Development and internal validation of a parsimonious model for neurological deterioration at 48 h after endovascular thrombectomy
- 1Department of Neurology, Shanxi Provincial People's Hospital, Shanxi Medical University
Abstract
Background: The incremental value of routine admission measurements for predicting early neurological deterioration after endovascular thrombectomy (EVT) remains unclear.
Methods: This retrospective single-center reanalysis used a 2020–2022 registry. END48 was defined as an increase of at least 4 National Institutes of Health Stroke Scale (NIHSS) points from admission to the fixed 48-h postoperative assessment. The conservative primary cohort excluded an unresolved longitudinal data segment. A revision-stage locked five-predictor logistic regression model included admission NIHSS scores, age, glucose levels, white blood cell count, and ln triglyceride value, with the NIHSS-only model serving as the comparator. Internal validation used 1,000 bootstrap samples and paired nested intervals of 1,000 outer × 100 inner samples.
Results: Of the 253 records representing 251 unique patients, 214 adults entered the primary cohort; 51 (23.8%) had END48. The primary model had an apparent area under the curve (AUC) of 0.651 and an optimism-corrected AUC of 0.601 (0.511 to 0.723). The corrected AUC difference vs. the NIHSS-only model was 0.038 (−0.025 to 0.171). The corrected Brier score was 0.180, compared to a prevalence-only reference of 0.182, resulting in a Brier skill score of 0.9%. The corrected primary calibration slope was 0.731 (0.519 to 0.892). Incremental net benefit was −0.015, 0.036, and 0.013 at 10%, 20%, and 30% thresholds, respectively; all corresponding 95% confidence intervals crossed zero. Post-hoc Riley assessments across plausible model-fit assumptions indicated that the available cohort was below the sample size desirable for model development. The reconstructed cohort sensitivity analysis did not change the overall interpretation.
Conclusion: Discrimination was limited, improvement in probability accuracy was marginal, and incremental discrimination and net benefit were uncertain. The model should be regarded as exploratory rather than clinically implementable; independent validation and improved outcome ascertainment are required.
Related
- Early neurological deterioration after endovascular treatment in acute ischemic stroke with large vessel occlusion.
- Endovascular thrombectomy activity as a candidate marker of organised acute stroke-service capacity: a global burden of disease-based European panel study
- Endovascular thrombectomy for mild vertebrobasilar artery occlusion: a hierarchical win-ratio analysis
- Technical considerations in endovascular thrombectomy for distal vessel occlusions
- Endovascular mechanical thrombectomy for the treatment of extensive postpartum cerebral venous sinus thrombosis: a case report
BibTeX
@article{ren2026development,
title = {{Development and internal validation of a parsimonious model for neurological deterioration at 48 h after endovascular thrombectomy}},
author = {Guoyong Ren and Chenyang Zhao and Yaxuan Sun},
journal = {Frontiers in Neurology},
year = {2026},
month = oct,
doi = {10.3389/fneur.2026.1888204},
url = {https://www.frontiersin.org/articles/10.3389/fneur.2026.1888204/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Development and internal validation of a parsimonious model for neurological deterioration at 48 h after endovascular thrombectomy
AU - Guoyong Ren
AU - Chenyang Zhao
AU - Yaxuan Sun
PY - 2026
DA - 2026/10/16
JO - Frontiers in Neurology
DO - 10.3389/fneur.2026.1888204
UR - https://www.frontiersin.org/articles/10.3389/fneur.2026.1888204/full
AB - Background: The incremental value of routine admission measurements for predicting early neurological deterioration after endovascular thrombectomy (EVT) remains unclear. Methods: This retrospective single-center reanalysis used a 2020–2022 registry. END48 was defined as an increase of at least 4 National Institutes of Health Stroke Scale (NIHSS) points from admission to the fixed 48-h postoperative assessment. The conservative primary cohort excluded an unresolved longitudinal data segment. A revision-stage locked five-predictor logistic regression model included admission NIHSS scores, age, glucose levels, white blood cell count, and ln triglyceride value, with the NIHSS-only model serving as the comparator. Internal validation used 1,000 bootstrap samples and paired nested intervals of 1,000 outer × 100 inner samples. Results: Of the 253 records representing 251 unique patients, 214 adults entered the primary cohort; 51 (23.8%) had END48. The primary model had an apparent area under the curve (AUC) of 0.651 and an optimism-corrected AUC of 0.601 (0.511 to 0.723). The corrected AUC difference vs. the NIHSS-only model was 0.038 (−0.025 to 0.171). The corrected Brier score was 0.180, compared to a prevalence-only reference of 0.182, resulting in a Brier skill score of 0.9%. The corrected primary calibration slope was 0.731 (0.519 to 0.892). Incremental net benefit was −0.015, 0.036, and 0.013 at 10%, 20%, and 30% thresholds, respectively; all corresponding 95% confidence intervals crossed zero. Post-hoc Riley assessments across plausible model-fit assumptions indicated that the available cohort was below the sample size desirable for model development. The reconstructed cohort sensitivity analysis did not change the overall interpretation. Conclusion: Discrimination was limited, improvement in probability accuracy was marginal, and incremental discrimination and net benefit were uncertain. The model should be regarded as exploratory rather than clinically implementable; independent validation and improved outcome ascertainment are required.
SN - 1664-2295
ER -