Abstract Body: Introduction: The National Institutes of Health Stroke Scale (NIHSS) provides a clinical measure of stroke severity. Molecular biomarkers that reflect severity of neurological injury may enhance the objectivity and accuracy of stroke severity assessment.
Objectives: This study aimed to discover plasma proteins indicative of stroke severity in patients with acute ischemic stroke (AIS) using aptamer-based proteomics and supervised machine learning algorithms.
Methods: We used clinical and proteomics data of AIS patients aged ≥18 years lodged within a prospective plasma repository from 2010 to 2014. We collected blood from each patient at hospital admission before administering any therapeutic intervention. Our outcome was differentially expressed levels of proteins in AIS patients classified by NIHSS scores. We classified AIS patients into mild NIHSS (0-7), moderate NIHSS (8-10), severe NIHSS (11-20), and critical NIHSS (21-42) subgroups. We performed aptamer-based proteomics using the plasma 7K SomaScan assay. For comparisons between the four NIHSS subgroups, we performed feature selection by sparse partial least squares discriminant analysis (sPLS-DA) using the MixOmics R package. We determined the area under the receiver operating characteristic curves (AUC-ROC) to classify the AIS-severity subgroups.
Results: We included 40 AIS patients (mean age 63.3 years, 45% males) classified into four subgroups: 10 mild NIHSS, 9 moderate NIHSS, 11 severe NIHSS, and 10 critical NIHSS (Figure 1). SomaScan quantified 7307 protein targets, including 6373 unique proteins. Using the sPLS-DA approach, we identified two components classifying critical NIHSS (component 1, 10 proteins) and mild and severe NIHSS (component 2, 35 proteins) from moderate NIHSS. The panel of 45 proteins from the two components had an AUC of 0.96 to classify mild NIHSS, 0.66 to classify moderate NIHSS, 0.96 to classify severe NIHSS, and AUC of 0.97 to classify critical NIHSS from other subgroups. The top 5 proteins for AIS risk stratification were SELENOW, ANGPTL4, FABP3, CFL2, and KDM8 (Figure 2).
Conclusions: Our study revealed distinct panels of protein biomarkers capable of classifying AIS patients into NIHSS-defined severity subgroups with high accuracy, particularly for mild, severe, and critical categories. These proteins may improve stroke severity assessment, especially in conditions where a clinical exam is limited, though further validation with larger cohorts is critically needed.