic, and non-Hispanic White), and between sexual and gender minority youth (identifying as lesbian, gay, bisexual, or transgender) and their heterosexual cisgender peers (data on sexual and gender minoritized group were not available in the CHOP-ED cohort). We examined differential associations of exposomic risk with suicide attempt by testing the interaction of ERS with race, ethnicity, or sexual or gender minoritized group, and conducted stratified analyses within subpopulations.

Sensitivity Analyses

To assess potential confounding of psychopathology on the association between exposomic risk and suicide attempt, we ran all main models from the ABCD cohort adjusting for general psychopathology assessed with the parent-report total Childhood Behavior Checklist score.²²

To address the possibility that our choice of P value threshold affected the results, we recalculated ERS using all risk or protective exposures that we conceived to have clinical significance (ie, odds ratio [OR] >1.2 or OR <0.8), regardless of P value.

Results

A total of 40 364 youth were included in this analysis: 11 564 from the ABCD study (mean [SD] age, 12.0 [0.7] years; 5530 female [47.8%]; 6034 male [52.2%]; 1154 environmental/lifestyle factors), 9000 from the MCS cohort (mean [SD] age, 17.2 [0.3] years; 4593 female [51.0%]; 4407 male [49.0%]; 2864 environmental/lifestyle factors), and 19 800 from the CHOP-ED cohort (mean [SD] age, 15.3 [1.5] years; 12 937 female [65.3%]; 6863 male [34.7%]; 36 environmental/lifestyle factors) (Figure 1 and Table 1). The total number of exposures included in the ExWAS were 447 in the ABCD study (1154 before data curation), 622 in the MCS cohort (2864 before data curation), and 17 in the CHOP-ED cohort (36 before data curation). Description of exposures is detailed in eMethods in Supplement 1 and eTables 1 to 3 in Supplement 2.

Figure 1. — ABCD indicates Adolescent Brain Cognitive Development study; CHOP-ED, Children’s Hospital of Philadelphia emergency department.

Table 1. Demographic Comparison of Those Who Attempted Suicide With Their Peers in the 3 Cohorts.

Characteristic	ABCD				CHOP-ED				MCS
Characteristic	Overall	No suicide attempt^a	Suicide attempt^b	P value^c	Overall	No suicide attempt	Suicide attempt	P value^c	Overall	No suicide attempt	Suicide attempt	P value^c
No.	11 564	11 220	344	NA	19 800	17 749	2051	NA	9000	8339	661	NA
Age, mean (SD), y^d	12.00 (0.66)	12.00 (0.66)	12.06 (0.64)	.11	15.30 (1.53)	15.33 (1.52)	15.07 (1.56)	<.001	17.17 (0.33)	17.17 (0.33)	17.18 (0.34)	.44
Female sex assigned at birth, No. (%)	5530 (47.8)	5364 (47.8)	166 (48.3)	.91	12937 (65.3)	11348 (63.9)	1589 (77.5)	<.001	4593 (51.0)	4125 (49.5)	468 (70.8)	<.001
American Indian or Alaska Native, No. (%)	401 (3.5)	385 (3.4)	16 (4.7)	.29	10 (0.1)	10 (0.1)	0 (0.0)	.61^f	NA	NA	NA	NA
Asian race, No. (%)	735 (6.4)	715 (6.4)	20 (5.8)	.76	505 (2.6)	447 (2.5)	58 (2.8)	.44	992 (11.0)	950 (11.4)	42 (6.4)	<.001
Black race, No. (%)	2443 (21.1)	2340 (20.9)	103 (29.9)	<.001	11130 (56.2)	9888 (55.7)	1242 (60.6)	<.001	288 (3.2)	276 (3.3)	12 (1.8)	.047
Hispanic ethnicity, No. (%)	2348 (20.6)	2272 (20.5)	76 (22.2)	.48	1405 (7.1)	1216 (6.9)	189 (9.2)	<.001	NA	NA	NA	NA
Multiracial, No. (%)	1410 (12.2)	1356 (12.1)	54 (15.7)	.05	239 (1.2)	209 (1.2)	30 (1.5)	.31	241 (2.7)	223 (2.7)	18 (2.7)	>.99
Native Hawaiian or Other Pacific Islanders, No. (%)	69 (0.6)	65 (0.6)	4 (1.2)	.30	9 (0.0)	5 (0.0)	4 (0.2)	.009^f	NA	NA	NA	NA
White race, No. (%)	8627 (74.6)	8391 (74.8)	236 (68.6)	.01	6479 (32.7)	5942 (33.5)	537 (26.2)	<.001	7432 (82.6)	6847 (82.1)	585 (88.5)	<.001
Other race, No. (%)^e	752 (6.5)	726 (6.5)	26 (7.6)	.49	1426 (7.2)	1246 (7.0)	180 (8.8)	.004	47 (0.5)	43 (0.5)	4 (0.6)	.78 ^f
Sexual or gender minority^g	804 (8.2)	724 (7.6)	80 (27.7)	<.001	NA	NA	NA	NA	1882 (20.9)	1569 (18.8)	313 (47.5)	<.001

