Predicting the First Onset of Suicidal Thoughts and Behaviors in Adolescents Using Multimodal Risk Factors: A Four-Year Longitudinal Study.

Abstract

OBJECTIVE Suicide is one of the leading causes of death among youth worldwide, yet existing studies that aimed to predict the first onset of suicidal thoughts or behaviors (STB) included a limited number of data modalities, and/or focused on adult populations. We aimed to prospectively predict first-onset STB across four-year follow-ups in adolescents using (1) an existing STB history classification model that was previously applied to baseline data, and (2) a new machine learning model with 195 biopsychosocial features. METHOD The study included 7,503 unrelated adolescents (54.5% female, aged 9-11 years at baseline) from the multisite, longitudinal Adolescent Brain Cognitive Development (ABCD) project. Our existing baseline STB history classification model was applied to predict longitudinal first-onset STB versus healthy controls and clinical controls (those with mental illness but no STB). A new elastic net logistic regression model with 195 features was trained on data from 14 sites (n=5,220) and the resulting top 15 features were validated in seven independent sites (n=2,283). RESULTS Our previously developed model to classify STB lifetime history also prospectively predicted first-onset STB with an area under the curve (AUC)[95%] of 0.73[0.70,0.75], p<.001 relative to healthy controls and AUC[95%] of 0.63[0.60,0.66], p<.001 compared to clinical controls. The newly trained model with top 15 features performed similarly with AUC[95%]=0.73[0.71,0.76], p<.001 and AUC[95%]=0.64 [0.60,0.66], p<.001 for the same comparison groups. The most consistent predictors across models included female sex, sleep disturbances, and maladaptive home and school environments. CONCLUSION Our models predicted first-onset STB in adolescents with moderate accuracy. Our study also confirmed the roles of well-established psychological risk factors for STB and identified several novel neurocognitive and brain imaging risk factors. Future studies should validate our models in large-scale diverse samples before clinical translation.