Bayesian attenuation of offset analgesia filters out random disturbances in noxious stimuli

Offset analgesia (OA), an endogenous pain inhibition after an abrupt decrease in noxious stimulation, provides a paradigm to study dynamic interaction between ascending and descending pain pathways. Previous studies assumed that this interaction follows deterministic dynamics. In contrast, a recent perspective views pain perception as a Bayesian process: a statistically optimal updating of pain predictions based on noisy sensory input.

We examined whether OA is driven by a deterministic interaction between ascending and descending pathways, or by a Bayesian process in which the brain updates pain perception by combining expectations with incoming signals.

We modified the conventional OA paradigm by adding high-frequency noise after an abrupt decrease in noxious stimulation and measured pain intensity responses in healthy participants. Pain reports were analyzed using 2 computational models: a deterministic dynamic equation model and a recursive Bayesian integration model. Hypothesis testing was conducted using model selection.

Offset analgesia was observed after reduction of noxious stimuli, but pain was disinhibited by high-frequency disturbances. The deterministic model predicted unbounded oscillations depending on disturbance sequence, whereas the Bayesian model predicted gradual OA attenuation by filtering out noise. Model selection favored the Bayesian model.

The brain dissociates noise from primary signals, achieving stable pain perception even in the presence of noisy inputs. Thus, OA reflects a stochastic integration between prediction and observation, with noise magnitude modulating pain intensity. Clinically, these results suggest that enhancing endogenous pain inhibition for chronic pain may be achieved through interventions targeting noise recognition mechanisms.