Identifying approaches to improve the accuracy of shock outcome prediction for out-of-hospital cardiac arrest

Extracting physiologic and clinical data from defibrillators for research purposes to improve treatment for patients in cardiac arrest

Background

A defibrillator should be connected to all patients receiving cardiopulmonary resuscitation (CPR) to allow early defibrillation. The defibrillator will collect signal data such as the electrocardiogram (ECG), thoracic impedance and end-tidal CO2, which allows for research on how patients demonstrate different responses to CPR. The aim of this review is to give an overview of methodological challenges and opportunities in using defibrillator data for research.

Methods

The successful collection of defibrillator files has several challenges. There is no scientific standard on how to store such data, which have resulted in several proprietary industrial solutions. The data needs to be exported to a software environment where signal filtering and classifications of ECG rhythms can be performed. This may be automated using different algorithms and artificial intelligence (AI). The patient can be classified being in ventricular fibrillation or -tachycardia, asystole, pulseless electrical activity or having obtained return of spontaneous circulation. How this dynamic response is time-dependent and related to covariates can be handled in several ways. These include Aalen's linear model, Weibull regression and joint models.

Conclusions

The vast amount of signal data from defibrillator represents promising opportunities for the use of AI and statistical analysis to assess patient response to CPR. This may provide an epidemiologic basis to improve resuscitation guidelines and give more individualized care. We suggest that an international working party is initiated to facilitate a discussion on how open formats for defibrillator data can be accomplished, that obligates industrial partners to further develop their current technological solutions.

A method to predict ventricular fibrillation shock outcome during chest compressions

BACKGROUND

Out-of-hospital ventricular fibrillation (VF) cardiac arrest is a leading cause of death. Quantitative analysis of the VF electrocardiogram (ECG) can predict patient outcomes and could potentially enable a patient-specific, guided approach to resuscitation. However, VF analysis during resuscitation is confounded by cardiopulmonary resuscitation (CPR) artifact in the ECG, challenging continuous application to guide therapy throughout resuscitation. We therefore sought to design a method to predict VF shock outcomes during CPR.

METHODS

Study data included 4577 5-s VF segments collected during and without CPR prior to defibrillation attempts in N = 1151 arrest patients. Using training data (460 patients), an algorithm was designed to predict the VF shock outcomes of defibrillation success (return of organized ventricular rhythm) and functional survival (Cerebral Performance Category 1-2). The algorithm was designed with variable-frequency notch filters to reduce CPR artifact in the ECG based on real-time chest compression rate. Ten ECG features and three dichotomous patient characteristics were developed to predict outcomes. These variables were combined using support vector machines and logistic regression. Algorithm performance was evaluated by area under the receiver operating characteristic curve (AUC) to predict outcomes in validation data (691 patients).

RESULTS

AUC (95% Confidence Interval) for predicting defibrillation success was 0.74 (0.71-0.77) during CPR and 0.77 (0.74-0.79) without CPR. AUC for predicting functional survival was 0.75 (0.72-0.78) during CPR and 0.76 (0.74-0.79) without CPR.

CONCLUSION

A novel algorithm predicted defibrillation success and functional survival during ongoing CPR following VF arrest, providing a potential proof-of-concept towards real-time guidance of resuscitation therapy.

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.

Fuzzy and Sample Entropies as Predictors of Patient Survival Using Short Ventricular Fibrillation Recordings during out of Hospital Cardiac Arrest

Optimal defibrillation timing guided by ventricular fibrillation (VF) waveform analysis would contribute to improved survival of out-of-hospital cardiac arrest (OHCA) patients by minimizing myocardial damage caused by futile defibrillation shocks and minimizing interruptions to cardiopulmonary resuscitation. Recently, fuzzy entropy (FuzzyEn) tailored to jointly measure VF amplitude and regularity has been shown to be an efficient defibrillation success predictor. In this study, 734 shocks from 296 OHCA patients (50 survivors) were analyzed, and the embedding dimension (m) and matching tolerance (r) for FuzzyEn and sample entropy (SampEn) were adjusted to predict defibrillation success and patient survival. Entropies were significantly larger in successful shocks and in survivors, and when compared to the available methods, FuzzyEn presented the best prediction results, marginally outperforming SampEn. The sensitivity and specificity of FuzzyEn were 83.3% and 76.7% when predicting defibrillation success, and 83.7% and 73.5% for patient survival. Sensitivities and specificities were two points above those of the best available methods, and the prediction accuracy was kept even for VF intervals as short as 2s. These results suggest that FuzzyEn and SampEn may be promising tools for optimizing the defibrillation time and predicting patient survival in OHCA patients presenting VF.

Does the choice of definition for defibrillation and CPR success impact the predictability of ventricular fibrillation waveform analysis?

