Pavlovian bias is associated with symptom severity but not diagnostic status in individuals with both anxious and non-anxious depression

Abstract

Human decision-making favors approach behavior to gain reward and avoidance behavior to prevent punishment. While adaptive in some cases, this Pavlovian bias can also interfere with instrumental motivation, leading to suboptimal decisions. As maladaptive avoidance is a known maintenance factor in anxiety and depression, this could be especially relevant to individuals with these disorders. To assess whether Pavlovian bias is altered in this population, we examined 106 healthy comparisons (HCs), 88 individuals with depression (Dep), and 184 with comorbid anxiety and depression (AnxDep) using an Orthogonalized Go/No-Go Task. Participants' choices and reaction times were modeled using a reinforcement learning and drift diffusion model (RL-DDM). As expected, results showed that accuracy was highest when Pavlovian and instrumental biases aligned and lowest when they conflicted. Linear models revealed no group differences in accuracy, reaction times, or any of the computational parameters. However, Pavlovian bias was positively associated with depression severity across individuals with both anxious and non-anxious depression. Anxiety sensitivity was also positively associated with Pavlovian bias in the AnxDep group specifically. Consistent with this, both depression severity and anxiety sensitivity in this group were also negatively associated with accuracy on the task when approach actions were required to avoid punishment. These results suggest that Pavlovian bias may contribute to symptom severity in both unique and overlapping ways within anxious vs. non-anxious depression. This may also specifically reflect suppression of approach behaviors when they would have adaptive value - potentially amplifying the avoidance behaviors known to maintain these disorders.

Fig. 1. Visual depiction of the Orthogonalized Go/No-Go Task.

After a fixation lasting between 750 ms and 1500 ms, participants observed one of four shapes. Participants could emit a “go” response by tapping the spacebar within 1500 ms of stimulus onset or could emit a “no-go” response by allowing the time to elapse. After a 2000ms fixation, participants were shown a green upward-facing arrow (win), yellow horizontal bar (neutral), or red downward-facing arrow (loss) as a probabilistic result of their choice.

Fig. 2. Accuracy and mean reaction time for each group and task condition.

Boxplots show the median and quartile values in addition to data points. The white circles and surrounding bars indicate the mean and standard error estimates. Accuracy was highest in the 'go to win' (GTW) condition, not significantly different between the 'go to avoid losing' (GTL) and 'no-go to avoid losing' (NTL) conditions, and lowest in the 'no-go to win' (NTW) condition. Reaction times were slower in the GTL condition than in the GTW condition, and faster for error responses in the NTW than NTL conditions (i.e., as reaction times cannot be calculated for no-go responses). This suggests less conflict between instrumental and Pavlovian control in the GTW than GTL condition, and more impulsively driven errors in the NTW than NTL condition – both consistent with the direction of Pavlovian bias.

Fig. 3. Regressions predicting GTL accuracy based on affective symptom measures within the combined clinical groups.

Top-left: Patient Health Questionnaire (PHQ) scores negatively related to accuracy in the 'go to avoid losing' (GTL) condition. Top-right: Overall Anxiety Severity and Impairment Scale (OASIS) scores also negatively related to GTL accuracy. Bottom-left: Anxiety Sensitivity Index (ASI) scores related to GTL accuracy in a group-specific manner. Bottom-right: the Brooding subscale of the Ruminative Response Scale (RRS) also related to GTL accuracy in a group-specific manner. Shaded areas indicate confidence intervals for each trend line. The terms p(main effect) and p(interaction) indicate the p-value associated with the predictor variable on the x-axis and the interaction between that variable and group, respectively. These models included age and sex as covariates.

Fig. 4. Computational parameter values separated by group.

These results supported the absence of a group difference for each computational parameter. Boxplots show the median and quartile values in addition to single data points. Overlaid white circles and error bars indicate means and standard errors.

Fig. 5. Regressions predicting Pavlovian bias based on affective symptom measures within the combined clinical groups.

Pavlovian bias was positively related to Patient Health Questionnaire (PHQ) scores (top-left). Groups differed in the association between Pavlovian bias and the Brooding subscale of the Ruminative Response Scale (RRS-Brooding; top-right), the physical anxiety subscale of the Anxiety Sensitivity Index (ASI-Physical; bottom-left), and the social anxiety subscale of the Anxiety Sensitivity Index (ASI-Social; bottom-right). Shaded areas indicate confidence intervals for each trend line. The terms p(main effect) and p(interaction) indicate the p-value associated with the predictor variable on the x-axis and the interaction between that variable and group, respectively. These models included age and sex as covariates. Note that the interaction effect of RRS-Brooding and group predicting Pavlovian bias did not survive correction for multiple comparisons.