Reliance on Prior Expectations in Psychosis: A Systematic Review and Meta-Analysis of Perceptual Tasks

This systematic review and meta-analysis of 27 studies involving over 1,900 participants found no evidence supporting the predictive processing theory that psychosis is characterized by a generalized aberrant reliance on prior expectations or specific associations with delusions and hallucinations.

The Gist

The Big Idea: The Brain's "Autopilot" vs. Reality

Imagine your brain is a smart GPS driving you through life. To save energy and react quickly, this GPS doesn't look at every single tree, pothole, or car in real-time. Instead, it relies on predictions (what it expects to see based on past experiences) to fill in the gaps.

• The "Prior": This is the GPS's prediction. If you usually drive down a street with a stop sign, your brain "predicts" a stop sign is there before you even see it.
• The "Sensory Evidence": This is what your eyes and ears actually see and hear right now.

In a healthy brain, these two things balance perfectly. If your GPS predicts a stop sign, but your eyes see a green light, the brain says, "Wait, my eyes are right; the GPS is wrong," and it updates the map.

The Theory of Psychosis:
Scientists have long suspected that in people with psychosis (like schizophrenia), this GPS is broken. The theory suggests that the brain relies too much on its predictions (priors) and ignores what the eyes and ears are actually telling it.

• Too strong a prior: You might "hear" a voice in the wind because your brain is so convinced someone is talking that it ignores the fact that it's just the wind.
• Too weak a prior: You might struggle to recognize a friend's face because your brain isn't using its "friend" database to help interpret the blurry image.

What Did This Study Do?

The authors of this paper decided to play the role of detectives. They didn't just look at one case; they gathered 34 different studies involving over 1,900 people (904 with psychosis and 1,039 healthy controls).

They asked a simple question: "Is the GPS in people with psychosis actually broken compared to healthy people?"

They looked at many different "driving tests" (perceptual tasks) where researchers tricked the brain to see how much it relied on its predictions versus reality.

The Surprising Verdict: "No Clear Evidence"

After crunching all the numbers, the answer was a bit of a letdown for the theory, but a huge step forward for science: They found no consistent evidence that people with psychosis rely on their "GPS predictions" differently than healthy people.

Here is the breakdown of their findings:

1. The "One-Size-Fits-All" Theory is Wrong:
   The idea that all people with psychosis have "stronger" or "weaker" predictions than everyone else is too simple. It's like saying, "All cars with engine trouble have flat tires." Sometimes the engine is fine, sometimes the brakes are bad, and sometimes the tires are fine. The study found that the "GPS" in psychosis is not universally broken in the same way.

2. The "Two-Level" Theory Didn't Work:
   Some scientists thought, "Maybe the problem is only at the bottom level (senses) or only at the top level (thoughts)." They hoped that if they separated the tests into "sensory" (like seeing a shape) and "cognitive" (like guessing a word), they would find a pattern.

   • The Result: Nope. The pattern didn't appear. The "GPS" wasn't broken in just one specific layer.

3. Symptoms Don't Match the Theory:
   They checked if people with hallucinations (hearing voices) or delusions (believing false things) had different "GPS" settings.

   • The Result: No connection. A person with severe hallucinations didn't necessarily have a "stronger" prior than someone without them. The broken GPS theory doesn't explain why these specific symptoms happen.

Why Are the Results So Mixed?

If the theory is so popular, why did this study find nothing? The authors suggest a few reasons, using a Cooking Analogy:

• Different Recipes: Every study used a different "recipe" to test the brain. Some used sounds, some used images, some used words. It's like trying to judge if a chef is bad at cooking by tasting one soup, one cake, and one salad. You can't compare them directly.
• The "Noise" Factor: Some studies found that the brain's predictions only go haywire when the world is very "noisy" or confusing (like trying to hear a whisper in a rock concert). In quiet, clear situations, the brain works fine.
• The "GPS" is Too Complex: The brain might not just have "strong" or "weak" predictions. It might be that the brain gets confused about how much to trust the prediction depending on the situation.

