Early adversity and individual differences in adolescent mental health

Change log
Uh, Stepheni 

Adolescence is a dynamic period of profound changes during which individuals are particularly susceptible to poor mental health. These biological, cognitive, and psychosocial changes – alongside the increased influence of environmental factors –interact and impact both immediate and long-term wellbeing. Early adversity, in particular, confers significant risks to adolescent mental health. Thus, adolescence is an important window for researchers to investigate processes underlying different outcomes. However, the heterogeneity across the various factors that influence adolescent mental health presents many empirical challenges. This thesis, therefore, applies novel multivariate approaches to explore the heterogeneity of individual differences, risk factors, and outcomes outlined in theoretical and empirical literature of adolescent mental health and development. The first empirical chapter entails a prospective longitudinal study that investigates distinct developmental trajectories and associated risk factors to self-harm. I first used unsupervised machine learning algorithms to identify distinct psychological subgroups of young people who self-harm. This was followed by a supervised machine learning algorithm to identify significant risk factors across a nine-year period for the different subgroups – ultimately distinguishing two different pathways to self-harm. The second empirical chapter provides a novel analytical pipeline to investigate the heterogeneity of overlapping types of adversity, and risks for internalising problems, in a large sample of adolescents. This pipeline involved three steps: constructing a topological representation (“map”) of co-occurring adversities experienced by adolescents; identifying whether poor mental health outcomes are co-localised with specific features of this adversity map; and applying a precision stratification method to investigate differences in corticolimbic connectivity associated with differential outcomes in participants matched for adversity profiles. This approach revealed an adversity profile highly linked with internalising problems; importantly, differential mental health outcomes in participants sharing this adversity profile were linked with differences in corticolimbic connectivity. The third and final empirical chapter targets a specific mechanism strongly implicated in adolescent mental health and development: implicit emotion regulation. I used a modified emotional Go/NoGo fMRI task, in addition to reproducible preprocessing pipelines and a nonparametric group analysis, to explore neural correlates of implicit emotion regulation and individual differences in childhood. The results showed multiple significant response inhibition effects (i.e., larger NoGo vs Go activation in the IFG, insula, and ACC) and valence effects in the putamen and pallidum. Though I did not find significant relationships between these neural responses and individual differences in mental health. This thesis offers advances towards capturing the heterogeneity across multiple levels of factors that interact and influence adolescent mental health. The multivariate and targeted approaches targeting the bio-psycho-social processes that influence mental health outcomes provide important avenues for future research and policies to promote mental wellbeing in adolescents.

Astle, Duncan
adolescence, adversity, development, heterogeneity, machine learning, mental health
Doctor of Philosophy (PhD)
Awarding Institution
University of Cambridge
Gates Cambridge Trust