Neurosurgery for treatment-resistant depression: valuation and decision making

Severe and enduring psychiatric illness is a major cause of long-term disability and reduced life expectancy. Within this group, treatment-resistant depression (TRD) remains a substantial therapeutic challenge. Ablative neurosurgery, defined here as anterior cingulotomy (ACING) which creates bilateral lesions in the dorsal cingulum bundle, and anterior capsulotomy (ACAPS) which creates bilateral lesions in the anterior thalamic radiation, can produce sustained clinical response in carefully selected patients, but the underlying mechanisms are not well understood. Here, we examined whether prior neurosurgical treatment is associated with differences in behavioural and neural processes implicated in TRD. Sixty-two participants completed a probabilistic reward-loss learning task during event-related fMRI: patients with a history of neurosurgical treatment, non-surgical TRD patients, and healthy controls. Behaviour was modelled using hierarchical Bayesian reinforcement learning drift-diffusion models, enabling decomposition of value learning, prediction-error updating, evidence accumulation, and non-decision processes. TRD was associated with impaired value discrimination, blunted reward prediction-error signals, increased loss-related responses, and slowed response dynamics. Behavioural accuracy and learning were preserved across groups, but reaction time analyses revealed graded slowing. Relative to non-surgical TRD, the surgical group showed differences in value-related processing but not in response slowing. Within the surgical cohort, clinical responders showed more efficient evidence accumulation and more normal value representation, whereas non-responders exhibited persistent abnormalities in reward valuation and a 'pessimistic' prediction-error profile. Neuroimaging analyses showed corresponding differences in reward- and loss-related circuits. Responders demonstrated increased reward prediction-error signalling in dorsal anterior cingulate cortex and enhanced reward-value signals in lateral orbitofrontal cortex, whereas non-responders showed persistent blunting of subcortical reward prediction-error responses associated with greater symptom severity. These findings suggest that clinical improvement following neurosurgical treatment is associated with relative normalisation of valuation-related neurophysiology, rather than changes in response slowing. More broadly, they highlight valuation processes as a key component of severe TRD and a potential target for future therapeutic interventions.