Subarachnoid Hemorrhage: Who Dies, and Why?
Subarachnoid Hemorrhage: Who Dies, and Why?
All spontaneous SAH patients admitted to the Neurological Intensive Care Unit at Columbia University Medical Center between July 1996 and January 2009 were offered enrollment in the Columbia University SAH Outcomes Project. Both patients with and those without a documented aneurysm were included in the analysis. The study was approved by the Columbia University Institutional Review Board, and in all cases written informed consent was obtained from the patient or a legally authorized representative. Patients were admitted to a dedicated neurological intensive care unit (ICU) and given treatment according to a standardized management protocol that has been described in detail previously; see Methods Supplement (Additional file 1 http://www.ccforum.com/content/19/1/309/additional) for details regarding inclusion and exclusion criteria, aneurysm management, and a description of our ICU management protocol and how it evolved over time.
Demographic data, social and medical history, and clinical features at onset were obtained shortly after admission. Neurological status was assessed with the Glasgow Coma Scale (GCS) and the Hunt-Hess scale. Physiologic stress was captured by using the Acute Physiology and Chronic Health Evaluation II (APACHE II) scale by calculating a physiological subscore after subtracting the GCS, age, and chronic health contributions to the total score. Admission computed tomography (CT) scans, and those with significant interval changes during hospitalization, were evaluated by using the Modified Fisher Scale, Hijdra SAH sum score, and Intraventricular Hemorrhage Score and for the presence of global cerebral edema. During weekly meetings, a review of the entire hospital course was conducted to document and adjudicate important procedures, events, and complications according pre-specified criteria. Delayed cerebral ischemia (DCI) was defined as neurological deterioration, cerebral infarction, or both, when the cause was felt to be vasospasm after careful exclusion of other causes.
In-hospital mortality was used as the primary outcome for this analysis to allow for uniform evaluation of the causes, mode (i.e., brain versus cardiac), and level of support related to the dying process. The principal mechanism of death or neurological devastation leading to withdrawal of care was identified by adjudication of the study and clinical team after review of all pertinent clinical and radiographic findings. Primary causes of death were divided into eight categories based on the underlying pathophysiological mechanism: (1) direct effect of the primary hemorrhage, (2) aneurysm rebleeding, (3) cerebral infarction from vasospasm, (4) refractory cerebral edema leading to brain stem herniation, (5) hydrocephalus, (6) operative complications, (7) medical complications (e.g., fatal arrhythmia, pulmonary embolism, or multisystem organ failure who initially had a good neurological prognosis), and (8) other. Although different mechanisms could overlap in the same patient, only the adjudicated primary cause of death was considered.
Do-not-resuscitate (DNR) status was systemically documented when instituted. Withdrawal of supportive care was defined as the cessation of intensive care support such as mechanical ventilation and the beginning of comfort care. The treating team actively withdrew life support only at the direction of family members on the basis of dismal expected prospects for recovery and the known wishes of the patient on the basis of previous written or verbal statements.
Continuous variables were dichotomized to relevant clinical cut-points. Continuous variables were assessed for normality by using the Kolmogorov-Smirnov test. Normally distributed data were reported as a mean ± standard deviation, and non-parametric data were reported as median and interquartile range (IQR). Univariable associations were tested by using chi-squared or Fisher's exact test for categorical variables, two-tailed t test for normally distributed continuous variables, and Mann–Whitney U test for non-normally distributed continuous variables. An initial multivariable analysis using logistic regression was performed to determine the relationship between premorbid demographic and admission clinical and radiographic variables and in-hospital mortality. For clinically intercorrelated variables that measure the same construct (e.g., admission clinical deficit measured by Hunt-Hess grade and GCS score), we selected the variable with the highest measure of association (odds ratio) and smallest P value to be included in the final model. After construction of the baseline model for prediction of mortality, adjusted odds ratios for specific hospital complications were calculated by adding each of these factors individually to the baseline model to evaluate their unique contribution. Receiver operating characteristic (ROC) analysis was used to assess the ability of various models to predict in-hospital mortality. Nagelkerke's R-squared was used to estimate the percentage of the variance predicted by the combination of variables into the predicted model. Finally, a Hosmer-Lemeshow test was used to test goodness-of-fit of the each model. Residual statistics analysis was performed to evaluate the influence of isolated points into the model. Potential multicollinearity between the parameters of the final regression model was performed by calculating tolerance and variance inflation factor (VIF) coefficients. Significance was set at 0.05 for all analyses. All analyses were performed with commercially available statistical software (SPSS version 18.0; SPSS Inc., now part of IBM Corporation, Armonk, NY, USA).
