Immune-inflammatory, coagulation, adhesion, and imaging biomarkers combined in machine learning models improve the prediction of death 1year after ischemic stroke.

Abstract

Some clinical, imaging, and laboratory biomarkers have been identified as predictors of prognosis of acute ischemic stroke (IS). The aim of this study was to evaluate the prognostic validity of a combination of clinical, imaging, and laboratory biomarkers in predicting 1-year mortality of IS. We evaluated 103 patients with IS within 24h of their hospital admission and assessed demographic data, IS severity using the National Institutes of Health Stroke Scale (NIHSS), carotid intima-media thickness (cIMT), and degree of stenosis, as well as laboratory variables including immune-inflammatory, coagulation, and endothelial dysfunction biomarkers. The IS patients were categorized as survivors and non-survivors 1year after admission. Non-survivors showed higher NIHSS and cIMT values, lower antithrombin, Protein C, platelet counts, and albumin, and higher Factor VIII, von Willebrand Factor (vWF), white blood cells, tumor necrosis factor (TNF)-α, interleukin (IL)-10, high-sensitivity C-reactive protein (hsCRP), and vascular cellular adhesion molecule 1 (VCAM-1) than survivors. Neural network models separated non-survivors from survivors using NIHSS, cIMT, age, IL-6, TNF-α, hsCRP, Protein C, Protein S, vWF, and platelet endothelial cell adhesion molecule 1 (PECAM-1) with an area under the receiving operating characteristics curve (AUC/ROC) of 0.975, cross-validated accuracy of 93.3%, sensitivity of 100% and specificity of 85.7%. In conclusion, imaging, immune-inflammatory, and coagulation biomarkers add predictive information to the NIHSS clinical score and these biomarkers in combination may act as predictors of 1-year mortality after IS. An early prediction of IS outcome is important for personalized therapeutic strategies that may improve the outcome of IS.