TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

benchmarks evals

The Recovery Model Faltered Where Patients Needed Certainty Most

External validation preserved average error but exposed systematic underprediction for severe upper-limb impairment.

Published Updated Story ID: mp-2026-08-03-010
Read the complete editionStory JSON

Summary

External validation preserved average error but exposed systematic underprediction for severe upper-limb impairment.

A stroke-recovery model trained in a Dutch cohort was tested in an independent Danish prospective cohort of 80 patients about two weeks after stroke. Median absolute error was statistically comparable to development results, and calibration was close to ideal above predicted Action Research Arm Test scores of 40. Lower ranges were systematically underpredicted, showing that additional predictors are needed before clinical implementation for severely impaired patients.

Why it matters

External validation preserved average error but exposed systematic underprediction for severe upper-limb impairment.

Limits and context

  • A stroke-recovery model trained in a Dutch cohort was tested in an independent Danish prospective cohort of 80 patients about two weeks after stroke.

Key claims

  1. External validation preserved average error but exposed systematic underprediction for severe upper-limb impairment.

    Qualification: A stroke-recovery model trained in a Dutch cohort was tested in an independent Danish prospective cohort of 80 patients about two weeks after stroke.

    Evidence: source-2026-08-03-010

Sources

  1. Neurorehabilitation and Neural Repair: stroke recovery model validationSAGE Publications · secondary reporting

Corrections

No corrections have been recorded for this story.