Quantifying Sub-Optimal Decision Making in Depression

• ¹ Queensland Institute of Medical Research, Complex Systems, Australia
• ² Black Dog Institute, Australia
• ³ University of New South Wales, School of Psychiatry, Australia
• ⁴ Royal Brisbane and Women’s Hospital, Australia

Healthy individuals integrate information and make inferences in a statistically near-optimal manner. Conversely, in depression, cognitive biases may result in sub-optimal inferences, particularly in the processing of emotionally-salient information. Despite this, almost all studies of decision-making in depression use standard statistical procedures such as the general linear model, and have thus neglected known links between signal detection and Bayesian inversion. We hypothesised that depression would be associated with increased perceptual variability around emotionally salient stimuli, and decreased capacity to identify emotionally neutral material. We employed hierarchical Bayes theory to test these hypotheses.

An emotional go/no-go task was administered to three groups of participants; a healthy control group and two groups with diagnosed depression (specifically, melancholic and non-melancholic depression). Participants were required to identify happy, sad or neutral words (signal trials) and ignore other stimuli (noise trials). Hierarchical signal detection theory (SDT) was then used to model the discriminability and bias of each of the three groups of participants. In particular, we compared the posterior means of discriminability and bias between groups as well as the variance of these posteriors.

Control participants were near-optimal for all stimuli types and readily discriminated signal and noise trials. The melancholic group responded more conservatively to positive signal blocks, with the non-melancholic group more liberal on both positive and negative signal blocks. Decreased precision to negative and positive blocks was observed in melancholic compared to non-melancholic participants. Overall, discriminability in melancholic depression was reduced, particularly to neutral signal blocks.

The results indicate that distinctions can be made between depressed and healthy individuals on the basis of perceptual integration of, and responses to, emotionally-salient and also neutral material, and supports the notion of processing biases in depression. The findings provide evidence towards the notion that different inferential cognitive processes underlie different depressive types.