Cognitive neuropsychiatry has increasingly embraced Bayesian brain frameworks to elucidate the pathogenesis of positive psychotic symptoms. Specifically, researchers frequently invoke predictive processing in psychosis as an overarching paradigm to describe how individuals balance internal expectations against incoming sensory stimuli. Canonical computational models argue that disrupted weighting between prior beliefs and sensory prediction errors generates delusions and hallucinations. However, empirical findings across behavioral paradigms remain strikingly inconsistent. While isolated tasks with modest sample sizes often claim strong evidence for aberrant prior weighting, broader replication attempts have yielded conflicting conclusions. A comprehensive multi-level meta-analysis now provides robust empirical clarity, demonstrating that current evidence does not support a generalised perceptual deficit in prior reliance among individuals with schizophrenia-spectrum disorders.

Theoretical Foundations of Predictive Processing in Psychosis

Predictive coding theory suggests that the healthy human brain functions as an active inference engine. Under this Bayesian model, sensory perception arises from a continuous comparison between top-down predictive priors and bottom-up sensory likelihoods. When discrepancies emerge, neural circuits register prediction errors to update subsequent beliefs. In computational psychiatry, investigators previously hypothesized that psychosis fundamentally distorts this delicate calibration. Consequently, some theorists proposed that overly precise priors cause the brain to override reality, thereby manifesting as hallucinations. Conversely, other frameworks posited that weakened high-level priors allow aberrant prediction errors to flood conscious perception, ultimately fostering persecutory delusions. These contrasting propositions created significant theoretical divergence across clinical neurosciences. Despite intense conceptual enthusiasm, studies frequently generalized findings from idiosyncratic behavioral paradigms to infer global sensory dysfunctions. As a result, psychiatric researchers faced uncertainty regarding whether aberrant prior reliance represents an authentic trait marker or merely a task-dependent artifact.

Methodological Rigour of the Multi-Level Meta-Analysis

To resolve these longstanding discrepancies, investigators conducted a rigorous random-effects, multi-level meta-analysis registered on PROSPERO. The research team systematically queried major biomedical databases, including Embase, MEDLINE, APA PsycINFO, and APA PsycArticles, spanning studies published between January 2005 and October 2024. Furthermore, the researchers utilized the Newcastle-Ottawa Scale to rigorously evaluate the methodological quality and risk of bias across all identified investigations. The final synthesis pooled 34 distinct effect sizes from 27 peer-reviewed comparative studies. In total, the dataset encompassed 904 adults with schizophrenia-spectrum psychosis and 1,039 matched healthy controls. By employing multi-level modeling, the authors successfully accounted for statistical dependencies arising from multiple effect sizes within identical cohorts. Moreover, the investigators extracted specific continuous behavioral metrics that explicitly captured participants' reliance on prior expectations across diverse sensory modalities, establishing an unprecedented benchmark for statistical precision in computational psychiatry.

Key Findings: Absence of Generalised Prior Reliance Deficits

The statistical synthesis delivered definitive findings that challenge prevailing computational assumptions. Across all perceptual paradigms, the pooled effect size indicated no significant difference in prior reliance between patients with schizophrenia-spectrum disorders and healthy controls. Specifically, the analysis revealed an effect size of g = 0.03 with a 95% confidence interval spanning -0.27 to 0.34, yielding a non-significant p-value of 0.818. Furthermore, secondary analyses directly tested associations between prior reliance and specific positive symptom dimensions. When examining delusions across six comparative results involving 183 patients, the correlation remained non-significant at r = -0.16. Similarly, the association with auditory and visual hallucinations across ten results comprising 370 individuals showed negligible correlation at r = 0.04. Therefore, pooled empirical data fail to substantiate the hypothesis that positive symptoms correspond to uniform shifts in prior weighting during perceptual processing.

