A rule for early outcome classification of out-of-hospital cardiac arrest patients presenting with ventricular fibrillation

Pre-Arrival Care of the Out-of-Hospital Cardiac Arrest Victim

Lay rescuers play a pivotal role in the recognition and initial management of out-of-hospital cardiac arrest. The provision of timely pre-arrival care by lay responders, including cardiopulmonary resuscitation and the use of automated external defibrillator before emergency medical service arrival, is important link in the chain of survival and has been shown to improve outcomes from cardiac arrest. Although physicians are not directly involved in bystander response to cardiac arrest, they play a key role in emphasizing the importance of bystander interventions.

Predicting defibrillation success in out-of-hospital cardiac arrested patients: Moving beyond feature design

OBJECTIVE

Optimizing timing of defibrillation by evaluating the likelihood of a successful outcome could significantly enhance resuscitation. Previous studies employed conventional machine learning approaches and hand-crafted features to address this issue, but none have achieved superior performance to be widely accepted. This study proposes a novel approach in which predictive features are automatically learned.

METHODS

A raw 4s VF episode immediately prior to first defibrillation shock was feed to a 3-stage CNN feature extractor. Each stage was composed of 4 components: convolution, rectified linear unit activation, dropout and max-pooling. At the end of feature extractor, the feature map was flattened and connected to a fully connected multi-layer perceptron for classification. For model evaluation, a 10 fold cross-validation was employed. To balance classes, SMOTE oversampling method has been applied to minority class.

RESULTS

The obtained results show that the proposed model is highly accurate in predicting defibrillation outcome (Acc = 93.6 %). Since recommendations on classifiers suggest at least 50 % specificity and 95 % sensitivity as safe and useful predictors for defibrillation decision, the reported sensitivity of 98.8 % and specificity of 88.2 %, with the analysis speed of 3 ms/input signal, indicate that the proposed model possesses a good prospective to be implemented in automated external defibrillators.

CONCLUSIONS

The learned features demonstrate superiority over hand-crafted ones when performed on the same dataset. This approach benefits from being fully automatic by fusing feature extraction, selection and classification into a single learning model. It provides a superior strategy that can be used as a tool to guide treatment of OHCA patients in bringing optimal decision of precedence treatment. Furthermore, for encouraging replicability, the dataset has been made publicly available to the research community.

Clinical Predictive Models of Sudden Cardiac Arrest: A Survey of the Current Science and Analysis of Model Performances

Background

More than 500 000 sudden cardiac arrests (SCAs) occur annually in the United States. Clinical predictive models (CPMs) may be helpful tools to differentiate between patients who are likely to survive or have good neurologic recovery and those who are not. However, which CPMs are most reliable for discriminating between outcomes in SCA is not known.

Methods and Results

We performed a systematic review of the literature using the Tufts PACE (Predictive Analytics and Comparative Effectiveness) CPM Registry through February 1, 2020, and identified 81 unique CPMs of SCA and 62 subsequent external validation studies. Initial cardiac rhythm, age, and duration of cardiopulmonary resuscitation were the 3 most commonly used predictive variables. Only 33 of the 81 novel SCA CPMs (41%) were validated at least once. Of 81 novel SCA CPMs, 56 (69%) and 61 of 62 validation studies (98%) reported discrimination, with median c‐statistics of 0.84 and 0.81, respectively. Calibration was reported in only 29 of 62 validation studies (41.9%). For those novel models that both reported discrimination and were validated (26 models), the median percentage change in discrimination was −1.6%. We identified 3 CPMs that had undergone at least 3 external validation studies: the out‐of‐hospital cardiac arrest score (9 validations; median c‐statistic, 0.79), the cardiac arrest hospital prognosis score (6 validations; median c‐statistic, 0.83), and the good outcome following attempted resuscitation score (6 validations; median c‐statistic, 0.76).

