Transition from a categorical to a dimensional approach has been proposed in the field of psychosis. However, whether key features of schizophrenia, such as cognitive deficits, really do lie along a linear continuum remains uncertain. To explore this, we compared for the first time verbal learning impairments in six entities of the psychosis spectrum using linear, nonlinear, and categorical models.

Studies involving verbal learning tests in familial high risk, clinical high risk, schizotypy/schizotypal, ultra-high risk, first episode of psychosis (FEP), and chronic schizophrenia populations were systematically searched in three databases in September 2024. Studies were included if they reported an immediate, delayed, or total recall measure in subclinical or clinical entities and healthy controls. The metafor package was used to compute effect sizes for the comparison between cases and control groups, categorized by psychosis entities. Model comparisons were also performed to compare linear, nonlinear, and categorical distributions of the effect sizes.

The meta-analysis aggregated a total of 262 studies in the psychosis spectrum. Effect sizes were moderate in at-risk populations (<0.50) and large in clinical populations (−1.00 for FEP and >1.00 for chronic schizophrenia). A nonlinear model best explained our data in immediate recall, while the results in delayed and total recall suggest the inferiority of linear models.

Our findings suggest a discontinuity in verbal learning between at-risk populations and clinical entities, challenging a purely linear dimensional model of cognitive impairment in the psychosis spectrum.