Outcome Prediction in Postanoxic Coma With Deep Learning*:

Focality in childhood absence epilepsy

BACKGROUND AND PURPOSE

Childhood absence epilepsy (CAE) has a typical electroencephalography (EEG) pattern of generalized 3 Hz spike and wave discharges (SWD). Focal interictal discharges were also documented in a small number of documents. The aim was to investigate the amplitudes of interictal 3 Hz SWD within the 1st second in drug-naïve CAE patients. In this way, areas with maximal electronegativity at the beginning of clinically generalized discharges will be documented.

METHODS

The EEG records of children with drug-naïve CAE were evaluated retrospectively by two child neurologists first for 3 Hz SWD. Then, a machine-learning model evaluated the amplitudes of 3 Hz in the 1st second of SWD. Maximum electronegativity areas were documented and classified as focal, bilateral, and generalized.

RESULTS

One hundred and twelve 3 Hz SWD were evaluated in 11 patients. Among discharges within the 1st second, maximum electronegativity areas were documented as focal for 44 (39.2%), bilateral for 8 (7.1%), generalized for 60 (53.5%). Among focal electronegativity areas, mostly right central, left occipital and midline parietal areas were documented in 12 (10.7%), 7 (6.2%), and 6 (5.3%), respectively. Eight (7.1%) of the maximum electronegativity areas were detected bilaterally, of which 7 (6.2%) originated from the frontopolar areas.

CONCLUSIONS

Focal maximal electronegativity areas were frequently observed in drug-naïve CAE patients, comprising approximately half of non-generalized discharges. Focal discharges might be misleading in diagnosis. Focal areas within the brain may be responsible for and contribute to absence seizures that appear bilaterally symmetrical and generalized clinically. Although its clinical implication is unknown, this warrants further study.

Artificial Intelligence and Machine Learning Applications in Critically Ill Brain Injured Patients

The utilization of Artificial Intelligence (AI) and Machine Learning (ML) is paving the way for significant strides in patient diagnosis, treatment, and prognostication in neurocritical care. These technologies offer the potential to unravel complex patterns within vast datasets ranging from vast clinical data and EEG (electroencephalogram) readings to advanced cerebral imaging facilitating a more nuanced understanding of patient conditions. Despite their promise, the implementation of AI and ML faces substantial hurdles. Historical biases within training data, the challenge of interpreting multifaceted data streams, and the "black box" nature of ML algorithms present barriers to widespread clinical adoption. Moreover, ethical considerations around data privacy and the need for transparent, explainable models remain paramount to ensure trust and efficacy in clinical decision-making.This article reflects on the emergence of AI and ML as integral tools in neurocritical care, discussing their roles from the perspective of both their scientific promise and the associated challenges. We underscore the importance of extensive validation in diverse clinical settings to ensure the generalizability of ML models, particularly considering their potential to inform critical medical decisions such as withdrawal of life-sustaining therapies. Advancement in computational capabilities is essential for implementing ML in clinical settings, allowing for real-time analysis and decision support at the point of care. As AI and ML are poised to become commonplace in clinical practice, it is incumbent upon health care professionals to understand and oversee these technologies, ensuring they adhere to the highest safety standards and contribute to the realization of personalized medicine. This engagement will be pivotal in integrating AI and ML into patient care, optimizing outcomes in neurocritical care through informed and data-driven decision-making.

Resting state EEG relates to short- and long-term cognitive functioning after cardiac arrest

BACKGROUND

Approximately half of cardiac arrest survivors have persistent cognitive impairment. Guidelines recommend early screening to identify patients at risk for cognitive impairment, but there is no consensus on the best screening method. We aimed to identify quantitative EEG measures relating with short- and long-term cognitive function after cardiac arrest for potential to cognitive outcome prediction.

METHODS

We analyzed data from a prospective longitudinal multicenter cohort study designed to develop a prediction model for cognitive outcome after cardiac arrest. For the current analysis, we used twenty-minute EEG registrations from 80 patients around one week after cardiac arrest. We calculated power spectral density, normalized alpha-to-theta ratio (nATR), peak frequency, and center of gravity (CoG) of this peak frequency. We related these with global cognitive functioning (scores on the Montreal Cognitive Assessment (MoCA)) at one week, three and twelve months follow-up with multivariate mixed effect models, and with performance on standard neuropsychological examination at twelve months using Pearson correlation coefficients.

RESULTS

Each individual EEG parameter related to MoCA at one week (βnATR = 7.36; P < 0.01; βpeak frequency = 1.73, P < 0.01; βCoG = -9.88, P < 0.01). The nATR also related with the MoCA at three months ((βnATR = 2.49; P 0.01). No EEG metrics significantly related to the MoCA score at twelve months. nATR and peak frequency related with memory performance at twelve months. Results were consistent in sensitivity analyses.

CONCLUSION

Early resting-state EEG parameters relate with short-term global cognitive functioning and with memory function at one year after cardiac arrest. Additional predictive values in multimodal prediction models need further study.

Electroencephalogram-based machine learning models to predict neurologic outcome after cardiac arrest: A systematic review

AIM OF THE REVIEW

The primary aim of this systematic review was to investigate the most common electroencephalogram (EEG)-based machine learning (ML) model with the highest Area Under Receiver Operating Characteristic Curve (AUC) in two ML categories, conventional ML and Deep Neural Network (DNN), to predict the neurologic outcomes after cardiac arrest; the secondary aim was to investigate common EEG features applied to ML models.

