Abstract T P314: Hemorrhagic Stroke Model to Predict In-House Mortality: A Single Center Cross-Validation Study

Abstract

Background: Although hemorrhagic strokes have high mortality rates, few models attempt to predict outcomes post ictus. A 2013 study by Smith and colleagues in JAHA used a national database to develop a risk score for predicting in-hospital mortality rates for stroke patients, assigning weighted point values to demographic and clinical variables based on a logistic regression model. We assessed the hypothesis that this risk score could be applied to a hemorrhagic population at a single tertiary center. Methods: The full cohort of 1872 patients consisted of subarachnoid hemorrhage (SAH) and intracerebral hemorrhage (ICH) patients admitted to a single tertiary center and enrolled into two separate prospective databases. There were 1560 consecutive patients enrolled in the SAH database between 1996 and 2013, and 312 consecutive patients enrolled in the ICH database between 2009 and 2013. Receiver operating characteristic analysis was performed on the entire cohort and subsets of SAH and ICH patients, and area under the curve (AUC) and 95% confidence intervals (CI) were generated. The analysis was repeated using only the combined age and admission National Institutes of Health Stroke Scale (NIHSS) components of the score. Delong’s test was used to compare AUCs. Results: The AUC and 95% CI for the overall cohort with the entire inclusion criteria was 0.89 (0.87-0.91), compared to 0.86 from the original study. Limited to ICH only and SAH only, the AUCs and 95% CIs were 0.90 (0.87-0.93) and 0.89 (0.86-0.91), respectively, compared to 0.82 and 0.89, respectively, in the original study. The combined age and NIHSS score alone had AUCs and 95% CIs of 0.89 (0.87-0.91), 0.90 (0.86-0.93), and 0.88 (0.86-0.91) for the overall cohort, ICH only, and SAH only, respectively. These AUCs were not significantly different from those associated with the total risk score. Conclusions: The risk score demonstrated outstanding discrimination for in-hospital mortality in this cohort. The combined age and NIHSS components alone showed similarly outstanding discrimination for in-hospital mortality, and held the added benefit of increased ease of calculation. Further studies are needed to assess the utility of this risk model in other populations.