Conclusions

Although only a small number of SCA CPMs have been rigorously validated, the ones that have been demonstrate good discrimination.

Combining multiple ECG features does not improve prediction of defibrillation outcome compared to single features in a large population of out-of-hospital cardiac arrests

Introduction

Quantitative electrocardiographic (ECG) waveform analysis provides a noninvasive reflection of the metabolic milieu of the myocardium during resuscitation and is a potentially useful tool to optimize the defibrillation strategy. However, whether combining multiple ECG features can improve the capability of defibrillation outcome prediction in comparison to single feature analysis is still uncertain.

Methods

A total of 3828 defibrillations from 1617 patients who experienced out-of-hospital cardiac arrest were analyzed. A 2.048-s ECG trace prior to each defibrillation without chest compressions was used for the analysis. Sixteen predictive features were optimized through the training dataset that included 2447 shocks from 1050 patients. Logistic regression, neural network and support vector machine were used to combine multiple features for the prediction of defibrillation outcome. Performance between single and combined predictive features were compared by area under receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and prediction accuracy (PA) on a validation dataset that consisted of 1381 shocks from 567 patients.

Results

Among the single features, mean slope (MS) outperformed other methods with an AUC of 0.876. Combination of complementary features using neural network resulted in the highest AUC of 0.874 among the multifeature-based methods. Compared to MS, no statistical difference was observed in AUC, sensitivity, specificity, PPV, NPV and PA when multiple features were considered.

Conclusions

In this large dataset, the amplitude-related features achieved better defibrillation outcome prediction capability than other features. Combinations of multiple electrical features did not further improve prediction performance.

Advanced life support therapy on out-of-hospital cardiac arrest patients: an engineering perspective

In the USA alone, several hundred thousand people die of sudden cardiac arrests each year. Basic life support, defined as chest compressions and ventilations, and early defibrillation are the only factors proven to increase the survival of patients with out-of-hospital cardiac arrest and are key elements in the chain of survival defined by the American Heart Association. The current cardiopulmonary resuscitation guidelines treat all patients the same but studies show a need for more individualization of treatment. This review focusses on ideas on how to strengthen the weak parts of the chain of survival including the ability to measure the effects of therapy, improve time efficiency and optimize the sequence and quality of the various components of cardiopulmonary resuscitation.

Quality management in resuscitation--towards a European cardiac arrest registry (EuReCa)

BACKGROUND

Knowledge about the epidemiology of cardiac arrest in Europe is inadequate.

AIM

To describe the first attempt to build up a Common European Registry of out-of-hospital cardiac arrest, called EuReCa.

METHODS

After approaching key persons in participating countries of the European Resuscitation Council, five countries or areas within countries (Belgium, Germany, Andalusia, North Holland, Sweden) agreed to participate. A standardized questionnaire including 28 items, that identified various aspects of resuscitation, was developed to explore the nature of the regional/national registries. This comprises inclusion criteria, data sources, and core data, as well as technical details of the structure of the databases.

RESULTS

The participating registers represent a population of 35 million inhabitants in Europe. During 2008, 12,446 cardiac arrests were recorded. The structure as well as the level of complexity varied markedly between the 5 regional/national registries. The incidence of attempted resuscitation ranged between registers from 17 to 53 per 100,000 inhabitants each year whilst the number of patients admitted to hospital alive ranged from 5 to 18 per 100,000 inhabitants each year. Bystander CPR varied 3-fold from 20% to 60%.

CONCLUSION

Five countries agreed to participate in an attempt to build up a common European Registry for out-of-hospital cardiac arrest. These regional/national registries show a marked difference in terms of structure and complexity. A marked variation was found between countries in the number of reported resuscitation attempts, the number of patients brought to hospital alive, and the proportion that received bystander CPR. At present, we are unable to explain the reason for the variability but our first findings could be a 'wake-up-call' for building up a high quality registry that could provide answers to this and other key questions in relation to the management of out-of-hospital cardiac arrest.