METHODS

Systematic search of medical literature from PubMed and engineering literature from Compendex up to June 2, 2023. One reviewer screened studies that used EEG-based ML models to predict the neurologic outcomes after cardiac arrest. Four reviewers validated that the studies met selection criteria. Nine variables were manually extracted. The top-five common EEG features were calculated. We evaluated each study's risk of bias using the Quality in Prognosis Studies guideline.

RESULTS

Out of 351 identified studies, 17 studies met the inclusion criteria. Random Forest (RF) (n = 7) was the most common ML model in the conventional ML category (n = 11), followed by Convolutional Neural Network (CNN) (n = 4) in the DNN category (n = 6). The AUCs for RF ranged between 0.8 and 0.97, while CNN had AUCs between 0.7 and 0.92. The top-three commonly used EEG features were band power (n = 12), Shannon's Entropy (n = 11), burst-suppression ratio (n = 9).

CONCLUSIONS

RF and CNN were the two most common ML models with the highest AUCs for predicting the neurologic outcomes after cardiac arrest. Using a multimodal model that combines EEG features and electronic health record data may further improve prognostic performance.

The International Cardiac Arrest Research Consortium Electroencephalography Database

OBJECTIVES

To develop the International Cardiac Arrest Research (I-CARE), a harmonized multicenter clinical and electroencephalography database for acute hypoxic-ischemic brain injury research involving patients with cardiac arrest.

DESIGN

Multicenter cohort, partly prospective and partly retrospective.

SETTING

Seven academic or teaching hospitals from the United States and Europe.

PATIENTS

Individuals 16 years old or older who were comatose after return of spontaneous circulation following a cardiac arrest who had continuous electroencephalography monitoring were included.

INTERVENTIONS

Not applicable.

MEASUREMENTS AND MAIN RESULTS

Clinical and electroencephalography data were harmonized and stored in a common Waveform Database-compatible format. Automated spike frequency, background continuity, and artifact detection on electroencephalography were calculated with 10-second resolution and summarized hourly. Neurologic outcome was determined at 3-6 months using the best Cerebral Performance Category (CPC) scale. This database includes clinical data and 56,676 hours (3.9 terabytes) of continuous electroencephalography data for 1,020 patients. Most patients died ( n = 603, 59%), 48 (5%) had severe neurologic disability (CPC 3 or 4), and 369 (36%) had good functional recovery (CPC 1-2). There is significant variability in mean electroencephalography recording duration depending on the neurologic outcome (range, 53-102 hr for CPC 1 and CPC 4, respectively). Epileptiform activity averaging 1 Hz or more in frequency for at least 1 hour was seen in 258 patients (25%) (19% for CPC 1-2 and 29% for CPC 3-5). Burst suppression was observed for at least 1 hour in 207 (56%) and 635 (97%) patients with CPC 1-2 and CPC 3-5, respectively.

CONCLUSIONS

The I-CARE consortium electroencephalography database provides a comprehensive real-world clinical and electroencephalography dataset for neurophysiology research of comatose patients after cardiac arrest. This dataset covers the spectrum of abnormal electroencephalography patterns after cardiac arrest, including epileptiform patterns and those in the ictal-interictal continuum.

The International Cardiac Arrest Research (I-CARE) Consortium Electroencephalography Database

Objective

To develop a harmonized multicenter clinical and electroencephalography (EEG) database for acute hypoxic-ischemic brain injury research involving patients with cardiac arrest.

Design

Multicenter cohort, partly prospective and partly retrospective.

Setting

Seven academic or teaching hospitals from the U.S. and Europe.

Patients

Individuals aged 16 or older who were comatose after return of spontaneous circulation following a cardiac arrest who had continuous EEG monitoring were included.

Interventions

not applicable.

Measurements and Main Results

Clinical and EEG data were harmonized and stored in a common Waveform Database (WFDB)-compatible format. Automated spike frequency, background continuity, and artifact detection on EEG were calculated with 10 second resolution and summarized hourly. Neurological outcome was determined at 3-6 months using the best Cerebral Performance Category (CPC) scale. This database includes clinical and 56,676 hours (3.9 TB) of continuous EEG data for 1,020 patients. Most patients died (N=603, 59%), 48 (5%) had severe neurological disability (CPC 3 or 4), and 369 (36%) had good functional recovery (CPC 1-2). There is significant variability in mean EEG recording duration depending on the neurological outcome (range 53-102h for CPC 1 and CPC 4, respectively). Epileptiform activity averaging 1 Hz or more in frequency for at least one hour was seen in 258 (25%) patients (19% for CPC 1-2 and 29% for CPC 3-5). Burst suppression was observed for at least one hour in 207 (56%) and 635 (97%) patients with CPC 1-2 and CPC 3-5, respectively.

Conclusions

The International Cardiac Arrest Research (I-CARE) consortium database provides a comprehensive real-world clinical and EEG dataset for neurophysiology research of comatose patients after cardiac arrest. This dataset covers the spectrum of abnormal EEG patterns after cardiac arrest, including epileptiform patterns and those in the ictal-interictal continuum.