Methods
Study Population
All spontaneous SAH patients admitted to the Neurological Intensive Care Unit at Columbia University Medical Center between July 1996 and January 2009 were offered enrollment in the Columbia University SAH Outcomes Project. Both patients with and those without a documented aneurysm were included in the analysis. The study was approved by the Columbia University Institutional Review Board, and in all cases written informed consent was obtained from the patient or a legally authorized representative. Patients were admitted to a dedicated neurological intensive care unit (ICU) and given treatment according to a standardized management protocol that has been described in detail previously; see Methods Supplement (Additional file 1 http://www.ccforum.com/content/19/1/309/additional) for details regarding inclusion and exclusion criteria, aneurysm management, and a description of our ICU management protocol and how it evolved over time.
Clinical and Radiological Assessment
Demographic data, social and medical history, and clinical features at onset were obtained shortly after admission. Neurological status was assessed with the Glasgow Coma Scale (GCS) and the Hunt-Hess scale. Physiologic stress was captured by using the Acute Physiology and Chronic Health Evaluation II (APACHE II) scale by calculating a physiological subscore after subtracting the GCS, age, and chronic health contributions to the total score. Admission computed tomography (CT) scans, and those with significant interval changes during hospitalization, were evaluated by using the Modified Fisher Scale, Hijdra SAH sum score, and Intraventricular Hemorrhage Score and for the presence of global cerebral edema. During weekly meetings, a review of the entire hospital course was conducted to document and adjudicate important procedures, events, and complications according pre-specified criteria. Delayed cerebral ischemia (DCI) was defined as neurological deterioration, cerebral infarction, or both, when the cause was felt to be vasospasm after careful exclusion of other causes.
Outcome Assessment
In-hospital mortality was used as the primary outcome for this analysis to allow for uniform evaluation of the causes, mode (i.e., brain versus cardiac), and level of support related to the dying process. The principal mechanism of death or neurological devastation leading to withdrawal of care was identified by adjudication of the study and clinical team after review of all pertinent clinical and radiographic findings. Primary causes of death were divided into eight categories based on the underlying pathophysiological mechanism: (1) direct effect of the primary hemorrhage, (2) aneurysm rebleeding, (3) cerebral infarction from vasospasm, (4) refractory cerebral edema leading to brain stem herniation, (5) hydrocephalus, (6) operative complications, (7) medical complications (e.g., fatal arrhythmia, pulmonary embolism, or multisystem organ failure who initially had a good neurological prognosis), and (8) other. Although different mechanisms could overlap in the same patient, only the adjudicated primary cause of death was considered.
Witholding or Withdrawal of Support
Do-not-resuscitate (DNR) status was systemically documented when instituted. Withdrawal of supportive care was defined as the cessation of intensive care support such as mechanical ventilation and the beginning of comfort care. The treating team actively withdrew life support only at the direction of family members on the basis of dismal expected prospects for recovery and the known wishes of the patient on the basis of previous written or verbal statements.
Statistical Analysis
Continuous variables were dichotomized to relevant clinical cut-points. Continuous variables were assessed for normality by using the Kolmogorov-Smirnov test. Normally distributed data were reported as a mean ± standard deviation, and non-parametric data were reported as median and interquartile range (IQR). Univariable associations were tested by using chi-squared or Fisher's exact test for categorical variables, two-tailed t test for normally distributed continuous variables, and Mann–Whitney U test for non-normally distributed continuous variables. An initial multivariable analysis using logistic regression was performed to determine the relationship between premorbid demographic and admission clinical and radiographic variables and in-hospital mortality. For clinically intercorrelated variables that measure the same construct (e.g., admission clinical deficit measured by Hunt-Hess grade and GCS score), we selected the variable with the highest measure of association (odds ratio) and smallest P value to be included in the final model. After construction of the baseline model for prediction of mortality, adjusted odds ratios for specific hospital complications were calculated by adding each of these factors individually to the baseline model to evaluate their unique contribution. Receiver operating characteristic (ROC) analysis was used to assess the ability of various models to predict in-hospital mortality. Nagelkerke's R-squared was used to estimate the percentage of the variance predicted by the combination of variables into the predicted model. Finally, a Hosmer-Lemeshow test was used to test goodness-of-fit of the each model. Residual statistics analysis was performed to evaluate the influence of isolated points into the model. Potential multicollinearity between the parameters of the final regression model was performed by calculating tolerance and variance inflation factor (VIF) coefficients. Significance was set at 0.05 for all analyses. All analyses were performed with commercially available statistical software (SPSS version 18.0; SPSS Inc., now part of IBM Corporation, Armonk, NY, USA).
Source...