OBJECTIVE: Most cardiac arrest patients who are successfully resuscitated are initially comatose due to hypoxic-ischemic brain injury. Quantitative electroencephalography (EEG) provides valuable prognostic information. However, prior approaches largely rely on snapshots of the EEG, without taking advantage of temporal information. METHODS: We present a recurrent deep neural network with the goal of capturing temporal dynamics from longitudinal EEG data to predict long-term neurological outcomes. We utilized a large international dataset of continuous EEG recordings from 1,038 cardiac arrest patients from seven hospitals in Europe and the US. Poor outcome was defined as a Cerebral Performance Category (CPC) score of 3-5, and good outcome as CPC score 0-2 at 3 to 6-months after cardiac arrest. Model performance is evaluated using 5-fold cross validation. RESULTS: The proposed approach provides predictions which improve over time, beginning from an area under the receiver operating characteristic curve (AUC-ROC) of 0.78 (95% CI: 0.72-0.81) at 12 hours, and reaching 0.88 (95% CI: 0.85-0.91) by 66 h after cardiac arrest. At 66 h, (sensitivity, specificity) points of interest on the ROC curve for predicting poor outcomes were (32,99)%, (55,95)%, and (62,90)%, (99,23)%, (95,47)%, and (90,62)%; whereas for predicting good outcome, the corresponding operating points were (17,99)%, (47,95)%, (62,90)%, (99,19)%, (95,48)%, (70,90)%. Moreover, the model provides predicted probabilities that closely match the observed frequencies of good and poor outcomes (calibration error 0.04). CONCLUSIONS AND SIGNIFICANCE: These findings suggest that accounting for EEG trend information can substantially improve prediction of neurologic outcomes for patients with coma following cardiac arrest.

BACKGROUND: Studies of neurologic outcomes have found conflicting results regarding differences between patients with substance-related cardiac arrests (SRCA) and non-SRCA. We investigate the effects of SRCA on severe cerebral edema development, a neuroimaging intermediate endpoint for neurologic injury. METHODS: 327 out-of-hospital comatose cardiac arrest patients were retrospectively analyzed. Demographics and baseline clinical characteristics were examined. SRCA categorization was based on admission toxicology screens. Severe cerebral edema classification was based on radiology reports. Poor clinical outcomes were defined as discharge Cerebral Performance Category scores>3. RESULTS: SRCA patients (N=86) were younger (P<0.001), and more likely to have non-shockable rhythms (P<0.001), be unwitnessed (P<0.001), lower Glasgow Coma Scale scores (P<0.001), absent brainstem reflexes (P<0.05) and develop severe cerebral edema (P<0.001) than non-SRCA patients (N=241). Multivariable analyses found younger age (P<0.001), female sex (P=0.008), non-shockable rhythm (P=0.01) and SRCA (P=0.05) to be predictors of severe cerebral edema development. Older age (P<0.001), non-shockable rhythm (P=0.02), severe cerebral edema (P<0.001), and absent pupillary light reflexes (P=0.004) were predictors of poor outcomes. SRCA patients had higher proportion of brain death (P<0.001) compared to non-SRCA deaths. CONCLUSIONS: SRCA results in higher rates of severe cerebral edema development and brain death. The absence of statistically significant differences in discharge outcomes or survival between SRCA and non-SRCA patients may be related to the higher rate of withdrawal of life-sustaining treatment (WLST) in the non-SRCA group. Future neuroprognostic studies may opt to include neuroimaging markers as intermediate measures of neurologic injury which are not influenced by WLST decisions.

OBJECTIVE: Electroencephalography (EEG) is an important tool for neurological outcome prediction after cardiac arrest. However, the complexity of continuous EEG data limits timely and accurate interpretation by clinicians. We develop a deep neural network (DNN) model to leverage complex EEG trends for early and accurate assessment of cardiac arrest coma recovery likelihood. METHODS: We developed a multiscale DNN combining convolutional neural networks (CNN) and recurrent neural networks (long short-term memory [LSTM]) using EEG and demographic information (age, gender, shockable rhythm) from a multicenter cohort of 1,038 cardiac arrest patients. The CNN learns EEG feature representations while the multiscale LSTM captures short-term and long-term EEG dynamics on multiple time scales. Poor outcome is defined as a Cerebral Performance Category (CPC) score of 3-5 and good outcome as CPC score 1-2 at 3-6 months after cardiac arrest. Performance is evaluated using area under the receiver operating characteristic curve (AUC) and calibration error. RESULTS: Model performance increased with EEG duration, with AUC increasing from 0.83 (95% Confidence Interval [CI] 0.79-0.87 at 12h to 0.91 (95%CI 0.88-0.93) at 66h. Sensitivity of good and poor outcome prediction was 77% and 75% at a specificity of 90%, respectively. Sensitivity of poor outcome was 50% at a specificity of 99%. Predicted probability was well matched to the observation frequency of poor outcomes, with a calibration error of 0.11 [0.09-0.14]. CONCLUSIONS: These results demonstrate that incorporating EEG evolution over time improves the accuracy of neurologic outcome prediction for patients with coma after cardiac arrest.

