(PS4-B37) Examining Latent State Inference Biases in Individuals with PTSD During Fear Learning

PTSD is characterized by higher threat expectancies and deficits in distinguishing between threat and safe contextual environments. One hypothesis regarding heightened threat expectancies and poor discrimination in PTSD is that these processes reflect biases in a specific type of inferential decision-making: latent state inferences; that is, forming abstract models of contexts and then correctly inferring when these abstract models are applicable. Applying a latent state paradigm, we can further examine how individuals with PTSD form and infer abstract models about fear associations. We hypothesize that PTSD is associated with a structured inference bias (i.e., over inferring high danger contexts) rather than a generic inference bias (i.e., global latent state misattributions). This study can aid in an improved understanding of how PTSD generalizes fear across different contexts.  A novel task, which we refer to as the Mystery Room task, was utilized that consists of two phases on consecutive days; latent state learning (Day 1) and latent state inference (Day 2). During the learning phase, participants are placed into four different ‘rooms’ distinguished by colors while being shown three neutral faces that are paired with a shock with varying probabilities depending on the room. Furthermore, of the four rooms with distinct face-shock probabilities, two rooms have a lower overall shock probability to represent a “low danger” context, while the other two have a higher overall shock probability to denote a “high danger” context. During latent state inference, participants are not explicitly shown what ‘room’ they are in, rather they are expected to infer which ‘room’ they are in based on observing face-shock outcomes probabilities. In each ‘mystery room’ on day 2, they provide ratings of the likelihood that they are in a given ‘room’. Data collection is on-going, and an interim analysis for this poster submission was completed with an n=33 (n=7 PTSD), with a final anticipated N = 200 (n=100 PTSD). Linear mixed effects models analyzed the day 2 inference ratings with a 2 (true room: low vs high danger) x 2 (inferred room: low vs high danger) x 2 (PTSD) mixed models. We observed a significant true x inferred x PTSD interaction, t=-3.22, p=.001, such that individuals with PTSD were more likely to infer high danger rooms when actually in low danger rooms, t=2.8, p=.005, compared to controls, and PTSD demonstrated no difference in inference during high danger rooms. These initial results suggest that heightened threat expectancies in PTSD may, at least partially, be due to biases in latent state inferences, with possible implications for trauma-focused psychotherapy such as exposure therapy and cognitive processing therapy.