Association of Genome-Wide Polygenic Scores for Multiple Psychiatric and Common Traits in Preadolescent Youths at Risk of Suicide

Abstract

Importance: Suicide is the second leading cause of death among youths worldwide, but no available means exist to identify the risk of suicide in this population.

Objective: To assess whether genome-wide polygenic scores for psychiatric and common traits are associated with the risk of suicide among preadolescent children and to investigate whether and to what extent the interaction between early life stress (a major environmental risk factor) and polygenic factors is associated with suicidal thoughts and behaviors among youths.

Design, setting, and participants: This cohort study analyzed the genotype-phenotype data of 11 869 preadolescent children aged 9 to 10 years from the Adolescent Brain and Cognitive Development study. Data were collected from September 1, 2016, to October 21, 2018, and analyzed from August 1, 2020, to January 3, 2021. Using machine learning approaches, genome-wide polygenic scores of 24 complex traits were estimated to investigate their phenome-wide associations and utility for assessing risk of suicidal thoughts and behaviors (suicidal ideation [active, passive, and overall] and suicide attempt).

Main outcomes and measures: Genome-wide polygenic scores were used to measure 24 traits, including psychiatric disorders, cognitive capacity, and personality and psychological characteristics. The Child Behavior Checklist was used to measure early life stress, and the Family Environment Scale was used to assess family environment. Suicidal ideation and suicide attempts were derived from the computerized version of the Kiddie Schedule for Affective Disorders and Schizophrenia.

Results: Among 11 869 preadolescent children in the US, complete data for phenotypic outcomes, genotypes, and covariates were available for 7140 participants in the multiethnic cohort (mean [SD] age, 9.9 [0.6] years; 3588 girls [50.3%]), including 925 participants with suicidal ideation and 63 participants with suicide attempts. Among those 7140 participants, 729 had African ancestry (self-reported race or ethnicity: 569 Black, 71 Hispanic, and 89 other), 276 had admixed American ancestry (self-reported race or ethnicity: 265 Hispanic, 3 White, and 8 other), 150 had East Asian ancestry (self-reported race or ethnicity: 67 Asian, 18 Hispanic, and 65 other), 5718 had European ancestry (self-reported race or ethnicity: 7 Asian, 39 Black, 1142 Hispanic, 3934 White, and 596 other), and 267 had other ancestries (self-reported race or ethnicity: 70 Asian, 13 Black, 126 Hispanic, 48 White, and 10 other). Three genome-wide polygenic scores were significantly associated (false discovery rate P < .05) with suicidal thoughts and behaviors among all participants: attention-deficit/hyperactivity disorder (odds ratio [OR], 1.12; 95% CI, 1.05-1.21; P = .001), schizophrenia (OR, 1.50; 95% CI, 1.17-1.93; P = .002), and general happiness (OR, 0.89; 95% CI, 0.83-0.96; P = .002). In the analysis including only children with European ancestry, 3 additional genome-wide polygenic scores with false discovery rate significance were associated with suicidal thoughts and behaviors: autism spectrum disorder (OR, 1.18; 95% CI, 1.06-1.31; P = .002), major depressive disorder (OR, 1.12; 95% CI, 1.04-1.21; P = .003), and posttraumatic stress disorder (OR, 1.12; 95% CI, 1.04-1.21; P = .004). A significant interaction between genome-wide polygenic scores and environment was found, with genetic risk factors for autism spectrum disorder and the level of early life stress associated with increases in the risk of overall suicidal ideation and overall suicidal thoughts and behaviors (OR, 1.20; 95% CI, 1.07-1.35; P = .002). A machine learning model using multitrait genome-wide polygenic scores and additional self-reported questionnaire data (Child Behavior Checklist and Family Environment Scale) produced a moderately accurate estimate of overall suicidal thoughts and behaviors (area under the receiver operating characteristic curve [AUROC], 0.77; 95% CI, 0.73-0.81; accuracy, 0.67) and suicidal ideation (AUROC, 0.76; 95% CI, 0.72-0.80; accuracy, 0.66) among children with European ancestry only. Among all children in the multiethnic cohort, the integrated model also outperformed the baseline model in estimating the risk of overall suicidal thoughts and behaviors (AUROC, 0.71; 95% CI, 0.67-0.75; accuracy, 0.68) and suicidal ideation (AUROC, 0.75; 95% CI, 0.71-0.78; accuracy, 0.67).