Electrical defibrillation: new technologies

Two-thirds of deaths from coronary disease occur in the pre-hospital phase and are caused by ventricular fibrillation or pulseless ventricular tachycardia, for which electrical defibrillation is the only effective treatment. The time delay between the onset of ventricular fibrillation and the administration of the first defibrillatory shock is the most important determinant for survival. To achieve the earliest defibrillation possible, rescuers others than physicians need to be able to initiate this treatment. The international scientific community strongly supports the concept of early defibrillation in the setting of a strong chain of survival. New technological developments of automated external defibrillators (AEDs) allowed the implementation of defibrillation by the first responding professional rescuer. As a consequence of the technological evolution in implantable defibrillators, much research has also been done on new defibrillation waveforms and alternative energy levels in external defibrillators. After initial animal research, human clinical investigation has shown that initial low energy (150J) nonprogressive (150J-150J-150J) impedance-adjusted biphasic waveform defibrillatory shocks for patients in out-of-hospital ventricular fibrillation are safe, acceptable and clinically effective. Reporting on outcome from cardiac arrest must be as uniform as possible to allow conclusions on performance of emergency medical service systems. The 'Utstein Style' nomenclature is a glossary of terms and a reporting guideline for uniform description of cardiac arrest, resuscitation, the emergency medical service (EMS) system and the outcome. Reports on experiences with AED programmes by traditional and non-traditional professional rescuers support the view that AEDs should not be implemented in EMS systems as an isolated intervention, but that efforts are equally needed to strengthen the other links of the chain of survival. The international scientific community (American Heart Association, International Liaison Committee on Resuscitation and European Resuscitation Council) have issued guidelines for the use of AEDs by EMS providers and first responders, and a universal treatment algorithm is proposed.

Practical issues in the evaluation of methods for the prediction of shock outcome success in out-of-hospital cardiac arrest patients

There is a need for robust, effective predictors of the outcome from shock for out-of-hospital cardiac arrest patients. Such technology would enable the emergency responder to provide a therapy tailored to the patient's needs. Here we report our most recent findings while dwelling intentionally on the rationale behind the decisions taken during system development. Specifically, we illustrate the need for sensible data selection, fully cross-validated results and the care necessary when evaluating system performance. We analyze 878 pre-shock ECG traces, all of at least 10 s duration from 110 patients with cardiac arrest of cardiac aetiology. The continuous wavelet transform was applied to preshock segments of ECG trace. Time-frequency markers are extracted from the transform and a linear threshold derived from a training set to provide high sensitivity prediction of successful defibrillation. These systems are then evaluated on a withheld test set. All experiments are cross-validated. When compared to popular Fourier-based techniques our wavelet transform method, COP (Cardioversion Outcome Predictor), provides a 10-20% improvement in performance with values of 66 +/- 4 specificity at 95 +/- 4 sensitivity, 61 +/- 4 specificity at 97 +/- 2 sensitivity and 56 +/- 1 specificity at 98 +/- 2 sensitivity achieved for datasets limited to 3, 6, and 9 shocks per patient, respectively. Thus, the assessment of the wavelet marker was associated with a high specificity value at or above 95% sensitivity in comparison to previously reported methods. Therefore, COP could provide an optimal index for the identification of patients for whom shocking would be futile, and for whom an alternative therapy could be considered.

Waveform analysis of ventricular fibrillation to predict defibrillation

PURPOSE OF REVIEW

Ventricular fibrillation occurs during many cases of cardiac arrest and is treated with rescue shocks. Coarse ventricular fibrillation occurs earlier after the onset of cardiac arrest and is more likely to be converted to an organized rhythm with pulses by rescue shocks. Less organized or fine ventricular fibrillation occurs later, has less power concentrated within narrow frequency bands and lower amplitude, and is less likely to be converted to an organized rhythm by rescue shocks. Quantitative analysis of the ventricular fibrillation waveform may distinguish coarse ventricular fibrillation from fine ventricular fibrillation, allowing more appropriate delivery of rescue shocks.

