(PS6-B32) Habit Learning and Goal-directed Control in Anorexia Nervosa

Background:  

Anorexia Nervosa (AN) is a psychiatric condition with marked persistence and suboptimal recovery rates across the lifespan; therefore, understanding its maintaining cognitive, behavioral, and neural mechanisms is of critical importance in its treatment. In service of furthering this understanding, this study aims to explore whether AN is maintained by goal-directed or habitual behavior, so that future treatments may target the corresponding maintaining neural mechanism. It’s been theorized that, while AN might initially begin with purposeful attempts to control shape and weight through dieting, as restrictive behavior is rewarded by weight loss, this behavior gets reinforced and repeated until ultimately becoming habitual. Once an ingrained habit, maladaptive behavior such as food restriction is incredibly resistant to change, which would explain some of the challenges in treating AN.

Previous research has sought to describe the balance between goal-directed versus habitual control in individuals with an eating disorder. These categories (goal-directed and habitual) arise from two systems for decision making known as model-based and model-free learning. These systems may be used to quantify goal-directed and habitual control in a reinforcement learning task. Similar to other compulsive disorders, existing data suggest persistence of habitual control and deficits in goal-directed control in AN in a mixed adolescent and adult sample. Exploring goal-directed versus habitual control in AN specifically in adolescence, which is the average onset of AN and a time when model-based learning is gradually developing, may provide further insight into the development of AN pathology.

Methods:  

We hypothesized that adolescents with anorexia would demonstrate a greater reliance on perseverative learning as compared to age-matched healthy adolescents, signaled by diminished goal-directed behavior as evidenced through the model-based parameter in a reinforcement learning task. Participants (N=40) completed a two-stage reinforcement learning task, adapted originally from Daw et al. (2011), designed to distinguish between model-based and model-free learning. Following previous analyses of the task, a generalized linear mix-effects regression analysis was applied to the data to examine how reward and transition type of the preceding trial predict first-stage choice in each trial. We will expand on this analysis by applying a computational model based on the “hybrid” model originally used for this task (Daw et al., 2011).

Results:  

Our findings failed to provide evidence for the hypothesis; we found no significant effect of the interaction between reward and transition type, which has previously been taken to indicate model-based learning, and no significant group effects. These findings suggest that adolescents might not experience deficits in goal-directed control; or that this task may not appropriately reflect habitual behavior.