OBJECTIVES: Determine the utility of the neurologic examination in comatose patients from nontraumatic causes in the modern era. DESIGN: Prospective observational study. SETTING: Single academic medical center. PATIENTS: Data from 500 patients in nontraumatic coma collected sequentially from 2000 to 2007 in the emergency department and neuroscience, medical, and cardiac intensive care units. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Clinical data were collected on days 0, 1, 3, and 7. Outcome was assessed at 6 months; good outcome was determined at two levels by modified Rankin Scale, ≤3 as independence and ≤4 as moderate but not severe disability. A classification and regression tree analysis was performed to determine prognostic variables, creating predictive algorithms of good vs. poor outcome for each day. Patients with coma attributable to subarachnoid hemorrhage (4/80; 5%) or global hypoxic-ischemic injury (20/202, 10%) were more likely to achieve good outcomes. The pupillary reflex was an important determinant, regardless of day or modified Rankin Scale cut point (mean odds ratio 12.51, range [6.01, 22.56] for modified Rankin Scale ≤3; mean odds ratio 19.26, range [5.38, 42.26] for modified Rankin Scale ≤4). A less robust effect was seen for oculocephalic reflexes (mean odds ratio 62.61, range [2.24, 177] for modified Rankin Scale ≤3; mean odds ratio 34.13, range [4.95, 89.93] for modified Rankin Scale ≤4). The motor response was selected as a predictor of outcome only on day 0 (odds ratio 2.35, 95% confidence interval 0.64-5.74 for modified Rankin Scale ≤3; odds ratio 2.1, 95% confidence interval 0.81-4.24 for modified Rankin Scale score ≤4). Age was not associated with outcome. CONCLUSIONS: The clinical neurologic examination remains central to determining prognosis in nontraumatic coma. Additional clinical and diagnostic variables may also aid in outcome prediction for specific disease states.

PURPOSE: To examine whether the severity and spatial distribution of reductions in apparent diffusion coefficient (ADC) are associated with clinical outcomes in patients who become comatose after cardiac arrest. MATERIALS AND METHODS: This was an institutional review board-approved, HIPAA-compliant retrospective study of 80 comatose patients with cardiac arrest who underwent diffusion-weighted magnetic resonance imaging. The need to obtain informed consent was waived except when follow-up phone calls were required; in those cases, informed consent was obtained from the families. Mean patient age was 57 years +/- 16 (standard deviation); 31 (39%) patients were women. ADC maps were semiautomatically segmented into the following regions: subcortical white matter; cerebellum; insula; frontal, occipital, parietal, and temporal lobes; caudate nucleus; putamen; and thalamus. Median ADCs were measured in these regions and in the whole brain and were compared (with a two-tailed Wilcoxon test) as a function of clinical outcome. Outcome was defined by both early eye opening in the 1st week after arrest (either spontaneously or in response to external stimuli) and 6-month modified Rankin scale score. RESULTS: Whole-brain median ADC was a significant predictor of poor outcome as measured by no eye opening (specificity, 100% [95% confidence interval {CI}: 86%, 100%]; sensitivity, 30% [95% CI: 18%, 45%]) or 6-month modified Rankin scale score greater than 3 (specificity, 100% [95% CI: 73%, 100%]; sensitivity, 41% [95% CI: 29%, 54%]), with patients with poor outcomes having significantly lower ADCs for both outcome measures (P

Neurological prognostication after cardiac arrest has always been challenging, and has become even more so since the advent of therapeutic hypothermia (TH) in the early 2000s. Studies in this field are prone to substantial biases--most importantly, the self-fulfilling prophecy of early withdrawal of life-sustaining therapies--and physicians must be aware of these limitations when evaluating individual patients. TH mandates sedation and prolongs drug metabolism, and delayed neuronal recovery is possible after cardiac arrest with or without hypothermia treatment; thus, the clinician must allow an adequate observation period to assess for delayed recovery. Exciting advances have been made in clinical evaluation, electrophysiology, chemical biomarkers and neuroimaging, providing insights into the underlying pathophysiological mechanisms of injury, as well as prognosis. Some clinical features, such as pupillary reactivity, continue to provide robust information about prognosis, and EEG patterns, such as reactivity and continuity, seem promising as prognostic indicators. Evoked potential information is likely to remain a reliable prognostic tool in TH-treated patients, whereas traditional serum biomarkers, such as neuron-specific enolase, may be less reliable. Advanced neuroimaging techniques, particularly those utilizing MRI, hold great promise for the future. Clinicians should continue to use all the available tools to provide accurate prognostic advice to patients after cardiac arrest.

