A Study on the Correlation Between Different Learning Abnormalities and Delusional-like Beliefs of Various Contents

Research Background

Delusions are a major feature of psychotic disorders (such as schizophrenia, bipolar disorder, depression, and certain neurological and autoimmune diseases). Although delusions clinically manifest as fixed erroneous beliefs, the relationship between the phenomenology of delusions with different contents and their underlying brain and psychological mechanisms remains unclear. The development of computational psychiatry has provided us with a new research approach to better explain the mechanisms of delusions. However, existing predictive error theories are still insufficient in explaining delusions of different contents.

Research Objectives

This study aims to understand the relationship between delusional-like beliefs and learning mechanisms by comparing learning in different cognitive tasks, particularly the role of predictive error signals in this process. Specifically, this study used probabilistic reversal learning (PRL) tasks and Kamin blocking tasks to explore the relationship between these tasks and delusional-like beliefs.

Paper Source

This research was conducted by multiple scholars including Rosa Rossi-Goldthorpe, Steven M. Silverstein, James M. Gold, and others, primarily from institutions such as Yale University, University of Rochester Medical Center, and University of Maryland School of Medicine. The paper was published in the journal “Brain” on April 18, 2024, as an open-access article.

Workflow

Research Subjects and Methods

The study included 452 individuals, with 181 in the clinical high-risk (CHR) group, 161 in the help-seeking control (HSC) group, and 110 in the healthy control (HC) group. Due to the COVID-19 pandemic, all data collection was conducted remotely. All behavioral tasks were implemented through an internet platform, with research assistants guiding participants throughout the tasks. The study received ethical approval from all participating institutions, and all participants provided written informed consent.

Behavioral Tasks

Blocking Task

The blocking task consisted of three phases: learning phase, blocking phase, and test phase. Participants had to learn the causal relationships of fictional patients’ allergies to various foods. The model updated the weight matrix using the classic Rescorla-Wagner (RW) learning rule, applied to each trial. For compound stimuli trials, weights were combinations of all presented cues.

Probabilistic Reversal Learning Task

In the PRL task, participants had to choose between three decks of cards to find the one with the highest probability of reward. The task included two phases with a total of 160 trials. The model was fitted using a Hierarchical Gaussian Filter (HGF), which describes participants’ beliefs about changes in the task environment through multiple levels of learning rate parameters.

Data Analysis

Repeated measures analysis of variance was used to compare win-switch rates (WSR) and HGF parameters between groups. To attempt to identify specific learning mechanisms associated with delusion content, the research team analyzed the relationships between behavioral performance and model parameters in both the blocking task and PRL task.

Main Results

Probabilistic Reversal Learning

Results showed that in the PRL task, high paranoia was significantly associated with irregular win-switching behavior (also known as erratic win-switching). In model parameters, paranoia was associated with lower learning flexibility in volatility processing.

Kamin Blocking

Highly paranoid individuals showed higher “allergic” responses in the blocking task, suggesting poor learning of blocked cues. They also showed deficits in learning control cues. In contrast, individuals with strong non-paranoid delusional-like beliefs showed abnormal learning of blocked cues but intact learning of control cues.

Conclusions

The conclusions of this study are as follows: 1. Paranoia and other types of delusional-like beliefs involve different learning mechanisms. This finding suggests that delusion content may form through different learning disabilities. 2. Paranoid individuals show poorer learning flexibility to changes in the task environment, while individuals with strong non-paranoid delusional-like beliefs exhibit specific selective learning deficits.

Scientific Value and Practical Applications

This study provides a new perspective for understanding different types of delusions, revealing the complex relationship between predictive errors and delusion content. By using behavioral tasks and computational models, this study not only provides theoretical support for explaining the mechanisms of delusions but also offers potential application value for developing future treatments targeting specific delusion contents.

Research Highlights

1. This study is the first to use two different cognitive tasks to explore the relationship between learning abnormalities and delusion content, revealing significant differences in learning mechanisms between paranoid and non-paranoid delusional-like beliefs.
2. The research combines behavioral data and computational models, providing a more comprehensive and in-depth understanding of the formation of delusion content.
3. The results indicate that paranoia is associated with lower flexibility in volatility processing, while non-paranoid delusional-like beliefs are associated with specific selective learning deficits, which has important implications for future treatments of delusions.

Future Prospects

Future research can further explore the differences between different delusion contents at various psychological and neural mechanism levels, and investigate the application prospects of computational models in predicting and intervening in delusions. Additionally, detailed classification of different symptom dimensions can provide more basis for individualized precision treatment.