Open in a new tab

Abbreviations: ABCD, Adolescent Brain Cognitive Development study; CHOP-ED, Children’s Hospital of Philadelphia emergency department; MCS, Millennium Cohort Study; NA, not applicable; SA, suicide attempt.

^{^a}

These participants responded no to history of suicide attempt in at least 1 of the 3 ABCD assessments and never responded yes.

^{^b}

These participants responded at least 1 yes to history of suicide attempt in 3 ABCD assessments.

^{^c}

Indicates t test or χ² test.

^{^d}

In the ABCD study, the average age in the 2-year follow-up (third assessment). In MCS, the average age at the seventh data collection sweep when suicide attempt was assessed.

^{^e}

Other race was not specified.

^{^f}

Indicates Fisher exact test.

^{^g}

In the ABCD study, values are from the 2-year follow-up (third assessment). In the MCS cohort, values are from the seventh data collection sweep at age 17 years.

Suicide Attempt Rates

Rate of suicide attempt was 3.0% (344 of 11 564) in ABCD study, 10.4% in the CHOP-ED cohort (2051 of 19 800), and 7.3% in the MCS cohort (661 of 9000) (Table 1).

ExWAS in the ABCD Study

ExWAS identified 99 exposures that significantly associated with suicide attempt after FDR correction (Figure 2), of which 73 were risk (OR >1) and 26 were protective (OR <1) exposures (eFigure 1 in Supplement 1). The exposure “survivor of crime, violence, or assault” showed the greatest risk association (OR, 7.3; 95% CI, 3.2-16.5; corrected P < .001), and the exposure “feels safe at school” showed the greatest protective association (OR, 0.3; 95% CI, 0.2-0.4; corrected P < .001). The 5 top-ranking risk and protective exposures are detailed in Figure 2 (full list in eTable 1 in Supplement 2).

ERS for Suicide Attempt in the ABCD Study

ERS was significantly associated with suicide attempt in the testing subsample (OR, 2.2; 95% CI, 2.0-2.6; P < .001) (model 2 in Table 2). This model explained 17.6% of the variance in suicide attempt (Nakagawa R²), significantly more than the 2.8% variance explained by demographics alone (model 1 in Table 2).

Table 2. Association of Exposomic Risk Scores (ERS) With Suicide Attempt (SA) in the Adolescent Brain Cognitive Development (ABCD) Study^a.

Predictors	Model 1 (n = 5719)^b		Model 2 (n = 5719)^c		Model 3 (n = 5424)^d		Model 4 (n = 5424)^e
Predictors	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value
Age	1.01 (1.00-1.02)	.18	0.98 (0.96-0.99)	.003	1.01 (1.00-1.02)	.18	0.98 (0.96-0.99)	.004
Male sex	0.83 (0.55-1.27)	.40	0.77 (0.53-1.13)	.19	0.83 (0.53-1.29)	.40	0.75 (0.50-1.13)	.17
Black race	1.57 (0.83-2.98)	.16	0.99 (0.55-1.78)	.97	1.35 (0.69-2.65)	.38	0.86 (0.46-1.60)	.63
Hispanic ethnicity	1.48 (0.89-2.49)	.13	1.29 (0.81-2.06)	.28	1.59 (0.93-2.71)	.09	1.37 (0.84-2.24)	.20
White race	0.84 (0.46-1.54)	.58	0.87 (0.50-1.50)	.61	0.69 (0.37-1.28)	.24	0.72 (0.41-1.28)	.27
ERS	NA	NA	2.23 (1.95-2.55)	<.001	NA	NA	2.25 (1.95-2.59)	<.001
Parental history of suicide/SA	NA	NA	NA	NA	3.94 (2.03-7.64)	<.001	2.68 (1.48-4.83)	.001
Marginal R², %	2.8		17.6		6.3		20.1

Open in a new tab

Abbreviations: NA, not applicable; OR, odds ratio.

