How do we learn to feel safe or threatened?

Background

Anxiety disorders are the most common type of mental disorder worldwide. People with anxiety disorders have a heightened sense of threat and danger that is excessive and not in proportion to the situation of concern. Exposure therapy is an effective psychological treatment for anxiety disorders. Exposure means "facing your fears", or systematically approaching a feared (but actually safe) situation to overcome excessive fear. Decades or research and clinical experience demonstrates that exposure therapy improves all types of anxiety disorder. Despite this, researchers are not sure how and why it works. The leading view currently is that exposure works by providing experiences where clients can learn that their fears don't come true. The technical term for when a fear doesn't come true is "threat prediction-error". Discrepancy between anxious predictions and actual outcome vary from being a lot worse than expected, as-expected, to a lot better than expected. Current theory posits that larger threat prediction error during an exposure will lead to more safety learning. Exposure therapists should thus try to create conditions where people are surprised by the outcome and see that their fears don’t come true to the greatest extent possible. Evidence supporting this theory of exposure therapy comes from basic fear learning research with animals and humans and using controlled experimental procedures in laboratory settings. This fear learning research often involves participants viewing trials of abstract geometric shapes on a computer screen. Some shapes pair with an electric shock or loud noise and some don't. Researchers measure fear to see how it changes through different stimuli combinations.

This research shows that fear in this kind of procedure reduces after threat prediction errors. Some researchers assume that the findings from this kind of research will translate to the "real-world" context of therapy in the clinic. However, findings may not transfer to the more complex and "messy" settings that characterise everyday life. We need more research to determine whether the threat prediction error model of exposure translates to real-world contexts outside of the laboratory. It is important to know the therapeutic mechanisms of our treatments. If we don't know the true treatment mechanisms, we may be subject clients to unnecessary procedures. Instead, we could drop unneeded procedures to focus on the real mechanisms of change. This study aims to address these gaps by exploring whether threat prediction error is associated with anxiety reduction through a procedure that resembles exposure therapy in the clinic. This procedure involves give six short speeches to a researcher over two days and rating anxiety, threat perception, and perception of threat outcome for each speech.

Research Questions / Hypotheses

Does overall threat prediction error, threat expectancy change, and learning rate prediction anxiety reduction during exposure? We thought that greater overall threat prediction error would be associated with more anxiety reduction.
For the same person, does threat prediction error and expectancy change from one exposure predict anxiety at the next exposure? We thought that it would.
Does prediction error relate to how surprising the exposure is? We thought exposures would be more surprising if there was greater prediction error.

Participants

One hundred and thirty people met with the researcher between May 2021 and October 2022, with 129 providing informed consent. Of these 129 participants, 2 withdrew before attempting any speeches and 2 were excluded post-hoc due to insufficient English proficiency that became evident during testing. The remaining 125 participants collectively gave 678 speeches out of a possible 750 (9.6% missing). A further 11 observations were re-coded as missing due to testing errors (0.99% of all scheduled observations). The final analysed dataset comprised of 667 observations (11% missing) provided by 125 participants.

Methods

We adapted the procedure from other research exploring speech performance anxiety (Chen et al., 2018; Harvey et al., 2000; Orr & Moscovitch, 2010). Participants were asked to complete two online testing sessions (one hour duration) separated by seven days and conducted by video teleconference using Zoom (https://zoom.us/). All measures were administered via Qualtrics (https://www.qualtrics.com/au/). Participants attended sessions alone and were asked to deliver six three-minute speeches on a pre-selected topic (such as “Should cigarettes be banned?”) to a single researcher (first author) over the two testing sessions (three speeches per session). Participants were given one minute to prepare what to say before starting each speech. At the start of each session, participants selected three topics rated between five and seven in familiarity on an 11-point Likert scale (0 = “Not at all familiar” to 10 = “Extremely familiar”) from a randomised list of 47 topics. The speech topics were thus constrained to be approximately moderately familiar to the participant, enough to speak for three minutes on the topic but not enough to be very “easy”. Constraining speech familiarity provides some control for task difficulty while reducing practice effects that would occur in repeating speeches on the same topic (Chen et al., 2018; Orr & Moscovitch, 2010). Selecting topics one session at a time also minimises and controls for preparation time. Participants completed questionnaires measureing threat predictions, outcomes, anxiety, and surprise intensity for each speech. Participants completed questionnaire measuring social anxiety symptom severity, mood, and stress at the beginning of each testing session.

Results

We examined between and within person relationships between learning indicators and outcomes over exposure using multilevel modelling. Average expectancy change and prediction error learning rate, but not average prediction error per se, was associated with more anxiety reduction across the exposure. Within person, anxiety was lower after exposures that triggered more expectancy change. Threat prediction error wasn’t linearly associated with with anxiety at the next exposure. Higher threat prediction error during an exposure was associated with greater subjective surprise for that exposure.

Implications

We concluded that threat prediction error is necessary but not sufficient for effective exposure. Success depends on how much the patient converts exposure-related prediction errors into threat expectancy change. This study suggests exposure-based CBT outcomes for anxiety may be enhanced by emphasising threat expectancy change rather than threat prediction error per se. Therefore, therapists should design exposures to generate maximal threat prediction error but should monitor changes to threat expectancy over repeated exposures to assess progress, rather than only tracking expectancy violations. Therapists should remain aware that large patient reported threat prediction errors may not translate to changes in threat beliefs and that change in threat beliefs seems important for therapeutic outcome. Future research should focus on ways to enhance the prediction-error learning rate from exposures. These results were presented at the AACBT National Conference in 2023 and has been submitted to a journal for publication. We plan to present the results at other relevant scientific conferences over the next year.