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Outcome prognostication in comatose patients after cardiac arrest (CA) remains to date a challenge. The major determinant of clinical outcome is the post-hypoxic/ischemic encephalopathy. Electroencephalography (EEG) is routinely used to assess neural functions in comatose patients. Currently, EEG-based outcome prognosis relies on visual evaluation by medical experts, which is time consuming, prone to subjectivity, and oblivious to complex patterns. The field of deep learning has given rise to powerful algorithms for detecting patterns in large amounts of data. Analyzing EEG signals of coma patients with deep neural networks with the goal of assisting in outcome prognosis is therefore a natural application of these algorithms. Here, we provide the first narrative literature review on the use of deep learning for prognostication after CA. Existing studies show overall high performance in predicting outcome, relying either on spontaneous or on auditory evoked EEG signals. Moreover, the literature is concerned with algorithmic interpretability, and has shown that largely, deep neural networks base their decisions on clinically or neurophysiologically meaningful features. We conclude this review by discussing considerations that the fields of artificial intelligence and neurology will need to jointly address in the future, in order for deep learning algorithms to break the publication barrier, and to be integrated in clinical practice.

Guidelines for Neuroprognostication in Comatose Adult Survivors of Cardiac Arrest

BACKGROUND

Among cardiac arrest survivors, about half remain comatose 72 h following return of spontaneous circulation (ROSC). Prognostication of poor neurological outcome in this population may result in withdrawal of life-sustaining therapy and death. The objective of this article is to provide recommendations on the reliability of select clinical predictors that serve as the basis of neuroprognostication and provide guidance to clinicians counseling surrogates of comatose cardiac arrest survivors.

METHODS

A narrative systematic review was completed using Grading of Recommendations Assessment, Development and Evaluation (GRADE) methodology. Candidate predictors, which included clinical variables and prediction models, were selected based on clinical relevance and the presence of an appropriate body of evidence. The Population, Intervention, Comparator, Outcome, Timing, Setting (PICOTS) question was framed as follows: "When counseling surrogates of comatose adult survivors of cardiac arrest, should [predictor, with time of assessment if appropriate] be considered a reliable predictor of poor functional outcome assessed at 3 months or later?" Additional full-text screening criteria were used to exclude small and lower-quality studies. Following construction of the evidence profile and summary of findings, recommendations were based on four GRADE criteria: quality of evidence, balance of desirable and undesirable consequences, values and preferences, and resource use. In addition, good practice recommendations addressed essential principles of neuroprognostication that could not be framed in PICOTS format.

RESULTS

Eleven candidate clinical variables and three prediction models were selected based on clinical relevance and the presence of an appropriate body of literature. A total of 72 articles met our eligibility criteria to guide recommendations. Good practice recommendations include waiting 72 h following ROSC/rewarming prior to neuroprognostication, avoiding sedation or other confounders, the use of multimodal assessment, and an extended period of observation for awakening in patients with an indeterminate prognosis, if consistent with goals of care. The bilateral absence of pupillary light response > 72 h from ROSC and the bilateral absence of N20 response on somatosensory evoked potential testing were identified as reliable predictors. Computed tomography or magnetic resonance imaging of the brain > 48 h from ROSC and electroencephalography > 72 h from ROSC were identified as moderately reliable predictors.

CONCLUSIONS

These guidelines provide recommendations on the reliability of predictors of poor outcome in the context of counseling surrogates of comatose survivors of cardiac arrest and suggest broad principles of neuroprognostication. Few predictors were considered reliable or moderately reliable based on the available body of evidence.

Time to Awakening and Self-Fulfilling Prophecies After Cardiac Arrest

OBJECTIVES

Withdrawal of life-sustaining therapies for perceived poor neurologic prognosis (WLST-N) is common after resuscitation from cardiac arrest and may bias outcome estimates from models trained using observational data. We compared several approaches to outcome prediction with the goal of identifying strategies to quantify and reduce this bias.

DESIGN

Retrospective observational cohort study.

SETTING

Two academic medical centers ("UPMC" and "University of Alabama Birmingham" [UAB]).

PATIENTS

Comatose adults resuscitated from cardiac arrest.

INTERVENTION

None.

MEASUREMENTS AND MAIN RESULTS

As potential predictors, we considered clinical, laboratory, imaging, and quantitative electroencephalography data available early after hospital arrival. We followed patients until death, discharge, or awakening from coma. We used penalized Cox regression with a least absolute shrinkage and selection operator penalty and five-fold cross-validation to predict time to awakening in UPMC patients and then externally validated the model in UAB patients. This model censored patients after WLST-N, considering subsequent potential for awakening to be unknown. Next, we developed a penalized logistic model predicting awakening, which treated failure to awaken after WLST-N as a true observed outcome, and a separate logistic model predicting WLST-N. We scaled and centered individual patients' Cox and logistic predictions for awakening to allow direct comparison and then explored the difference in predictions across probabilities of WLST-N. Overall, 1,254 patients were included, and 29% awakened. Cox models performed well (mean area under the curve was 0.93 in the UPMC test sets and 0.83 in external validation). Logistic predictions of awakening were systematically more pessimistic than Cox-based predictions for patients at higher risk of WLST-N, suggesting potential for self-fulfilling prophecies to arise when failure to awaken after WLST-N is considered as the ground truth outcome.

CONCLUSIONS

Compared with traditional binary outcome prediction, censoring outcomes after WLST-N may reduce potential for bias and self-fulfilling prophecies.

