Excessive Deliberation in Social Anxiety: Using Neuroeconomic Applications to Improve Characterizations, Diagnostic Criteria, and Treatment Options for Social Anxiety Disorder.

Abstract

Recent work has suggested that mental health disorders may be understood in terms of dysfunction in the brain’s goal-directed Reinforcement-Learning (RL) architecture. More specifically, recent work has linked the compulsive aspect of drugs of abuse, and other disorders such as OCD and General Anxiety Disorder, to excessive use of automatic (model-free) over deliberative (model-based) action evaluation. In addition, a converse pathology of excess deliberation (model-based action evaluation) has been suggested to relate to other disorders (e.g. rumination in mood disorders). By assessing how symptoms of social anxiety disorder (SAD) predict model-based (vs model-free) learning in a socially framed RL task, we seek to investigate this hypothesis. In this experiment, we used a data-driven approach, employing large-scale online testing of 489 participants from a general population sample (Amazon Mechanical Turk). Social Anxiety Disorder (SAD) is then particularly interesting, and practical, to focus on both because of its prevalence in the Turk population, and due to it’s specific susceptibility to socially framed tasks. Therefore, subjects completed the Liebowitz Social Anxiety Scale (LSAS), and played 80 rounds of a competitive economic game, the Patent Race, against a computerized opponent. Previous research has captured human choices and neural responses on such tasks with a class of computational models referred to as the Experience Weighted Attraction (EWA) model. The EWA model nests two RL strategies, learning action values by a weighted combination of model-free reward sampling, vs. model-based learning (marginalizing) of the opponent’s move distribution. Estimating the parameters of the EWA model that best fit each subject’s choices, we extract parameters that characterize prior belief decay rate, belief updating, and learning rate. In accordance with our hypothesis, we found that self-reported social anxiety is related to a model parameter, delta, thought to reflect higher order belief learning selectively associated with increased use of model-based evaluation (P < .01; 6% increase in model-based learning per 1 SD increase in LSAS). These results are consistent with prior models of SAD that suggest the root of the disorder lies in excessive post-event processing that leads to paralysis in social situations, characterized by overthinking, heightened levels of mentalizing, and upward counterfactual thoughts. Yet, these results are unique in that they associate clinical symptomology with greater computational ability. By grounding the adverse symptoms of SAD in well-delineated neuro-computational mechanisms, these results illuminate the mental processes specific to SAD, and offer a rare example of enhanced function in disease.