Predicting Poor Outcome Before Endovascular Treatment in Patients With Acute Ischemic Stroke

Abstract

Background: Although endovascular treatment (EVT) has greatly improved outcomes in acute ischemic stroke, still one third of patients die or remain severely disabled after stroke. If we could select patients with poor clinical outcome despite EVT, we could prevent futile treatment, avoid treatment complications, and further improve stroke care. We aimed to determine the accuracy of poor functional outcome prediction, defined as 90-day modified Rankin Scale (mRS) score ≥5, despite EVT treatment. Methods: We included 1,526 patients from the MR CLEAN Registry, a prospective, observational, multicenter registry of ischemic stroke patients treated with EVT. We developed machine learning prediction models using all variables available at baseline before treatment. We optimized the models for both maximizing the area under the curve (AUC), reducing the number of false positives. Results: From 1,526 patients included, 480 (31%) of patients showed poor outcome. The highest AUC was 0.81 for random forest. The highest area under the precision recall curve was 0.69 for the support vector machine. The highest achieved specificity was 95% with a sensitivity of 34% for neural networks, indicating that all models contained false positives in their predictions. From 921 mRS 0-4 patients, 27-61 (3-6%) were incorrectly classified as poor outcome. From 480 poor outcome patients in the registry, 99-163 (21-34%) were correctly identified by the models. Conclusions: All prediction models showed a high AUC. The best-performing models correctly identified 34% of the poor outcome patients at a cost of misclassifying 4% of non-poor outcome patients. Further studies are necessary to determine whether these accuracies are reproducible before implementation in clinical practice.

Beyond Accuracy: Investigating the Potential Clinical Utility of Predicting Functional Dependency and Severe Disability or Death in Successfully Reperfused Patients using Machine Learning

“…Results of primary analysis suggest that the stacked ensemble method performs best with regards to discrimination, calibration and clinical utility. In accordance with previous feature importance analyses reported by Ramos et al 39 for ML-based prediction of poor outcome (mRS 5–6) and the analysis by Xu et al 40 on predictors of futile recanalization, we found age on admission, NIHSS on admission, and pre stroke independence to be most predictive. In the secondary analysis, all methods showed a considerable drop in performance when used to predict mRS 5-6.…”

“…The exact cut-off deemed useful in a clinical scenario may also depend on other factors including health care resources, cultural perception and available information on the patients good will. In summary, the appealing ROC-AUC reported here and by others 39,43 , should thus be handled cautiously with regards to the clinical utility in excluding patients from reperfusion therapies.…”

Association of Ischemic Core Imaging Biomarkers With Post-Thrombectomy Clinical Outcomes in the MR CLEAN Registry

“…If imaging biomarkers could identify patients with a high mortality risk or identify patients who are likely to be severely disabled despite successful EVT, this could potentially support physicians in their decision to treat or not to treat a patient. Consequently, this would enable an individualized, patient- and physiology-based stroke treatment selection rather than selection based on time only ( 13 , 14 ).…”

Moving Toward Explainable Decisions of Artificial Intelligence Models for the Prediction of Functional Outcomes of Ischemic Stroke Patients

“…However, our work showed no difference in the predictive performance of LR models to gradient boosting and ANN models trained on patient demographics, comorbidities, baseline NIHSS and the presence of IV-rtPA treatment (27). Similarly, another study that compared LR to RF, SVM, gradient boosting, and ANN classifiers reported similar performance between all classifiers, with ANN achieving the highest AUC (0.81) (26) showing that while more advanced AI models might outperform rulebased scores, they perform comparably to traditional ML methods.…”

“…The second approach involves using more advanced AI models, such as artificial neural networks (ANN) and gradient boosting to process the information available at admission (20,(25)(26)(27). For example, when comparing models based on patient demographics and baseline clinical information, an ANN model performed better than the rule-based ASTRAL score for predicting 90-days mRS (AUC 0.84 vs AUC 0.89) (25).…”