London, United Kingdom 2013 Poster Session Red Cerebrovasc Dis 2013; 35 (suppl 3)1-854 407 229 Brain imaging Prediction of Post-anoxic Coma outcome using Decision Tree Analysis T.G. PHAN1, S. SINGHAL2, H. Ma3, B. CLISSOLD4, J. LY5 STROKE AND AGING RESEARCHc, CLAYTON, AUSTRALIA1, STROKE AND AGING RESEARCH, CLAYTON, AUSTRALIA2, STROKE AND AGING RESEARCH, CLAYTON, 3, STROKE AND AGING RESEARCH, CLAYTON, AUSTRALIA4, STROKE AND AGING RE-SEARCH, CLAYTON, AUSTRALIA5 Background and purpose: There are very few publications in the literature which address the role of MR imaging in the prediction of coma outcome. Recursive partitioing is a decision tree analysis which generate clinically intuitative pathway for arriving at a decision and is based on partitioning predictors into binary caetgory. In this exploratory study, we evaluate the use of recursive parti-tioning to predict poor outcome following cardiorespiratory arrest. Methods: The inclusion criteria were: age ≥17 years, cardio-respiratory arrest and a coma on admission (2003-2011). Ischemic in-jury was manually segmented from MRI scans obtained >24 hours after cardio-respiratory arrest. Poor outcome was defined as modified Rankin score 5-6. The data was split into a training set (4/5) and validation set (1/5). Recursive partitioning was used to develop a clinical model (R Foundation). The accuracy of the model was measured using the area under the receiver operating characteris-tics curve (AUC). Results: Forty patients were included in this retrospective series(mean age ± SD = 51.5 ± 18.9 years). Seventeen patients were assigned a good outcome ( modified Rankin score mRS of 0-4 at 3 months); 23 patients were assigned a bad outcome (mRS of 5 or 6 at 3 months). The initial recursive partitioning model for Day 2 were Glasgow Coma Score (GCS) dichotmoised at 7 and age dichotomised at 68 year old. However when infarct volume was entered into the model, the final model was infarct volume (≥ 6 ml), age (≥ 68 year old). This model had AUC of 0.88. At day 7, GCS (≤10) was the only predictor of poor outcome. This model had AUC of 0.86. Conclu-sion: Decision tree analysis may be used to help with developing pathway for predicting coma out-come. In our small dataset, infarct volume may have a role in prediction of outcome at day 2. 230 Brain imaging Relating ASPECTS infarct location to stroke disability in the NINDS rt-PA trial: proof of con-cept study using penalized logistic regression. T.G. PHAN1, A. DEMCHUK2, V. SRIKANTH3, B. SILVER4, S. PATEL5, P.A. BARBER6, S.R. LEVINE7, M. HILL8, The NINDS rt-PA Stroke Study Group9 STROKE AND AGING RESEARCH, CLAYTON, AUSTRALIA1, HOTCHKISS BRAIN IN-STITUTE, CALGARY, CANADA2, STROKE AND AGING RESEARCH, CLAYTON, AUSTRA-LIA3, BROWN UNIVERSITY, RHODE ISLAND, USA4, HENRY FORD HOSPITAL, WAYNE STATE, USA5, HOTCHKISS BRAIN INSTITUTE, CALGARY, CANADA6, THE STATE UNI-VERSITY OF NEW YORK, BROOKLYN, USA7, HOTCHKISS BRAIN INSTITUTE, CALGARY, CANADA8, The NINDS rt-PA Stroke Study Group, , USA9 Background and Purpose: The Alberta stroke programme early CT score (ASPECTS) on head CT representing the extent of early brain ischemia has been shown to be useful for predicting stroke outcome. In this exploratory analysis we hypothesized that infarct location as represented by the in-dividual ASPECTS region may be independently related to disability. Methods: ASPECTS scores from the baseline CT scans of The National Institute of Neurological Disorders (NINDS) rt-PA Stroke Study were obtained. The readers were blind to outcome and clinical information except symptom laterality. Due to the collinearity (relatedness) between the ASPECTS regions, we used penalized logistic regression (PLR) to determine the independent associations of exposures (recom-binant tissue plasminogen activator (rt-PA), demographic variables (age and sex), and imaging (AS-PECTS location) with poor outcome as defined by modified Rankin Score (mRS) of ≥ 2. Penalized logistic regression is an analytic tool which can handle data when collinearity is present. It provides results in terms of β coefficient related to specific infarct locations in a manner that is intuitively un-derstood by clinicians. Results: In 607/624 subjects with ASPECTS readings, variables significantly associated with poor outcome included: interactions between ASPECTS M6 region (primary motor cortex/parietal lobe and adjacent white matter) and age (p = 0.004), lentiform nucleus and age (p= 0.007), and blood sugar level and age (p = 0.01). The model suggested that as the patient become older, involvement of either M6 or lentiform nucleus slightly increased the odds of disability. How-ever, the predominant effect was driven by rt-PA which reduced the odds of poor disability (β coef-ficient -0.515, p=0.003). This may explain why certain patients have potentially smaller gains from rt-PA treatments. Conclusion: At an older age, specific infarct locations are associated with a greater degree of disability.
Karger_ESC London_2013
To see the actual publication please follow the link above