Predictors of Outcome Clusters in Patients With Unruptured Intracranial Aneurysms Treated With Microsurgery: An Unsupervised Machine Learning Analysis

未破裂颅内动脉瘤患者显微外科手术治疗预后分组的预测因素:一项无监督机器学习分析

阅读:1

Abstract

BACKGROUND AND OBJECTIVES: Identifying surgical candidates who are prone to poor outcomes is crucial for adapting treatment and ensuring optimal outcomes. Our aim was to use unsupervised machine learning to reveal patient outcome subgroups and identify predictors of attribution to these subgroups in patients with unruptured intracranial aneurysms. METHODS: We retrospectively reviewed 400 patients who underwent microsurgical treatment for unruptured aneurysms at a large quaternary center from January 1, 2014, to December 31, 2020. Evaluated outcomes included new neurological deficits at discharge and poor neurological outcome at follow-up. Outcomes were clustered using a k-means model after the number of optimal clusters was obtained using a total within sum of squares algorithm. Uniform manifold approximation and projection was used to project the outcomes into a 2-dimensional space. On cluster determination, multivariate regression with a P-inclusion of ≤.20 was used. RESULTS: Three unique clusters were identified. Cluster 1 represented patients who had no poor neurological outcome (0/142, 0%) and no neurological deficits at discharge (0%). Cluster 2 represented patients with a high burden of new neurological deficit at discharge (76/143 [53%]) but no poor neurological outcome (0%). Cluster 3 represented patients with a new neurological deficit at discharge (52/115 [45%], P < .001) and a poor neurological outcome (47/115 [41%], P < .001). The mean (SD) population, hypertension, age, size of aneurysm, earlier subarachnoid hemorrhage from another aneurysm, site of aneurysm score was highest in cluster 3 vs cluster 2 vs cluster 1 (5.03 [2.83] vs 4.71 [2.56] vs 3.91 [2.66], P < .001). On multivariate analysis, diabetes and hyperlipidemia significantly predicted nonmembership in cluster 1. CONCLUSION: Unsupervised machine learning grouped outcomes into 3 distinct, clinically meaningful clusters. Vascular comorbidities, rather than aneurysm characteristics, significantly predicted attribution to outcome clusters.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。