13:30 - 15:00
Mon-A7-Talk II-
Mon-Talk II-
Room: A7
Chair/s:
Franziska Knolle
Computational psychiatry provides a direct approach for the investigation and development of mechanistic explanations for psychiatric illnesses, through the mathematical description of processes underlying behaviour. Alterations in decision-making, a cognitive process relevant for the successful interaction with the environment, have been reported in disorders including psychosis, obsessive-compulsive disorder (OCD), and eating disorders, and have been linked to the development of symptoms. Importantly, we describe a transdiagnostic approach using similar tasks and models to show that specific alterations in mathematical parameters express disease-specific dysfunction and symptom associations. Kelly Diederen (King’s College London) used a novel gamified task in conjunction with computational modelling to demonstrate that decision-confidence and belief-updating can be measured at scale using online assessment, and that these processes are altered in people at increased risk of psychosis. Elisabeth Sterner (LMU/TUM) will show that deficits in punishment learning in early psychosis are linked to increased forgetting and reduced confidence in policy selection using an Active Inference model of the Go/NoGo-Task. Pritha Sen (LMU/TUM) will present first-time imaging data investigating model-based (MB) and model-free (MF) decision making in OCD using hierarchical Bayesian modelling of the Two-Step task. Computational result show differences between patients and controls with links to symptom strength. Margaret Westwater (Oxford/Yale) will report data from both laboratory-based and online assessments of learning under uncertainty, which used computational modelling to
demonstrate that impaired learning in individuals with eating disorders is linked to altered reward sensitivity. Franziska Knolle (TUM/Cambridge) will discuss the effect of dopaminergic treatment on probabilistic reward learning in OCD using Rescorla-Wagner-modelling, showing that exaggerated cingulate reward prediction errors in patients are remediated by dopaminergic modulation. This state-of-the-art symposium demonstrates that computational models of decision making provide mechanistic explanations for dysfunctions underlying three common psychiatric conditions: psychosis, OCD, and eating disorders, and may provide starting points for treatment development.
Back
WITHDRAWN A computational account of learning and decision-making under uncertainty in eating disorders WITHDRAWN
Mon-A7-Talk II-04
Presented by: Margaret Westwater
Margaret Westwater 1, 2, Hisham Ziauddeen 2, Kelly Diederen 3, Paul Fletcher 2
1 Department of Psychiatry, University of Oxford, Oxford, UK, 2 Department of Psychiatry, University of Cambridge, Cambridge, UK, 3 Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK
Background: Eating disorders (EDs) are characterised by abnormal food intake, including energy restriction despite starvation (in anorexia nervosa; AN) or binge-eating despite satiety (in bulimia nervosa; BN). These behavioural patterns indicate alterations in reward value in EDs. Affected individuals show reduced reactivity to rewards (e.g., money), yet it remains unknown if such alterations persist under conditions of uncertainty that approximate real-world environments. We therefore examined decision-making under uncertainty in two independent cohorts of ED participants.

Methods: Eighty-five women (n=22 AN, n=33 BN, n=30 controls) and 299 adults (n=201 ED, n=98 controls) were recruited to a laboratory-based and online study, respectively. Participants performed a probabilistic reversal learning task, which involved making ‘risky’ or ‘less risky’ gambles in response to a cue. One cue was associated with a monetary gain on 80% of trials; the second cue led to a monetary loss 80% of the time. Two unannounced contingency reversals required participants to update their choice behaviour throughout the task. Linear mixed-effects models assessed group differences in performance, and a series of hierarchical reinforcement learning models were fit to examine differences in model parameter estimates.

Results: Across both cohorts, individuals with EDs had reduced learning, indexed as the proportion of optimal responses throughout the task (all p’s<0.05). Impaired performance in EDs was related to increased choice temperature; however, this effect was only statistically significant in the online cohort (p=.004).

Conclusion: Individuals with EDs showed impaired learning under uncertainty, related to lower reward sensitivity, which may explain the persistence of maladaptive eating behaviour.


Keywords: Anorexia nervosa, bulimia nervosa, reinforcement learning, reward sensitivity
Back