External Validation of a Dynamic Prediction Model for Upper Limb Function After Stroke.
Iris C Brunner, Eleni-Rosalina Andrinopoulou, Ruud Selles and 2 others
PMID 38482101WHAT IT FOUND
This stroke prediction model is not accurate enough to use at 2 weeks post-stroke or for patients with severe arm impairment.
It works best for those with mild impairment and only when predictions are made 3 months after the stroke, limiting its early clinical usefulness.
Key findings
01The model's prediction error was largest at baseline (2 weeks post-stroke) and for patients with severe upper limb impairment (ARAT 0-22), with a median absolute error of 30 points.
02Prediction accuracy substantially improved by 3 months post-stroke, with the median absolute error for severely impaired patients dropping to 7 points.
03The authors concluded the model has limited clinical usability due to poor accuracy at 2 weeks and for patients with severe upper limb impairments.
STILL TO COME
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The validation cohort differed from the original development cohort: it included both ischemic and hemorrhagic strokes, and had fewer assessments (3 vs a median of 6 in the development cohort). The model sometimes predicted scores outside the valid ARAT range (below 0 or above 57), which had to be artificially truncated for analysis. The study did not collect data between 2 weeks and 3 months, missing a critical window for early prediction refinement. Patients with severe impairment showed highly variable recovery patterns that the model failed to capture accurately at baseline.
Declared interests
No conflicts of interest or funding sources were declared in the provided text.
The easy way to misread this
Do not assume that a later, more accurate prediction (at 3 months) means the model is useful for early rehabilitation planning. The primary clinical need for prediction is at 2 weeks to guide initial therapy, and the model is explicitly not accurate enough for this purpose, especially in severe cases.