Abstract 405: A Machine Learning-Based Prediction of Cardiac Arrest Outcome Using a Large Multi-Center Database

Fever after therapeutic hypothermia – Does rebound pyrexia matter?

Risk factors for neurological disability outcomes in patients under extracorporeal membrane oxygenation following cardiac arrest: An observational study

Computational signatures for post-cardiac arrest trajectory prediction: Importance of early physiological time series

Background: With the recent introduction of therapeutic hypothermia, the application of sedation becomes necessary in cardiac arrest patients. Therefore early prediction of neurologic outcome in these patients is challenging problem. Nevertheless neurologic examination including the Glasgow coma scale should be basis for decision making. So, we investigated the usefulness of the Glasgow motor score before induction of therapeutic hypothermia for outcome prediction in survivors of cardiac arrest treated with therapeutic hypothermia. Methods: Between March 2009 and July 2011, patients who successfully resuscitated from cardiac arrest and treated with mild therapeutic hypothermia were included in this study. Data were collected by local post-cardiac arrest care registry. Neurologic outcome was assessed with Cerebral Performance Category (CPC) at hospital discharge and good neurologic outcome was defined CPC score of 1 and 2. Results: Total 89 patients were treated with therapeutic hypothermia during study period. Among these patients, 39 were good neurologic outcome and 50 were poor neurologic outcome. The value for prediction of good outcome was as follows: sensitivity 53.8%, specificity 96.0%, PPV 91.3%, NPV 72.7% and AUC 0.749. Conclusions: The results of this study show that the Glasgow motor score > 1 before induction of therapeutic hypothermia is an independent predictor of good outcome in patients treated with therapeutic hypothermia. Thus, the patients with the Glasgow motor score > 1 soon after ROSC should be treated aggressively including therapeutic hypothermia.

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. 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. 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). These findings suggest that accounting for EEG trend information can substantially improve prediction of neurologic outcomes for patients with coma following cardiac arrest.

Few studies have demonstrated the prognostic potential of neutrophil gelatinase-associated lipocalin (NGAL) in post-cardiac arrest patients. This study evaluated the usefulness of plasma NGAL in predicting neurologic outcome and mortality in out-of-hospital cardiac arrest (OHCA) patients treated with targeted temperature management (TTM). A prospective observational study was conducted between October 2013 and April 2016 at a single tertiary hospital. We enrolled 75 patients treated with TTM and collected their demographic data, cardiopulmonary resuscitation-related information, data on plasma NGAL concentration, and prognostic test results. Plasma NGAL was measured at 4 hours after return of spontaneous circulation (ROSC). The primary endpoint was the neurologic outcome at discharge and the secondary outcome was 28-day mortality. Neurologic outcomes were analyzed using a stepwise multivariate logistic regression while 28-day mortality was analyzed using a stepwise Cox regression. The predictive performance of plasma NGAL for neurologic outcome was measured by the area under the receiver operating characteristic curve and the predictability of 28-day mortality was measured using Harrell C-index. We also compared the predictive performance of plasma NGAL to that of other traditional prognostic modalities for outcome variables. Thirty patients (40%) had good neurologic outcomes and 53 (70.7%) survived for more than 28 days. Plasma NGAL in patients with good neurologic outcomes was 122.7 ± 146.7 ng/ml, which was significantly lower than that in the poor neurologic outcome group (307.5 ± 269.6 ng/ml; P < .001). The probability of a poor neurologic outcome was more than 3.3-fold in the NGAL >124.3 ng/ml group (odds ratio, 3.321; 95% confidence interval [CI], 1.265–8.721]). Plasma NGAL in the survived group was significantly lower than that in the non-survived group (172.7 ± 191.6 vs 379.9 ± 297.8 ng/ml; P = .005). Plasma NGAL was significantly correlated with 28-day mortality (hazard ratio 1.003, 95% CI 1.001–1.004; P < .001). The predictive performance of plasma NGAL was not inferior to that of other prognostic modalities except electroencephalography. Plasma NGAL is valuable for predicting the neurologic outcome and 28-day mortality of patients with OHCA at an early stage after ROSC.This study was registered at ClinicalTrials.gov on November 19, 2013 (Identifier: NCT01987466).

