Using Radiological Data to Estimate Ischemic Stroke Severity

Abstract

Goal: Risk-adjusted post-stroke mortality has been proposed for use as a measure of stroke care quality. Although valid measures of stroke severity (e.g., the NIH Stroke Scale) are not typically available in administrative datasets, radiology reports are often available within electronic health records. We sought to examine whether admission head computed tomography data could be used estimate stroke severity. Materials and Methods:Using chart review data from a cohort of acute ischemic stroke patients (1998)(1999)(2000)(2001)(2002)(2003), we developed a radiographic measure (Brain Imaging Score) of stroke severity in a two-thirds development set and assessed in a one-third validation set. The retrospective NIH Stroke Scale was dichotomized as: mild/moderate (<10) and severe (≥10). We compared the association of this radiographic score with NIH Stroke Scale and in-hospital mortality at the patient-level. Findings:Among 1348 stroke patients, 86.5% had abnormal findings on initial head computed tomography. The c-statistic for the Brain Imaging Score for modeling severe stroke (development, 0.581; validation, 0.579) and in-hospital mortality (development, 0.623; validation, 0.678) were generated. Conclusion:Although the c-statistics were only moderate, the BIS provide significant risk stratification information with a two-variable score. Until administrative data routinely includes a valid measure of stroke severity, radiographic data may provide information for use in risk adjustment.4