Eating disorder symptoms are associated with altered reinforcement learning in the general population

Eating disorders (EDs) are characterised by abnormal food intake, including energy restriction despite starvation or episodic binge-eating, which may reflect alterations in the reward value of food. Affected individuals show impaired learning from non-food rewards (e.g., money), yet the generalisability of such perturbations across the spectrum of disordered eating and to disorder-salient stimuli remains poorly understood. Therefore, we examined associations between ED symptoms and decision-making during reinforcement learning with food and non-food reward in a population-based sample. We recruited four hundred adults (n=250 female, M(SD)=29.5(6.6) years) to an online study, during which they performed two runs of a probabilistic instrumental learning task with either food or monetary reward. On each trial, participants chose between two cues with the aim of maximising the amount of money or food earned throughout the task, where cues were probabilistically associated with either wins, losses or neutral outcomes. A reinforcement learning model was fit to trial-wise choices to generate parameter estimates of learning rate and inverse choice temperature. Increased ED symptomatology was related to poorer task accuracy on loss relative to win trials across both reward types. Computational modelling revealed that greater ED symptom scores were associated with decreased inverse choice temperature for loss versus win trials. Our findings identify a link between ED symptomatology and impaired reinforcement learning from punishment that appears to generalise across food and monetary reinforcers. This may explain, in part, the challenges with modifying behaviour in patients who are struggling with EDs despite serious negative health outcomes.