Patients admitted to a cardiac intensive care unit are often unconscious with uncertain prognosis. Automated infrared pupillometry for neurological assessment in the intensive care unit may provide early prognostic information. This study aimed to determine the prognostic value of automated pupillometry in different subgroups of patients in a cardiac intensive care unit with 30-day mortality as the primary endpoint and neurological outcome as the secondary endpoint. A total of 221 comatose patients were divided into three groups: out-of-hospital cardiac arrest, in-hospital cardiac arrest and others (i.e. patients with cardiac diagnoses other than cardiac arrest). Automated pupillometry was serially performed until discharge or death and pupil measurements were analysed using the neurological pupil index algorithm. We applied receiver operating characteristic curves in univariable and multivariable logistic regression models and a calculated Youden index identified neurological pupil index cut-off values at different specificities. In out-of-hospital cardiac arrest patients higher neurological pupil index values were independently associated with lower 30-day mortality. The univariable model for 30-day mortality had an area under the curve of 0.87 and the multivariable model achieved an area under the curve of 0.94. The Youden index identified a neurological pupil index cut-off in out-of-hospital cardiac arrest patients of 2.40 for a specificity of 100%. For patients with in-hospital cardiac arrest and other cardiac diagnoses, we found no association between neurological pupil index values and 30-day mortality, and the univariable models showed poor predictive values. Automated infrared pupillometry has promising predictive value after out-of-hospital cardiac arrest, but poor predictive value in patients with in-hospital cardiac arrest or cardiac diagnoses unrelated to cardiac arrest. Our data suggest a possible neurological pupil index cut-off of 2.40 for poor outcome in out-of-hospital cardiac arrest patients.

Background: Progressive hypoxic brain injury after cardiac arrest can affect serum osmolality (SOsm) and urine osmolality (UOsm) in out-of-hospital cardiac arrest (OHCA) patients. This study aimed to examine the association between the ratios of UOsm to SOsm (USRs) within 3 days after return of spontaneous circulation (ROSC) and neurological outcomes in OHCA patients. Methods: The present study was prospective observational study and included OHCA patients with targeted temperature management (TTM) at Chonnam National University Hospital in Gwangju, Korea, between January 2016 and December 2022. We collected SOsm and UOsm at admission and at 24 h, 48 h, and 72 h after ROSC. We assessed the predictive performance of USRs to determine neurological outcome at discharge. Results: A total of 319 patients were included. UOsm levels and USRs at admission and 24 h, 48 h, and 72 h after ROSC in patients with poor outcome were lower than those in patients with good outcome. In multivariable analysis, the USRs at 24 h (odd ratio [OR], 0.363; 95% confidence interval [CI], 0.221-0.594), 48 h (OR, 0.451; 95% CI, 0.268-0.761), and 72 h (OR, 0.559; 95% CI, 0.357-0.875) after ROSC were independently associated with poor neurological outcome at discharge. The area under the receiver operating characteristic curves of USRs at admission and 24 h, 48 h, and 72 h after ROSC for predicting neurological outcome at discharge were 0.615 (95% CI, 0.559-0.669), 0.711 (95% CI, 0.658-0.760), 0.724 (95% CI, 0.671-0.772), and 0.751 (95% CI, 0.699-0.797). Conclusions: The USRs within 3 days of ROSC were associated with poor neurologic outcome at discharge in OHCA patients.

