External Validation of a Dynamic Prediction Model for Upper Limb Function After Stroke.

Area of Science:

• Neuroscience
• Rehabilitation Medicine
• Clinical Prediction Modeling

Background:

• Accurate prediction of upper limb (UL) function post-stroke is crucial for rehabilitation planning.
• Existing dynamic prediction models require external validation in diverse cohorts.

Purpose of the Study:

• To externally validate a dynamic prediction model for predicting upper limb function at 6 months post-stroke.
• To assess the model's prediction accuracy across different severity levels and time points.

Main Methods:

• External validation using data from a prospective Danish cohort study (N=80).
• Prediction of Action Research Arm Test (ARAT) scores at 6 months using baseline (2 weeks) and 3-month post-stroke data.
• Assessment of prediction accuracy for mild, moderate, and severe UL impairment categories.

Main Results:

• The model performed best for patients with mild UL impairment (median error 3) and worst for severe impairment (median error 30) at baseline.
• Prediction accuracy significantly improved when 3-month data was included compared to 2-week baseline data.

Conclusions:

• The dynamic prediction model has limited clinical usability for early prediction (2 weeks) and for patients with severe UL impairments.
• Refinement of the model using biomarker data may enhance its predictive capabilities.