552

Abstract

Introduction: Therapeutic hypothermia (TH) is now widely used worldwide in the management of post-cardiac arrest encephalopathy. Due to the use of limited ICU resources, there has also been a search for predictors with the required specificity as early as possible after cardiac arrest so futile treatment can be terminated; however, the reliability of predictors that can identify TH candidates remains unknown. Hypothesis: We can develop a prediction rule to predict the neurological outcome of TH patients; clinical predictors are obtained before patients underwent TH. Methods: We performed a single-center retrospective cohort study over 12 years (2001-2012). Inclusion criteria were out-of-hospital cardiac arrest of cardiac origin, successful return of spontaneous circulation and induction of therapeutic hypothermia. Data were abstracted from the charts included demographics, characteristics of cardiac arrest, clinical variables on hospital arrival and subsequent interventions. Outcome measure was neurological status at hospital discharge (cerebral-performance category [CPC]). Backward elimination of non-significant variables was used to identify the important predictors in a stepwise logistic regression model. The beta-coefficient of each of the predictors was used to build a score. We tested the discriminative power of prediction scores using receiver-operator characteristic (ROC) curve analysis with the area under the curve (AUC). Results: Of 72 eligible patients, 44 (61.1%) had favorable outcome (CPC1-2) and 28 (38.9%) had poor outcome (CPC3-5). Four variables were identified as important predictors of poor outcome (age >= 60 yrs, cardiac arrest in the home, absence of pupillary reflex on hospital arrival and CPR duration >= 20 min). The integer score of 2 was assigned for age >= 60 yrs and absence of pupillary reflex, and the integer score of 1 assigned for cardiac arrest in the home and CPR duration >= 20 min, weighted by beta-coefficient values. ROC analysis showed an AUC was 0.92 (95% confidence interval [CI], 0.87-0.98) and identified a prediction score >= 5 as an optimal cut-off point, with a sensitivity of 0.61 (95% CI, 0.50-0.64), specificity of 0.98 (95% CI, 0.91-1.00) and positive predictive value of 0.94 (95% CI, 0.78-0.99), negative predictive value of 0.80 (95% CI, 0.74-0.81) for predicting poor outcome. Conclusions: Our prediction rule was valid to identify futility of TH in patients after cardiac arrest. This tool requires external validation study.