The Takeaway for the Future

This paper is like a reality check for the scientific community.

• Don't Overgeneralize: We can't just say, "Psychosis is caused by too much prediction." It's likely much more nuanced.
• Need Better Tools: Future research needs to stop using small, simple tests and start using bigger, more complex tests that look at how the brain changes its mind over time.
• Hope for Patients: While this study didn't find the "smoking gun" for the cause of psychosis, it helps scientists stop chasing a dead end. Now they can focus on finding the real, more complex reasons why the brain's GPS gets lost, which could lead to better treatments and therapies in the future.

In short: The brain's "prediction machine" is fascinating, but this study proves it's not the simple broken engine we thought it was in people with psychosis. The mystery is still unsolved, but now we know we need a more sophisticated map to find the answer.

Technical Summary

1. Problem Statement

The predictive processing theory of psychosis posits that positive symptoms (hallucinations and delusions) arise from an imbalance in the weighting of prior expectations (priors) versus sensory evidence. The prevailing hypothesis suggests that individuals with schizophrenia-spectrum disorders exhibit aberrant reliance on priors—either over-weighting them (leading to hallucinations/delusions) or under-weighting them (leading to sensory instability).

However, the existing literature presents conflicting results:

• Some studies report stronger priors in psychosis.
• Others report weaker priors.
• Some find no significant difference compared to healthy controls.
• Findings often disagree on whether specific symptoms (delusions vs. hallucinations) are linked to specific types of priors (cognitive vs. perceptual).

These discrepancies are attributed to small sample sizes, diverse task paradigms, and a lack of large-scale synthesis. The authors aim to determine if a generalized deficit in prior reliance exists across perceptual tasks and whether a hierarchical model (distinguishing between low-level perceptual and high-level cognitive priors) can reconcile conflicting findings.

2. Methodology

Study Design & Protocol

• Type: Systematic Review and Multi-level Meta-Analysis.
• Registration: Pre-registered in PROSPERO (CRD42024574379).
• Search Strategy: Comprehensive search of Embase, MEDLINE, APA PsycINFO, and APA PsycArticles (Jan 1, 2005 – Oct 31, 2024).
• Inclusion Criteria:
  • Population: Adults (≥18) with schizophrenia-spectrum psychosis (medicated or unmedicated) vs. healthy controls (HC). Excluded: At-risk mental state (ARMS), drug-induced psychosis, or neurological disorders.
  • Intervention: Tasks with controlled manipulation of priors (explicit or implicit induction of expectations).
  • Outcomes: Quantitative, continuous, behavioral measures of reliance on priors (e.g., accuracy, response bias). Imaging and metacognitive tasks were excluded.
• Exclusion: Studies without direct prior manipulation, reviews, case reports, and non-English papers.

Data Analysis

• Sample: 34 results from 27 studies (Total $N=1,943$: 904 patients, 1,039 controls).
• Statistical Model: Three-level random-effects meta-analysis using R (metafor and dmetar packages).
  • Level 1: Sampling variance (participants).
  • Level 2: Within-study variance (multiple tasks/experiments per study).
  • Level 3: Between-study variance.
• Effect Size: Hedges' g (standardized mean difference). Positive values indicate stronger priors in patients; negative values indicate weaker priors.
• Sub-group Analyses:
  • Prior Type: Coded as Perceptual (implicit, low-level sensory constraints) vs. Cognitive (explicit, high-level beliefs/schemas).
  • Symptom Correlation: Meta-analyses of correlations between prior strength and hallucinations and delusions (using Pearson's r).
  • Meta-Regression: Tested if PANSS-Positive scores moderated the effect size.
• Quality Assessment: Newcastle-Ottawa Scale (NOS); sensitivity analyses excluded high-risk-of-bias studies (e.g., Colbert et al., 2010).