BACKGROUND

Quantitative analysis of ventricular fibrillation (VF), such as amplitude spectral area (AMSA), predicts shock outcomes. However, there is no uniform definition of shock/cardiopulmonary resuscitation (CPR) success in out-of-hospital cardiac arrest (OHCA). The objective of this study is to investigate post-shock rhythm variations and the impact of shock/CPR success definition on the predictability of AMSA.

METHODS

A total of 554 shocks from 257 OHCA patients with VF as initial rhythm were analyzed. Post-shock rhythms were analyzed every 5s up to 120s and annotated as VF, asystole (AS) and organized rhythm (OR) at serial time intervals. Three shock/CPR success definitions were used to evaluate the predictability of AMSA: (1) termination of VF (ToVF); (2) return of organized electrical activity (ROEA); (3) return of potentially perfusing rhythm (RPPR).

RESULTS

Rhythm changes occurred after 54.5% (N=302) of shocks and 85.8% (N=259) of them occurred within 60s after shock delivery. The observed post-shock rhythm changes were (1) from AS to VF (24.9%), (2) from OR to VF (16.1%), and (3) from AS to OR (12.1%). The area under the receiver operating characteristic curve (AUC) for AMSA as a predictor of shock/CPR success reached its maximum 60s post-shock. The AUC was 0.646 for ToVF, 0.782 for ROEA, and 0.835 for RPPR (p<0.001) respectively.

CONCLUSIONS

Post-shock rhythm is unstable in the first minute after the shock. The predictability of AMSA varies depending on the definition of shock/CPR success and performs best with the return of potentially perfusing rhythm endpoint for OHCA.

Combining Amplitude Spectrum Area with Previous Shock Information Using Neural Networks Improves Prediction Performance of Defibrillation Outcome for Subsequent Shocks in Out-Of-Hospital Cardiac Arrest Patients

Objective

Quantitative ventricular fibrillation (VF) waveform analysis is a potentially powerful tool to optimize defibrillation. However, whether combining VF features with additional attributes that related to the previous shock could enhance the prediction performance for subsequent shocks is still uncertain.

Methods

A total of 528 defibrillation shocks from 199 patients experienced out-of-hospital cardiac arrest were analyzed in this study. VF waveform was quantified using amplitude spectrum area (AMSA) from defibrillator's ECG recordings prior to each shock. Combinations of AMSA with previous shock index (PSI) or/and change of AMSA (ΔAMSA) between successive shocks were exercised through a training dataset including 255shocks from 99patientswith neural networks. Performance of the combination methods were compared with AMSA based single feature prediction by area under receiver operating characteristic curve(AUC), sensitivity, positive predictive value (PPV), negative predictive value (NPV) and prediction accuracy (PA) through a validation dataset that was consisted of 273 shocks from 100patients.

Results

A total of61 (61.0%) patients required subsequent shocks (N = 173) in the validation dataset. Combining AMSA with PSI and ΔAMSA obtained highest AUC (0.904 vs. 0.819, p<0.001) among different combination approaches for subsequent shocks. Sensitivity (76.5% vs. 35.3%, p<0.001), NPV (90.2% vs. 76.9%, p = 0.007) and PA (86.1% vs. 74.0%, p = 0.005)were greatly improved compared with AMSA based single feature prediction with a threshold of 90% specificity.

Conclusion

In this retrospective study, combining AMSA with previous shock information using neural networks greatly improves prediction performance of defibrillation outcome for subsequent shocks.

Integration of Attributes from Non-Linear Characterization of Cardiovascular Time-Series for Prediction of Defibrillation Outcomes

Objective

The timing of defibrillation is mostly at arbitrary intervals during cardio-pulmonary resuscitation (CPR), rather than during intervals when the out-of-hospital cardiac arrest (OOH-CA) patient is physiologically primed for successful countershock. Interruptions to CPR may negatively impact defibrillation success. Multiple defibrillations can be associated with decreased post-resuscitation myocardial function. We hypothesize that a more complete picture of the cardiovascular system can be gained through non-linear dynamics and integration of multiple physiologic measures from biomedical signals.

Materials and Methods

Retrospective analysis of 153 anonymized OOH-CA patients who received at least one defibrillation for ventricular fibrillation (VF) was undertaken. A machine learning model, termed Multiple Domain Integrative (MDI) model, was developed to predict defibrillation success. We explore the rationale for non-linear dynamics and statistically validate heuristics involved in feature extraction for model development. Performance of MDI is then compared to the amplitude spectrum area (AMSA) technique.

