Development and Assessment of a Computer Algorithm for Stroke Vascular Localization Using Components of the National Institutes of Health Stroke Scale

Background The National Institutes of Health Stroke Scale (NIHSS) was not intended to be used to determine the stroke's vascular distribution. The aim of this study was to develop, assess the reliability, and validate a computer algorithm based on the NIHSS for this purpose. Methods Two cohorts of patients with ischemic stroke having similar distributions of Oxfordshire localizations (total anterior, partial anterior, lacunar, and posterior circulation) based on neuroimaging were identified. The first cohort (n = 40) was used to develop a computer algorithm for vascular localization using a modified version of the NIHSS (NIHSS-Localization [NIHSS-Loc]) that included the laterality of selected deficits; the second (n = 20) was used to assess the reliability of algorithm-based localizations compared to those of 2 vascular neurologists. The validity of the algorithm-based localizations was assessed in comparison to neuroimaging. Agreement was assessed using the unweighted kappa (κ) statistic. Results Agreement between the 2 raters using the standard NIHSS was slight to moderate (κ =.36, 95% confidence interval [CI].10-.61). Inter-rater agreement significantly improved to the substantial to almost perfect range using the NIHSS-Loc (κ =.88, 95% CI.73-1.00). Agreement was perfect when the 2 raters entered the data into the NIHSS-Loc computer algorithm (κ = 1.00, 95% CI 1.00-1.00). Agreement between the algorithm localization and neuroimaging results was fair to moderate (κ =.59, 95% CI.35-.84) and not significantly different from the localizations of either rater using the NIHSS-Loc. Conclusion A computerized, modified version of the standard NIHSS can be used to reliably and validly assign the vascular distribution of an acute ischemic stroke.