The authors studied the clinical and neuroimaging features of cerebellar hematomas to predict poor outcome using comprehensive statistical models.

We retrospectively reviewed clinical and neuroimaging features in 94 patients with spontaneous cerebellar hematomas to identify predictive features for a poor neurologic outcome, defined as death or dismissal to long-term care facility. Data were analyzed using chi square and Fisher's exact test with calculation of odd's ratios together with 95% confidence intervals.

Clinical and neuroradiologic predictors for a poor outcome at p < 0.05 were admission systolic blood pressure > 200 mm Hg, hematoma size >3cm, visible brain stem distortion, and acute hydrocephalus. Presenting findings predicting subsequent death at p < 0.05 were abnormal corneal and oculocephalic responses, Glasgow coma sum score less than 8, motor response less than localization to pain, acute hydrocephalus and intraventricular hemorrhage.

A tree-based analysis model using binary recursive partitioning showed that cornea reflex, hydrocephalus, doll's eyes, age, and size were the most important discriminating factors. Absent corneal reflexes on admission highly predicts poor outcome (86 percent, confidence limits 67-96 percent). When a cornea reflex is present, acute hydrocephalus predicts poor outcome but only when doll's eyes are additionally absent.