Computational modeling for addiction medicine: From cognitive models to clinical applications.

Decision-making tasks that have good ecological validity, such as simulated gambling tasks, are complex, and performance on these tasks represents a synthesis of several different underlying psychological processes, such as learning from experience, and motivational processes such as sensitivity to reward and punishment. Cognitive models can be used to break down performance on these tasks into constituent processes, which can then be assessed and studied in relation to clinical characteristics and neuroimaging outcomes. Whether it will be possible to improve treatment success by targeting these constituent processes more directly remains unexplored. We review the development and testing of the Expectancy-Valence and Prospect-Valence Learning models from the past 10 years or so using simulated gambling tasks, in particular the Iowa and Soochow Gambling Tasks. We highlight the issues of model generalizability and parameter consistency, and we describe findings obtained from these models in clinical populations including substance use disorders. We then suggest future directions for this research that will help to bring its utility to broader research and clinical applications.