Gait asymmetry phenotypes in subacute stroke: A cluster analysis of spatiotemporal parameters

Gait asymmetry is a hallmark of post-stroke walking impairment, yet single-parameter measures may miss distinct multivariate patterns. Whether asymmetry-based phenotypes carry prognostic information independent of baseline clinical severity has not been tested.

This retrospective cohort study included 245 individuals with subacute stroke who underwent instrumented walkway gait analysis. Seven variables-six asymmetry indices and leg-length-normalised walking velocity-were entered into k-means clustering. Cluster number was selected by three convergent criteria; stability was assessed by 1000 bootstrap resamples and cross-validation against partitioning around medoids. Functional Ambulation Category at discharge was modelled with cumulative link ordinal regression, with threshold-specific sensitivity analyses.

Three clusters were identified (average silhouette width 0.308; bootstrap Jaccard ≥0.918): Cluster 1 (n = 70, "Stance-Phase Compensatory"), Cluster 2 (n = 112, "Near-Symmetric Fast") and Cluster 3 (n = 63, "Swing-Phase Dominant"). Unadjusted change in the Functional Ambulation Category differed among clusters (H = 18.20, ε² = 0.067, P < 0.001), but cluster membership was not independently associated with the outcome after adjustment for its baseline value, age, onset time and lower-extremity Motricity Index (Cluster 2: odds ratio 0.94, 95% confidence interval 0.40-2.20; Cluster 3: 0.93, 0.48-1.79). Cluster effects were non-significant at every threshold in sensitivity analyses (all P ≥ 0.11).

Three gait asymmetry phenotypes exist in subacute stroke, differentiated by swing-stance phase distribution. Their value appears descriptive-characterising qualitatively different compensation strategies-rather than prognostic; this requires prospective testing.