A five-predictor model using routine admission measurements (NIHSS, age, glucose, white blood cell count, ln triglyceride) to predict neurological deterioration at 48 h after thrombectomy showed limited discrimination, marginal gains in probability accuracy, and uncertain incremental value over an NIHSS-only model, so it should be regarded as exploratory rather than clinically implementable.
About one in four patients in the primary cohort had neurological deterioration at 48 h (END48) after endovascular thrombectomy.
The five-predictor model had limited discrimination after correction for optimism.
The incremental discrimination of the five-predictor model over an NIHSS-only model was uncertain.
Improvement in probability accuracy over a prevalence-only reference was marginal.
The model showed overfitting in calibration, with a corrected slope below 1.
Incremental net benefit over the NIHSS-only model was not distinguishable from zero at the tested decision thresholds.
The cohort was smaller than desirable for developing a prediction model.
A sensitivity analysis using a reconstructed cohort did not change the overall interpretation.
The authors conclude the model is exploratory and requires independent validation and better outcome ascertainment.
This study asked whether simple measurements taken when a stroke patient is admitted, such as stroke severity score, age, blood sugar, white blood cell count, and triglycerides, could help predict who will get worse neurologically within 48 hours after clot-removal surgery (thrombectomy). The researchers looked back at records from 214 patients at one hospital; about 24% of them (51 patients) worsened by at least 4 points on a standard stroke scale. The model they built did only modestly better than chance at telling which patients would worsen, and after statistical correction for overfitting its performance dropped further. It improved the accuracy of predicted probabilities by less than 1% over simply using the overall rate of worsening for everyone. Adding the extra measurements to stroke severity alone did not clearly improve prediction or clinical usefulness, as the uncertainty ranges included no benefit. This research suggests that these routine admission measurements, in this small single-hospital dataset, are not enough to reliably predict early worsening after thrombectomy. The authors describe the model as exploratory, not ready for clinical use, and note that larger studies, independent validation in other hospitals, and better outcome data are needed.
Ren G, Zhao C, Sun Y. (2026). Development and internal validation of a parsimonious model for neurological deterioration at 48 h after endovascular thrombectomy.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1888204