Predicting outcome in hyper-acute stroke - validation of a prognostic model in the Third International Stroke Trial (IST3)

Abstract

Background and Purpose: Models are used to adjust for case mix and to stratify treatment allocation in clinical trials and can, if accurate enough, be used to aid decision making in individual patients. We aimed to validate, in patients assessed within six hours of onset, a previously described six simple variable (SSV) model that was developed in stroke patients assessed sub-acutely. The explanatory variables in the model are age, living alone, independent pre-stroke, Glasgow Coma Scale verbal score, ability to lift arms, ability to walk.

Methods: The six variables were collected at randomisation into the IST3 trial of recombinant tissue plasminogen activator in ischaemic stroke. We assessed survival to 30 days and functional status at six months using the Oxford Handicap scale. We constructed receiver operator characteristic (ROC) curves to establish the model’s discriminatory performance and tested its calibration by charting predicted versus actual outcomes.

Results: 537 patients (mean age 74 years) were included of whom 422 (79%) survived 30 days and 179 (33%) were alive and independent at 6 months. The SSV model had an area under the ROC curve of 0.73 for 30-day survival and 0.82 for independent survival at six months. Calibration was satisfactory.

Conclusions: This study confirms the external validity of the SSV model in an ischaemic stroke population assessed within 6 hours of symptom onset. The SSV model comprising easily collected variables can therefore be used to stratify patients in hyper-acute stroke trials, but probably is not accurate enough for decision-making in individual patients.