Abstract TP245: Neuroimaging biomarkers moderately predict the early neurological improvement or worsening after thrombectomy in patients with medium vessel stroke

Introduction: Recent trials have not shown a clear benefit of endovascular therapy (EVT) for stroke patients with medium vessel occlusion (MeVO). Neuroimaging markers were successfully utilized in the past as a predictor of outcome in chronic neurological conditions. The goal of this study is to identify neuroimaging parameters along with the machine learning tools that can be utilized to predict early neurological improvement or worsening within 7 days after thrombectomy in patients with MeVO. Methods: A prospectively collected, multicenter stroke registry is utilized to select consecutive patients who received thrombectomy for MeVO. Patients without complete data were excluded. Logistic regression analysis is utilized to develop predictive models to identify factors affecting early neurological improvement or worsening. R 2 McF is utilized to select models with strong predictive value, as by convention, R 2 McF values of 0.2-0.4 are considered acceptable models and >0.4 are considered strong models. The omnibus likelihood ratio is utilized to select factors with a strong contribution to this model. Results and Conclusions: A total of 38 patients were included. Documented neurological exams were utilized to assess for signs of qualitative early neurological improvement or worsening. When comparing groups with different early neurological outcomes, no statistically significant difference was found in medical history, demographics, and initial laboratory findings except platelets (p≤0.05). On correlation of early neurological changes with different factors (p≤0.05), early neurological changes had a moderately positive correlation with 90-day mRS (Pearson’s r = 0.420) and a moderately negative correlation with acute care length of stay (Pearson’s r = - 0.393) ( Table 1 ). When these factors are compared using multinomial logistic regression, the medical history, platelets, highest troponin, LDL, and HgbA1c significantly contributed to the models' ability to predict the likelihood of early neurological changes (R 2 McF =0.999) ( Table 2 ). The commonly used neuroimaging parameters like ASPECTS score, core volume, and penumbra volume moderately contributed to models' ability to predict the likelihood of early neurological changes (R 2 McF =0.424) ( Table 3 ). These results provide preliminary data for a larger-scale prediction machine learning tool to aim to reliably predict post-thrombectomy outcomes for patients with MeVO.