[Prediction of outcome in putaminal hemorrhage]

We obtained information on patients with putaminal hemorrhage (3,638 medically treated cases and 3,372 surgically treated cases). With these data, we have developed easily applicable and clinically useful prediction models for both mortality and activities of daily living (ADL) at three months after admission. We derived these models by Hayashi's discriminant technique for categorical data. Before their derivation, variables generally available on the initial day of hospitalization and possibly related to outcome were examined individually with tests of homogeneity. Neurological grading (NG), age, deformity of cisterns around midbrain, midline shift, hematoma volume, and motor paralysis were identified as candidate predictors of both mortality and ADL. CT classification was also added to the predictors of ADL. Initially, all these factors were included in tentative models, and then those which were found not to contribute substantially to the prediction were deleted. Ultimately, we proposed two models, one for mortality and the other for ADL. The mortality model used NG, hematoma volume, and deformity of cisterns around midbrain as predictors, and the ADL model used age, NG, CT classification, hematoma volume, and motor paralysis. Average correct classification rates were 87% (medical) and 69% (surgical) for mortality, and 64% (medical) and 60% (surgical) for ADL. With these models, we evaluated indications of both kinds of treatment by comparing each predicted ADL of the medically treated cases with that of the surgically treated cases.