BackgroundThere are conflicting data regarding sex-based differences in the outcomes of out-of-hospital cardiac arrest (OHCA) patients, and whether the specific sex advantage is age-specific remains unclear. We assessed the impact of the interactions between sex and age on the neurological outcomes of OHCA patients receiving targeted temperature management (TTM).MethodsData collected from 2007 to 2012 for a multicenter, registry-based study of the Korean Hypothermia Network were analyzed. We used a multivariate logistic regression model with an interaction term (age × sex) as the final model for the outcomes. To evaluate the association between sex and outcome in specific age groups, all patients were divided into specific age subgroups, and the adjusted ORs and 95% CIs of good neurological outcomes for males were calculated for each age group. Finally, the ORs of a good neurological outcome for the specific age groups compared with the 50- to 59-year-old group were calculated for both sexes.ResultsIn the interaction analysis, age was a negative prognostic factor (OR, 0.95 [95% CI, 0.93-0.98]), whereas sex was not associated with neurological outcomes (OR, 3.74 [95% CI, 0.85–16.35]), and reproductive age in females (age, < 50 years) was also not associated with good neurological outcomes. After the patients were divided into five age groups, sex was not an independent predictor of neurological outcomes across all age groups. Patients of both sexes aged < 40 years had significantly better outcomes than patients in the 50- to 59-year-old group (males, OR, 4.03 [95% CI, 1.86–8.73]; females, OR, 10.34 [95% CI, 1.99–53.85]). Males aged ≥ 70 years had significantly poorer neurological outcomes than those in the 50- to 59-year-old group (OR, 0.15 [95% CI, 0.07–0.32]), but this outcome was not observed for females (OR, 0.78 [95% CI, 0.20–3.14]).ConclusionsSex did not influence the neurological outcomes of TTM-treated OHCA patients. In contrast to the outcomes in males, the neurological outcomes of females worsened from 18 to 59 years of age and then remained constant.

Part 6: Defibrillation

Outcome prediction after cardiac arrest is important to decide on continuation or withdrawal of intensive care. Neuron-specific enolase is an easily available, observer-independent prognostic biomarker. Recent studies have yielded conflicting results on its prognostic value after targeted temperature management. We analyzed neuron-specific enolase serum concentrations 3 days after nontraumatic in-hospital cardiac arrest and out-of-hospital cardiac arrest and outcome of patients from five hospitals in Germany, Austria, and Italy. Patients were treated at 33°C for 24 hours. Cerebral Performance Category was evaluated upon ICU discharge. We performed case reviews of good outcome patients with neuron-specific enolase greater than 90 μg/L and poor outcome patients with neuron-specific enolase less than or equal to 17 μg/L (upper limit of normal). A neuron-specific enolase serum concentration greater than 90 μg/L predicted Cerebral Performance Category 4-5 with a positive predictive value of 99%, false positive rate of 0.5%, and a sensitivity of 48%. All three patients with neuron-specific enolase greater than 90 μg/L and Cerebral Performance Category 1-2 had confounders for neuron-specific enolase elevation. An neuron-specific enolase serum concentration less than or equal to 17 μg/L excluded Cerebral Performance Category 4-5 with a negative predictive value of 92%. The majority of 14 patients with neuron-specific enolase less than or equal to 17 μg/L who died had a cause of death other than hypoxic-ischemic encephalopathy. Specificity and sensitivity for prediction of poor outcome were independent of age, sex, and initial rhythm but higher for out-of-hospital cardiac arrest than for in-hospital cardiac arrest patients. High neuron-specific enolase serum concentrations reliably predicted poor outcome at ICU discharge. Prediction accuracy differed and was better for out-of-hospital cardiac arrest than for in-hospital cardiac arrest patients. Our "in-the-field" data indicate 90 μg/L as a threshold associated with almost no false positives at acceptable sensitivity. Confounders of neuron-specific enolase elevation should be actively considered: neuron-specific enolase-producing tumors, acute brain diseases, and hemolysis. We strongly recommend routine hemolysis quantification. Neuron-specific enolase serum concentrations less than or equal to 17 μg/L argue against hypoxic-ischemic encephalopathy incompatible with reawakening.

