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

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.

Ventricular fibrillation waveform measures combined with prior shock outcome predict defibrillation success during cardiopulmonary resuscitation

AIM

Amplitude Spectrum Area (AMSA) and Median Slope (MS) are ventricular fibrillation (VF) waveform measures that predict defibrillation shock success. Cardiopulmonary resuscitation (CPR) obscures electrocardiograms and must be paused for analysis. Studies suggest waveform measures better predict subsequent shock success when combined with prior shock success. We determined whether this relationship applies during CPR.

METHODS

AMSA and MS were calculated from 5-second pre-shock segments with and without CPR, and compared to logistic models combining each measure with prior return of organized rhythm (ROR).

RESULTS

VF segments from 692 patients were analyzed during CPR before 1372 shocks and without CPR before 1283 shocks. Combining waveform measures with prior ROR increased areas under receiver operating characteristic curves for AMSA/MS with CPR (0.66/0.68 to 0.73/0.74, p<0.001) and without CPR (0.71/0.72 to 0.76/0.76, p<0.001).

CONCLUSIONS

Prior ROR improves prediction of shock success during CPR, and may enable waveform measure calculation without chest compression pauses.

Amplitude Spectrum Area to Guide Defibrillation

Background—

This study sought to validate the ability of amplitude spectrum area (AMSA) to predict defibrillation success and long-term survival in a large population of out-of-hospital cardiac arrests.

Methods and Results—

ECGs recorded by automated external defibrillators from different manufacturers were obtained from patients with cardiac arrests occurring in 8 city areas. A database, including 2447 defibrillations from 1050 patients, was used as the derivation group, and an additional database, including 1381 defibrillations from 567 patients, served as validation. A 2-second ECG window before defibrillation was analyzed, and AMSA was calculated. Univariable and multivariable regression analyses and area under the receiver operating characteristic curve were used for associations between AMSA and study end points: defibrillation success, sustained return of spontaneous circulation, and long-term survival. Among the 2447 defibrillations of the derivation database, 26.2% were successful. AMSA was significantly higher before a successful defibrillation than a failing one (13±5 versus 6.8±3.5 mV-Hz) and was an independent predictor of defibrillation success (odds ratio, 1.33; 95% confidence interval, 1.20–1.37) and sustained return of spontaneous circulation (odds ratio, 1.22; 95% confidence interval, 1.17–1.26). Area under the receiver operating characteristic curve for defibrillation success prediction was 0.86 (95% confidence interval, 0.85–0.88). AMSA was also significantly associated with long-term survival. The following AMSA thresholds were identified: 15.5 mV-Hz for defibrillation success and 6.5 mV-Hz for defibrillation failure. In the validation database, AMSA ≥15.5 mV-Hz had a positive predictive value of 84%, whereas AMSA ≤6.5 mV-Hz had a negative predictive value of 98%.

Conclusions—

In this large derivation-validation study, AMSA was validated as an accurate predictor of defibrillation success. AMSA also appeared as a predictor of long-term survival.

A support vector machine for predicting defibrillation outcomes from waveform metrics

BACKGROUND

Algorithms to predict shock success based on VF waveform metrics could significantly enhance resuscitation by optimising the timing of defibrillation.

OBJECTIVE

To investigate robust methods of predicting defibrillation success in VF cardiac arrest patients, by using a support vector machine (SVM) optimisation approach.

METHODS

Frequency-domain (AMSA, dominant frequency and median frequency) and time-domain (slope and RMS amplitude) VF waveform metrics were calculated in a 4.1Y window prior to defibrillation. Conventional prediction test validity of each waveform parameter was conducted and used AUC>0.6 as the criterion for inclusion as a corroborative attribute processed by the SVM classification model. The latter used a Gaussian radial-basis-function (RBF) kernel and the error penalty factor C was fixed to 1. A two-fold cross-validation resampling technique was employed.

RESULTS

A total of 41 patients had 115 defibrillation instances. AMSA, slope and RMS waveform metrics performed test validation with AUC>0.6 for predicting termination of VF and return-to-organised rhythm. Predictive accuracy of the optimised SVM design for termination of VF was 81.9% (± 1.24 SD); positive and negative predictivity were respectively 84.3% (± 1.98 SD) and 77.4% (± 1.24 SD); sensitivity and specificity were 87.6% (± 2.69 SD) and 71.6% (± 9.38 SD) respectively.

CONCLUSIONS

AMSA, slope and RMS were the best VF waveform frequency-time parameters predictors of termination of VF according to test validity assessment. This a priori can be used for a simplified SVM optimised design that combines the predictive attributes of these VF waveform metrics for improved prediction accuracy and generalisation performance without requiring the definition of any threshold value on waveform metrics.

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.

Part 8: adult advanced cardiovascular life support: 2010 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care

The goal of therapy for bradycardia or tachycardia is to rapidly identify and treat patients who are hemodynamically unstable or symptomatic due to the arrhythmia. Drugs or, when appropriate, pacing may be used to control unstable or symptomatic bradycardia. Cardioversion or drugs or both may be used to control unstable or symptomatic tachycardia. ACLS providers should closely monitor stable patients pending expert consultation and should be prepared to aggressively treat those with evidence of decompensation.