Conclusions and relevance: In this cohort study of preadolescent youths in the US, higher genome-wide polygenic scores for psychiatric disorders, such as attention-deficit/hyperactivity disorder, autism spectrum disorder, posttraumatic stress disorder, and schizophrenia, were significantly associated with a greater risk of suicidal ideation and suicide attempt. The findings and quantitative models from this study may help to identify children with a high risk of suicide, potentially assisting with early screening, intervention, and prevention.

Figure 1.. Study Flow Diagram

The study initially assessed 11 869 preadolescent children aged 9 to 10 years recruited from the Adolescent Brain and Cognitive Development study. After an initial quality control assessment, complete data including phenotypic outcome, genotype, and covariate data were available for 7140 children in the multiethnic cohort (5718 of whom had European ancestry only). Continuous variables for those individuals were imputed and used in the models. For the primary association analysis, individuals with missing continuous data were removed, and the integrative multimodal data of 6592 children in the multiethnic cohort (5374 of whom had European ancestry only) were included. FES indicates Family Environment Scale; and KSADS-Comp, Kiddie Schedule for Affective Disorders and Schizophrenia, computerized version. ^aIncludes 59 children in the multiethnic cohort (43 of whom had European ancestry only) who attempted suicide. ^bIncludes 1652 children in the multiethnic cohort (1351 of whom had European ancestry only) who were missing data from the KSADS-Comp.

Figure 2.. Manhattan Plot of Association Between 24 Genome-Wide Polygenic Scores and Suicidal Thoughts and Behaviors Among Children With European Ancestry Only

The analysis included 5374 children aged 9 to 10 years in the multiethnic cohort. The blue line represents a false discovery rate–corrected P value of .05. The dotted horizontal lines indicate different levels of statistical significance (negative log P values of 1, 3, 5, and 7, from bottom to top of plot). Each triangle represents the effect direction (positive or negative) and the effect size of each association, with inverted triangles indicating negative direction and effect size. The Manhattan plot of the association between 24 genome-wide polygenic scores and suicidal thoughts and behaviors among 6592 children in the multiethnic cohort is available in eFigure 2 in the Supplement. AD indicates Alzheimer disease; ADHD, attention-deficit/hyperactivity disorder; ASD, autism spectrum disorder; BPD, bipolar disorder; EA, educational attainment; GH, general happiness; MDD, major depressive disorder; PTSD, posttraumatic stress disorder; SCZ, schizophrenia.

Figure 3.. Performance of Machine Learning Models Based on Genome-Wide Polygenic Scores and Cognitive, Psychological, Behavioral, Environmental, and Familial Factors Among Children With European Ancestry Only

Receiver operating characteristic (ROC) curves of the models. A total of 5718 children with European ancestry were included in the analysis. A, The area under the ROC (AUROC) of the best model was 0.77 (95% CI, 0.73-0.81; accuracy, 0.67; positive predictive value [PPV], 0.65; negative predictive value [NPV], 0.71). B, The AUROC of the best model was 0.76 (95% CI, 0.72-0.80; accuracy, 0.66; PPV, 0.64; NPV, 0.69). C, The AUROC of the best model was 0.93 (95% CI, 0.87-0.99; accuracy, 0.84; PPV, 0.83; NPV, 0.85) using the elastic net model. The model was also evaluated using data from the entire sample of 7140 children, with the results available in eTable 7 in the Supplement. GPS indicates genome-wide polygenic score.