Genome-wide Polygenic Scores for Multiple Psychiatric and Common Traits Identify Preadolescent Youth with Risk for Suicide

Abstract Importance. Suicide is the second leading cause of death in children worldwide but no available means exist to identify the risk in youth. Objective. To predict the risk of suicide in children and to investigate whether and to what extents genetic factors and a major environmental risk factor, early life stress(ELS), influence youth suicide. Design, Setting and Participants. We analyzed the genotype-phenotype data of 11,869 preadolescent children ages 9- to 10-year-old from the Adolescent Brain and Cognitive Development (ABCD) study. We estimated genome-wide polygenic scores (GPSs) of 25 complex traits to investigate their phenome-wide associations and predictive utility with suicidality (suicidal ideation and attempt) with machine learning approaches. Predictors. GPSs of 25 traits including psychiatric disorders, personality, cognitive capacity, and psychological traits. Parent Child Behavior Checklist to measure ELS in youth and Youth Family Environment Scale to assess family environment. Main outcomes and Measures. Records of suicidal ideation and attempt of the participants were derived from the computerized version of Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS). Results. We identified three GPSs associated with youth suicidality in multiethnic (n = 7,206) and European-ancestry (n = 5,749) participants: ADHD (P = 3.48x10− 4; odds ratio = 1.13 in multiethnic participants, P = 5.60x10− 5, OR = 1.25 in European-ancestry participants), general happiness (P = 1.43x10− 3; OR = 0.89 in multiethnic, P = 8.61x10− 4, OR = 0.89 in European) and autism spectrum disorder(ASD) (P = 1.81x10− 3; OR = 1.15 in multiethnic, P = 1.26x10− 3, OR = 1.18 in European). We also found a significant GPS-by-environment interaction between the effects of genetic risk factors for ASD and the level of ELS in increasing the risk for suicidal ideation (P = 1.36x10− 2, OR = 1.12 in multiethnic, P = 1.39x10− 3, OR = 1.19 in European). A machine learning model trained on the same data showed moderately accurate prediction of children with overall suicidal ideation with a test ROC-AUC of 0.727 (0.746 in European), and with suicidal attempts with a test ROC-AUC of 0.641 (0.975 in European) in held-out samples. Conclusions and Relevance. This study provides the first quantitative account of polygenic and environmental factors of suicidality in a large, representative population of preadolescent youth. It thus shows the potential utility of the GPSs in identifying a child with high risk for suicidality for early screening, intervention, and prevention.