Reinforcement Learning Performance in Early Schizophrenia Patients and Their Nonpsychotic Siblings

Abstract

Background: While it is well established that schizophrenia is highly heritable, the genetic architecture of the disorder remains unclear. Reinforcement learning deficits, which have been consistently shown to be present in schizophrenia patients, may represent a potential endophenotypic marker for the disorder. This study aims to compare early schizophrenia patients and their nonpsychotic siblings in reinforcement learning performance. Methods: Thirty-four patents with DSM-IV diagnosis of schizophrenia, schizoaffective disorder, or schizophreniform disorder with illness duration fewer than 5 years were recruited in specialized early intervention service in Hong Kong. Thirty-four nonpsychotic siblings of the patients and 33 healthy controls were also enrolled. Each participant completed a computerized Go/No Go task, in which they had to decide whether or not to choose each stimulus that had different reinforcement probabilities. The task comprised 3 training blocks and 1 transfer/test phase presenting novel combinations of previously learned stimuli. A battery of cognitive assessments was administered to patients, siblings, and controls. Omnibus repeated measures analysis of variance (ANOVA) and 1-way ANOVAs were used to compare reinforcement learning measures of the 3 groups. Results: The 3 groups were comparable in age, gender ratio, and educational level. An omnibus repeated measures ANOVA revealed significant group differences in response accuracy across the training phase, F(2, 98) = 3.23, P = .04). One-way ANOVAs with post hoc Tukey’s honest significant difference tests revealed that patients (M = 0.81, SD = 0.19) performed significantly worse than controls (M = 0.92, SD = 0.11) in reward-driven reinforcement learning in Block 3 (P = .02), and on average, both patients (M = 0.81, SD = 0.13) and siblings (M = 0.82, SD = 0.13) were significantly less accurate in reward-driven learning than controls (M = 0.90, SD = 0.11) across the training blocks (patients vs controls: P = .02; siblings vs. controls: P = .04). There was no group difference in learning measures in the transfer/test phase. Conclusion: Our preliminary results indicate that both early schizophrenia patients and their nonpsychotic siblings exhibit poorer performance on gradual reward-driven (i.e., positive) reinforcement learning in the training phase relative to healthy controls. As the study is ongoing, a comprehensive data analysis with a larger sample size will provide more definitive results regarding whether reinforcement learning may represent a potentially useful endophenotype for schizophrenia research.