A generalizable computational mechanism underlying the interaction between momentary craving and decision-making

Substance craving and maladaptive choices are intertwined across addictive disorders. However, the computational mechanisms connecting craving and decision-making remain elusive. Here, we tested a hypothesis that momentary craving and value-based decision-making influence each other during substance-related reinforcement learning. We measured momentary craving as two groups of human participants (alcohol drinkers and cannabis users; total n=132) performed a reinforcement learning task in which they received group-specific addictive cue or monetary rewards. Using computational modeling, we found that, across both groups, momentary craving biased learning rate related to substance-associated prediction errors (RPEs), but not monetary RPEs. Additionally, expected values and RPEs jointly influenced elicited craving across reward types and participant groups. Alcohol and cannabis users also differed in the extent to which their craving and decision-making influenced each other, suggesting important computational divergence between the two groups. Finally, regressions incorporating model-derived parameters best predicted substance use severity in the alcohol, but not cannabis group, supporting the utility of using these model-based parameters in making clinical predictions for selective substance groups. Together, these findings provide a computational mechanism for the interaction between substance craving and maladaptive choices that is generalizable across addictive domains.