Abstract
Mental health conditions such as attention-deficit/hyperactivity disorder (ADHD) and mood disorders show marked symptom heterogeneity, complicating diagnosis and treatment. Computational psychiatry offers a way forward by using mathematical models, such as sequential sampling models, applied to trial-by-trial behavior in well-defined neurocognitive tasks, to infer latent mechanisms underlying behavior. In ADHD, this approach has revealed consistent alterations in information integration (reduced drift rates) in attention-demanding tasks and also indicates that combinations of different model parameters (increased drift rate and longer nondecision time) distinguish the different neurocomputational mechanisms that underlie symptom dimensions. Early work in ADHD also suggests that drift rate predicts illness trajectories and provides insights into treatment response. Yet current applications remain preliminary, limited by task constraints, assumptions in model specification, and questions of reliability and generalizability of the derived parameters. Integrating mechanistic modeling with naturalistic tasks, physiological measures, and longitudinal designs may help to disentangle context-specific from generalizable processes. Ultimately, shifting from symptom descriptions to mechanistic models of belief and behavioral adaptation in dynamic environments may pave the way for next-generation assessments in ADHD, and help to support interventions that are ecologically valid, developmentally informed, and adaptive to patients' changing needs across time and context.