^{^a}

Four mixed-effects logistic regression models were run to examine the association of exposomic risk (independent variable) with suicide attempt (dependent variable) in relation to demographics and family history. All models took the 4-level hierarchical and 3-assessment longitudinal nature of the ABCD data into account (the data was first nested according to participant, then according to family, and finally according to the research site). All models assumed a similar time trend across all participants and only allowed random intercepts in all mixed models.

^{^b}

Base model with demographic characteristics.

^{^c}

Model 2 tests association of ERS with suicide attempt over and above demographic characteristics.

^{^d}

Model 3 tests association of parental history of suicide or suicide attempt over and above demographic characteristics.

^{^e}

Model 4 tests association of ERS with suicide attempt over and above parental history of suicide or suicide attempt and of demographic characteristics.

Compared with a clinical risk indicator (parental suicide/attempted suicide) that explained 6.3% of the variance (model 3 in Table 2), the addition of ERS significantly increased the variance explained to 20.1% (model 4 in Table 2).

External Validation of Approach in the CHOP-ED and MCS Cohorts

We generalized our approach in 2 external cohorts with different demographic characteristics (Table 1). Cohorts also differed in the number of exposures they collected (36 in the CHOP-ED cohort and 2864 in the MCS cohort, compared with 1154 in the ABCD study). We included 17 exposures in CHOP-ED ExWAS and 622 exposures in MCS ExWAS (after data curation). ExWAS revealed 15 and 197 exposures associated with suicide attempt in the CHOP-ED and MCS cohorts, respectively.

Association of ERS with suicide attempt in the independent testing subsamples of the CHOP-ED and MCS cohorts was consistent with ABCD findings (in the CHOP-ED cohort: OR, 2.3; 95% CI, 2.2-2.4; P < .001 and in the MCS cohort: OR, 2.4; 95% CI, 2.1-2.7; P < .001) (Figure 3A and eTable 4 in Supplement 1). Similar to ABCD findings, addition of ERS to base models that included only demographics increased the variance explained in suicide attempt substantially (in the CHOP-ED cohort, from 2.5%-19.3%; in the MCS cohort, from 5.6%-22.6%) (Figure 3B and eTable 4 in Supplement 1). In each of the 3 cohorts, the variance explained in suicide attempt by the ERS was greater than that explained by individual risk exposures that emerged as the top-ranking exposure in the ExWAS, as revealed by delta R² values between 8.9% and 17.0% in favor of the ERS (eTable 5 in Supplement 1).

Figure 3. — In each cohort, we tested associations of exposomic scores (independent variable) with suicide attempt (dependent variable) adjusting for demographic characteristics. Analyses were done in independent testing subsamples within each of the 3 cohorts after an exposome-wide association study was run in an independent discovery subsample to minimize overfitting. A, Effect size of the association of exposomic score with suicide attempt in each of the 3 cohorts. B, Added variance explained by exposomic scores compared with variance explained solely by demographic characteristics in all 3 cohorts. C, Association of exposomic risk scores with suicide attempt in each of 5 different exposomic scores quintiles in relation to the median quintile. ABCD indicates Adolescent Brain Cognitive Development study; CHOP-ED, Children’s Hospital of Philadelphia emergency department; MCS, Millennium Cohort Study.

To appreciate the risk classification properties of the ERS, we categorized ERS using quintiles and measured the case-control ORs using the middle quintile (median ERS) as reference. Compared with the median, the top fifth quintile was associated with suicide attempt with a large effect size in all 3 cohorts (OR, 4.3; 95% CI, 2.6-7.2 in the ABCD study; OR, 5.8; 95% CI, 4.7-7.1 in the CHOP-ED cohort; OR, 3.8; 95% CI, 2.7-5.3 in the MCS cohort) (Figure 3C).

Last, we examined the difference in classification of those who attempted suicide in each of the 3 cohorts using the AUROC. In each cohort, the AUROC of models that included ERS was significantly better than models that only included demographics (in the ABCD study, 0.82 vs 0.61; in the CHOP-ED cohort, 0.78 vs 0.61; in the MCS cohort, 0.76 vs 0.61; all P <.001).

