BACKGROUND Prediction of patient outcome is an important aspect of the
management and study of aneurysmal subarachnoid hemorrhage (SAH). In
the present study, we evaluated the prognostic value of two multivaria
te approaches to risk classification, Classification and Regression Tr
ees (CART) and multiple logistic regression, and compared them with th
e best single predictor of outcome, level of consciousness. METHODS Da
ta prospectively collected in the first Cooperative Aneurysm Study of
intravenous nicardipine after aneurysmal SAH (NICSAH I, n = 885) were
used to develop the prediction models. Low-, medium-, and high-risk gr
oups for unfavorable outcome were devised using CART and a stepwise lo
gistic regression analysis. Admission factors incorporated into both c
lassification schemes were: level of consciousness, age, location of a
neurysm (basilar versus other), and the Glasgow Coma Score. The CART p
rediction tree also branched on a dichotomy of admission glucose level
. The two multivariate classifications were then compared with a predi
ction scheme based on the single best performing prognostic factor, le
vel of consciousness in an independent series, NICSAH II (n = 353), an
d also in the original training dataset. RESULTS A similar discriminat
ion of risk was achieved by the three classification systems in the te
sting sample (NICSAH II). The 8%, 19%, and 52% rates of unfavorable ou
tcome obtained from low-, medium-, and high-risk groups defined by LOC
approximated those obtained using the more complex multivariate syste
ms.CONCLUSION Although multivariate classification systems are useful
to characterize the relationship of multiple risk factors to outcome,
the simple clinical measure LOC is favored as a concise and practical
classification for predicting the probability of unfavorable outcome a
fter aneurysmal SAH. (C) 1998 by Elsevier Science Inc.