Predicting stroke outcome: Guy's prognostic score in practice.

This study aimed to cross validate the Guy's prognostic score in a new sample of patients of all ages admitted to hospital with a stroke and to devise and test a simpler version (the G-score). 361 consecutive acute patients with stroke who had been admitted to the acute hospitals in two adjacent health districts in East London were recruited and followed up for six months after the stroke. The G-score was derived by simplifying the weights used in calculating the Guy's score. With the conventional threshold of 0 for the Guy's score and 3 (out of 7) for the G-score, the sensitivity of both scores for predicting a bad outcome (death or Barthel score < 13 out of 20 at six months) was 0.72 and specificity was 0.63. The likelihood ratio for the Guy's score was 1.97 and for the G-score 1.95. Both versions of the score performed better than conscious level alone at predicting the outcome (sensitivity 0.47, specificity 0.73, likelihood ratio 1.74). Similar data are presented for different thresholds and prior probabilities. In addition, the G-score permits direct estimation of 95% confidence intervals for the probability of a bad outcome for five grades of stroke severity. Outcome prediction with multivariate techniques has the potential to improve and inform clinical decision making. The G-score should be used to define stroke severity for individual patients and for case mix adjustment.