Difficult-to-Treat Anxiety: A Neurocomputational Framework

Anxiety disorders, affecting approximately 1 in 9 individuals globally, impose significant socioeconomic and health burdens, with many individuals failing to achieve symptom remission despite standard treatments. Difficult-to-treat anxiety (DTA) encompasses a broad spectrum of persistent anxiety disorders that remain refractory to conventional interventions, necessitating a shift from rigid response-based criteria to a mechanistically driven framework that integrates computational psychiatry and systems neuroscience. Dysregulated approach-avoidance decision making, where heightened punishment sensitivity, inflexible belief updating, and uncertainty misestimation drive persistent avoidance behaviors and reinforce maladaptive anxiety cycles, is central to DTA. Computational modeling of reinforcement learning tasks reveals exaggerated Pavlovian biases and impaired exploratory learning, while predictive processing models highlight overestimation of threat and rigidity in safety learning, perpetuating chronic anxiety. Neural dysfunction in default mode and negative affective networks, characterized by hyperstable attractor states in the amygdala and impaired top-down regulation by the prefrontal cortex, further sustains maladaptive anxiety states. Novel interventions that target these dysfunctions-such as neuromodulation, precision pharmacotherapy, and personalized digital therapeutics-offer potential breakthroughs in managing DTA. In this review, we synthesize current evidence on computational, neural, and behavioral mechanisms that underlie DTA and propose an integrative, process-targeted approach to assessment and treatment. Future research must refine biomarker-driven subtyping and individualized interventions, moving beyond trial-and-error approaches toward mechanistically informed precision psychiatry for persistent anxiety disorders.