Abstract
OBJECTIVE: Suicide is a leading cause of death among adolescents, with complex neurobiological origins. This study aimed to develop a prediction model for future suicide attempts in adolescents assigned female at birth (AFAB) with varying histories of lifetime non-suicidal self-injury (NSSI), utilizing multi-unit data across three relevant Research Domain Criteria (RDoC) domains. METHOD: Participants (N = 164; ages 12-16) were enrolled in a 3-year longitudinal study with annual assessments. Of these, 121 completed at least one follow-up, during which 21 reported a suicide attempt. Binomial Elastic-Net regression was used to identify baseline RDoC, clinical, and demographic predictors applying 100 iterations of 80/20 training-test splits and optimizing the penalty term via 10-fold cross-validation. RESULTS: RDoC predictors of future suicide attempts included (1) Threat: higher amygdala-medial prefrontal cortex (mPFC) resting-state functional connectivity, lower amygdala activation and poorer accuracy of emotion matching in threat task; (2) Cognitive control: greater putamen activation during negative inhibition, lower dorsal frontal cortex activation during positive inhibition, slower reactions to negative no-inhibition, thinner mPFC; (3) Self-knowledge: lower anterior Self Network activation to negative stimuli, thinner posterior Self Network, and faster reactions to positive stimuli. Clinical and demographic predictors included higher depression and NSSI severity, psychotropic medication use, histories of neglect and abuse, and identifying as gender-minority. CONCLUSION: While supporting past research implicating well-known clinical and demographic factors, current findings suggest novel neurobiological markers of suicide risk in adolescents, which could represent targets for interventions to prevent suicide attempts in adolescents, if confirmed in future research with larger samples.