Results

358 defibrillations were evaluated (218 unsuccessful and 140 successful). Non-linear properties (Lyapunov exponent > 0) of the ECG signals indicate a chaotic nature and validate the use of novel non-linear dynamic methods for feature extraction. Classification using MDI yielded ROC-AUC of 83.2% and accuracy of 78.8%, for the model built with ECG data only. Utilizing 10-fold cross-validation, at 80% specificity level, MDI (74% sensitivity) outperformed AMSA (53.6% sensitivity). At 90% specificity level, MDI had 68.4% sensitivity while AMSA had 43.3% sensitivity. Integrating available end-tidal carbon dioxide features into MDI, for the available 48 defibrillations, boosted ROC-AUC to 93.8% and accuracy to 83.3% at 80% sensitivity.

Conclusion

At clinically relevant sensitivity thresholds, the MDI provides improved performance as compared to AMSA, yielding fewer unsuccessful defibrillations. Addition of partial end-tidal carbon dioxide (PetCO₂) signal improves accuracy and sensitivity of the MDI prediction model.

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.

Beyond ventricular fibrillation analysis: comprehensive waveform analysis for all cardiac rhythms occurring during resuscitation

AIM

To propose a method which analyses the electrocardiogram (ECG) waveform of any cardiac rhythm occurring during resuscitation and computes the probability of that rhythm converting into another with better prognosis (Pdes).

METHODS

Rhythm transitions occurring spontaneously or due to defibrillation were analyzed. For each possible rhythm, ventricular fibrillation/ventricular tachycardia (VF/VT), pulseless electrical activity (PEA), pulse-generating rhythm (PR) and asystole (AS), the desired and undesired transitions were defined. ECG segments corresponding to the last 3s of rhythms prior to transition were used to extract waveform features. For each rhythm type, waveform features were combined into a logistic regression model to develop a rhythm specific classifier of desired transitions. This model was the monitoring function for the Pdes. The capacity of each rhythm specific classifier to discriminate between desired and undesired transitions was evaluated in terms of area under the curve (AUC). Pdes was integrated into a state sequence representation, which structures the information of cardiac arrest episodes, to analyze the effect of therapy on patient. As a case study, the effect of optimal/suboptimal cardiopulmonary resuscitation (CPR) on Pdes was analyzed. The mean Pdes was computed for the pre- and post-CPR intervals which presented the same underlying rhythm. The relationship between the optimal/suboptimal CPR and increase/decrease of Pdes was analyzed.

RESULTS

The AUC was 0.80, 0.79, 0.73 and 0.61 for VF/VT, PEA, PR and AS respectively. The Pdes quantified the probability of every rhythm of the episode developing to a better state, and the evolution of Pdes was coherent with the provided therapy. The case study indicated, for most rhythms, that positive trends in the dynamic behaviour could be associated with optimal CPR, whereas the opposite seemed true for negative trends.

CONCLUSION

A method for continuous ECG waveform analysis covering all cardiac rhythms during resuscitation has been proposed. This methodology can be further developed to be used in retrospective studies of CPR techniques, and, in the future, for potentially monitoring in real time the probability of survival of patients being resuscitated.

Non-linear dynamical signal characterization for prediction of defibrillation success through machine learning

Background

Ventricular Fibrillation (VF) is a common presenting dysrhythmia in the setting of cardiac arrest whose main treatment is defibrillation through direct current countershock to achieve return of spontaneous circulation. However, often defibrillation is unsuccessful and may even lead to the transition of VF to more nefarious rhythms such as asystole or pulseless electrical activity. Multiple methods have been proposed for predicting defibrillation success based on examination of the VF waveform. To date, however, no analytical technique has been widely accepted. We developed a unique approach of computational VF waveform analysis, with and without addition of the signal of end-tidal carbon dioxide (PetCO2), using advanced machine learning algorithms. We compare these results with those obtained using the Amplitude Spectral Area (AMSA) technique.

Methods

A total of 90 pre-countershock ECG signals were analyzed form an accessible preshosptial cardiac arrest database. A unified predictive model, based on signal processing and machine learning, was developed with time-series and dual-tree complex wavelet transform features. Upon selection of correlated variables, a parametrically optimized support vector machine (SVM) model was trained for predicting outcomes on the test sets. Training and testing was performed with nested 10-fold cross validation and 6–10 features for each test fold.

Results

The integrative model performs real-time, short-term (7.8 second) analysis of the Electrocardiogram (ECG). For a total of 90 signals, 34 successful and 56 unsuccessful defibrillations were classified with an average Accuracy and Receiver Operator Characteristic (ROC) Area Under the Curve (AUC) of 82.2% and 85%, respectively. Incorporation of the end-tidal carbon dioxide signal boosted Accuracy and ROC AUC to 83.3% and 93.8%, respectively, for a smaller dataset containing 48 signals. VF analysis using AMSA resulted in accuracy and ROC AUC of 64.6% and 60.9%, respectively.

Conclusion

We report the development and first-use of a nontraditional non-linear method of analyzing the VF ECG signal, yielding high predictive accuracies of defibrillation success. Furthermore, incorporation of features from the PetCO2 signal noticeably increased model robustness. These predictive capabilities should further improve with the availability of a larger database.

