Document resource
Introduction Moyamoya disease is an isteno-occlusive arteriopathy resulting in both ischemic and hemorrhagic stroke. The mainstay of treatment is extracranial to intracranial bypass. However, determinants of robust revascularization remain in question. In this study, we sought to use predict revascularization following bypass surgery for moyamoya disease utilized a novel application of supervising clustering techniques.Methods A retrospective cohort study of adult patients who underwent direct or indirect bypass surgery for moyamoya disease was performed. Clinical and demographic factors, angiographic severity, pre-operative measures of cerebrovascular reserve (CVR) on Blood Oxygenation Level-Dependent (BOLD) MRI, revascularization on follow-up angiography via Matsushima grading, and post-operative follow-up CVR measures were collected. Prediction of change between pre and post-operative CVR (ΔCVR) was performed via partial least squares regression, and feature importance assessed using Variable Importance in Projection (VIP) scores. Prediction of Matsushima grading as a binary (A and B vs C) was performed using Firth’s bias-reduced logistic regression and partial least squared discriminant analysis. P-values of <0.05 were considered significant.Results A total of 44 patients with 57 hemispheres who underwent bypass surgery were analyzed. The mean age was 44 years old and was predominantly female (90%). Mean Suzuki score was 3.5 pre-operatively. For post-operative CVR prediction, the 6-feature PLS regression model incorporating age, sex, Suzuki score, and three pre-operative CVR measures was able to predict ΔCVR delay (R 2: 0.33, Pearson r: 0.59, P<0.001) and ΔCVR (R2: 0.191, Pearson r=0.45, P<0.001). VIP score analysis showed pre-operative CVR delay and raw CVR as the most significant predictors (VIP score >1). Using only pre-operative CVR values also resulted in robust prediction of ΔCVR delay (R2: 0.42, Pearson r: 0.649, p<0.001) and ΔCVR (R2: 0.193, Pearson r: 0.448, p<0.001). For good Matsushima grade prediction, Firth logistic regression using age and Suzuki grading achieved the best AUC of 0.749. PLS-DA resulted in an AUC of 0.702 with a p-value of 0.015. Utilizing CVR as a predictor resulted in a worse performance of 0.316. In predicting poor Matsushima grade, no model achieved an AUC greater than 0.53.Conclusion Pre-operative CVR values reliably predict change in quantitative hemodynamic reserve following bypass surgery for moyamoya disease, however, are less robust in predicting angiographic outcomes compared to pre-operative clinical factors such as age and Suzuki grading. Utilizing clinical and angiographic characteristics as complementary factors to quantitative measures of CVR can aid in patient selection and prediction of good outcomes.Disclosures L. Velagapudi: None. S. Sundrani: None. C. Baker: None. K. Raygor: None.Abstract E-214 Figure 1