Artificial Intelligence in Resuscitation: A Scoping Review

Introduction: Cardiac arrest is a significant cause of premature mortality and severe disability. Despite the death rate steadily decreasing over the previous decade, only 22% of survivors achieve good clinical status and only 25% of patients survive until their discharge from the hospital. The objective of this scoping review was to review relevant AI modalities and the main potential applications of AI in resuscitation. Methods: We conducted the literature search for related studies in PubMed, EMBASE, and Google Scholar. We included peer-reviewed publications and articles in the press, pooling and characterizing the data by their model types, goals, and benefits. Results: After identifying 268 original studies, we chose 59 original studies (reporting 1,817,419 patients) to include in the qualitative synthesis. AI-based methods appear to be superior to traditional methods in achieving high-level performance. Conclusion: AI might be useful in predicting cardiac arrest, heart rhythm disorders, and post-cardiac arrest outcomes, as well as in the delivery of drone-delivered defibrillators and notification of dispatchers. AI-powered technologies could be valuable assistants to continuously track patient conditions. Healthcare professionals should assist in the research and development of AI-powered technologies as well as their implementation into clinical practice.

Auditory stimulation and deep learning predict awakening from coma after cardiac arrest

Assessing the integrity of neural functions in coma after cardiac arrest remains an open challenge. Prognostication of coma outcome relies mainly on visual expert scoring of physiological signals, which is prone to subjectivity and leaves a considerable number of patients in a 'grey zone', with uncertain prognosis. Quantitative analysis of EEG responses to auditory stimuli can provide a window into neural functions in coma and information about patients' chances of awakening. However, responses to standardized auditory stimulation are far from being used in a clinical routine due to heterogeneous and cumbersome protocols. Here, we hypothesize that convolutional neural networks can assist in extracting interpretable patterns of EEG responses to auditory stimuli during the first day of coma that are predictive of patients' chances of awakening and survival at 3 months. We used convolutional neural networks (CNNs) to model single-trial EEG responses to auditory stimuli in the first day of coma, under standardized sedation and targeted temperature management, in a multicentre and multiprotocol patient cohort and predict outcome at 3 months. The use of CNNs resulted in a positive predictive power for predicting awakening of 0.83 ± 0.04 and 0.81 ± 0.06 and an area under the curve in predicting outcome of 0.69 ± 0.05 and 0.70 ± 0.05, for patients undergoing therapeutic hypothermia and normothermia, respectively. These results also persisted in a subset of patients that were in a clinical 'grey zone'. The network's confidence in predicting outcome was based on interpretable features: it strongly correlated to the neural synchrony and complexity of EEG responses and was modulated by independent clinical evaluations, such as the EEG reactivity, background burst-suppression or motor responses. Our results highlight the strong potential of interpretable deep learning algorithms in combination with auditory stimulation to improve prognostication of coma outcome.

Can we learn from hidden mistakes? Self-fulfilling prophecy and responsible neuroprognostic innovation

A self-fulfilling prophecy (SFP) in neuroprognostication occurs when a patient in coma is predicted to have a poor outcome, and life-sustaining treatment is withdrawn on the basis of that prediction, thus directly bringing about a poor outcome (viz. death) for that patient. In contrast to the predominant emphasis in the bioethics literature, we look beyond the moral issues raised by the possibility that an erroneous prediction might lead to the death of a patient who otherwise would have lived. Instead, we focus on the problematic epistemic consequences of neuroprognostic SFPs in settings where research and practice intersect. When this sort of SFP occurs, the problem is that physicians and researchers are never in a position to notice whether their original prognosis was correct or incorrect, since the patient dies anyway. Thus, SFPs keep us from discerning false positives from true positives, inhibiting proper assessment of novel prognostic tests. This epistemic problem of SFPs thus impedes learning, but ethical obligations of patient care make it difficult to avoid SFPs. We then show how the impediment to catching false positive indicators of poor outcome distorts research on novel techniques for neuroprognostication, allowing biases to persist in prognostic tests. We finally highlight a particular risk that a precautionary bias towards early withdrawal of life-sustaining treatment may be amplified. We conclude with guidelines about how researchers can mitigate the epistemic problems of SFPs, to achieve more responsible innovation of neuroprognostication for patients in coma.

Diagnostic and prognostic EEG analysis of critically ill patients: A deep learning study

Visual interpretation of electroencephalography (EEG) is time consuming, may lack objectivity, and is restricted to features detectable by a human. Computer-based approaches, especially deep learning, could potentially overcome these limitations. However, most deep learning studies focus on a specific question or a single pathology. Here we explore the potential of deep learning for EEG-based diagnostic and prognostic assessment of patients with acute consciousness impairment (ACI) of various etiologies. EEGs from 358 adults from a randomized controlled trial (CERTA, NCT03129438) were retrospectively analyzed. A convolutional neural network was used to predict the clinical outcome (based either on survival or on best cerebral performance category) and to determine the etiology (four diagnostic categories). The largest probability output served as marker for the confidence of the network in its prediction ("certainty factor"); we also systematically compared the predictions with raw EEG data, and used a visualization algorithm (Grad-CAM) to highlight discriminative patterns. When all patients were considered, the area under the receiver operating characteristic curve (AUC) was 0.721 for predicting survival and 0.703 for predicting the outcome based on best CPC; for patients with certainty factor ≥ 60 % the AUCs increased to 0.776 and 0.755 respectively; and for certainty factor ≥ 75 % to 0.852 and 0.879. The accuracy for predicting the etiology was 54.5 %; the accuracy increased to 67.7 %, 70.3 % and 84.1 % for patients with certainty factor of 50 %, 60 % and 75 % respectively. Visual analysis showed that the network learnt EEG patterns typically recognized by human experts, and suggested new criteria. This work demonstrates for the first time the potential of deep learning-based EEG analysis in critically ill patients with various etiologies of ACI. Certainty factor and post-hoc correlation of input data with prediction help to better characterize the method and pave the route for future implementations in clinical routine.

