The conceptualization, measurement, and critical appraisal of computational models of anhedonia in depression

Anhedonia, a cardinal feature of depressive disorders, is classically defined as an inability to experience pleasure, but modern definitions also encompass deficits in anticipatory pleasure and motivation. Validated scales and behavioural tasks have evolved alongside these conceptual shifts, and when integrated with computational modelling, may help reveal mechanisms underlying anhedonia. Reinforcement learning (RL) is the dominant computational framework for studying anhedonia, but fails to fully capture anticipation and motivation. We reviewed the operationalization and measurement of anhedonia from a historical perspective, and conducted a scoping review and critical appraisal of 37 past studies that included computational models of anhedonia using a structured appraisal guide to assess face, construct and predictive validity. We focussed on generative models (i.e., models that can simulate behavioural data) that were paired with measures of anhedonia in both clinical and non-clinical samples. Model types include RL models, RL integrated with functional magnetic resonance imaging, RL integrated with electroencephalography, and models of decision making, effort expenditure, selective attention, and self-referential/memory processes. Our review suggests that anhedonia-related deficits span not only reward processing, but also executive and sensory processing. Models generally demonstrated face validity, lacked predictive validity, and showed construct validity for cognitive-behavioural, but not neurobiological, domains. We propose an integrative, systems neuroscience-inspired approach, which aligns with multidimensional definitions of anhedonia.