Wearable-Based Digital Metrics Reveal Dynamic Dissociations Between Upper Limb Capacity and Real-World Use Over the First Six Months After Stroke

Technical Summary: Wearable-Based Digital Metrics Reveal Dynamic Dissociations Between Upper Limb Capacity and Real-World Use Over the First Six Months After Stroke

Problem Statement
Post-stroke rehabilitation traditionally relies on clinical measures of upper limb "capacity" (what a patient can do under standardized conditions) to assess recovery. However, functional independence in daily life depends on "performance" (what a patient actually does spontaneously). While previous studies have established moderate-to-strong cross-sectional correlations between capacity and performance, the longitudinal dynamics of these constructs remain incompletely understood. Specifically, it is unclear whether improvements in clinical capacity systematically translate into increased real-world arm use over time, and whether the relationship between these two constructs evolves or decouples during the subacute recovery phase. Furthermore, existing evidence is largely derived from Western cohorts, limiting generalizability to Asian populations, which exhibit different stroke epidemiology (e.g., higher rates of hemorrhagic stroke).

Methodology
This study utilized data from the TAILOR study, a prospective longitudinal observational cohort of 72 Asian stroke survivors recruited from the Tan Tock Seng Hospital Rehabilitation Centre in Singapore. Participants were assessed from the early subacute phase (2–5 weeks post-stroke) through six months post-stroke.

• Capacity Measurement: Upper limb activity capacity was quantified using the Action Research Arm Test (ARAT), a standardized clinical scale measuring maximal ability across grasp, grip, pinch, and gross movement.
• Performance Measurement: Real-world arm use was objectively quantified using wrist-worn inertial measurement units (IMUs) worn continuously for 48 hours. Data were processed to classify functional arm movements versus non-functional movements based on activity counts and forearm elevation. Three multidimensional metrics were derived:
  • Functional Arm Use Ratio (FAUR): The ratio of paretic to non-paretic arm use duration (symmetry).
  • Functional Arm Use Duration (FAUD): The percentage of awake time spent in functional use.
  • Functional Arm Use Intensity (FAUI): The mean movement magnitude during functional use.
• Predictors: A comprehensive set of baseline predictors was collected, including physiological markers (Fugl-Meyer Assessment, MEP status, muscle tone), demographic factors (age, gender), cognitive screening (MoCA), and contextual factors (stroke type, inpatient duration, hand dominance).
• Analysis:
  • Association: Spearman's rank correlation assessed cross-sectional associations. Longitudinal coupling was modeled using ordinary least squares (OLS) regression to determine if the rate of change in performance predicted the rate of change in capacity over time.
  • Trajectories: Mixed-effects models (MEMs) characterized population-level recovery trajectories and identified plateau times (defined as reaching 90% of the asymptotic value).
  • Individual Improvement: Simulation-based posterior predictive distributions were used to classify individual improvements (significantly greater than zero) without relying on arbitrary Minimum Clinically Important Difference (MCID) thresholds.
  • Predictor Modeling: A nested cross-validated approach with L2-penalized logistic regression identified stable baseline predictors for improvements in capacity and performance metrics.

Key Contributions and Results

1. Dynamic Decoupling of Capacity and Performance:
   While capacity (ARAT) and performance metrics were moderately to strongly correlated at individual time points ($\rho_s = 0.45–0.78$), the longitudinal coupling between their changes weakened significantly over time. The association was positive early in recovery but became indistinguishable from zero from approximately two months post-stroke onward. This suggests a progressive dissociation where gains in clinical capacity do not necessarily drive concurrent gains in real-world use.

2. Divergent Recovery Trajectories and Plateaus:
   Population-level modeling revealed that performance metrics plateaued earlier than clinical capacity.

   • FAUR (Ratio) and FAUI (Intensity) plateaued at approximately 2.2 and 2.1 months, respectively.
   • ARAT (Capacity) plateaued later, at 2.7 months.
   • FAUD (Duration) plateaued last, at 3.5 months.
     This indicates that the balance of arm use (symmetry) and movement intensity stabilize before the total duration of use or maximal clinical capacity reaches a plateau.

3. Individual-Level Dissociation:
   Improvements in capacity and performance were frequently dissociated at the individual level.

   • 19% of participants improved in capacity (ARAT) but not in performance (FAUR).
   • 15% improved in performance (FAUR) but not in capacity.
   • Only 12% showed improvement across all four metrics (ARAT, FAUR, FAUD, FAUI).
     Notably, participants improving in performance without capacity gains tended to have more severe baseline motor impairments, suggesting compensatory behavioral adaptations (e.g., incorporating the paretic arm into stabilizing tasks) rather than motor recovery.

4. Distinct Baseline Predictors:
   The determinants of improvement differed significantly between constructs:

   • Capacity Improvement: Predicted primarily by younger age and intact corticospinal tract integrity (positive MEP).
   • Performance Improvement: Predicted by different factors. FAUR improvement was associated with longer inpatient rehabilitation duration. FAUD improvement was negatively associated with the paretic arm being the dominant limb. FAUI improvement was linked to baseline motor function (FMA-UE) and negatively associated with finger muscle tone (MAS).

Significance and Claims
The authors claim that these findings highlight a critical limitation in current rehabilitation paradigms: improvements in clinical capacity do not automatically translate into increased real-world arm use. The study demonstrates that capacity and performance are distinct constructs with different recovery dynamics, plateau times, and underlying predictors.

The paper argues that real-world arm use is influenced by behavioral and contextual factors (such as task efficiency and rehabilitation duration) that are not captured by impairment-focused clinical assessments. Consequently, the authors suggest that rehabilitation strategies should be personalized to target upper-limb performance in daily life, potentially requiring interventions beyond standard motor training (e.g., constraint-induced movement therapy or home-based technology) to ensure that gains in capacity are utilized in the real world. The study underscores the added value of wearable-derived digital metrics in revealing these dissociations, which would remain undetectable using clinical assessments alone.