Transdiagnostic mental health symptom dimensions predict use of flexible model-based inference in complex environments

A key goal within computational psychiatry is to identify mechanisms underpinning symptom dimensions that cut across diagnostic categories. Complex, naturalistic, situations, such as the inference of others’ mental states and how to act accordingly (e.g., social interactions), is a common junction at which mental health symptoms emerge. Such patterns may reflect the breakdown of the fundamental computational processes that ordinarily underpin these behaviours, with the use of flexible goal-directed decision-making being a prime candidate. Here, we used a validated, naturalistic threat inference task to assess the use of goal-directed decision-making in complex interactive decision problems. Participants (n=1025) completed this task alongside a battery of self-report measures of mental health symptoms and neurodevelopmental characteristics. By performing hierarchical dimensionality reduction on symptom measures, we found that task behaviour was most clearly associated with fine-grained symptom dimensions, where higher scores on an inattentive/neurodevelopmental dimension predicted more accurate inferences and higher scores on an externalising dimension predicted poorer performance. Using computational modelling, we show that these associations are mediated by the degree to which individuals use goal-directed decision-making to make inferences about the predator’s behaviour. Our results suggest that symptoms and traits that manifest in complex, social environments may result from alterations in the use of complex computational mechanisms.