iSPAN: Improved prediction of outcomes post thrombectomy with Machine Learning

BackgroundThis study aimed to develop and evaluate a machine learning model and a novel clinical score for predicting outcomes in stroke patients undergoing endovascular thrombectomy. MethodsThis retrospective study included all patients aged over 18 years with an anterior circulation stroke treated at a thrombectomy centre from 2010 to 2020. External validation data was obtained. The primary outcome variable was day 90 mRS [&ge;]3. Existing clinical scores (SPAN and PRE) and Machine Learning (ML) models were compared. A novel clinical score (iSPAN) was derived by adding an optimised weighting of the most important ML features to the SPAN and compared results. Results812 patients were initially included (397 female, average age 73), 62 for external validation. The best performing clinical score and ML model were SPAN and XGBoost (sensitivity specificity and accuracy 0.967, 0.290, 0.628 and 0.783, 0.693, 0.738 respectively). A significant difference was found overall and XGBoost was more accurate than SPAN (p< 0.0018). The most important features were Age, mTICI and total number of passes. The addition of 11 points for mTICI of [&le;] 2B and [&ge;] 3 points for 3 passes to the SPAN achieved the best accuracy and was used to create the iSPAN. iSPAN was not significantly less accurate than XGBoost (p>0.5). In the external validation set, iSPAN and SPAN achieved sensitivity, specificity, and accuracy of (0.735, 0.862, 0.79) and (0.471, 0.897, 0.67), respectively. ConclusionsiSPAN incorporates machine-derived features to achieve better predictions compared to existing scores. It is not inferior to the XGB model and is externally generalisable. Key PointsO_LIAn XGB model performed better than existing scores and other tested models for prognostication post EVT. C_LIO_LIIt identified mTICI and number of passes as important and modifiable factors. C_LIO_LIIntegrating these into the SPAN score (iSPAN) was not inferior to the XGB model and is generalisable and easier to use and interpret. C_LI