Machine learning and clinical neurophysiology

Clinical neurophysiology constructs a wealth of dynamic information pertaining to the integrity and function of both central and peripheral nervous systems. As with many technological fields, there has been an explosion of data in neurophysiology over recent years, and this requires considerable analysis by experts. Computational algorithms and especially advances in machine learning (ML) have the ability to assist with this task and potentially reveal hidden insights. In this update article, we will provide a brief overview where such technology is being applied in clinical neurophysiology and possible future directions.

Outcome Prediction of Postanoxic Coma: A Comparison of Automated Electroencephalography Analysis Methods

Background

To compare three computer-assisted quantitative electroencephalography (EEG) prediction models for the outcome prediction of comatose patients after cardiac arrest regarding predictive performance and robustness to artifacts.

Methods

A total of 871 continuous EEGs recorded up to 3 days after cardiac arrest in intensive care units of five teaching hospitals in the Netherlands were retrospectively analyzed. Outcome at 6 months was dichotomized as “good” (Cerebral Performance Category 1–2) or “poor” (Cerebral Performance Category 3–5). Three prediction models were implemented: a logistic regression model using two quantitative features, a random forest model with nine features, and a deep learning model based on a convolutional neural network. Data from two centers were used for training and fivefold cross-validation (n = 663), and data from three other centers were used for external validation (n = 208). Model output was the probability of good outcome. Predictive performances were evaluated by using receiver operating characteristic analysis and the calculation of predictive values. Robustness to artifacts was evaluated by using an artifact rejection algorithm, manually added noise, and randomly flattened channels in the EEG.

Results

The deep learning network showed the best overall predictive performance. On the external test set, poor outcome could be predicted by the deep learning network at 24 h with a sensitivity of 54% (95% confidence interval [CI] 44–64%) at a false positive rate (FPR) of 0% (95% CI 0–2%), significantly higher than the logistic regression (sensitivity 33%, FPR 0%) and random forest models (sensitivity 13%, FPR, 0%) (p < 0.05). Good outcome at 12 h could be predicted by the deep learning network with a sensitivity of 78% (95% CI 52–100%) at a FPR of 12% (95% CI 0–24%) and by the logistic regression model with a sensitivity of 83% (95% CI 83–83%) at a FPR of 3% (95% CI 3–3%), both significantly higher than the random forest model (sensitivity 1%, FPR 0%) (p < 0.05). The results of the deep learning network were the least affected by the presence of artifacts, added white noise, and flat EEG channels.

Conclusions

A deep learning model outperformed logistic regression and random forest models for reliable, robust, EEG-based outcome prediction of comatose patients after cardiac arrest.

Comparison of Machine Learning Methods for Predicting Outcomes After In-Hospital Cardiac Arrest

OBJECTIVES

Prognostication of neurologic status among survivors of in-hospital cardiac arrests remains a challenging task for physicians. Although models such as the Cardiac Arrest Survival Post-Resuscitation In-hospital score are useful for predicting neurologic outcomes, they were developed using traditional statistical techniques. In this study, we derive and compare the performance of several machine learning models with each other and with the Cardiac Arrest Survival Post-Resuscitation In-hospital score for predicting the likelihood of favorable neurologic outcomes among survivors of resuscitation.

DESIGN

Analysis of the Get With The Guidelines-Resuscitation registry.

SETTING

Seven-hundred fifty-five hospitals participating in Get With The Guidelines-Resuscitation from January 1, 2001, to January 28, 2017.

PATIENTS

Adult in-hospital cardiac arrest survivors.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Of 117,674 patients in our cohort, 28,409 (24%) had a favorable neurologic outcome, as defined as survival with a Cerebral Performance Category score of less than or equal to 2 at discharge. Using patient characteristics, pre-existing conditions, prearrest interventions, and periarrest variables, we constructed logistic regression, support vector machines, random forests, gradient boosted machines, and neural network machine learning models to predict favorable neurologic outcome. Events prior to October 20, 2009, were used for model derivation, and all subsequent events were used for validation. The gradient boosted machine predicted favorable neurologic status at discharge significantly better than the Cardiac Arrest Survival Post-Resuscitation In-hospital score (C-statistic: 0.81 vs 0.73; p < 0.001) and outperformed all other machine learning models in terms of discrimination, calibration, and accuracy measures. Variables that were consistently most important for prediction across all models were duration of arrest, initial cardiac arrest rhythm, admission Cerebral Performance Category score, and age.

CONCLUSIONS

The gradient boosted machine algorithm was the most accurate for predicting favorable neurologic outcomes in in-hospital cardiac arrest survivors. Our results highlight the utility of machine learning for predicting neurologic outcomes in resuscitated patients.

Predicting Neurological Outcome from Electroencephalogram Dynamics in Comatose Patients after Cardiac Arrest with Deep Learning

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.

