Changes in reward-specific learning parameters across adolescence and associations with depression risk

Background: Reward learning is thought to undergo refinement in adolescence, but little is known about how computational components of reinforcement learning develop. Given that adolescence is a sensitive period for reward system plasticity with associated vulnerability for depression, it is important to understand developmental trajectories of different reinforcement learning parameters in normative development and in youth at risk for depression.Methods: Youth aged 9-17 years completed the Play-or-Pass Iowa Gambling Task (PoP-IGT) across five timepoints. We calculated task metrics using a traditional scoring approach—yielding summary scores for good deck play, bad deck play, and net play—and a computational modeling approach—yielding parameters for reward learning rate, punishment learning rate, win frequency sensitivity, and go bias. We examined normative developmental trajectories for each traditional and computational performance metric using multilevel models. We also examined whether maternal history of depression was associated with individual differences in trajectories. Results: Youth showed significant age-related change in several metrics including increasing net play (p < .01), a measure of overall good performance. Further, exploratory analyses found that youth showed significant developmental change in reward-specific learning parameters including increasing win frequency sensitivity (FDR < .005) and decreasing reward learning rate (FDR < .001). In line with hypotheses, youth with a high-risk for depression showed lower reward learning rates in early adolescence (p = .041). Conclusion: Developmental changes in traditional and computational metrics are consistent with the optimization of learning from rewards across adolescence. Observed ages-related changes in reward-related computational parameters specifically are consistent with heightened adolescent reward system plasticity. Additionally, there was support for our hypothesis that maternal history of depression may exert a unique effect on reward learning specifically, but further research across additional reward learning tasks is needed.