441. New Learning or Unlearning: Computational Heterogeneity in Extinction Predicts the Recovery of Threat Responses
Section snippets
Background
Anxiety disorders are characterized by persistent and debilitating fear. Exposure therapies, based on the principles of extinction learning, are effective for many. However, for some, extinguished threat responses reemerge, a phenomenon that compromises the efficacy of such therapies. Here, we propose that this heterogeneity might stem from qualitative individual differences in the nature of extinction learning that can be captured by a computational process model.
Methods
We fit to participants’ physiological conditioning data a model (Gershman et al., 2010) positing that learners attempt to segment their experience into “states” or “latent causes” that capture regularity in the configuration of observed stimuli (cue and reinforcement). The model can distinguish those who appear to cluster CS and US observations during threat learning and extinction into a single state, effectively updating the original threat association, or into two states reflecting separate
Results
Although spontaneous recovery of extinguished threat responses is a well-documented phenomenon at the group level, there was substantial heterogeneity in our sample. The model distinguished those whose data was best fit by a one-state unlearning-like representation from those who appeared to infer distinct threat and safety states. Strikingly, only the group of “two-state” learners exhibited spontaneous recovery the following day.
Conclusions
Qualitative variation in extinction learning may have important implications for understanding vulnerability and resilience to fear-related psychiatric disorders. Moreover, developmental changes in extinction learning may provide insight into the typical onset and peak prevalence of anxiety in adolescence.
Keywords
Extinction Learning and Recall, Computational Modeling