Abstract

The relationship between clinical eating disorder symptom severity and balance of model-based (MB) and model-free (MF) control is unclear, and these traits predictive capacity is untested in this population. In 25 healthy controls (HCs), 25 subjects with binge eating disorder (BED), and 25 subjects with bulimia nervosa (BN), we show an inverse relationship between symptom severity and MB (though not MF) control. However, trial-by-trial behavioural data discriminated BN from other groups (area under receiver operating characteristic curve of 0.78; 95% CI=0.64-0.92) based primarily on impaired MF control. Our data--including analyses of reaction time and theory-driven computational modeling--support the hypothesis that among pathological binge eating groups, BN may be characterized by impaired value function learning. Our results suggest that trial-by-trial analysis of behavioural data may provide unique insights into the BN phenotype, which may thus be computationally distinct from the related disorder of BED and the HC state.