Current learning theories of anxiety propose that disrupted fear learning underlie anxiety disorders. This suggests that treatments like exposure therapy work by changing learned threat values. However, empirical data does not support these models where changes in fear conditioning lead to symptom changes and later reductions of avoidance behavior. Instead, recent findings suggest that avoidance behavior can be a mechanism on its own, and reductions in avoidance behavior both precede and predict improvements in anxiety symptoms. To target avoidance, a working theory of the processes underlying avoidance behavior is needed. Recently, Markov Decision Process (MDP) models have been used to measure threat expectancy, which has the advantage of capturing the difference between avoidance and fear learning as mechanisms in anxiety disorders. Initial MDP models that demonstrate the rise of avoidance behavior show promise; however, additional non-behavioral information is needed to fit these models in humans.
The researcher's main hypothesis is that MDP models, augmented with decoded neural signals of threat imminence, can characterize and modify anxiety disorder-related avoidance behavior in and outside of the laboratory.
Aim 1: Adults unselected for psychopathology (120, assessed twice): develop a brain signature of threat imminence (from a predictive model independently trained on fMRI data from other threat imminence tasks) and combine with task behavior to create an MDP model of avoidance behavior.
Aims 2 and 3: A separate set of participants with clinical anxiety and maladaptive avoidance (N=163, assessed four times) will complete an MDP-based learning task assessing avoidance during fMRI scanning and quantify differences in MDP-modeled behavior and test if the magnitude of these task-based differences predicts between-and within-person differences in the severity of real-world avoidance behavior.
Multivariate predictors of functional magnetic resonance imaging (fMRI) data can decode latent values and enhance the computational fit of the MDP models. To identify these latent values, previously validated neural signatures of a threat-imminence model will be used. In the threat-imminence model, the same brain regions (e.g., amygdala, vmPFC), but different ensembles and neural populations, are involved in different stages of threat imminence, and these patterns do not differ between humans with different levels of clinical anxiety, allowing for a neurobiologically supported approach to creating latent values of threat. Therefore, researchers aim to use MDP models, augmented with decoded neural signals of threat imminence, to characterize and support a new mechanism (avoidance behavior) of symptom change in anxiety and modify anxiety disorder-related avoidance behavior.