RECENT FINDINGS

A variety of studies in animals and humans indicate that there is underlying structure within the ventricular fibrillation waveform. Highly organized or coarse ventricular fibrillation is characterized by large power contributions from a few component frequencies and higher amplitude. Amplitude, decomposition into power spectra, or probability-based, nonlinear measures all can quantify the organization of human ventricular fibrillation waveforms. Clinical data have accumulated that these quantitative measures, or combinations of these measures, can predict the likelihood of rescue shock success, restoration of circulation, and survival to hospital discharge.

SUMMARY

Many quantitative ventricular fibrillation measures could be implemented in current generations of monitors/defibrillators to assist the timing of rescue shocks during clinical care. Emerging data suggest that a period of chest compressions or reperfusion can increase the likelihood of successful defibrillation. Therefore, waveform-based prediction of defibrillation success could reduce the delivery of failed rescue shocks.

The prediction of defibrillation outcome using a new combination of mean frequency and amplitude in porcine models of cardiac arrest

We estimated the predictive power with respect to defibrillation outcome of ventricular fibrillation (VF) mean frequency (FREQ), mean peak-to-trough amplitude (AMPL), and their combination. We examined VF electrocardiogram signals of 64 pigs from 4 different cardiac arrest models with different durations of untreated VF, different durations of cardiopulmonary resuscitation, and use of different drugs (epinephrine, vasopressin, N-nitro-L-arginine methyl ester, or saline placebo). The frequency domain was restricted to the range from 4.33 to 30 Hz. In the 10-s epoch between 20 and 10 s before the first defibrillation shock, FREQ and AMPL were estimated. We introduced the survival index (SI; 0.68 Hz(-1). FREQ + 12.69 mV(-1). AMPL) by use of multiple logistic regression. Kruskal-Wallis nonparametric one-way analysis was used to analyze the different porcine models for significant difference. The variables FREQ, AMPL, and SI were compared with defibrillation outcome by means of univariate logistic regression and receiver operating characteristic curves. SI increased predictive power compared with AMPL or FREQ alone, resulting in 89% sensitivity and 86% specificity. The probabilities of predicting defibrillation outcome for FREQ, AMPL, and SI were 0.85, 0.89 and 0.90, respectively. FREQ, AMPL, and SI values were not sensitive in regard to the four different cardiac arrest models but were significantly different for vasopressin and epinephrine animals.

IMPLICATIONS

We present a retrospective data analysis to evaluate the predictive power of different ventricular fibrillation electrocardiogram variables in pigs with respect to defibrillation outcome. We showed that our combination of variables leads to an improved forecast, which may help to reduce harmful unsuccessful defibrillation attempts.

Prevention of deterioration of ventricular fibrillation by basic life support during out-of-hospital cardiac arrest

Survival of cardiac arrest is improved by basic life support (BLS). This study investigated the relationship between ventricular fibrillation (VF) characteristics and survival. In a 2-year prospective study out-of-hospital witnessed non-traumatic cardiac arrests were observed. The probabilities of recording VF, asystole or other rhythms in relation to BLS and the time to the rhythm recording were analyzed with logistic regression. Amplitude and baseline crossings of VF were related to survival, using linear regression analysis. In 873 patients, the probability to record VF decreased per minute (OR 0.92, 95%CI 0.89-0.95) and of asystole increased (OR 1.13, 95%CI 1.09-1.18) as time from collapse elapsed. BLS reduced these trends significantly for VF (OR 0.97, 95%CI 0.94-0.99) and asystole (OR 1.09, 95%CI 1.05-1.13). This effect was not observed for other rhythms. The amplitude of VF decreased in time; significantly less for patients who received BLS than for those who did not (p=0.009). Survival significantly decreased with lower amplitude of VF (OR 0.23 per mV, 95%CI 0.07-0.79) and with less baseline crossings (OR 0.80 per baseline crossings per second, 95%CI 0.71-0.91). Our study demonstrated that BLS and VF as initial rhythm, considered being "baseline" predictors in survival models, were proved not independent of each other. The decrease of VF amplitude and increase in prevalence of asystole is slowed significantly by BLS. Predicting survival from VF amplitude and baseline crossings alone is limited.

