Beyond reward learning deficits: Exploration-exploitation instability reveals computational heterogeneity in value-based decision making in early psychosis.

Psychosis spectrum illnesses are characterized by impaired goal-directed behavior and significant neurophysiological heterogeneity. To investigate the neurocomputational underpinnings of this heterogeneity, 75 participants with Early Psychosis (EP) and 68 controls completed a dynamic decision-making task. Consistent with prior studies, EP exhibited more choice switching, not explained by reward learning deficits, but instead by increased transition to exploration from exploitation. Bayesian modeling implicated elevated uncertainty intolerance and decision noise as independent contributors to suboptimal transition dynamics across individuals, which identified three computational subtypes with unique cognitive and symptom profiles. Replicating prior studies, a high decision-noise subtype emerged showing learning deficits and worse negative symptoms; our analyses further uncovered a normative subtype with worse mood symptoms and a novel uncertainty-intolerance subtype with higher hospitalization rates. These specific microcognitive disruptions underlying the distinct neurocomputational subtypes are individually measurable and may have the potential for targeted interventions.