3. Key Contributions

• First Large-Scale Meta-Analysis: This is the first study to aggregate data across diverse perceptual paradigms to test the predictive processing hypothesis in psychosis with sufficient statistical power.
• Hierarchical Testing: It empirically tests the "two-level hierarchy" theory (Haarsma et al., 2020), which suggests that weaker perceptual priors might be compensated by stronger cognitive priors in psychosis.
• Symptom Specificity: It moves beyond general group comparisons to specifically investigate the link between prior reliance and distinct positive symptoms (hallucinations vs. delusions).
• Methodological Rigor: By using a three-level model, the study accounts for the non-independence of multiple effect sizes within single studies, reducing the risk of false positives common in previous literature.

4. Key Results

A. General Reliance on Priors (Patients vs. Controls)

• Main Finding: There was no significant difference in the overall reliance on priors between patients with psychosis and healthy controls.
  • Effect Size: $g = 0.03$ ($95\%$ CI $[-0.27, 0.34]$; $p = .818$).
• Heterogeneity: Significant heterogeneity was observed ($I^2 = 70.37\%$), driven primarily by between-study differences.
• Prediction Interval: The 95% prediction interval $[-1.51, 1.58]$ indicates that future studies could find increased, decreased, or no difference, suggesting no consistent general deficit.
• Sensitivity: Excluding outliers (e.g., Fogelson et al., 2011) or high-bias studies did not alter the null result.

B. Prior Type (Perceptual vs. Cognitive)

• Sub-group Analysis: The distinction between perceptual and cognitive priors did not explain the heterogeneity.
  • Perceptual priors: $g = -0.01$ ($p = .979$).
  • Cognitive priors: $g = 0.08$ ($p = .758$).
  • Difference between types: $F(1,32) = 0.1$, $p = .758$.
• Implication: The "compensatory hierarchy" theory (weak perceptual + strong cognitive priors) was not supported by the aggregate data.

C. Relationship with Symptoms

• Hallucinations: No significant correlation between prior strength and hallucination severity.
  • $r = 0.04$ ($95\%$ CI $[-0.28, 0.36]$; $p = .780$).
• Delusions: No significant correlation between prior strength and delusion severity.
  • $r = -0.16$ ($95\%$ CI $[-0.51, 0.19]$; $p = .293$).
• Moderation: Prior type did not moderate the relationship between priors and symptoms. Meta-regression using PANSS-Positive scores also showed no significant effect.

5. Significance and Conclusion

Theoretical Implications
The study challenges the generalized predictive processing deficit hypothesis. It suggests that psychosis is not characterized by a universal over- or under-reliance on priors across all perceptual domains. Instead, predictive processing alterations may be:

• Task-specific: Dependent on specific stimulus features (e.g., sensory uncertainty, signal-to-noise ratio).
• Nuanced: Potentially related to specific computational objects (e.g., numeric vs. causal priors) rather than broad perceptual/cognitive categories.
• State/Trait Dependent: The study included mixed active/remitted patients; future work is needed to determine if deficits are state-dependent.

Clinical and Research Directions

• Caution in Generalization: Researchers should avoid extrapolating results from single tasks to a broad theory of psychosis.
• Refined Paradigms: Future studies should focus on:
  • Uncertainty: Investigating how patients adjust priors under varying levels of sensory uncertainty.
  • Task Specificity: Isolating features that trigger differential processing (e.g., voice detection in noise).
  • Longitudinal Designs: Tracking predictive processing changes across illness stages (ARMS to FEP to chronic).
• Intervention: The lack of a generalized deficit suggests that cognitive interventions targeting "prior weighting" broadly may need to be more specific to the symptom profile and task context.

Conclusion
This meta-analysis provides robust evidence against a generalized alteration in prior reliance in psychosis. The conflicting results in the literature are likely due to task-specific factors and the complexity of the predictive processing hierarchy, rather than a single, unified mechanism of aberrant prior weighting.