A Big-Data Variational Bayesian Framework for Supporting the Prediction of Functional Outcomes in Wake-Up Stroke Patients

Prognosis in Wake-up ischemic stroke (WUS) is important for guiding treatment and rehabilitation strategies, in order to improve recovery and minimize disability. For this reason, there is growing interest on models to predict functional recovery after acute ischemic events in order to personalize the therapeutic intervention and improve the final functional outcome. The aim of this preliminary study is to evaluate the possibility to predict a good functional outcome, in terms of modified Rankin Scale (mRS ≤ 2), in thrombolysis treated WUS patients by Bayesian analysis of clinical, demographic and neuroimaging data at admission. The study was conducted on 54 thrombolysis treated WUS patients. The Variational Bayesian logistic regression with Automatic Relevance Determination (VB-ARD) was used to produce model and select informative features to predict a good functional outcome (mRS ≤ 2) at discharge. The produced model showed moderately high 10 × 5-fold cross validation accuracy of 71% to predict outcome. The sparse model highlighted the relevance of NIHSS at admission, age, TACI stroke syndrome, ASPECTs, ischemic core CT Perfusion volume, hypertension and diabetes mellitus. In conclusion, in this preliminary study we assess the possibility to model the prognosis in thrombolysis treated WUS patients by using VB ARD. The identified features related to initial neurological deficit, history of diabetes and hypertension, together with necrotic tissue relate ASPECT and CTP core volume neuroimaging features, were able to predict outcome with moderately high accuracy.