Predicting mortality after intracerebral hemorrhage: comparison of scoring systems and influence of withdrawal of care.

Objective: The Broderick system and the intracerebral hemorrhage (ICH) score are two systems for predicting 30-day mortality in patients with spontaneous ICH. No previous study has compared two ICH scoring systems in an independent patient cohort. Our purpose was to externally validate and directly compare these two systems and evaluate the effect of withdrawal of care on system performance.

Methods: In all, 307 consecutive patients admitted with ICH between 1998 and 2002 were evaluated. Broderick exclusion criteria were used, resulting in a cohort of 241 patients. Admission Glasgow Coma Scale score, ICH volume, 30-day mortality, and day-30 location were collected. The sensitivity, specificity, receiver operator characteristic curves, and model explained variance (R2) of the two systems were directly compared. The statistical performances of both systems were then compared in subsets that included or excluded patients from whom care was withdrawn.

Results: Overall mortality was 76 of 241 (31.5%). The ICH score had significantly higher sensitivity (66% v 45%, P = .001) and higher receiver operator characteristic curves (0.814 v 0.773, P < .001) for predicting 30-day mortality. The model R2 and specificity were not significantly different between systems. Both systems were significantly associated with 30-day location stratified as home, acute rehabilitation, skilled nursing facility, or death (ICH score Chi square = 79.28, P < .001; Broderick Chi square = 60.63, P < .001). Inclusion or exclusion of patients who had supportive care withdrawn did not significantly affect overall statistical performance.

Conclusions: The ICH score performed significantly better than the Broderick system. Overall statistical performance of both systems was not influenced by withdrawal of care.