Abstract / Summary
Abstract Purpose: Longitudinal recovery from post-stroke ataxia remains poorly characterized. We examined whether recovery could be characterized by model-derived initial severity (), asymptotic proportional recovery coefficient (r), and recovery time constant ({tau}), and whether trajectories differed across these parameters. Methods: Longitudinal Scale for the Assessment and Rating of Ataxia (SARA) scores from 80 patients with post-stroke ataxia were analyzed using a Bayesian nonlinear mixed-effects model. Trajectory classes were identified using a latent class mixed model. Walking independence was analyzed with interval censoring. Posterior-propagated regressions included , age, sex, and Trail Making Test Part A (TMT-A), with baseline Berg Balance Scale (BBS) included in a functional sensitivity model. Results: Three trajectory classes were identified (n=55, 19, and 6). Class 1 had lower initial severity and shorter {tau}. Class 2 had intermediate initial severity and {tau} broadly comparable to that of Class 1. Class 3 had the greatest initial severity and markedly longer {tau}. Posterior summaries of r overlapped, whereas model-implied proportional improvement by 180 days differed across classes (approximately 0.80, 0.78, and 0.53). Walking independence was observed in 52/55, 18/19, and 0/6 participants. A 10-point higher baseline BBS score was associated with a 15.2-day shorter {tau} (95% credible interval, -27.3 to -5.2). Poorer TMT-A performance was directionally associated with longer {tau}, with substantial uncertainty; associations with r remained uncertain. Conclusions: Recovery trajectories were heterogeneous, particularly in initial severity and timing. Separating recovery magnitude from timing may be more clinically informative than initial severity alone. External validation is needed before patient-level prognostic use.