Mapping the risk and protective factors of depression

: A machine learning approach

Abstract

Understanding the aetiological trajectory of depression has been a complex challenge in
epidemiology in recent years, with various hypotheses of depression causes being challenged and discussed. Despite decades of research in psychiatric science dedicated to
understanding the determinants of depression, progress towards the reduction of its prevalence
and illness burden in the population has been slow and relatively unsuccessful.
Recent research has indicated that rather than being explained by a single unifying hypothesis, depression is multifactorial, heterogeneous, and characterised by diverse antecedents.
In order to comprehensively understand depression’s determinants, computational approaches
which utilise a broad and deep range of health-related data should be applied to this complex
healthcare challenge.
Mapping the longitudinal dynamics of depression exposures paves the way for an enhanced
understanding of when to optimally intervene to improve mental health in young people. Examination of different domains relating to depression risk may help us to understand which
phenotypic domains (psychological, familial, lifestyle, biological, behavioural) should be preferentially intervened on to improve clinical outcomes.
Additionally, understanding the predictive power of interventions related to depression symptom management is a novel way to interrogate how interventions for depression map on to the
symptom trajectory of individuals.
Using data from the Avon Longitudinal Study of Parents and Children (ALSPAC), this
thesis examines how modelling the risk and protective factors of depression using interpretable
machine learning (ML) can be used to understand which factors are most predictive of early
adulthood depression symptoms. I build and evaluate eight Random Forest (RF) prediction models
which achieved F1 scores ranging from 0.46 to 0.66. Of the higher performing models, social
connectedness, experiencing a sense of purpose, self-perception of health and the psychological
impact of trauma were amongst some of the most predictive factors of experiencing depression in
young adults.

Date of Award	10 Dec 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Rebecca M Pearson (Supervisor) & Ryan McConville (Supervisor)