Wearable-based digital metrics reveal dynamic dissociations between upper limb capacity and real-world use over the first six months after stroke

Abstract Background After stroke, upper limb recovery is typically assessed using clinical measures of capacity, yet independence in daily life primarily depends on real-world performance. Methods Here, we modeled the longitudinal relationship between activity capacity (Action Research Arm Test, ARAT) and performance (wrist-worn wearable sensors for 48 h) from the early subacute phase to six months post-stroke in 72 patients undergoing conventional rehabilitation. Results While capacity and performance were correlated (Spearman’s $$\rho _s$$ = 0.45–0.78, p < 0.001), their longitudinal coupling weakened over time and became negligible at approximately 2 months post-stroke. Paretic arm use relative to the non-paretic arm (use ratio) plateaued at 2.2 months post-stroke, and ARAT plateaued at 2.7 months post-stroke. At the individual level, 52 $$\%$$ of the participants who improved on ARAT did not improve their use ratio. Conclusions Upper-limb activity capacity and performance do not necessarily co-evolve during stroke recovery. These findings support considering both constructs as complementary targets for personalized rehabilitation.