Abstract
Interoception, the perception of the internal state of the body, has been shown to be closely linked to emotions and mental health. Anxiety is a potentially debilitating mental health disorder that is thought to involve alterations in interoceptive processing at many levels, from sensation to metacognition to learning. This project aimed to develop a computational model of interoceptive learning by incorporating measures of uncertainty and valence that are thought to be affected by anxiety. The model was tested with data from a pilot study of 16 healthy participants, which will be used to inform multiple longitudinal studies assessing the effect of both pharmacotherapy and exercise on anxiety and interoception. These pilot participants completed a number of questionnaires about anxiety and interoception as well as completing a Breathing Learning Task (BLT) which measured interoceptive learning by asking participants to predict the occurrence of breathing resistances. The task was an extension of that used in a previous study (Harrison et al., 2021), where previously binary predictions were altered to continuous predictions that encoded participant certainty about the outcome. The model was developed to fit a learning rate to each participant using continuous response data, and was subsequently extended to test whether learning rates differed according to stimuli valence. Model simulation of the task and parameter recovery was successful using this continuous response data. The empirical BLT data demonstrated excellent replicability to that observed previously, and the model fit to continuous data closely captured participant behaviour at the group level, with some variability in individual participant fits. The model extension to estimate different learning rates for negative (i.e. breathing resistance) and positive (i.e. no breathing resistance) trials indicated that learning rates did not significantly differ according to stimuli valence. While no relationship was able to be conclusively determined between learning rates and anxiety levels in these 16 pilot participants, this project provided the necessary model validation for the planned longitudinal studies on anxiety.