484 Predictors of Outcome Clusters for Anterior Circulation Mechanical Thrombectomy: An Unsupervised Machine Learning Analysis of Prospective Data. (April 2023)

• Record Type: Journal Article
• Authors: Naik, Anant; Catapano, Joshua; Hartke, Joelle Nicole; Koester, Stefan Wolfgang; Winkler, Ethan A.; Jadhav, Ashutosh; Ducruet, Andrew F.; Albuquerque, Felipe
• Is Part Of: Neurosurgery. Volume 69(2023)Supplement 1
• Page Start: 104
• Page End: 105
• Publication Date: 2023-04
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1227/neu.0000000000002375_484
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract : INTRODUCTION: Endovascular thrombectomy for select patients with anterior circulation stroke can be lifesaving if done in an appropriate therapeutic window. However, studies have demonstrated that failure of treatment ranges from 30-70%. Early determination of patients that are most likely to benefit from early revascularization is of pivotal importance. METHODS: We prospectively collected data from a high-volume stroke center for patients receiving MT for anterior circulation strokes. Outcomes evaluated included final Thrombolysis in Cerebral Infarction (TICI) score, the number of passes, complications from endovascular intervention, discharge NIHSS. Outcomes were clustered using a k-means model after the number of optimal clusters were obtained using a silhouette algorithm. Uniform manifold projections (UMAP) were used to project the four outcomes into a 2-D space. Upon cluster determination, multivariate regression was used to determine predictors of cluster membership. Univariate comparisons leveraged chi-square and t-tests. RESULTS: 187 consecutive patients were selected for analysis. Included patients had an average age 69.96 (SD – 15.14). Two distinct clusters were obtained, confirmed by the Silhouette model (1B). Cluster 2 was found to be associated with improved outcomes on univariate regression. Multivariate regression was used to identify predictors of cluster membership. ICA occlusion (OR = 7.63 (1.40 – 48.6), p = 0.02). A one-point increase in admission NIHSS was associated with an 11% increase in cluster 2 risk. CONCLUSIONS: In this study, we consolidate 4 outcomes using unsupervised machine learning. We found that cluster 2 patients did worse. We also demonstrated that ICA occlusion, Admission NIHSS, and smoking status were significantly associated with cluster 2 membership. Machine learning can be used to identify novel morphological characteristics of specific patient groups.