Objective: External validation of prognostic models is crucial but rarely done. Our aim was to externally validate a prognostic model to predict 60-day case fatality after aneurysmal subarachnoid hemorrhage developed from the International Subarachnoid Aneurysm Trial in a retrospective unselected cohort of subarachnoid hemorrhage patients.
Design: The model's predictors were age, aneurysm size, Fisher grade, and World Federation of Neurological Surgeons grade. Two versions of the model were validated: one with World Federation of Neurological Surgeons grade scored at admission and the other with World Federation of Neurological Surgeons grade at treatment decision. The outcome was 60-day case fatality. Performance of the model was assessed by studying discrimination, expressed by the area under the receiver operating characteristic curve, and calibration.
Setting: University hospital.
Patients: We analyzed data from 307 consecutive aneurysmal subarachnoid hemorrhage patients admitted between 2007 and 2011 (validation cohort).
Interventions: None.
Measurements and main results: The observed 60-day case fatality rate was 30.6%. Discrimination was good, and differed between the model with World Federation of Neurological Surgeons grade at treatment decision (area under the receiver operating characteristic curve, 0.89) and at admission (area under the receiver operating characteristic curve, 0.82). Mean predicted probabilities were lower than observed: 17.0% (model with World Federation of Neurological Surgeons grade at admission) and 17.7% (model with World Federation of Neurological Surgeons grade at treatment decision).
Conclusions: The model discriminated well between patients who died or survived within 60 days. In addition, we found that using World Federation of Neurological Surgeons grade at moment of treatment decision of the ruptured aneurysm improved model performance. However, since predicted probabilities were much lower than observed probabilities, the International Subarachnoid Aneurysm Trial prediction model needs to be adapted to be used in clinical practice.