Machine Learning-Based Identification of Target Groups for Thrombectomy in Acute Stroke

Area of Science:

• Neuroendovascular interventions
• Stroke outcome prediction
• Machine learning in medicine

Background:

• Endovascular thrombectomy (EVT) is a key treatment for large-vessel occlusion (LVO) stroke.
• Uncertainty exists regarding EVT's benefit in patients excluded from randomized controlled trials (RCTs).
• Identifying specific patient subgroups who benefit from EVT is crucial for clinical practice.

Purpose of the Study:

• To systematically identify LVO stroke patients underrepresented in RCTs who may benefit from EVT.
• To assess the importance of reperfusion level on functional outcome prediction using machine learning.
• To determine if reperfusion's importance differs across patient subgroups and stroke types.

Main Methods:

• Machine learning models were used to predict functional outcomes in LVO stroke patients.
• Two cohorts were analyzed: a real-world registry (N=5235) and RCT data (N=1488).
• The predictive importance of reperfusion level was compared to EVT treatment allocation and across different patient subgroups.

Main Results:

• Reperfusion level was a significant predictor of outcome, comparable in importance to EVT treatment allocation in RCT data.
• The predictive importance of reperfusion was magnified in patient groups underrepresented in RCTs (e.g., low NIHSS, M2 occlusions, low ASPECTS).
• Reperfusion level was equally important for anterior and vertebrobasilar LVO strokes.

Conclusions:

• Reperfusion level is a critical factor for predicting functional outcomes after EVT in LVO stroke.
• This highlights potential EVT benefit in specific patient groups, including those with vertebrobasilar strokes, low NIHSS, low ASPECTS, and M2 occlusions.
• Machine learning analysis of reperfusion can identify target populations for EVT beyond traditional RCT criteria.