A Comparison of ‘Pruning’ During Multi-Step Planning in Depressed and Healthy Individuals

Paul Faulkner, Quentin J. M. Huys, Daniel Renz, Neir Eshel, Stephen Pilling, Peter Dayan, Jonathan P. RoiserbioRxiv 2020 · DOI

Abstract

Background: Real-life decisions are often complex because they involve making sequential choices that constrain future options. We have previously shown that to render such multi-step decisions manageable, people “prune” (i.e. selectively disregard) branches of decision trees that contain negative outcomes. We have theorized that sub-optimal pruning contributes to depression by promoting an oversampling of branches that result in unsavoury outcomes, which results in a negatively-biased valuation of the world. However, no study has tested this theory in depressed individuals.

Methods: Thirty unmedicated depressed and 31 healthy participants were administered a sequential reinforcement-based decision-making task to determine pruning behaviours, and completed measures of depression and anxiety. Computational, Bayesian and frequentist analyses examined group differences in task performance and relationships between pruning and depressive symptoms.

Results: Consistent with prior findings, participants robustly pruned branches of decision trees that began with large losses, regardless of the potential utility of those branches. However, there was no group difference in pruning behaviours. Further, there was no relationship between pruning and levels of depression/anxiety.

Limitations: The relatively small sample size limited the examination of individual differences. The use of other heuristics that are used to render complex decisions feasible, such as memoization and fragmentation, were not examined.

Conclusions: We found no evidence that sub-optimal pruning is evident in depression. Future research could determine whether maladaptive pruning behaviours are observable in specific sub-groups of depressed patients (e.g. in treatment-resistant individuals), or whether misuse of other heuristics may contribute to depression.

TL;DR

When faced with complex, multi-step choices, our brains simplify the problem by "pruning"—reflexively ignoring any path that starts with a significant loss. This study tested the theory that depressed individuals fail to prune effectively, leading them to over-ruminate on negative outcomes, but found that both depressed and healthy individuals actually use this mental shortcut in remarkably similar ways.

Introduction: The Computational Cost of Planning

Real-world planning is computationally expensive. Consider a scenario where you must make eight sequential decisions, each with two choices; you are immediately faced with $2^8$ (256) possible paths. To manage this "combinatorial explosion," the human brain relies on heuristics—mental shortcuts that simplify complex decision trees.

One such heuristic is aversive pruning. When a specific branch of a decision tree begins with a large penalty, we often stop evaluating that branch entirely. This reflexive "cutting" of the tree happens even if that initial loss is followed by a massive reward that would make the path optimal in the long run. Researchers have theorized that a breakdown in this pruning mechanism might be a core component of depression. If an individual fails to "prune" negative paths, they may oversample "unsavoury" outcomes during mental simulation, leading to the negatively biased view of the world characteristic of Major Depressive Disorder (MDD).

Experimental Design and Modeling

To test this theory, researchers recruited 30 unmedicated individuals with MDD and 31 healthy controls. Participants performed a sequential reinforcement-learning task involving transitions between six "states," represented as boxes on a screen.

The sequential decision-making task structure and interface.

In the task, participants moved between states using two keys. Each move resulted in a deterministic reward or loss (e.g., +140, +20, -20, or -140 points). In "plan-ahead" trials, participants were required to commit to a sequence of 2–4 moves simultaneously, forcing them to mentally calculate the best path before acting.

The researchers employed Bayesian model comparison to determine which mathematical strategy best described participant behavior. They tested eight models, ranging from a simple "Lookahead" (calculating all possibilities) to complex "Pruning" models. The winning model, Pruning ‘rho’, included parameters for:

  • General Pruning ($\gamma_G$): A baseline tendency to stop searching deeper into the tree.
  • Specific Pruning ($\gamma_S$): The specific tendency to stop searching a branch after encountering a large loss (-140).
  • Sensitivity ($\rho$): Individual sensitivity to different reward/loss magnitudes, ensuring pruning was distinguished from simple loss aversion.