Artificial intelligence, machine learning, and deep learning for clinical outcome prediction

AI is a broad concept, grouping initiatives that use a computer to perform tasks that would usually require a human to complete. AI methods are well suited to predict clinical outcomes. In practice, AI methods can be thought of as functions that learn the outcomes accompanying standardized input data to produce accurate outcome predictions when trialed with new data. Current methods for cleaning, creating, accessing, extracting, augmenting, and representing data for training AI clinical prediction models are well defined. The use of AI to predict clinical outcomes is a dynamic and rapidly evolving arena, with new methods and applications emerging. Extraction or accession of electronic health care records and combining these with patient genetic data is an area of present attention, with tremendous potential for future growth. Machine learning approaches, including decision tree methods of Random Forest and XGBoost, and deep learning techniques including deep multi-layer and recurrent neural networks, afford unique capabilities to accurately create predictions from high dimensional, multimodal data. Furthermore, AI methods are increasing our ability to accurately predict clinical outcomes that previously were difficult to model, including time-dependent and multi-class outcomes. Barriers to robust AI-based clinical outcome model deployment include changing AI product development interfaces, the specificity of regulation requirements, and limitations in ensuring model interpretability, generalizability, and adaptability over time.

Predicting neurological outcome in comatose patients after cardiac arrest with multiscale deep neural networks

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.

Hierarchical Cluster Analysis Identifies Distinct Physiological States After Acute Brain Injury

BACKGROUND

Analysis of intracranial multimodality monitoring data is challenging, and quantitative methods may help identify unique physiological signatures that inform therapeutic strategies and outcome prediction. The aim of this study was to test the hypothesis that data-driven approaches can identify distinct physiological states from intracranial multimodality monitoring data.

METHODS

This was a single-center retrospective observational study of patients with either severe traumatic brain injury or high-grade subarachnoid hemorrhage who underwent invasive multimodality neuromonitoring. We used hierarchical cluster analysis to group hourly values for heart rate, mean arterial pressure, intracranial pressure, brain tissue oxygen, and cerebral microdialysis across all included patients into distinct groups. Average values for measured physiological variables were compared across the identified clusters, and physiological profiles from identified clusters were mapped onto physiological states known to occur after acute brain injury. The distribution of clusters was compared between patients with favorable outcome (discharged to home or acute rehab) and unfavorable outcome (in-hospital death or discharged to chronic nursing facility).

RESULTS

A total of 1704 observations from 20 patients were included. Even though the difference in mean values for measured variables between patients with favorable and unfavorable outcome were small, we identified four distinct clusters within our data: (1) events with low brain tissue oxygen and high lactate-to-pyruvate ratio-values (consistent with cerebral ischemia), (2) events with higher intracranial pressure values without evidence for ischemia (3) events which appeared to be physiologically "normal," and (4) events with high cerebral lactate without brain hypoxia (consistent with cerebral hyperglycolysis). Patients with a favorable outcome had a greater proportion of cluster 3 (normal) events, whereas patients with an unfavorable outcome had a greater proportion of cluster 1 (ischemia) and cluster 4 (hyperglycolysis) events (p < 0.0001, Fisher-Freeman-Halton test).

CONCLUSIONS

A data-driven approach can identify distinct groupings from invasive multimodality neuromonitoring data that may have implications for therapeutic strategies and outcome predictions. These groupings could be used as classifiers to train machine learning models that can aid in the treatment of patients with acute brain injury. Further work is needed to replicate the findings of this exploratory study in larger data sets.

Chasing Certainty After Cardiac Arrest: Can a Technological Innovation Solve a Moral Dilemma?

When information on a coma patient’s expected outcome is uncertain, a moral dilemma arises in clinical practice: if life-sustaining treatment is continued, the patient may survive with unacceptably poor neurological prospects, but if withdrawn a patient who could have recovered may die. Continuous electroencephalogram-monitoring (cEEG) is expected to substantially improve neuroprognostication for patients in coma after cardiac arrest. This raises expectations that decisions whether or not to withdraw will become easier. This paper investigates that expectation, exploring cEEG’s impacts when it becomes part of a socio-technical network in an Intensive Care Unit (ICU). Based on observations in two ICUs in the Netherlands and one in the USA that had cEEG implemented for research, we interviewed 25 family members, healthcare professionals, and surviving patients. The analysis focuses on (a) the way patient outcomes are constructed, (b) the kind of decision support these outcomes provide, and (c) how cEEG affects communication between professionals and relatives. We argue that cEEG can take away or decrease the intensity of the dilemma in some cases, while increasing uncertainty for others. It also raises new concerns. Since its actual impacts furthermore hinge on how cEEG is designed and implemented, we end with recommendations for ensuring responsible development and implementation.

Quantitative analysis of EEG reactivity for neurological prognostication after cardiac arrest

OBJECTIVE

To test whether 1) quantitative analysis of EEG reactivity (EEG-R) using machine learning (ML) is superior to visual analysis, and 2) combining quantitative analyses of EEG-R and EEG background pattern increases prognostic value for prediction of poor outcome after cardiac arrest (CA).

METHODS

Several types of ML models were trained with twelve quantitative features derived from EEG-R and EEG background data of 134 adult CA patients. Poor outcome was a Cerebral Performance Category score of 3-5 within 6 months.

RESULTS

The Random Forest (RF) trained on EEG-R showed the highest AUC of 83% (95-CI 80-86) of tested ML classifiers, predicting poor outcome with 46% sensitivity (95%-CI 40-51) and 89% specificity (95%-CI 86-92). Visual analysis of EEG-R had 80% sensitivity and 65% specificity. The RF was also the best classifier for EEG background (AUC 85%, 95%-CI 83-88) at 24 h after CA, with 62% sensitivity (95%-CI 57-67) and 84% specificity (95%-CI 79-88). Combining EEG-R and EEG background RF classifiers reduced the number of false positives.