Differential Exposure and Associations of Exposomic Risk With Suicide Attempt

In the ABCD study, we found greater ERS among non-Hispanic Black youth and among Hispanic youth compared with non-Hispanic White youth, and greater ERS among sexual and gender minoritized youth compared with their peers (eFigure 2 in Supplement 1). Similarly, we observed disparities in the CHOP-ED (eTable 6 in Supplement 1) and MCS (eTable 7 in Supplement 1) cohorts with generally greater ERS among marginalized youth.

We did not find evidence for differential association of exposomic risk with suicide attempt across race or ethnicity in any of the cohorts (for all ERS × Black race and ERS × ethnicity interactions) (eTable 8 in Supplement 1). We observed a significant ERS × sexual and gender minoritized group interaction in association with suicide attempt in the ABCD study (OR, 0.7; 95% CI, 0.6-1.0; P = .03), whereby the association of ERS with suicide attempt was slightly less strong among sexual and gender minoritized youth (OR, 1.9; 95% CI, 1.4-2.5) compared with their heterosexual cisgender peers (OR, 2.2; 95% CI, 1.9-2.6). This ERS × sexual and gender minoritized group interaction was not observed in MCS. Stratified models are detailed in eTables 9 to 11 in Supplement 1.

Sensitivity Analyses

We ran 2 sensitivity analyses on the ABCD study data. First, addition of youth overall psychopathology to models using exposomic risk (independent variable) explaining suicide attempt (dependent variable) did not change the associations between ERS and suicide attempt (OR, 2.0; 95% CI, 1.7-2.3 compared with OR, 2.2; 95% CI, 2.0-2.6 in the main analysis model without psychopathology) (eTable 12 in Supplement 1).

Second, calculation of ERS using thresholds based on effect size (187 risk exposures with OR >1.2 and 113 protective factors with OR <0.8) regardless of P value resulted in similar association of ERS with suicide attempt (OR, 2.4; 95% CI, 2.1-2.8 compared with OR, 2.2; 95% CI, 2.0-2.6 in the main analysis), explaining similar variance in suicide attempt (20.6% vs 17.6% in the main analysis) (eTable 13 in Supplement 1).

Discussion

Using ExWAS of individual-level risk and protective environmental and lifestyle factors, we describe the development and application of exposomic scores to quantify youth suicide attempt risk. Our results suggest that a single measure representing aggregate exposomic risk consistently explained substantial variance in suicide attempt across 3 independent youth cohorts. These findings add to the existing literature on individual risk and protective factors for youth suicide attempt risk^33,34 and to the emerging use of exposomic scores in other medical conditions.³⁵ The generalization of our approach in different developmental epochs of adolescence, from early adolescence when suicide attempt rates were 3% (ABCD study) to late adolescence when suicide attempt rates were 7% (MCS cohort) and 10% (CHOP-ED cohort), suggests that exposomic scores may be relevant for risk classification across developmental periods. Our work addresses a major challenge of generalizability of quantifying suicide risk across different settings,³⁶ suggesting that dimensionality reduction techniques of exposures may have the potential to advance suicide research and suicide prevention efforts.

Across all 3 cohorts, exposures with the greatest risk associations with suicide attempt have strong grounding in youth suicide risk literature, with history of assault (physical or sexual³) and being bullied⁴ or cyberbullied⁵ by peers emerging among the top-ranking risk exposures, supporting the internal validity of our approach. Additionally, we reported consistently increased ERS among minoritized youth, aligning with previous works on individual adversities that are more prevalent among marginalized US youth populations,^37,38 further supporting internal validity of the scores. As for external validity, exposomic risk explained between 17% and 22% of the variance in suicide attempts across all 3 cohorts, despite substantial differences in age, setting, ascertainment method, and geographic location. Importantly, the similarity in variance explained by the ERS was observed despite notable differences in the number and granularity of exposures included in the ExWAS. ABCD and MCS ExWAS included hundreds of measures (447 and 622 exposures, respectively) assessed repeatedly at multiple levels of exposure (school, household, neighborhood, lifestyle), whereas the CHOP-ED cohort included only 17 exposures that composed a list of clinically relevant behavioral health risk factors. The ability to generate exposomic scores that explain substantial variance in suicide attempt using a relatively small set of exposures supports the translational potential of our approach in less resourced settings, such as in low-income and middle-income countries, where tools that can promote suicide prevention efforts are in great need.⁸

