Identifying heterogeneous patterns of real-world upper limb use beyond motor severity: a cross-sectional sensor-based study in people with subacute stroke

Abstract Background Upper limb (UL) activity after stroke varies widely, even among individuals with similar motor deficits. Discrepancies between clinical assessments and real-world UL activity may reflect attentional, behavioral, and dominance-related factors. Understanding these patterns is essential for tailoring individualized rehabilitation strategies. We aimed to identify subgroups of people with stroke based on the relationship between motor deficits and real-world UL activity and to determine clinical and behavioral variables associated with cluster membership. Methods In this cross-sectional study, 101 people with stroke wore bilateral accelerometers for 24 hours to measure UL activity in free-living conditions. Motor deficits were assessed using the Fugl-Meyer Assessment for the UL (FMA-UL). Gaussian mixture modeling classified participants based on FMA-UL scores and magnitude ratio. Inter-cluster comparisons were conducted using the Kruskal–Wallis test with Dunn’s test for post-hoc pairwise comparisons. Categorical variables were analyzed using Fisher’s exact test. Decision tree analysis (C5.0) identified clinical variables associated with subgroup membership. Results Six subtypes were identified, showing heterogeneous UL activity patterns across motor deficit levels. In the moderate-deficit groups, one subtype had marked asymmetry despite comparable motor scores, suggesting learned non-use. Among participants with mild impairments, one group exhibited pronounced asymmetry linked to UL inattention, whereas others showed more balanced bilateral engagement or paretic-side-dominant activity patterns. Key clinical variables distinguishing clusters included the Action Research Arm Test, Motor Activity Log, UL inattention, and paretic-side dominance. Conclusion Real-world UL activity after stroke is shaped by factors beyond motor deficits. Combining accelerometer metrics with clinical and cognitive assessments can identify clinically meaningful subtypes, which may inform personalized rehabilitation.