Evaluating the Hierarchical Model of Predictive Inference

Faced with divergent empirical results, several neuroscientists previously hypothesized that predictive processing deficits operate within a multi-tiered hierarchy. Under this formulation, lower sensory tiers might exhibit weakened structural priors, whereas higher cognitive tiers compensate through overly rigid priors. To empirically evaluate this popular reconciliatory proposal, the meta-analysts performed a detailed subgroup moderation analysis comparing lower-level perceptual tasks against higher-level contextual paradigms. Nevertheless, the statistical output demonstrated that a two-level hierarchical classification failed to reconcile the observed data variance. The subgroup analysis yielded an F-ratio of 0.1 with an associated p-value of 0.758, confirming that hierarchical stratification does not resolve the disparate literature. Thus, neither a unitary deficit nor a simple two-tiered hierarchical disruption explains task outcomes across schizophrenia cohorts. Consequently, computational psychiatrists must reconsider sweeping assumptions regarding hierarchical Bayesian processing failures in severe psychiatric illness.

Clinical Implications for Psychiatric Assessment and Research

These findings present vital practical implications for both clinical researchers and practicing psychiatrists. Because computational tasks do not demonstrate reproducible group-level deficits, clinicians cannot reliably employ current prior-reliance paradigms as objective diagnostic biomarkers. Moreover, psychiatric scholars must exercise caution when extrapolating overarching pathophysiological theories from isolated behavioral tasks. Modest sample sizes and task-specific idiosyncrasies have frequently fueled premature generalizations across psychiatric literature. In addition, these results emphasize that schizophrenia-spectrum disorders encompass profound neurobiological heterogeneity that resists reductive computational classifications. Rather than attributing psychosis to uniform prior weighting anomalies, researchers must investigate multidimensional interactions across attention, neuromodulation, and cognitive flexibility. Therefore, future computational research must prioritize standardized testing batteries and preregistered multi-center replications before proposing unified theoretical doctrines.

Translational Insights for Contemporary Clinical Practice

From a bedside perspective, these meta-analytic conclusions remind clinicians that patient symptomatology rarely mirrors simplified laboratory simulations. Psychotic phenomena present with complex affective, cognitive, and narrative dimensions that standard perceptual tasks fail to capture entirely. Therefore, comprehensive diagnostic interviews, longitudinal psychiatric assessments, and validated psychometric tools remain the undisputed gold standards for clinical management. Although computational psychiatry provides valuable mathematical tools to conceptualize brain dynamics, clinicians must avoid overinterpreting experimental cognitive markers as definitive pathological indices. Moving forward, cross-disciplinary research should integrate neurochemical imaging with nuanced longitudinal phenotyping to illuminate why individual patients experience hallucinations and delusions. Ultimately, empirical scrutiny strengthens psychiatric science by dismantling unverified assumptions and redirecting clinical efforts toward clinically validated diagnostic and therapeutic interventions.

What is the predictive processing theory of psychosis?

The predictive processing theory posits that perception relies on combining prior beliefs with incoming sensory data. In psychosis, theorists hypothesized that an imbalance between these predictions and sensory prediction errors generates delusions and hallucinations, either through overly rigid priors overriding sensory reality or fragile priors amplifying meaningless perceptual noise.

Did the meta-analysis identify any prior weighting deficits in schizophrenia?

No, the multi-level meta-analysis found no significant difference in prior reliance between individuals with schizophrenia-spectrum disorders and healthy controls. Across 27 studies, the pooled effect size was negligible and statistically non-significant, demonstrating that patients do not exhibit a generalised perceptual deficit in updating or maintaining prior expectations.

Can hierarchical Bayesian models reconcile conflicting findings in psychosis research?

Current empirical evidence suggests they cannot. When researchers tested a two-level hierarchical model separating lower sensory priors from higher cognitive priors, the moderation analysis showed no significant explanatory power. Consequently, hierarchical stratification does not adequately account for the conflicting findings observed across diverse behavioral paradigms.

Disclaimer: This content is for informational and educational purposes only. It does not constitute medical advice, diagnosis, or treatment. Refer to the latest local and national guidelines for clinical practice.

References

Miller-Silva C et al. No evidence for a generalised predictive processing deficit in psychosis: A meta-analysis. Psychol Med. 2026 Sep 09. doi: 10.1017/S0033291726105455. PMID: 42711797.

Sterzer P, Adams RA, Fletcher P, et al. The Predictive Coding Account of Psychosis. Biol Psychiatry. 2018;84(9):634-643.

Corlett PR, Horga G, Fletcher PC, Alderson-Day B, Schmack K, Deserno L. Hallucinations and Strong Priors. Trends Cogn Sci. 2019;23(2):114-127.