Three key findings are worth highlighting in the clinical context. First, our finding in the ABCD study that the exposomic score explains 13.8% variance in suicide attempt over and above parental history of suicide/suicide attempt emphasizes the clinical potential of ERS relative to a clinical benchmark of risk assessment (family history). More work is needed on how ERS compares with other commonly used clinical risk factors. Second, our finding in the CHOP-ED cohort suggests that we can use clinically obtained data to create a score that identifies youth at approximately 6-fold greater odds than the median odds of having made a suicide attempt (ie, youth at the top quintile of the ERS distribution). This finding is critical for risk assessment because past suicide attempt is the leading risk factor to suicide mortality.³⁹ More research is needed to test implementation of platforms that can calculate weighted ERS in real time and make these scores accessible for clinicians in real-world clinical settings. Third, our approach allows identification of exposures associated with suicide attempt that are included in the exposomic score, which can improve interpretability for patients and clinicians, and might consequently help increase acceptance of exposomic scores as decision support tools in the future.⁴⁰ Additionally, more community engagement work is needed to address the ethical challenges of using exposomic scores for risk classification in clinical settings.⁴¹

Limitations

Several limitations should be noted. First, this study leveraged data from different cohorts using different measures that could not be harmonized; hence, we do not claim that the findings replicated across settings, rather, we hope this work provides a blueprint for a reproducible approach to suicide risk classification that can be generalized across studies. Nonetheless, our findings that the variance explained in suicide attempt in each cohort was similar strongly supports the notion that our approach can be generalizable even when measurement of exposures across settings largely differs. Second, we used self-report of suicide attempt as the primary outcome, which may be underreported.⁴² Third, the number of youths with suicide attempt in some of our subpopulation analyses was quite small. Fourth, our calculation of exposome score does not directly account for the correlations among exposures; however, we did apply a threshold for correlations to be below 0.9 before running the ExWAS to account for cases of high collinearity. Notably, we had recently described another method of generating exposome scores using factor analysis and bifactor modeling that captures the correlated nature of exposures,⁴³ but this approach is more computationally heavy and is less interpretable. Another limitation is that the exposome score calculation does not account for the CIs of the associations of exposure with suicide attempt; hence, exposures with similar coefficients will receive similar weights despite differences in the confidence of the estimated association. Last, our approach does not address the causal effects of environmental exposures. We emphasize that our aim was to highlight the potential use of exposomic scores and their generalizability for risk classification (ie, explain variance) and not for identification of environmental and lifestyle causes for suicide risk, which will require other statistical methods.⁴⁴

Conclusions

In this cohort study, we applied a novel reproducible approach to study aggregate contribution of environmental and lifestyle factors to youth suicide attempt in 3 different cohorts, from early to late adolescence. Exposomic scores consistently explain substantial variance in suicide attempt over and above demographic variables and an established clinical indicator of suicide risk—family history. We suggest that exposomic scores could be implemented alongside existing clinical risk factors and potentially alongside clinical,³⁶ genetic,⁴⁵ or geocoded derived data⁴⁶ to optimize individual youth suicide attempt risk classification. Future clinical research should test the relevance of exposomic scores in low- or middle-income countries, address implementation of exposomic scores in clinical settings, and test their contribution to youth suicide prevention.

Supplement 1.

eMethods. Study Participants, Choice of Exposures Included in ExWAS, Handling of Longitudinal Data, ExWAS Methods, Calculation of Exposomic Risk Scores in all 3 Cohorts

eTable 4. Association of Exposomic Risk Scores With Suicide Attempt in CHOP-ED and MCS

eTable 5. Difference in Variance Explained in Suicide Attempt When Substituting Exposomic Risk Score With a Single Risk Exposure Identified in the ExWAS in Each of the 3 CohortsError! Not a valid link.eTable 6. Differences in Exposomic Risk Scores Across Minority Groups in CHOP-ED

eTable 7. Differences in Exposomic Risk Scores Across Minority Groups in MCS

eTable 8. Interaction of Exposomic Risk Scores With Minority Indicators in all 3 Cohorts

eTable 9. Stratified Analyses of the Association of Exposomic Risk Score With Suicide Attempt Across Subpopulations in ABCD

eTable 10. Stratified Analyses of the Association of Exposomic Risk Score With Suicide Attempt Across Subpopulations in CHOP-ED