Analysing ventricular fibrillation ECG-signals and predicting defibrillation success during cardiopulmonary resuscitation employing N(alpha)-histograms

Mean fibrillation frequency may predict defibrillation success during cardiopulmonary resuscitation (CPR). N(alpha)-histogram analysis should be investigated as an alternative. After 4 min of cardiac arrest, and 3 versus 8 min of CPR, 25 pigs received either vasopressin or epinephrine (0.4, 0.4, and 0.8 U/kg vasopressin versus 45, 45, and 200 microg/kg epinephrine) every 5 min with defibrillation at 22 min. Before defibrillation, the N(alpha)-parameter histogramstart/histogramwidth and the mean fibrillation frequency in resuscitated versus non-resuscitated pigs were 2.9+/-0.4 versus 1.7+/-0.5 (P=0.0000005); and 9.5+/-1.7 versus 6.9+/-0.7 (P=0.0003). During the last minute prior to defibrillation, histogramstart/histogramwidth of > or =2.3 versus mean fibrillation frequency > or =8 Hz predicted successful defibrillation with subsequent return of a spontaneous circulation for more than 60 min with sensitivity, specificity, positive predictive value and negative predictive value of 94 versus 82%, 96 versus 89%, 98 versus 93% and 90 versus 74%, respectively. We conclude, that N(alpha)-analysis was superior to mean fibrillation frequency analysis during CPR in predicting defibrillation success, and distinction between vasopressin versus epinephrine effects.

Algorithms to analyze ventricular fibrillation signals

Prediction of the success of defibrillation to avoid myocardial injury and performance feedback during CPR requires algorithms to analyze ventricular fibrillation signals. This report reviews investigations on different parameters of ventricular fibrillation electrocardiographic signals, including amplitude, frequency, bispectral analysis, amplitude spectrum area, wavelets, nonlinear dynamics, N(alpha) histograms, and combinations of several of these parameters. To date, no satisfactory methods have been found that cope with CPR artifacts and show adequate predictive power of successful defibrillation. The usual limitations of the studies are the small number of subjects, which precludes separation into training and test data. Because many investigations are animal studies of untreated short ventricular fibrillation, the results may be different for prolonged ventricular fibrillation in humans. The universality of threshold values has to be examined, and promising new parameters have to be monitored over longer time periods and analyzed for the effects of chest compressions, ventilation, and concomitant vasopressor therapy.

"Probability of successful defibrillation" as a monitor during CPR in out-of-hospital cardiac arrested patients

The frequency spectrum of the ECG in ventricular fibrillation (VF) correlates with myocardial perfusion and might predict defibrillation success defined as return of spontaneous circulation (ROSC). The predictive power increases when more spectral variables are combined, but the complex information can be difficult to handle during the intensity of CPR. We therefore developed a method for expressing this multidimensional information in a single reproducible variable reflecting the probability of defibrillation success. This is based on the highest performing predictor for ROSC after 883 shocks given to 156 patients with VF. This was a combination of two decorrelated spectral features based on a principal component analysis of an original feature set with information on centroid frequency, peak power frequency, spectral flatness and energy. The function "Probability of defibrillation success" (P(ROSC)(v)) was developed by a 2-dimensional histogram technique. P(ROSC)(v) discriminated between shocks followed by ROSC and No-ROSC (P<0.0001). The present methodology indicates a possible way to develop a CPR monitor.