Reduction of CPR artifacts in the ventricular fibrillation ECG by coherent line removal

Background

Interruption of cardiopulmonary resuscitation (CPR) impairs the perfusion of the fibrillating heart, worsening the chance for successful defibrillation. Therefore ECG-analysis during ongoing chest compression could provide a considerable progress in comparison with standard analysis techniques working only during "hands-off" intervals.

Methods

For the reduction of CPR-related artifacts in ventricular fibrillation ECG we use a localized version of the coherent line removal algorithm developed by Sintes and Schutz. This method can be used for removal of periodic signals with sufficiently coupled harmonics, and can be adapted to specific situations by optimal choice of its parameters (e.g., the number of harmonics considered for analysis and reconstruction). Our testing was done with 14 different human ventricular fibrillation (VF) ECGs, whose fibrillation band lies in a frequency range of [1 Hz, 5 Hz]. The VF-ECGs were mixed with 12 different ECG-CPR-artifacts recorded in an animal experiment during asystole. The length of each of the ECG-data was chosen to be 20 sec, and testing was done for all 168 = 14 × 12 pairs of data. VF-to-CPR ratio was chosen as -20 dB, -15 dB, -10 dB, -5 dB, 0 dB, 5 dB and 10 dB. Here -20 dB corresponds to the highest level of CPR-artifacts.

Results

For non-optimized coherent line removal based on signals with a VF-to-CPR ratio of -20 dB, -15 dB, -10 dB, -5 dB and 0 dB, the signal-to-noise gains (SNR-gains) were 9.3 ± 2.4 dB, 9.4 ± 2.4 dB, 9.5 ± 2.5 dB, 9.3 ± 2.5 dB and 8.0 ± 2.7 (mean ± std, n = 168), respectively. Characteristically, an original VF-to-CPR ratio of -10 dB, corresponds to a variance ratio var(VF):var(CPR) = 1:10. An improvement by 9.5 dB results in a restored VF-to-CPR ratio of -0.5 dB, corresponding to a variance ratio var(VF):var(CPR) = 1:1.1, the variance of the CPR in the signal being reduced by a factor of 8.9.

Discussion

The localized coherent line removal algorithm uses the information of a single ECG channel. In contrast to multi-channel algorithms, no additional information such as thorax impedance, blood pressure, or pressure exerted on the sternum during CPR is required. Predictors of defibrillation success such as mean and median frequency of VF-ECGs containing CPR-artifacts are prone to being governed by the harmonics of the artifacts. Reduction of CPR-artifacts is therefore necessary for determining reliable values for estimators of defibrillation success.

Conclusions

The localized coherent line removal algorithm reduces CPR-artifacts in VF-ECG, but does not eliminate them. Our SNR-improvements are in the same range as offered by multichannel methods of Rheinberger et al., Husoy et al. and Aase et al. The latter two authors dealt with different ventricular rhythms (VF and VT), whereas here we dealt with VF, only. Additional developments are necessary before the algorithm can be tested in real CPR situations.

Development of the probability of return of spontaneous circulation in intervals without chest compressions during out-of-hospital cardiac arrest: an observational study

BACKGROUND

One of the factors that limits survival from out-of-hospital cardiac arrest is the interruption of chest compressions. During ventricular fibrillation and tachycardia the electrocardiogram reflects the probability of return of spontaneous circulation associated with defibrillation. We have used this in the current study to quantify in detail the effects of interrupting chest compressions.

METHODS

From an electrocardiogram database we identified all intervals without chest compressions that followed an interval with compressions, and where the patients had ventricular fibrillation or tachycardia. By calculating the mean-slope (a predictor of the return of spontaneous circulation) of the electrocardiogram for each 2-second window, and using a linear mixed-effects statistical model, we quantified the decline of mean-slope with time. Further, a mapping from mean-slope to probability of return of spontaneous circulation was obtained from a second dataset and using this we were able to estimate the expected development of the probability of return of spontaneous circulation for cases at different levels.

RESULTS

From 911 intervals without chest compressions, 5138 analysis windows were identified. The results show that cases with the probability of return of spontaneous circulation values 0.35, 0.1 and 0.05, 3 seconds into an interval in the mean will have probability of return of spontaneous circulation values 0.26 (0.24-0.29), 0.077 (0.070-0.085) and 0.040(0.036-0.045), respectively, 27 seconds into the interval (95% confidence intervals in parenthesis).

CONCLUSION

During pre-shock pauses in chest compressions mean probability of return of spontaneous circulation decreases in a steady manner for cases at all initial levels. Regardless of initial level there is a relative decrease in the probability of return of spontaneous circulation of about 23% from 3 to 27 seconds into such a pause.