22. European Stroke Conference 884 Intracerebral/subarachnoid haemorrhage and venous diseases A Prediction Score for Hematoma Expansion Following Acute Intracerebral Hemorrhage H.B. Brouwers1, Y. Chang2, G.J. Falcone3, A.M. Ayres4, K.A. McNamara5, A. Vashkevich6, T.W. Battey7, V. Valant8, K. Schwab9, N.S. Rost10, J.M. Romero11, A. Viswanathan12, S.M. Greenberg13, J. Rosand14, J.N. Goldstein15 Massachusetts General Hospital, Harvard Medical School, Boston, USA1, Massachusetts Gen-eral Hospital, Harvard Medical School, Boston, USA2, Massachusetts General Hospital, Harvard Medical School, Boston, USA3, Massachusetts General Hospital, Harvard Medical School, Boston, USA4, Massachusetts General Hospital, Harvard Medical School, Boston, USA5, Massachusetts General Hospital, Harvard Medical School, Boston, USA6, Massachusetts General Hospital, Har-vard Medical School, Boston, USA7, Massachusetts General Hospital, Harvard Medical School, Boston, USA8, Massachusetts General Hospital, Harvard Medical School, Boston, USA9,Massachu-setts General Hospital, Harvard Medical School, Boston, USA10, Massachusetts General Hospital, Harvard Medical School, Boston, USA11, Massachusetts General Hospital, Harvard Medical School, Boston, USA12, Massachusetts General Hospital, Harvard Medical School, Boston, USA13, Massa-chusetts General Hospital, Harvard Medical School, Boston, USA14, Massachusetts General Hospi-tal, Harvard Medical School, Boston, USA15 BACKGROUND: Hematoma expansion (HE) following acute intracerebral hemorrhage (ICH) is common and associated with poor outcome. Ongoing clinical trials therefore focus on restricting HE. To optimize patient selection for such trials, we aimed to develop a prediction score for HE. METHODS: We performed a prospective cohort study of consecutive primary ICH patients with available baseline and follow-up CTs. HE was assessed using semi-automated software and defined as 6 mL or 33% growth. Our cohort was randomly divided into a 2/3 development and 1/3 valida-tion dataset. Covariates were tested for association with HE, using uni- and multivariable logistic regression. A prediction model was derived based on regression estimates and subsequently tested in the validation cohort. RESULTS: 817 patients were included: 544 in the development cohort and 273 in the validation cohort. Overall, HE occurred in 156 patients (19%). In multivariable analysis, warfarin use (OR 2.09 95%CI 1.24-3.52, p = 0.006), shorter time to CT (≤6 vs. >6 hours; OR 2.14 95%CI 1.18-3.90, p = 0.013), baseline ICH volume (<30, 30-60, >60 mL; OR 1.90 95%CI 1.01-3.55, p = 0.045), and the CT angiography spot sign (OR 3.59 95%CI 1.77-7.29, p = 0.0004) were predic-tive of HE. The derived prediction score (0-9) showed a linear relation with the probability of HE and mortality (discharge and 90 days) in both cohorts. The probability of HE in the validation cohort was 0%, 10%, 39%, and 54% for categorized scores 0, 1-3, 4-6, and 7-9, respectively. The c-statis-tics was 0.69 for the development cohort and 0.78 for the validation cohort. CONCLUSION: Warfarin use, presentation within 6 hours after ICH, larger ICH volumes, and CT angiography spot sign are independent predictors of HE and mortality. Based on these findings, we developed a prediction score for HE, which performed well in the validation cohort. Our results open a path for individualized treatment and trial design in ICH, aimed at patients at highest risk of HE. 814 © 2013 S. Karger AG, Basel Scientific Programme 885 Intracerebral/subarachnoid haemorrhage and venous diseases Prediction of disability outcome following ICH: a decision tree analysis T.G. PHAN1, J. CHEN2, V. SRIKANTH3, H. MA4, B. CLISSOLD5, J. LY6 VISTA Collaboration. UNITED KINGDOM STROKE AND AGING RESEARCH, CLAYTON, AUSTRALIA1, STROKE AND AGING RE-SEARCH, CLAYTON, AUSTRALIA2, STROKE AND AGING RESEARCH, CLAYTON, AUS-TRALIA3, STROKE AND AGING RESEARCH, CLAYTON, AUSTRALIA4, STROKE AND AG-ING RESEARCH, CLAYTON, AUSTRALIA5, STROKE AND AGING RESEARCH, CLAYTON, ARGENTINA6 Background: Regression models for predicting poor outcome (n=81-374) following intracerebral hemorrhage (ICH) have high specificity but low sensitivity and consequently do not perform well clinically. Decision tree analysis is a powerful data driven method, based on partitioning every pre-dictors into binary category. It generates intuitive pathway for decision making. Our aim was to develop a model to predict disability outcome following ICH using binary and ordinal decision tree analysis. Methods: The Virtual International Stroke Trials Archive (VISTA) was searched for patients with re-cordings of ICH volume measured from CT scans and of outcome. We used recursive partitioning to develop a predictive model (version 2.14.0, The R Foundation for Statistical Computing). Disability was defined as mild if modified Rankin score (mRS) at 90 days was 0-2, moderate if mRS was 3-4 and severe if mRS was 5-6. The data were randomly divided into a training data set (4/5 of data) and validation data (1/5 of data). The area under (AUC) the receiver operating characteristic curve was used to calculate the accuracy of the model. Results: VISTA contained 957 eligible patients, average age 66±12 years (64% males). The bina-ry decision tree showed that older age (>66.5 year old), higher serum glucose (6.35 mmol/L) and mid-line shift (>12 mm) or higher ICH volume (>9.7 ml) discriminated between mild disability and moderate to severe disability. The AUC for the training data was 0.85 and for the validation data was 0.60. The ordinal decision tree showed that low ICH volume (<27 ml), younger patient (<66.5 year old) and low serum glucose (<6.35 mmol/L) predicted the disability categories (see attached figure explaining the ordinal tree). The AUC for the training data was 0.80 and for validation data was 0.82. Conclusion: The ordinal recursive partitioning method had better reproducibility of results for the validation dataset compared to the binary method. Figure 1
Karger_ESC London_2013
To see the actual publication please follow the link above