SA28. Reinforcement Learning Impairment Medication-Naive 
First-Episode Psychosis Patients

Background: Reinforcement learning impairment has been observed in chronic schizophrenia and first-episode psychosis. It is, however, known that antipsychotic medication treatment may modulate reinforcement learning performance and hence confound the study results via its pharmacological action on dopamine receptor blockade. Very few studies have been conducted to delineate the illness effect from antipsychotic treatment on reinforcement learning impairment by investigating medication-naïve, first-episode samples. The current study aimed to examine reinforcement learning performance in a representative cohort of antipsychotic-naïve, first-episode psychosis patients. Methods: Thirty-three patients aged between 15 and 40 years presenting with first-episode DSM -IV nonaffective psychotic disorder to specialized early intervention service in Hong Kong and 34 healthy controls matched with age, gender, and educational level were recruited. Each subject completed a computerized reinforcement learning task which has been studied in chronic and clinically stabilized, first-episode schizophrenia patients. In this task, subjects were presented with 4 pairs of stimuli, 1 pair at a time. Choices from each pair resulted in probabilistic outcomes with different reward contingencies (e.g., reward gaining or loss avoidance as positive outcomes). The task comprised 4 raining blocks and 1 test/transfer phase presenting novel combinations of previously learned stimuli. Omnibus repeated-measures analysis of variance (ANOVA) and 1-way ANOVAs were used to compare reinforcement learning performance of patients and controls. Results: Patients exhibited significantly lower win-stay rate in Block 1 ( P < .001) and across training phase ( P = .01) than controls. Repeated-measures ANOVA revealed significant main effect of group ( F (1, 67) = 13.54, P < .01) with patients showing significantly lower accuracy in average positive (reward-driven; P < .01) and negative (punishment-driven; P = .01) reinforcement learning across training phase than controls. There was no significant main effect of group in test/transfer phase. Conclusion: Our results indicate that antipsychotic-naïve, first-episode psychosis patients exhibited mild deficits in rapid positive reinforcement learning as well as in gradual positive and negative reinforcement learning but not value-guided decision-making capacity. The current study thus provides evidence in support of the presence of reinforcement learning impa