The Mechanics of Aversive Pruning

The study confirmed that pruning is a robust human behavior. Both depressed and healthy participants routinely missed optimal paths if they required taking an initial -140 hit.

Visualization of aversive pruning in a decision tree.

As shown in the figure above, aversive pruning (orange bars) effectively shuts down the evaluation of branches starting with a large negative value. Interestingly, the study successfully decoupled pruning from simple loss aversion. Pruning is not just a dislike of losses; it is a structural failure to even consider paths containing them. Once the pruning mechanism was accounted for, participants were actually found to be more sensitive to large rewards (+140) than large losses (-140), suggesting a "risk-seeking" element in their search strategy.

Experimental Findings

Contrary to the initial hypothesis, the study found no significant differences in pruning behavior between depressed individuals and healthy controls.

  • Specific Pruning ($\gamma_S$): Healthy = 0.630 vs. Depressed = 0.617 ($p = 0.750$).
  • General Pruning ($\gamma_G$): Healthy = 0.123 vs. Depressed = 0.136 ($p = 0.266$).

Furthermore, there was no significant correlation between the severity of depressive symptoms (measured by the Beck Depression Inventory) and the degree of pruning. Bayesian analyses provided moderate-to-strong evidence for the null hypothesis, suggesting that the lack of difference was a genuine absence of effect rather than a result of low statistical power.

Discussion and Future Directions

The results suggest that the "negatively-biased valuation" seen in depression does not stem from a failure to prune decision trees. However, several caveats remain. The participants in this study were mostly in the "mild to moderate" range of depression and were unmedicated. It is possible that sub-optimal pruning only emerges in more severe or treatment-resistant cases.

Additionally, the study focused exclusively on pruning. Other heuristics, such as memoization (recalling previous solutions) or fragmentation (breaking deep sequences into smaller chunks), might still be where depressive biases manifest. While behavioral performance was identical, future fMRI studies could investigate whether the underlying neural circuitry (such as the subgenual anterior cingulate cortex) operates differently in depressed individuals to achieve these same pruning results.

In conclusion, while aversive pruning is a powerful and robust heuristic for managing complex decisions, it does not appear to be the primary driver of negative bias in mild-to-moderate depression.

Summary updated

Abstract

Abstract Background Real-life decisions are often complex because they involve making sequential choices that constrain future options. We have previously shown that to render such multi-step decisions manageable, people “prune” (i.e. selectively disregard) branches of decision trees that contain negative outcomes. We have theorized that sub-optimal pruning contributes to depression by promoting an oversampling of branches that result in unsavoury outcomes, which results in a negatively-biased valuation of the world. However, no study has tested this theory in depressed individuals. Methods Thirty unmedicated depressed and 31 healthy participants were administered a sequential reinforcement-based decision-making task to determine pruning behaviours, and completed measures of depression and anxiety. Computational, Bayesian and frequentist analyses examined group differences in task performance and relationships between pruning and depressive symptoms. Results Consistent with prior findings, participants robustly pruned branches of decision trees that began with large losses, regardless of the potential utility of those branches. However, there was no group difference in pruning behaviours. Further, there was no relationship between pruning and levels of depression/anxiety. Limitations The relatively small sample size limited the examination of individual differences. The use of other heuristics that are used to render complex decisions feasible, such as memoization and fragmentation, were not examined. Conclusions We found no evidence that sub-optimal pruning is evident in depression. Future research could determine whether maladaptive pruning behaviours are observable in specific sub-groups of depressed patients (e.g. in treatment-resistant individuals), or whether misuse of other heuristics may contribute to depression.
BibTeX
@misc{faulkner2020comparison,
  author = {Paul Faulkner and Quentin J. M. Huys and Daniel Renz and Neir Eshel and Stephen Pilling and Peter Dayan and Jonathan P. Roiser},
  title = {{A Comparison of ‘Pruning’ During Multi-Step Planning in Depressed and Healthy Individuals}},
  year = {2020},
  doi = {10.1101/2020.05.08.084509},
  howpublished = {bioRxiv},
}
Figure