Artikel
A new approach for predicting outcome and in-hospital mortality in primary intracerebral hemorrhage with associated intraventricular hemorrhage – a comparison between decision tree and logistic regression analysis
Eine neuer Ansatz zur Vorhersage des Outcomes und der 30-Tages-Mortalität bei Patienten mit primärer intrazerebraler Blutung und sekundärer intraventrikulärer Blutung. Ein Vergleich zwischen Entscheidungsbaumanalyse und logistischer Regression
Suche in Medline nach
Autoren
Veröffentlicht: | 30. Mai 2008 |
---|
Gliederung
Text
Objective: Spontaneous primary intracerebral hemorrhage (ICH) with associated intraventricular hemorrhage (IVH) accounts for the highest in-hospital mortality after stroke. The prediction of outcome, especially 30-day mortality or unfavourable outcome at 6 months, has not been clearified in studies such as the STICH-trial.
Methods: A total of 104 ICH/IVH patients (Mean age 63±11,3; male : female 1:0,7) were retrospectively analyzed. All patients received an external ventricular drainage (EVD). Age, gender, initial Glasgow Coma Score (GCS), initial blood pressure and laboratory data were recorded. On the initial CAT-Scan the volume of the hematoma, severity of IVH, midline shift, the origin of the hematoma, evans-ratio, size of temporal horns and hydrocephalus were determined. For the prediction of hydrocephalus a non-evaluated score was used that categorize hydrocephalus into absent, beginning, moderate and severe. Univariate and multivariat logistic regression analyses were used to identify prognostic predictors for 30-day mortality and outcome. Outcome was determined by the modified Ranking Scale (mRS) and the Barthel-Index 6 months after hemorrhage. Decision tree analysis was performed with the C5 software. The developed models from the multivariat logistic regression and the decision tree analyses were tested with prospective data from 43 patients.
Results: The most significant predictors of 30-day mortality in our data were severe hydrocephalus, volume of hematoma, midline shift and initial GCS. In the decision tree analysis volume of parenchymal hemorrhage >60 cm3, severe hydrocpehalus, GCS <6 and age >70 years were the strongest predictors of 30-day mortality. Significant prognostic factors for unfavourable outcome were age, volume of hematoma, GCS and the presence of hydrocephalus. Using decision tree analysis we found that GCS <8, Age <60 years and volume of the parenchymal hemorrhage >34 cm3 were the best predictors of bad outcome. Comparing the predictive accuracy of the C5 model and the logistic regression model the decision tree analysis was better for both, the 30-day mortality and outcome at 6 months. The results were confirmed in the prospective group.
Conclusions: GCS, age, hematoma volume and hydrocephalus are reliable predictors of clinical outcome. Decision tree models defined realistic cutoff values for these parameters.