Development and validation of a web-based machine learning model for predicting early neurological deterioration following stroke thrombolysis: A multicenter study

Published in Journal of Medical Internet Research, 27, e77858 (2025)

Background: Early neurological deterioration (END) after intravenous thrombolysis can significantly worsen outcomes in acute ischemic stroke, but reliable early-risk tools are limited.
Objective: Build and validate a high-performance END prediction model to support personalized, risk-stratified post-thrombolysis management.
Methods: Multicenter cohort of 1,927 patients (development: 1,361; external validation: 566); evaluated 27 clinical parameters and developed ENDRAS with 6 routinely available variables.
Results: XGBoost-based ENDRAS achieved strong discrimination (AUC 0.988, 95% CI 0.983–0.993), enabled low-/high-risk triage at a 29% threshold, and was deployed as a web calculator with <0.02s inference time.
Conclusion: ENDRAS translates prediction into actionable clinical pathways for targeted monitoring and intervention, with further prospective validation still needed.

China Clinical Trial Registry ChiCTR2400085504
https://www.chictr.org.cn/showproj.html?proj=231133