Computational Models Describe Individual Differences in Cognitive Function and Their Relationships to Mental Health Symptoms

Cognitive alterations have long been reported in patients with mental health disorders, though with inconsistent results. These inconsistencies are likely due to highly heterogeneous diagnostic categories used for recruitment, and imprecise cognitive task measures. This thesis addresses the former by measuring symptoms with continuous questionnaire scales, and the latter by using theory- driven computational models that summarise participant behaviour using a small number of mechanistic parameters. This methodology is applied within the realm of attention set shifting and risky decision making to improve understanding of cognition in mental health, using large samples collected online. Following a general introduction (Chapter 1), Chapter 2 describes the computational approach employed in subsequent experimental chapters. In Chapter 3, we develop models of CANTAB IED (Intra-Extra Dimensional Set Shifting Task) to explore how learning and attention processes lead to differences in attention set shifting ability, and to investigate their relationship with symptoms of compulsivity. The second study (Chapter 4) applies the computational approach to risky decisions with CANTAB CGT (Cambridge Gamble Task) and explores the relationship between model parameters and symptoms of depression and anxiety. The final experimental chapter (Chapter 5) examines whether specific symptoms of anxiety are related to changes in risky decision making, focusing on the relationship between catastrophising and probability weighting. Overall, the computational approach offers increased precision when examining behavioural data. In several chapters we identify moderate relationships between model parameters and demographic variables such as age, gender, and level of education, which often exceed associations with traditional model- agnostic measures. However, relationships with mental health symptoms are minimal in the general population datasets tested here. The general discussion (Chapter 6) considers these findings in relation to the wider field of computational psychiatry, discussing both the limitations of the work presented and possible future directions