OBJECTIVE: To build new algorithms for prognostication of comatose cardiac arrest patients using clinical examination, and investigate whether therapeutic hypothermia influences the value of the clinical examination. METHODS: From 2000 to 2007, 500 consecutive patients in non-traumatic coma were prospectively enrolled, 200 of whom were post-cardiac arrest. Outcome was determined by modified Rankin Scale (mRS) score at 6 months, with mRS≤3 indicating good outcome. The clinical examination was performed on days 0, 1, 3 and 7 post-arrest, and clinical variables analyzed for importance in prognostication of outcome. A classification and regression tree analysis (CART) was used to develop a predictive algorithm. RESULTS: Good outcome was achieved in 9.9% of patients. In CART analysis, motor response was often chosen as a root node, and spontaneous eye movements, pupillary reflexes, eye opening and corneal reflexes were often chosen as splitting nodes. Over 8% of patients with absent or extensor motor response on day 3 achieved a good outcome, as did 2 patients with myoclonic status epilepticus. The odds of achieving a good outcome were lower in patients who suffered asystole (OR 0.187, 95% CI: 0.039-0.875, p=0.033) compared with ventricular fibrillation or non-perfusing ventricular tachycardia, but some still achieved good outcome. The absence of pupillary and corneal reflexes on day 3 remained highly reliable for predicting poor outcome, regardless of therapeutic hypothermia utilization. CONCLUSION: The clinical examination remains central to prognostication in comatose cardiac arrest patients in the modern area. Future studies should incorporate the clinical examination along with modern technology for accurate prognostication.

BACKGROUND: The role of neuroimaging in assessing prognosis in comatose cardiac survivors appears promising, but little is known regarding the import of particular spatial patterns. We report a specific spatial imaging abnormality on magnetic resonance imaging (MRI) that portends a poor prognosis: bilateral hippocampal hyperintensities on diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) sequences. METHODS: Eighty sequential comatose cardiac arrest patients underwent MRI scans. Qualitative and quantitative regional analyses were performed. Patients were categorized as HIPPO(+) (n = 18) or HIPPO(-) (n = 62) based on whether they had bilateral hippocampal hyperintensities. Poor outcome was defined by a modified Rankin Scale (mRS) score ≥4 at 6 months. RESULTS: Patients with bilateral hippocampal abnormalities had a higher frequency of poor outcome (P = .032). HIPPO(+) patients suffered more severe cerebral injury, with lower whole brain apparent diffusion coefficient values (P = .043) and a greater number of affected regions on DWI (P = .001) and FLAIR (P = .001) than HIPPO(-) patients. The hippocampal approach was 100% specific for a poor prognosis; only 1 patient survived and remained in a vegetative state. CONCLUSIONS: Bilateral hippocampal hyperintensities on MRI may be a specific imaging finding that is indicative of poor prognosis in patients who suffer global hypoxic-ischemic injury. More research on the prognostic significance of this and similar neuroimaging patterns is indicated.

BACKGROUND AND PURPOSE: Early assessment of the likelihood of neurological recovery in comatose cardiac arrest survivors remains challenging. We hypothesize that quantitative noncontrast computed tomography (NCCT) combined with neurological assessments, are predictive of outcome. METHODS: We analyzed data sets acquired from comatose cardiac arrest patients who underwent CT within 72 hours of arrest. Images were semiautomatically segmented into anatomic regions. Median Hounsfield units (HU) were measured regionally and in the whole brain (WB). Outcome was based on the 6-month modified Rankin Scale (mRS) score. Logistic regression was used to combine Glasgow Coma Scale (GCS) score measured on Day 3 post arrest (GCS_Day3) with imaging to predict poor outcome (mRS>4). RESULTS: WB HU (P=0.02) and the ratio of HU in the putamen to the posterior limb of the internal capsule (PLIC) (P=0.004) from 175 datasets from 151 patients were univariate predictors of poor outcome. Thirty-three patients underwent hypothermia treatment. Multivariate analysis showed that combining median HU in the putamen (P=0.0006) and PLIC (P=0.007) was predictive of poor outcome. Combining WB HU and GCS_Day3 resulted in 72% [61% to 80%] sensitivity and 100% [73% to 100%] specificity for predicting poor outcome in 86 patients with measurable GCS_Day3. This was an improvement over prognostic performance based on GCS_Day3≤8 (98% sensitive but 71% specific). DISCUSSION: Combining density changes on CT with GCS_Day3 may be useful for predicting poor outcome in comatose cardiac arrest patients who are neither rapidly improving nor deteriorating. Improved prognostication with CT compared with neurological assessments can be achieved in patients treated with hypothermia.