Cardiac arrest (CA) is a serious health concern that often results in mortality or severe neurological dysfunction in the case of survival. Our aim was to explore the neurological prognostic factors in patients with CA. This retrospective observational study included adult patients with CA. We investigated serum neuron-specific enolase (NSE), S100 calcium-binding protein β (S100β), and indices and parameters at 1, 3, 5, 7 and intensive care unit (ICU) discharge days after CA. The primary study endpoint was the Cerebral Performance Category (CPC) scale score at ICU discharge, which was dichotomized as good neurological outcome (CPC 1–2: full recovery or moderate disability) and poor neurological outcome (CPC 3–5: severe disability, vegetative state, or death). Of the 191 adult patients with CA, 42 (22%) had good neurological outcomes, and 149 (78%) had poor neurological outcomes. NSE at 1,3,5,7 and ICU discharge days showed excellent predictive accuracy for neurological outcomes (area under the curve [AUC]: 0.666, 0.716, 0.870, 0.739, and 0.901, respectively). However, S100β exhibited general predictive power (AUC: 0.666, 0.573, 0.607, 0.594, 0.727). Finally, the early warning model, which combined day 1 NSE, day 1 S100β, cardiac arrest time, SOFA scores, APACHE II scores, and age, was used to screen CA patients with poor neurological prognosis at early stages and had an AUC of 0.792. Serum concentrations of NSE and S100β were significantly elevated in CA patients and could be prognostic biomarkers to predict neurological outcomes. Day 1 NSE and S100β combined with multiple indicators could be a decent early warning model for poor neurological prognosis in patients with CA.

The ultimate goal of cardiopulmonary resuscitation (CPR) is restoring spontaneous circulation and minimizing neurologic deficits. Since two human studies presented improved neurologic outcome and reduced mortality after cardiac arrest in 2002,[1,2] therapeutic hypothermia (TH) has been recommended consistently in the international CPR guidelines for post-cardiac arrest care.[3-5] TH improves the neurologic outcome due to attenuation of the inflammatory response in the brain.[1] On the other hand, TH can have systemic adverse effect such as high infection rate. Geurts et al.[6] emphasized in a meta-analysis of 23 studies that TH was a risk factor of both pneumonia and sepsis after return of spontaneous circulation in cardiac arrest patients. Even if TH is not used, infection is more common in post-cardiac arrest care,[7] and pneumonia is the most common type of infection in out-of-hospital cardiac arrest (OHCA).[8] The cause of this high incidence of pneumonia in OHCA is that factors such as loss of airway protection, changed mental status, pulmonary contusion by chest compression, emergent airway access, and mechanical ventilation increase the risk of pulmonary infection.[7] Some studies have suggested that post-resuscitation pneumonia could be divided into early-onset and late-onset pneumonia according to onset time and prevalent pathogens; however, the onset time varied from three to seven days depending on the study.[8-11] Perbet et al.[11] reported that TH was an independent risk factor of early-onset pneumonia (EOP), which prolonged mechanical ventilation support and intensive care unit (ICU) stay with unchanged neurologic outcome and ICU mortality in a retrospective and large (641 patients) cohort study. Therefore, intensivists should be aware of management of EOP while conducting HT. Prophylactic antibiotics decreased the incidence of EOP in comatose patients with a variety of causes such as head trauma, intracranial hemorrhage, stroke, or cardiac arrest.[12] In a recently published study, prophylactic antibiotics reduced the incidence of pneumonia in cardiac arrest survivors undergoing TH, but they did not reduce patient mortality.[13] However, the researchers did not distinguish earlyor late-onset pneumonia. Kim et al.[14] have reported that prophylactic antibiotics in OHCA patients undergoing TH does not reduce the incidence of EOP. The authors insisted that the study evaluated the effect of prophylactic antibiotics on EOP for the first time. The incidence of EOP was 29.2% and 30.0% in prophylactic antibiotics and non-antibiotics, respec-

