Introduction:
Mechanical thrombectomy for acute ischemic stroke (AIS) is the current standard of care based on level 1 evidence from multiple randomized controlled trials. Recently, real-world indications for mechanical thrombectomy (MT) has extended beyond the inclusion criteria used in the majority of trials including elderly patients. We have recently developed a machine-learning based tool, SPOT, to optimize selection of elderly patients for MT based on single-center data. Here, we use a large cohort of international multicenter patients who underwent MT for AIS to externally validate SPOT.
Methods:
Patients who underwent MT for AIS at 12 comprehensive stroke centers in the US and Europe between 01/2013 and 12/2018 were reviewed. Patients age 80 years or older were included for validation of SPOT. SPOT is designed based on a combination of decision trees and linear regression models to provide binary output of predicted good (mRS 0-2) or poor outcome (mRS 3-6) after MT. SPOT uses admission variables: age, gender, comorbidities, admission NIHSS, baseline mRS score, ASPECT score and whether IV-tPA was administered. Predicted outcome was compared to actual outcome recorded at 90-days after treatment. A receiver operating characteristic curve was used to evaluate the accuracy of SPOT, and the negative predictive value was computed. The rate of post-procedural hemorrhage and mortality were compared between patients predicted by SPOT to have good versus poor outcome.
Results:
A total of 3,228 patients underwent MT for AIS during the study duration, of which 647 patients were at least 80 years of age or older and were included in the study. The average age was 85±5 years, and 65% were females. The median mRS score at 90 days was 4, and 21.3% had a good outcome (mRS 0-2). Of patients predicted by SPOT to have a poor outcome, 90% had a poor outcome. The area under the ROC curve was 0.7. The mortality rate in patients predicted by SPOT to have poor outcome had twice higher mortality than those predicted to have good outcome (55% vs 27%, p<0.001).
Conclusions:
Based on multicenter validation, SPOT presents a clinical decision in aid in assisting for exclusion of elderly patients unlikely to benefit from MT for AIS with a 90% negative predictive value.
A radiomics model combined with XGBoost may improve the accuracy of distinguishing between mediastinal cysts and tumors: a multicenter validation analysis