eTable 11. Stratified Analyses of the Association of Exposomic Risk Score With Suicide Attempt Across Subpopulations in MCS

eTable 12. Association of Exposomic Risk Scores With Suicide Attempt Adjusting for Overall Psychopathology in ABCD

eTable 13. Association of Exposomic Risk Scores With Suicide Attempt Using All Risk and Protective Exposures With Small or Larger Effect Size (Exposures With Odds Ratio>1.2 or Odds Ratio<0.8), and Not Just Statistically Significant Exposures as in Main Analyses)

eFigure 1. Forest Plot Depicting Effect Sizes of Risk and Protective Exposures Associated With Suicide Attempt in ABCD Study

eFigure 2. Disparities in Exposomic Risk Scores for Suicide Attempt in ABCD Study

jamapsychiatry-e241887-s001.pdf^{(721.2KB, pdf)}

Supplement 2.

eTable 1. Exposures Included in ABCD ExWAS and Their Association With Suicide Attempts

eTable 2. Exposures Included in CHOP-ED ExWAS and Their Association With Suicide Attempts

eTable 3. Exposures Included in MCS ExWAS and Their Association With Suicide Attempts

jamapsychiatry-e241887-s002.xlsx^{(207.9KB, xlsx)}

Supplement 3.

Data Sharing Statement.

