Objective: To develop an abbreviated and practical neurologic scale that could assist emergency medical services or triage personnel in identifying patients with stroke.
Methods: A prospective, observational, cohort study was performed at university-based EDs. Participants were 74 patients treated in a thrombolytic stroke trial and 225 consecutive non-stroke patients evaluated during 4 random 12-hour shifts in the ED. Scores on the NIH Stroke Scale were obtained for all patients by physicians. Items of this scale were modified and recoded to a binomial (normal or abnormal) scale. Serial univariate analyses using χ2 were performed to rank items. Recursive partitioning was then performed to develop the decision rule for predicting the presence of stroke.
Results: Three items identified 100% of patients with stroke: facial palsy, motor arm, and dysarthria. An Abbreviated NIH Stroke Scale based on these items had a sensitivity of 100% and a specificity of 92%. A proposed Out-of-hospital NIH Stroke Scale consisting of facial palsy, motor arm, and a combination of dysarthria and best language items (abnormal speech) had a sensitivity of 100% and a specificity of 88%.
Conclusion: Using the derivation data set, a proposed Out-of-hospital NIH Stroke Scale had a high sensitivity and specificity for identifying patients with stroke when performed by physicians in this group of 299 ED patients. Prospective studies of other health care professionals using the scale in the out-of-hospital arena are needed.