Inflammation‐Derived and Clinical Indicator‐Based Predictive Model for Ischemic Stroke Recovery

Abstract

Background Neuroinflammatory responses are closely associated with poststroke prognosis severity. This study aimed to develop a predictive model, combining inflammation‐derived markers and clinical indicators, for distinguishing functional outcomes in patients with subacute ischemic stroke. Methods and Results Based on activities of daily living assessments, ischemic stroke participants were categorized into groups with little effective (LE) recovery and obvious effective (OE) recovery. Initial biocandidates were identified by overlapping differentially expressed proteins from proteomics of clinical serum samples (5 LE, 5 OE, and 6 healthy controls) and differentially expressed genes from an RNA sequence of the ischemic cortex in middle cerebral artery occlusion mice (n=3). Multidimensional validations were conducted in ischemia–reperfusion models and a clinical cohort (15 LE, 11 OE, and 18 healthy controls). Models of robust biocandidates combined with clinical indicators were developed with machine learning in the training data set and prediction in another test data set (15 LE and 11 OE). We identified 194 differentially expressed proteins (LE versus healthy controls) and 174 differentially expressed proteins (OE versus healthy controls) in human serum, and 5121 differentially expressed genes (day 3) and 5906 differentially expressed genes (day 7) in middle cerebral artery occlusion mice cortex. Inflammation‐derived biomarkers TIMP1 (tissue inhibitor metalloproteinase‐1) and galactosidase‐binding protein LGLAS3 (galectin‐3) exhibited robust increases under ischemic injury in mice and humans. TIMP1 and LGALS3 coupled with clinical indicators (hemoglobin, low‐density lipoprotein cholesterol, and uric acid) were developed into a combined model for differentiating functional outcome with high accuracy (area under the curve, 0.8). Conclusions The combined model is a valuable tool for evaluating prognostic outcomes, and the predictive factors can facilitate development of better treatment strategies.