CONCLUSIONS

Quantitative EEG-R using ML predicts poor outcome with higher specificity, but lower sensitivity compared to visual analysis of EEG-R, and is of some additional value to ML on EEG background data.

SIGNIFICANCE

Quantitative EEG-R using ML is a promising alternative to visual analysis and of some added value to ML on EEG background data.

Artificial Intelligence shaping the future of neurology practice

Neurology practice has faced many challenges since Jean-Martin Charcot established its sacred tenets. Artificial Intelligence (AI) promises to revolutionize the time-tested neurology practice in unimaginable ways. AI can now diagnose stroke from CT/MRI scans, detect papilledema and diabetic retinopathy from retinal scans, interpret electroencephalogram (EEG) to prognosticate coma, detect seizure well before ictus, predict conversion of mild cognitive impairment to Alzheimer's dementia, classify neurodegenerative diseases based on gait and handwriting. Clinical practice would likely change in near future to accommodate AI as a complementary tool. The clinician should be prepared to change the perception of AI from nemesis to opportunity.

Artificial intelligence in critical care: Its about time!

Currently, most critical care information is not expressed automatically at a granular level, rather is continually assessed by overindulged Intensive Care Unit (ICU) staff. Furthermore, due to different confounding morbidities and the uniqueness of the ICU setting, it is difficult to protocolize treatment regimens in the ICU. In highly complex ICU setting where man and resource management becomes extremely challenging, definite advancements are required to implement Artificial Intelligence (AI) for prognosticating the course of the disease to aid in informed decision-making. AI is the intelligence of a computer or computer-supervised robot to execute a piece of work commonly associated with intelligent beings, wherein the machines go beyond the realms of normal information processing by adding the characteristics of learning, sound reasoning, and weighting of the inputs. AI recognizes circuitous, relational time-series blueprint within datasets and this reasoning of analysis transcends conventional threshold-based analysis adapted in ICU protocols. AI works on the principle of a more complex form of Machine Learning by Artificial Neural Networks (ANN). These information-processing paradigms use multidimensional arrays called tensors which aid in 'learning' and 'weighting' all the information made available to it, thereby converting normal machine learning into Deep Learning. Here, the use of AI for data mining in complex ICU settings for protocol formulation and temporal representation and reasoning is discussed.

EEG functional connectivity contributes to outcome prediction of postanoxic coma

OBJECTIVE

To investigate the additional value of EEG functional connectivity features, in addition to non-coupling EEG features, for outcome prediction of comatose patients after cardiac arrest.

METHODS

Prospective, multicenter cohort study. Coherence, phase locking value, and mutual information were calculated in 19-channel EEGs at 12 h, 24 h and 48 h after cardiac arrest. Three sets of machine learning classification models were trained and validated with functional connectivity, EEG non-coupling features, and a combination of these. Neurological outcome was assessed at six months and categorized as "good" (Cerebral Performance Category [CPC] 1-2) or "poor" (CPC 3-5).

RESULTS

We included 594 patients (46% good outcome). A sensitivity of 51% (95% CI: 34-56%) at 100% specificity in predicting poor outcome was achieved by the best functional connectivity-based classifier at 12 h after cardiac arrest, while the best non-coupling-based model reached a sensitivity of 32% (0-54%) at 100% specificity using data at 12 h and 48 h. Combination of both sets of features achieved a sensitivity of 73% (50-77%) at 100% specificity.

CONCLUSION

Functional connectivity measures improve EEG based prediction models for poor outcome of postanoxic coma.

SIGNIFICANCE

Functional connectivity features derived from early EEG hold potential to improve outcome prediction of coma after cardiac arrest.

A matter of timing: EEG monitoring for neurological prognostication after cardiac arrest in the era of targeted temperature management

Neuromonitoring with electroencephalography (EEG) is an essential tool in neurological prognostication post-cardiac arrest. EEG allows reliable and real-time assessment of early changes in background patterns, development of seizures and epileptiform activity, as well as testing for background reactivity to stimuli despite use of sedation or targeted temperature management. Delayed emergence of consciousness post-cardiac arrest is common, therefore longitudinal monitoring of EEG allows the detection of trends indicative of neurological improvement before coma recovery can be observed clinically. In this review, we summarize essential recent literature in EEG monitoring for neurological prognostication post-cardiac arrest in the context of targeted temperature management, with a particular focus on the importance of the evolution of EEG patterns in the first few days following resuscitation.

Use of Machine Learning to Screen for Acute Respiratory Distress Syndrome Using Raw Ventilator Waveform Data

To develop and characterize a machine learning algorithm to discriminate acute respiratory distress syndrome from other causes of respiratory failure using only ventilator waveform data.

Design

Retrospective, observational cohort study.

Setting

Academic medical center ICU.

Patients

Adults admitted to the ICU requiring invasive mechanical ventilation, including 50 patients with acute respiratory distress syndrome and 50 patients with primary indications for mechanical ventilation other than hypoxemic respiratory failure.

Interventions

None.

