Abstract 1580: Prediction of In-Hospital Mortality in Ischemic Stroke Using Data From Get With the Guidelines Stroke
Introduction: There is increasing interest in mortality as a measure of quality of care in stroke. We used Get With the Guidelines-Stroke program data to derive and validate prediction models for a patient’s risk of in-hospital ischemic stroke mortality.
Methods: Between 10/2001–12/2007 there were 1,036 hospitals that contributed complete data on 274,988 ischemic stroke patients. The sample was randomly divided into a derivation (60%) and validation (40%) dataset. Logistic regression models were used to determine characteristics that independently predict patient mortality and to assign point scores for prediction. To determine the effect of the NIH stroke scale score (NIHSS) on model discrimination, we also separately derived and validated a model in the 109,187 patients with NIHSS recorded (39.7%).
Results: In-hospital mortality was 5.51% overall and 5.19% in the subset where NIHSS was recorded. Points were assigned for the following predictors in the overall model: age (0 –28 points), method of arrival at the hospital (e.g. via ambulance vs. other, 0 –70 pts), atrial fibrillation (27 pts), previous stroke (4 pts), previous MI (11 pts), carotid stenosis (8 pts), diabetes (4 pts), peripheral vascular disease (12 pts), hypertension (6 pts), dyslipidemia (20 pts), smoking (8 pts), and weekend or night admission (6 pts). The c-statistic, a measure of model discrimination, was 0.73. There was better discrimination in a model derived from the subset with NIHSS recorded (c-statistic=0.84); a model with NIHSS alone provided nearly as good discrimination (c-statistic=0.83). The models validated well (Table⇓).
Conclusions: Predictive models for a patient’s in-hospital ischemic stroke mortality, using patient demographics and past medical history, produce fair discrimination. Incorporating stroke severity scores (NIHSS) substantially improved the discrimination. Stroke severity should be considered when adjusting hospital mortality for patient case mix.
This research has received full or partial funding support from the American Heart Association, National Center.