Predicting defibrillation success

PURPOSE OF REVIEW

Ventricular fibrillation is the primary rhythm in many cardiac arrest patients. Since the late 1980s, the surface electrocardiogram of ventricular fibrillation has been subjected to analysis to obtain reliable information about the likelihood of successful countershock and to estimate the duration of cardiac arrest. Considerable efforts were made in the past 2 years to further improve the predictive power of rescue shock measures.

RECENT FINDINGS

In a retrospective clinical study, ventricular fibrillation single feature analysis was not able to reliably estimate duration between cardiac arrest onset and initial electrocardiogram. Combining ventricular fibrillation features in the time and frequency domain by employing neural networks did not further improve the best single feature prediction power taken from higher ventricular fibrillation frequency bands. Cardioversion outcome prediction based on the wavelet technique increased the specificity up to 66% at the 95% sensitivity level.

SUMMARY

Recent results question the ventricular fibrillation feature analysis as a reliable tool to estimate the duration of human cardiac arrest. Animal and clinical studies confirmed that ventricular fibrillation waveform analysis contains information to reliably predict the countershock success rate and further improved countershock outcome prediction. Prospective clinical studies are highly warranted to demonstrate that ventricular fibrillation waveform analysis definitely improves survival after cardiac arrest.

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

BACKGROUND

Analysis of the electrocardiogram (ECG) can predict if a cardiac arrest patient in ventricular fibrillation is likely to have a return of spontaneous circulation if defibrillated. The accuracy of such methods determines how useful it is clinically and for retrospective analysis.

METHODS AND RESULTS

We wanted to identify if there is a potential of improving prediction accuracy by adding peri-arrest factors to an ECG-based prediction system, or constructing a prediction system that adapts to each patient. Therefore, we analysed shock outcome prediction data with a mixed effects logistic regression model to identify if there are random effects (unexplained variation between patients) influencing the prediction accuracy. We also added information about the patients' age, sex and presenting rhythm, ambulance response time and presence of bystander CPR to the model to try to improve it by reducing the random effects. For all the six predictive features analysed random effects where present, with p-values below 10(-3). The random effect size was 73-189% of the feature effect size. Adding the peri-arrest factors to the best ECG-based model gave no significant improvement.

CONCLUSIONS

The presence of random effects shows that the shock outcome prediction accuracy can be improved by explaining more of the variation between patients, for example using the approaches outlined above, and that there is within-patient correlation between samples that should be accounted for when evaluating prediction accuracy. The specific peri-arrest factors tested here did not significantly improve prediction accuracy, but other factors should be explored.

Wavelet transform analysis predicts outcome of DC cardioversion for atrial fibrillation patients

The aim of this study was to examine whether wavelet transform analysis of the electrocardiogram (ECG) can improve the prediction of the maintenance of sinus rhythm in patients with atrial fibrillation (AF) after external DC cardioversion. We examined a variety of wavelet transform-based statistical markers as potential candidates for the prediction of patient status post-cardioversion. Considering a 'success' as a patient who remains in normal sinus rhythm for one month post cardioversion and 'failure' as a patient who does not, it was shown the proposed non-parametric classification system can achieve 89% specificity at 100% sensitivity using a non-parametric classification method.

Outcome prediction for guidance of initial resuscitation protocol: Shock first or CPR first

BACKGROUND

Ventricular fibrillation (VF) is treated optimally with a defibrillation shock shortly after patient collapse, but may benefit from initial cardiopulmonary resuscitation (CPR) if the shock is delayed. An objective measure of potential responsiveness to defibrillation could help decide optimal initial therapy.

METHODS AND RESULTS

a new electrocardiogram (ECG) analysis algorithm was compared with response interval (call-to-shock) for prediction of patient outcome in a population of 87 VF patients in the Rochester, Minnesota area. In a retrospective analysis, both call-to-shock interval (p = 0.009) and ECG analysis (p < 0.001) predicted neurologically intact survival, with ECG analysis the stronger predictor (p = 0.034). When applied to advising initial patient treatment, ECG analysis compared favorably with the call-to-shock interval. Using a 7 min call-to-shock time criterion, 69% of patients would receive shocks first treatment using ECG analysis versus 67% using the call-to-shock interval (p = NS), 94% of survivors would retain successful shocks first treatment versus 85% (p = NS), and 48% of non-survivors receive alternate CPR-first treatment versus 45% (p = NS). Similarly, no significant differences were observed between ECG analysis and call-to-shock interval using an 8 min criterion.

CONCLUSIONS

Both call-to-shock interval and a real-time ECG analysis are predictive of patient outcome. The ECG analysis is more predictive of neurologically intact survival. Moreover, the ECG analysis is dependent only upon the patient's condition at the time of treatment, with no need for knowledge of the response interval, which may be difficult to estimate at the time of treatment.