Leveraging Latent Factors and Machine Learning to Forecast Internalizing Psychopathology in Emerging Adulthood

点击翻译按钮获取中文摘要

Project Summary Mood and anxiety disorders are common and highly comorbid conditions with peak incidence in emerging adulthood (~ages 18-23). Developmental psychopathology models suggest that vulnerability to internalizing disorders in emerging adults is driven by interactions between still maturing self-regulatory abilities (as executive function [EF] continues to mature into young adulthood), and individual differences in reward and threat sensitivity. Together, this highlights the importance of complex neurocognitive profiles consisting of abnormalities across these three RDoC constructs for internalizing disorders. However, prior research has largely investigated these constructs individually, in relation to individual disorders or symptom dimensions. Given the high co-occurrence and complex multi-causality of internalizing psychopathology, the critical next step is to build a framework for understanding how these neurocognitive dimensions interact to predict transdiagnostic person-specific symptom trajectories. The proposed study aims to advance this precision medicine goal, by evaluating how the neurocognitive dimensions of EF, reward and threat sensitivity interact to produce risk phenotypes; and by using machine learning techniques to identify the most parsimonious set of risk markers (across units of analysis) that forecast psychopathology. This longitudinal study will recruit a final sample of 480 emerging adults during the transition to college, when stress and psychopathology risk increase, to test risk pathways for transdiagnostic (common across internalizing symptoms) and specific (anhedonia, anxious arousal, mania) internalizing dimensions, using a methodologically rigorous latent variable approach. Our first aim is to test interactions among the neurocognitive dimensions of executive function, threat sensitivity and reward sensitivity as risk mechanisms for transdiagnostic and specific internalizing symptom profiles and trajectories. We hypothesize that poor EF is a transdiagnostic risk factor, with specific symptom profile depending on threat (contributing to anxious arousal) and reward (contributing to anhedonia or mania) sensitivity, and different maladaptive behaviors (e.g., social withdrawal vs. risky behavior). Our second aim is to perform automated risk profiling, using machine learning to determine most parsimonious set of units that predict outcome– a key objective for future clinical translation for screening for internalizing psychopathology risk. Strengths of this approach include model-driven dimensional constructs of cognitive control, negative and positive valence, spanning units of analysis (physiology, behavior, self-report) at a critical developmental risk period. The robust sample size enables a rigorous statistical modeling approach and testing of moderating influences (e.g., sex). By elucidating interactions between neurocognitive risks and the specific biobehavioral mechanisms involved, we can make a novel impact that will be critical for developing translational tools that predict person-specific symptom trajectories, informing future diagnostic systems (RDoC priority) and personalizing promising interventions (which risk mechanisms to target for intervention, and for whom).