Post-thrombectomy models for outcome prediction in ischemic stroke: systematic review and external validation

Area of Science:

• Neurology
• Medical Informatics
• Clinical Prediction Modeling

Background:

• Outcome prediction models for ischemic stroke patients undergoing endovascular thrombectomy (EVT) benefit from post-procedural predictors.
• Systematic review and external validation of existing models were performed.

Purpose of the Study:

• To systematically review and externally validate outcome prediction models for EVT patients.
• To assess the impact of post-procedural predictors on model performance.

Main Methods:

• Systematic search of multiple databases for EVT outcome prediction models including post-procedural predictors.
• Quality assessment using a shortened PROBAST+AI checklist.
• External validation in 1417 EVT patients from three RCTs (MR CLEAN MED, NOIV, LATE).
• Performance evaluation using discrimination (C-statistic) and calibration.

Main Results:

• 54 studies included: 30 regression-based, 24 machine learning (ML).
• Post-procedural predictors like Thrombolysis in Cerebral Infarction score and NIHSS were common.
• 12 regression models validated (C-statistics 0.67-0.90); ML models were not validated.
• Models with continuous post-procedural NIHSS demonstrated superior discrimination.

Conclusions:

• Regression-based models offer moderate to excellent performance for predicting EVT outcomes.
• ML models faced validation challenges due to data accessibility and reporting.
• Models incorporating strong clinical predictors, particularly continuous post-procedural NIHSS, yield better predictive accuracy.