jamapsychiatry-e241887-s003.pdf^{(69.6KB, pdf)}

References

1.US Centers for Disease Control and Prevention . Injury center, Accessed February 6, 2024. https://www.cdc.gov/injury/wisqars/leadingcauses.html
2.Turecki G, Brent DA, Gunnell D, et al. Suicide and suicide risk. Nat Rev Dis Primers. 2019;5(1):74. doi: 10.1038/s41572-019-0121-0 [DOI] [PubMed] [Google Scholar]
3.Angelakis I, Austin JL, Gooding P. Association of childhood maltreatment with suicide behaviors among young people: a systematic review and meta-analysis. JAMA Netw Open. 2020;3(8):e2012563-e2012563. doi: 10.1001/jamanetworkopen.2020.12563 [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Holt MK, Vivolo-Kantor AM, Polanin JR, et al. Bullying and suicidal ideation and behaviors: a meta-analysis. Pediatrics. 2015;135(2):e496-e509. doi: 10.1542/peds.2014-1864 [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Arnon S, Brunstein Klomek A, Visoki E, et al. Association of cyberbullying experiences and perpetration with suicidality in early adolescence. JAMA Netw Open. 2022;5(6):e2218746. doi: 10.1001/jamanetworkopen.2022.18746 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.APA PsycNet . Ecological systems theory. Accessed December 13, 2022. https://psycnet.apa.org/record/2004-22011-010
7.Bronfenbrenner U. Ecological models of human development. Read Dev Child. 1994;2(1):37-43. doi: 10.5281/zenodo.5651995 [DOI] [Google Scholar]
8.O’Connor RC, Worthman CM, Abanga M, et al. Gone too soon: priorities for action to prevent premature mortality associated with mental illness and mental distress. Lancet Psychiatry. 2023;10(6):452-464. doi: 10.1016/S2215-0366(23)00058-5 [DOI] [PubMed] [Google Scholar]
9.Barzilay R, Pries LK, Moore TM, et al. Exposome and trans-syndromal developmental trajectories toward psychosis. Biol Psychiatry Glob Open Sci. 2022;2(3):197-205. doi: 10.1016/j.bpsgos.2022.05.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Mullins N, Bigdeli TB, Børglum AD, et al. ; M.R.C.Psych; Dr.Med.Sc; M.R.C.Psych; M.R.C.Psych; Dipl.-Psych; M.R.C.Psych; Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium; Bipolar Disorder Working Group of the Psychiatric Genomics Consortium; Schizophrenia Working Group of the Psychiatric Genomics Consortium . GWAS of suicide attempt in psychiatric disorders and association with major depression polygenic risk scores. Am J Psychiatry. 2019;176(8):651-660. doi: 10.1176/appi.ajp.2019.18080957 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Vermeulen R, Schymanski EL, Barabási AL, Miller GW. The exposome and health: where chemistry meets biology. Science. 2020;367(6476):392-396. doi: 10.1126/science.aay3164 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Wild CP. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Prev. 2005;14(8):1847-1850. doi: 10.1158/1055-9965.EPI-05-0456 [DOI] [PubMed] [Google Scholar]
13.Park SK, Tao Y, Meeker JD, Harlow SD, Mukherjee B. Environmental risk score as a new tool to examine multipollutants in epidemiologic research: an example from the NHANES study using serum lipid levels. PLoS One. 2014;9(6):e98632. doi: 10.1371/journal.pone.0098632 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Wray NR, Lin T, Austin J, et al. From basic science to clinical application of polygenic risk scores: a primer. JAMA Psychiatry. 2021;78(1):101-109. doi: 10.1001/jamapsychiatry.2020.3049 [DOI] [PubMed] [Google Scholar]
15.Juarez PD, Matthews-Juarez P. Applying an exposome-wide (ExWAS) approach to cancer research. Front Oncol. 2018;8(AUG):313. doi: 10.3389/fonc.2018.00313 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Guillien A, Lepeule J, Seyve E, et al. Profile of exposures and lung function in adults with asthma: An exposome approach in the EGEA study. Environ Res. 2021;196:110422. doi: 10.1016/j.envres.2020.110422 [DOI] [PubMed] [Google Scholar]
17.Liang H, Zhang Y, Zhang S, et al. Association between environmental composite quality index score and obesity in children and their family: a cross-sectional study in Northeast China. Chemosphere. 2022;297:134204. doi: 10.1016/j.chemosphere.2022.134204 [DOI] [PubMed] [Google Scholar]
18.Guloksuz S, van Os J, Rutten BPFF. The Exposome Paradigm and the Complexities of Environmental Research in Psychiatry. Vol 75. American Medical Association; 2018, doi: 10.1001/jamapsychiatry.2018.1211. [DOI] [PubMed] [Google Scholar]
19.Lin BD, Pries LK, Sarac HS, et al. Nongenetic factors associated with psychotic experiences among UK Biobank participants: exposome-wide analysis and mendelian randomization analysis. JAMA Psychiatry. 2022;79(9):857-868. doi: 10.1001/jamapsychiatry.2022.1655 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Meza JI, Patel K, Bath E. Black youth suicide crisis: prevalence rates, review of risk and protective factors, and current evidence-based practices. Focus (Am Psychiatr Publ). 2022;20(2):197-203. doi: 10.1176/appi.focus.20210034 [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Wallace ER, O’Neill S, Lagdon S. Risk and protective factors for suicidality among lesbian, gay, bisexual, transgender, and queer (LGBTQ+) young people, from countries with a high global acceptance index (GAI), within the context of the socio-ecological model: a scoping review. J Adolesc. 2024. Published online February 19, 2024. doi: 10.1002/jad.12308 [DOI] [PubMed] [Google Scholar]
22.Karcher NR, Barch DM. The ABCD study: understanding the development of risk for mental and physical health outcomes. Neuropsychopharmacology. 2021;46(1):131-142. doi: 10.1038/s41386-020-0736-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Connelly R, Platt L. Cohort profile: UK Millennium Cohort Study (MCS). Int J Epidemiol. 2014;43(6):1719-1725. doi: 10.1093/ije/dyu001 [DOI] [PubMed] [Google Scholar]
