Search | Search by Center | Search by Source | Keywords in Title
Sico J, Phipps MS, Chumbler NR, Bravata DM. Radiographic Surrogates of Ischemic Stroke Severity for Use in Risk Adjusting In-Hospital Mortality. [Abstract]. Stroke; A Journal of Cerebral Circulation. 2010 Feb 22; 41:e221.
Background: Several organizations have proposed using post-stroke mortality as a measure of stroke care quality. However, given the strong association between stroke severity and post-stroke mortality, a method of adjusting for stroke severity is needed if post-stroke mortality is to be compared across hospitals. The NIH Stroke Scale (NIHSS) is a valid measure of stroke severity, but is not generally available from administrative data. Within the Department of Veterans Affairs (VA) system, neurological examination data cannot be text mined, but radiology reports are available for text mining. Therefore, a stroke = severity system that is based on radiology reports may provide a means of risk adjusting post-stroke mortality in the VA. Objective: We sought to develop a surrogate for ischemic stroke severity based on admission brain imaging data. Methods: We used data from a retrospective cohort study that included ischemic stroke patients within 48 hours of symptom onset at five hospitals (1998-2003). The admission brain imaging data (either computed tomography or magnetic resonance imaging) were abstracted from radiology reports. Admission neurologic examination data were used to calculate the NIHSS. Three radiographic surrogates of stroke severity were assessed: (1) the presence or absence of edema; (2) the sum of the number of lobes with an infarct; and (3) the modified Oxfordshire classification. We compared each of the radiographic systems with the NIHSS and with in-hospital mortality. Results: Among 1363 stroke patients, 13% had a normal brain image, 87% had any abnormality, and 0.07% had no admission brain imaging. The NIHSS ranged from 2 to 38 (median 7). The in-hospital mortality rate was 6.5%. The correlations between the three systems and the NIHSS varied: edema (r 0.14), sum of lobes (r 0.09), and Oxfordshire (r 0.18). The NIHSS provided robust risk adjustment for in-hospital mortality (c-statistic, 0.81). The Oxfordshire provided considerable risk adjustment for in-hospital mortality (c-statistic, 0.69). Conclusion: The Oxfordshire approach, although only modestly correlated with baseline stroke severity, provides a simple means of scoring brain imaging data for risk adjusting in-hospital mortality. This system can be used for text mining radiology reports for the purpose of risk adjusting post-stroke mortality.