Confidence and psychosis: a neuro-computational account of contingency learning disruption by NMDA blockade

Fabien Vinckier, Raphaël Gaillard, Stefano Palminteri, Lionel Rigoux, Alexandre Salvador, Alex Fornito, R. Adapa, Marie‐Odile Krebs, Mathias Pessiglione, Paul C. FletcherMolecular 2015 · DOI

TL;DR

This study uses ketamine, a drug that mimics some aspects of psychosis, to investigate how confidence influences learning in unstable environments. They found that ketamine disrupts this process, leading to a failure to adapt behavior based on learned regularities, suggesting a potential computational basis for early psychosis symptoms.

What Problem Does This Paper Address?

Psychiatry faces a significant challenge in understanding the computational and neurobiological underpinnings of psychotic symptoms like hallucinations and delusions. While reinforcement learning theory offers a promising framework, the precise mechanisms by which the brain extracts and utilizes environmental regularities, especially in dynamic and uncertain situations, remain unclear. This paper aims to bridge this gap by investigating how confidence in learned information influences decision-making and learning, and how this process is affected by NMDA receptor blockade, a pharmacological model for early psychosis. The researchers hypothesize that a core deficit in early psychosis is a disruption in how confidence is updated and used to guide behavior, leading to a persistent doubt that hinders the stabilization of learned policies.

Methods

The study employed a double-blind, placebo-controlled, crossover design with healthy volunteers. Participants underwent functional magnetic resonance imaging (fMRI) while performing a probabilistic instrumental learning task. This task involved learning cue-outcome contingencies that were unstable and reversed periodically, forcing participants to adapt their strategies.

Key methodological components include:

  • Pharmacological Intervention: Participants received either ketamine (an NMDA receptor antagonist known to induce psychosis-like experiences) or a placebo.
  • Probabilistic Instrumental Learning Task: Participants learned to associate abstract cues with different reward probabilities and magnitudes. The task was designed with "risky" (higher variance, higher potential reward) and "less risky" (lower variance, lower potential reward) options, and the optimal choice depended on the cue. Crucially, the contingencies between cues and outcomes were reversed multiple times, creating an unstable environment.
  • Computational Modeling: A hierarchical Bayesian approach was used to model participants' behavior. This involved fitting various reinforcement learning models, including those with a "meta-cognitive" level that incorporated confidence to modulate learning rate and choice temperature. Bayesian model selection was used to identify the best-fitting model for both placebo and ketamine conditions.
  • fMRI Data Analysis: Standard fMRI preprocessing and statistical analysis were performed using SPM5. General linear models (GLMs) were used to identify brain regions whose activity correlated with computational variables derived from the best-fitting model, such as confidence-modulated learning rate and choice temperature.

Datasets

Not discussed in the paper. The study used data collected from participants during the experimental task.

Key Insights and Ideas

  • Confidence as a Regulator of Learning: In stable environments, confidence in learned contingencies allows individuals to downregulate learning and exploit established policies. This study demonstrates that this adaptive mechanism is crucial for navigating unstable environments.
  • Ketamine Disrupts Confidence-Based Adaptation: NMDA receptor blockade with ketamine specifically impaired the ability to adjust learning and decision-making based on confidence. This suggests that the drug interferes with the brain's ability to rely on learned information when uncertainty is low.
  • Persistent Doubt as a Hallmark of Early Psychosis: The findings suggest that a core feature of early psychosis might be a persistent state of doubt, where confidence in learned environmental regularities is compromised. This doubt prevents the stabilization of behavioral strategies, leading to suboptimal performance.
  • Fronto-Parietal Network's Role in Confidence Modulation: The study identified a fronto-parietal network, including the dorsomedial prefrontal cortex (dmPFC), as crucial for integrating confidence into decision-making and learning. Ketamine disrupted the confidence-modulated activity in this network.
  • Computational Psychiatry as a Tool: The research highlights the power of computational psychiatry, combining pharmacological models with sophisticated modeling techniques, to uncover the neural and computational mechanisms underlying psychiatric disorders.

Key Results

  • Behavioral: While overall choice and motor impulsivity did not differ between ketamine and placebo conditions, participants under ketamine showed a reduced ability to optimize their behavior at the end of learning blocks, where confidence should be high. This suggests a failure to stabilize behavior and exploit learned contingencies.
  • Computational Modeling: The best model indicated that confidence modulated both learning rate and choice temperature. However, ketamine significantly reduced the weight of confidence on these parameters, meaning confidence had less influence on how participants learned and made choices.
  • Neuroimaging:
    • Under placebo, brain activity in a fronto-parietal network (including dmPFC) reflected confidence-modulated choice temperature.
  • Under placebo, the ventromedial prefrontal cortex (vmPFC) and bilateral posterior insula showed increased responses to confirmatory outcomes and decreased responses to contradictory outcomes as confidence accumulated. This effect was significantly reduced under ketamine.
  • Ketamine significantly reduced the positive correlation between confidence and choice temperature in the fronto-parietal network.

Limitations and Future Work

  • Sample Size: The authors acknowledge that the experiment was conducted with a limited number of participants, which may affect the statistical power for detecting correlations between subtle psychopathology and behavioral/neural measures.
  • Generalizability: The study used a pharmacological model of psychosis in healthy volunteers. While informative, it does not fully replicate the complex neurobiological and experiential aspects of clinical psychosis.
  • Future Directions: The authors suggest further research to explore the kinetics of these neuro-cognitive perturbations in relation to psychotic symptoms, investigate the link between confidence monitoring and conscious processing, and examine how these mechanisms might be modulated by individual differences in baseline beliefs or introspective abilities. They also propose that this approach can be extended to study other psychiatric disorders characterized by altered belief updating and decision-making under uncertainty.

Summary updated

Abstract

A state of pathological uncertainty about environmental regularities might represent a key step in the pathway to psychotic illness. Early psychosis can be investigated in healthy volunteers under ketamine, an NMDA receptor antagonist. Here, we explored the effects of ketamine on contingency learning using a placebo-controlled, double-blind, crossover design. During functional magnetic resonance imaging, participants performed an instrumental learning task, in which cue-outcome contingencies were probabilistic and reversed between blocks. Bayesian model comparison indicated that in such an unstable environment, reinforcement learning parameters are downregulated depending on confidence level, an adaptive mechanism that was specifically disrupted by ketamine administration. Drug effects were underpinned by altered neural activity in a fronto-parietal network, which reflected the confidence-based shift to exploitation of learned contingencies. Our findings suggest that an early characteristic of psychosis lies in a persistent doubt that undermines the stabilization of behavioral policy resulting in a failure to exploit regularities in the environment.
BibTeX
@inproceedings{vinckier2015confidence,
  author = {Fabien Vinckier and Raphaël Gaillard and Stefano Palminteri and Lionel Rigoux and Alexandre Salvador and Alex Fornito and R. Adapa and Marie‐Odile Krebs and Mathias Pessiglione and Paul C. Fletcher},
  title = {{Confidence and psychosis: a neuro-computational account of contingency learning disruption by NMDA blockade}},
  booktitle = {Molecular},
  year = {2015},
  doi = {10.1038/mp.2015.73},
}
Figure