Measurements and Main Results

Pressure and flow time series data from mechanical ventilation during the first 24-hours after meeting acute respiratory distress syndrome criteria (or first 24-hr of mechanical ventilation for non-acute respiratory distress syndrome patients) were processed to extract nine physiologic features. A random forest machine learning algorithm was trained to discriminate between the patients with and without acute respiratory distress syndrome. Model performance was assessed using the area under the receiver operating characteristic curve, sensitivity, specificity, positive predictive value, and negative predictive value. Analyses examined performance when the model was trained using data from the first 24 hours and tested using withheld data from either the first 24 hours (24/24 model) or 6 hours (24/6 model). Area under the receiver operating characteristic curve, sensitivity, specificity, positive predictive value, and negative predictive value were 0.88, 0.90, 0.71, 0.77, and 0.90 (24/24); and 0.89, 0.90, 0.75, 0.83, and 0.83 (24/6).

Conclusions

Use of machine learning and physiologic information derived from raw ventilator waveform data may enable acute respiratory distress syndrome screening at early time points after intubation. This approach, combined with traditional diagnostic criteria, could improve timely acute respiratory distress syndrome recognition and enable automated clinical decision support, especially in settings with limited availability of conventional diagnostic tests and electronic health records.

Big data, machine learning and artificial intelligence: a neurologist's guide

Modern clinical practice requires the integration and interpretation of ever-expanding volumes of clinical data. There is, therefore, an imperative to develop efficient ways to process and understand these large amounts of data. Neurologists work to understand the function of biological neural networks, but artificial neural networks and other forms of machine learning algorithm are likely to be increasingly encountered in clinical practice. As their use increases, clinicians will need to understand the basic principles and common types of algorithm. We aim to provide a coherent introduction to this jargon-heavy subject and equip neurologists with the tools to understand, critically appraise and apply insights from this burgeoning field.

Proceedings from the Neurotherapeutics Symposium on Neurological Emergencies: Shaping the Future of Neurocritical Care

Effective treatment options for patients with life-threatening neurological disorders are limited. To address this unmet need, high-impact translational research is essential for the advancement and development of novel therapeutic approaches in neurocritical care. "The Neurotherapeutics Symposium 2019-Neurological Emergencies" conference, held in Rochester, New York, in June 2019, was designed to accelerate translation of neurocritical care research via transdisciplinary team science and diversity enhancement. Diversity excellence in the neuroscience workforce brings innovative and creative perspectives, and team science broadens the scientific approach by incorporating views from multiple stakeholders. Both are essential components needed to address complex scientific questions. Under represented minorities and women were involved in the organization of the conference and accounted for 30-40% of speakers, moderators, and attendees. Participants represented a diverse group of stakeholders committed to translational research. Topics discussed at the conference included acute ischemic and hemorrhagic strokes, neurogenic respiratory dysregulation, seizures and status epilepticus, brain telemetry, neuroprognostication, disorders of consciousness, and multimodal monitoring. In these proceedings, we summarize the topics covered at the conference and suggest the groundwork for future high-yield research in neurologic emergencies.

Adaptive Sedation Monitoring From EEG in ICU Patients With Online Learning

Sedative medications are routinely administered to provide comfort and facilitate clinical care in critically ill ICU patients. Prior work shows that brain monitoring using electroencephalography (EEG) to track sedation levels may help medical personnel to optimize drug dosing and avoid the adverse effects of oversedation and undersedation. However, the performance of sedation monitoring methods proposed to date deal poorly with individual variability across patients, leading to inconsistent performance. To address this challenge we develop an online learning approach based on Adaptive Regularization of Weight Vectors (AROW). Our approach adaptively updates a sedation level prediction algorithm under a continuously evolving data distribution. The prediction model is gradually calibrated for individual patients in response to EEG observations and routine clinical assessments over time. The evaluations are performed on a population of 172 sedated ICU patients whose sedation levels were assessed using the Richmond Agitation-Sedation Scale (scores between -5 = comatose and 0 = awake). The proposed adaptive model achieves better performance than the same model without adaptation (average accuracies with tolerance of one level difference: 68.76% vs. 61.10%). Moreover, our approach is shown to be robust to sudden changes caused by label noise. Medication administrations have different effects on model performance. We find that the model performs best in patients receiving only propofol, compared to patients receiving no sedation or multiple simultaneous sedative medications.

Quantitative and Qualitative EEG as a Prediction Tool for Outcome and Complications in Acute Stroke Patients

Currently, the relevance of EEG measurements in acute stroke patients is considered low in clinical practice. However, recent studies on the predictive value of EEG measurements after stroke for various outcomes may increase the role of EEG in patients with stroke. We aimed to review the current literature on the utility of EEG measurements after stroke as a tool to predict outcome and complications, focusing on studies in which the EEG measurement was performed in the acute phase after the event and in which long-term outcome measures were reported. In our literature review, we identified 4 different outcome measures (functional outcome, mortality, development of post-stroke cognitive decline, and development of post-stroke epilepsy) where studies on the utility of acute EEG measurements exist. There is a large body of evidence for the prediction of functional outcome, in which a multitude of associated quantitative and qualitative EEG parameters are described. In contrast, only few studies focus on mortality as outcome parameter. We found studies of high methodical quality on the prediction of post-stroke cognitive decline, though the number of patients in these studies often was small. The role of EEG as a prediction tool for seizures and epilepsy after stroke could increase after a recently published study, especially if its result can be incorporated into already existing post-stroke epilepsy prediction tools. In summary, EEG is useful for the prediction of functional outcome, mortality, development of post-stroke cognitive decline and epilepsy, even though there is a discrepancy between the large amount of studies on EEG in acute stroke patients and its underuse in clinical practice.