Abstract Body: Introduction: Mechanical thrombectomy (MT) devices were fundamentally designed for the treatment of cardioembolic (CE) strokes despite that 10-20% of large vessel occlusions (LVO) are caused by large artery atherosclerosis (LAA), which may not respond comparably. Therefore, the differentiation between LAA and non-LAA LVO may be critical for selecting the ideal MT technique. Here, we construct an algorithm for use in the emergent setting pre-operatively to differentiate between LAA and non-LAA LVO using clinical and radiographic features.

Methods: Using our prospectively collected stroke database we identified all middle cerebral artery (MCA) occlusions treated with MT with confirmed etiology as either LAA or non-LAA (CE, cryptogenic) according to TOAST criteria. Stroke risk factors and CT angiography (CTA) of the head and neck features were compared between LAA and non-LAA etiologies in a univariate analysis. An algorithm was then constructed using these variables in a multivariable logistic regression with corresponding score estimates for each variable to assess how well the model distinguished between LAA and non-LAA groups. The sensitivity and specificity were computed for each sum score to identify the ideal cut-off point for differentiating LAA from non-LAA LVOs.

Results: Final analysis included 33 LAA, 207 CE, and 120 cryptogenic strokes. Patients presenting with LAA LVOs were significantly less likely to have atrial fibrillation (9.1% vs 53.8%; p<0.0001), EF<35% (9.1% vs 27.8%; p=0.0208) and more likely to present with progressive or fluctuating symptoms (21.2% vs 4.6%; p=0.0018), intracranial multi-vessel atherosclerotic disease on CTA (84.8% vs 37%; p<0.0001), tapered appearance of occlusion with associated collaterals (60.6% vs 0.9%; p<0.0001), and ICA bulb plaque with high-risk features (87.9% vs 37.6%; p<0.0001) (Table 1). The AUC was 0.9650 (95% CI=0.9332, 0.9968) and the highest combination of sensitivity and specificity (97% and 88%, respectively) was associated with a cut-off score of 9 (Tables 2 and 3).

Conclusion: Our scoring system reliably differentiates between non-LAA and LAA LVO with high sensitivity and specificity based on 6 clinical and radiographic features. This will need to be validated with an external patient dataset and then in a prospective study.