24.White J, Borgia S, Rehkopf DH. Socioeconomic inequalities in the risk of suicide attempts among sexual minority adolescents: findings from the UK’s Millennium Cohort Study. Lancet Reg Health Eur. 2022;26:100570. doi: 10.1016/j.lanepe.2022.100570 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Quimby EG, Brogan L, Atte T, Diamond G, Fein JA. Evaluating adolescent substance use and suicide in the pediatric emergency department. Pediatr Emerg Care. 2022;38(2):e595-e599. doi: 10.1097/PEC.0000000000002453 [DOI] [PubMed] [Google Scholar]
26.Bennett DA. How can I deal with missing data in my study? Aust N Z J Public Health. 2001;25(5):464-469. doi: 10.1111/j.1467-842X.2001.tb00294.x [DOI] [PubMed] [Google Scholar]
27.Kim JH. Multicollinearity and misleading statistical results. Korean J Anesthesiol. 2019;72(6):558-569. doi: 10.4097/kja.19087 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Hua LL, Lee J, Rahmandar MH, Sigel EJ; Committee on Adolescence; Council on Injury, Violence, and Poison Prevention . Suicide and suicide risk in adolescents. Pediatrics. 2024;153(1):e2023064800. doi: 10.1542/peds.2023-064800 [DOI] [PubMed] [Google Scholar]
29.Pries LK, Erzin G, van Os J, et al. Predictive performance of exposome score for schizophrenia in the general population. Schizophr Bull. 2021;47(2):277-283. doi: 10.1093/schbul/sbaa170 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Nakagawa S, Johnson PCD, Schielzeth H. The coefficient of determination R² and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. J R Soc Interface. 2017;14(134):20170213. doi: 10.1098/rsif.2017.0213 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Nagelkerke NJD. A note on a general definition of the coefficient of determination. Biometrika. 1991;78(3):691-692. doi: 10.1093/biomet/78.3.691 [DOI] [Google Scholar]
32.Gaylor EM, Krause KH, Welder LE, et al. Suicidal thoughts and behaviors among high school students—youth risk behavior survey, US, 2021. MMWR Suppl. 2023;72(1):45-54. doi: 10.15585/mmwr.su7201a6 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Carballo JJ, Llorente C, Kehrmann L, et al. ; STOP Consortium . Psychosocial risk factors for suicidality in children and adolescents. Eur Child Adolesc Psychiatry. 2020;29(6):759-776. doi: 10.1007/s00787-018-01270-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Janiri D, Doucet GE, Pompili M, et al. Risk and protective factors for childhood suicidality: a US population-based study. Lancet Psychiatry. 2020;7(4):317-326. doi: 10.1016/S2215-0366(20)30049-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Ding E, Wang Y, Liu J, Tang S, Shi X. A review on the application of the exposome paradigm to unveil the environmental determinants of age-related diseases. Hum Genomics. 2022;16(1):54. doi: 10.1186/s40246-022-00428-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.King CA, Brent D, Grupp-Phelan J, et al. ; Pediatric Emergency Care Applied Research Network . Prospective development and validation of the computerized adaptive screen for suicidal youth. JAMA Psychiatry. 2021;78(5):540-549. doi: 10.1001/jamapsychiatry.2020.4576 [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Argabright ST, Visoki E, Moore TM, et al. Association between discrimination stress and suicidality in preadolescent children. J Am Acad Child Adolesc Psychiatry. 2022;61(5):686-697. doi: 10.1016/j.jaac.2021.08.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Gordon JH, Tran KT, Visoki E, et al. The role of individual discrimination and structural stigma in the mental health of sexual minority youth. J Am Acad Child Adolesc Psychiatry. 2024;63(2):231-244. doi: 10.1016/j.jaac.2023.05.033 [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Favril L, Yu R, Geddes JR, Fazel S. Individual-level risk factors for suicide mortality in the general population: an umbrella review. Lancet Public Health. 2023;8(11):e868-e877. doi: 10.1016/S2468-2667(23)00207-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Wysocki O, Davies JK, Vigo M, et al. Assessing the communication gap between AI models and healthcare professionals: explainability, utility and trust in AI-driven clinical decision-making. Artif Intell. 2023;316:103839. doi: 10.1016/j.artint.2022.103839 [DOI] [Google Scholar]
41.Smith WR, Appelbaum PS, Lebowitz MS, et al. The ethics of risk prediction for psychosis and suicide attempt in youth mental health. J Pediatr. 2023;263:113583. doi: 10.1016/j.jpeds.2023.113583 [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Deming CA, Harris JA, Castro-Ramirez F, et al. Inconsistencies in self-reports of suicidal ideation and attempts across assessment methods. Psychol Assess. 2021;33(3):218-229. doi: 10.1037/pas0000976 [DOI] [PubMed] [Google Scholar]
43.Moore TM, Visoki E, Argabright ST, et al. Modeling environment through a general exposome factor in two independent adolescent cohorts. Exposome. 2022;2(1):osac010. doi: 10.1093/exposome/osac010 [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Hernán MA. The C-word: scientific euphemisms do not improve causal inference from observational data. Am J Public Health. 2018;108(5):616-619. doi: 10.2105/AJPH.2018.304337 [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Barch DM, Hennefield L, Herzberg MP. What makes a useful “predictor” of risk for suicide attempt? JAMA Psychiatry. 2022;79(10):948-950. doi: 10.1001/jamapsychiatry.2022.2031 [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Xiao Y, Mann JJ, Chow JCC, et al. Patterns of social determinants of health and child mental health, cognition, and physical health. JAMA Pediatr. 2023;177(12):1294-1305. doi: 10.1001/jamapediatrics.2023.4218 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials