Beyond proportional recovery in wake-up stroke: unsupervised recovery clusters based on the NIHSS.

The analysis included only 201 of the 503 patients in the source dataset, and only those with complete NIHSS data, a zero Level-of-Consciousness Vigilance score, onset NIHSS greater than 4, and a non-zero NIHSS at 22-36 h, so it may not apply to patients with lower initial NIHSS scores or missing early scores. The clustering used only the total NIHSS recovery ratio, not individual NIHSS items, lesion features, or other clinical measures, so it cannot identify which specific impairments shaped a patient's trajectory. For patients with NIHSS below 6 at 22-36 h, the models and prediction intervals did not adequately discriminate between individuals, so the clusters are less reliable in low-impairment stroke. The NIHSS ceiling effect may hide variability, especially near full recovery. The clusters were not validated in an independent cohort, and the authors state replication in independent cohorts is needed. The study is a secondary clustering analysis of trial data, not a treatment trial, so it cannot show that any therapy caused recovery. The sample was limited to wake-up stroke patients with unknown onset time, so it may not apply to all stroke types.

The paper reports funding from HORIZON EUROPE Framework Programme and HORIZON EUROPE European Research Council. No author competing interests are listed in the provided text.

Do not treat an early recovery ratio greater than 0.3 as a proven prediction rule. It was an association in one wake-up stroke sample, and the clusters were not validated in an independent cohort.