Developing predictive models of excellent and devastating outcome after stroke

Background: models to predict functional status post-stroke have utility in balancing groups in randomised trials, for outcome comparison between stroke centres and may assist in outcome prediction. This study aimed to develop models of both excellent [modified Rankin score (mRS) 0-1] and devastating outcomes (mRS of 5-6).Methods: patients admitted with ischaemic or haemorrhagic stroke in 2001-02 to the Halifax Infirmary, Canada, were enrolled. Sixteen clinical variables from the first neurological assessment and six radiological variables from the acute CT scan were used to the model outcome at 6 months. Results: five hundred and thirty-eight stroke patients were enrolled. Thirty per cent had an excellent outcome and 30% had a devastating outcome. Three models of the excellent outcome were developed [area under the receiver operator curve (AUC) 0.866-882] including the variables age, pre-stroke functional status, stroke severity, ability to lift both arms, walk independently, normal verbal Glasgow Coma Scale and leukoaraiosis. Predictive models of the devastating outcome (AUC of 0.859-0.874) included additional variables living alone pre-stroke and total anterior circulation stroke. The simplest models of both outcomes were externally validated (AUC of 0.856-0.885).Conclusion: this study demonstrates new externally validated predictive models of both excellent and devastating outcomes. Leukoaraiosis was the only independent radiological predictor of both outcomes. Living alone pre-stroke predicted devastating outcome post-stroke.