In a recent issue of World Journal of Surgery, Moriwaki et al. [1] present data regarding patients in cardiopulmonary arrest from blunt trauma. As discussed by the authors, resuscitation of a patient in cardiopulmonary arrest has significant cost [2, 3], and presents considerable risk to both prehospital providers and the public [4, 5]. While not specifically discussed by the authors, survival without functional neurological outcome incurs additional costs, with prolonged admissions to acute care and rehabilitation facilities and a lifetime of required supportive care. Therefore, prehospital termination of resuscitation (TOR) protocols have the potential to contribute to public health through conservation of limited financial healthcare resources and prevention of further injuries by reducing the unnecessary use of warning lights and sirens. Moriwaki et al. [1] recommend 20 min of resuscitation prior to TOR. While we appreciate that in this study of 477 patients in cardiopulmonary arrest from blunt trauma there were 13 survivors with a 2.7% survival rate, in analysis of the data presented in the authors’ Table 1, it is evident that only 3 of these patients survived in a non-vegetative state, for a survival rate with good neurological outcome of 0.6%. Additionally, and perhaps more importantly, from the authors’ Fig. 8, it is clear that a cluster of 12 survivors achieved return of spontaneous circulation (ROSC) in under 10 min, and a single isolated survivor achieved ROSC after 17 min. No information is provided for this outlier. The initial rhythm, vital signs, and mechanism of injury are not described. Furthermore, it is unclear if this outlier was one of the three patients listed in the authors’ Table 1 with a good neurological outcome. In the development of a healthcare system, it is crucial to make decisions based on population-based data and not on individual outcomes. Moriwaki et al. [1] recommend TOR after 20 min of resuscitation based on the survival of one patient who had 17 min of resuscitation before achieving ROSC. However, their data plainly demonstrate that 10 min of resuscitation will capture the population of patients in cardiopulmonary arrest from blunt trauma. Therefore, in analysis of the Moriwaki et al. [1] data, we recommend a population-based resuscitation time of 10 min prior to TOR.

We sought to determine the outcome of suicidal hanging and the impact of targeted temperature management (TTM) on hanging-induced cardiac arrest (CA) through an Eastern Association for the Surgery of Trauma (EAST) multicenter retrospective study. We analyzed hanging patient data and TTM variables from January 1992 to December 2015. Cerebral performance category score of 1 or 2 was considered good neurologic outcome, while cerebral performance category score of 3 or 4 was considered poor outcome. Classification and Regression Trees recursive partitioning was used to develop multivariate predictive models for survival and neurologic outcome. A total of 692 hanging patients from 17 centers were analyzed for this study. Their overall survival rate was 77%, and the CA survival rate was 28.6%. The CA patients had significantly higher severity of illness and worse outcome than the non-CA patients. Of the 175 CA patients who survived to hospital admission, 81 patients (46.3%) received post-CA TTM. The unadjusted survival of TTM CA patients (24.7% vs 39.4%, p < 0.05) and good neurologic outcome (19.8% vs 37.2%, p < 0.05) were worse than non-TTM CA patients. However, when subgroup analyses were performed between those with an admission Glasgow Coma Scale score of 3 to 8, the differences between TTM and non-TTM CA survival (23.8% vs 30.0%, p = 0.37) and good neurologic outcome (18.8% vs 28.7%, p = 0.14) were not significant. Targeted temperature management implementation and post-CA management varied between the participating centers. Classification and Regression Trees models identified variables predictive of favorable and poor outcome for hanging and TTM patients with excellent accuracy. Cardiac arrest hanging patients had worse outcome than non-CA patients. Targeted temperature management CA patients had worse unadjusted survival and neurologic outcome than non-TTM patients. These findings may be explained by their higher severity of illness, variable TTM implementation, and differences in post-CA management. Future prospective studies are necessary to ascertain the effect of TTM on hanging outcome and to validate our Classification and Regression Trees models. Therapeutic study, level IV; prognostic study, level III.