Abstract 204: Validity of Etiologic Stroke Classification: Comparison of Different Classification Systems
Background and Purpose: The goal of etiologic ischemic stroke classification is to generate determined subtypes with discrete phenotypic, therapeutic, and prognostic characteristics. Although several new etiologic classification systems have been developed in recent years, their ability to unambiguously generate discrete subtypes is not known. We prospectively examined predictive ability of TOAST, Causative Classification of Stroke System (CCS), and ASCO (based on grade-1 definitions) for various hard stroke features in 2284 consecutive patients within the context of an NIH-funded study.
Methods: Three raters blinded to each other’s assignments as well as to the study end points performed subtype assignments based on information available at discharge. The primary validation end-point was 90-day stroke recurrence. Secondary validation end-points were admission infarct volume on DWI, admission NIHSS score, and 90-day mortality. We assessed predictive ability of each system by ROC analysis and determined within-category / between-category variance by calculating F values from ANOVA for continuous variables.
Results: CCS demonstrated larger area under the ROC curve (AUC, 0.71, 95%CI: 0.66-0.75) for stroke recurrence when compared to ASCO (0.66, 95%CI: 0.61-0.71, p=0.04) and TOAST (0.61, 95%CI: 0.55-0.67, p<0.01). The difference in AUC between ASCO and TOAST for recurrent stroke was borderline (p=0.052). There was no difference in AUC for 90-day mortality among the systems [0.67 (95%CI, 0.63-0.70) for CCS, 0.67 (95%CI, 0.64-0.70) for ASCO, and 0.65 (95%CI, 0.62-0.68) for TOAST]. The size of undetermined category was 33% with CCS, 40% with ASCO, and 52% with TOAST (p<0.01). CCS demonstrated larger F values for admission NIHSS (41.8) and infarct volume on DWI (12.0) than ASCO (38.1, 11.4 respectively) and TOAST (37.7, 9.5 respectively).
Conclusion: This study demonstrates that etiologic stroke subtypes confer predictive information for early stroke recurrence regardless of the classification system used to identify them. Our results also indicate that CCS exhibits greater capacity to generate homogenous categories with different features and assigns a larger proportion of strokes into determined subtypes as compared to other systems.
Author Disclosures: E.M. Arsava: None. R. Avery: None. M. Sorgun: None. O. Pontes Neto: None. K. Park: None. G. Kim: None. H. Ay: Research Grant; Significant; NIH: R01-NS059710.
- © 2015 by American Heart Association, Inc.