Recovery Trajectories in Post-stroke Ataxia: Modeling a Bayesian Nonlinear Mixed-effects Model

PurposeThe prognosis of post-stroke ataxia remains controversial. It is unclear whether the proportional recovery rule (PRR) established for hemiparesis applies to ataxia, given that cerebellar plasticity suggests trajectories may not depend solely on initial severity. This study was conducted to quantitatively decompose longitudinal ataxia recovery trajectories into proportional recovery coefficient (r) and time constant ({tau}) using a Bayesian nonlinear mixed-effects model, and elucidate their independent determinants and associations with functional walking independence. MethodsWe analyzed longitudinal SARA scores of 80 subacute patients with stroke to estimate individual initial severity (), r, and {tau}. Recovery patterns were clustered based on these parameters. We analyzed the attainment of independent walking using the Kaplan-Meier method and identified predictors via hierarchical multiple regression analysis. ResultsThree distinct clusters were identified. The moderate group (younger, preserved attention) achieved rapid improvement and early walking independence. In contrast, the severe group showed a significantly prolonged time constant ({tau}) but maintained a high proportional recovery coefficient (r), ultimately achieving walking independence in over 90% of cases. Regression analysis revealed a dissociation: biological age constrained the recovery ceiling (r), while attentional function independently regulated recovery speed ({tau}). ConclusionsRecovery from post-stroke ataxia bifurcates into rapid neurological restoration and a delayed process driven by compensatory learning. Especially in severe cases, long-term learning using attentional resources is crucial. These findings challenge prognosis prediction based solely on initial severity, supporting stratified rehabilitation strategies tailored to individual recovery ceilings and learning speeds.