Computational Psychiatry: Using Computational Modeling to Elucidate Anxiety and Depression-Related Deficits in Decision Making Under Uncertainty

Our lab’s mission over the next 5-10 years is to computationally characterize the aspects of decision-making that are disrupted in anxiety and depression, differentiating those common to both anxiety and depression versus unique to one or the other (or indeed unique to a given sub-dimension of anxious or depressive symptomatology). We are also interested in extending this work to developing a computational model of PTSD. This program of research is being conducted in collaboration with Professor Peter Dayan at the Max Planck Institute in Tubingen. I especially welcome inquiries from potential graduate students and post-docs interested in this line of work. A road-map for our work in these areas can be found in these two theoretical and review papers:

Bishop, S.J., Gagne, C. (2018) Anxiety, Depression, and Decision Making: A Computational Perspective. Annu Rev Neurosci. doi: 10.1146/annurev-neuro-080317-062007. Epub 2018. Apr 25. PubMed PMID: 29709209 pdf

Gagne, C., Dayan, P., Bishop, S.J. (2018) When planning to survive goes wrong: predicting the future and replaying the past in anxiety and PTSD, Current Opinion in Behavioral Sciences, 24, 89-95. ISSN 2352-1546, pdf

Decision-making under second order uncertainty

Our initial work within the field of computational psychiatry has focused on the question of whether individuals with high trait propensity to anxiety struggle to adapt their decision-making to second order uncertainty. We have since extended this to ask whether the deficit observed is unique to anxiety or common to both anxiety and depression.

In every-day life, our decision-making is impacted by many sources of uncertainty. We can try to estimate the probabilities that specific actions will result in particular outcomes and weigh these relative probabilities against the value we place on the potential outcomes. This process is complicated by the possibility that action-outcome contingencies may be volatile, i.e. changing across time. Further, we may have insufficient information to precisely estimate an outcome’s probability. Both volatility and missing information lead to what has become known as second-order uncertainty – difficulty in precisely estimating the 1st order contingency. In initial work, we have shown that individuals high in trait anxiety struggle to adapt their learning rate to changing levels of volatility (Browning et al., Nature Neuroscience, 2015).

From Browning et al. Nature Neuroscience, 2015.

From Browning et al. Nature Neuroscience, 2015.

Difficulties in adapting learning rate to changing levels of volatility could conceivably reflect a generic deficit in learning flexibility. Hence, in ongoing work we are addressing whether high trait anxious individuals also show suboptimal decision-making when second-order uncertainty is created by missing information but there is no learning component to the task (Lawrance et al., in prep). We are also replacing our reliance on a single measure of trait anxiety with dimensional reduction of symptomatology (using bifactor modeling) across multiple measures of anxiety and depression. This enables us to address whether deficits in decision-making under uncertainty are linked to shared or unique components of anxious and depressive symptomatology. Here, our early work (Gagne et al., In Prep) suggests that impoverished adaptation of learning under volatility is not unique to anxiety or depression but loads heavily on a common symptomatology dimension.

From Gagne et al, (In Prep; please do not reproduce without permission). Individuals with low scores on the general factor (common to anxiety and depression) show elevated learning rates under volatility, especially following optimal outcomes. This is not observed or individuals with high scores on the general factor.

In other work we are also looking at effort-based decision-making and model-based decision-making.