Predicting defibrillation success by ‘genetic’ programming in patients with out-of-hospital cardiac arrest

doi:10.1016/S0300-9572(03)00030-3

Resuscitation

Volume 57, Issue 2, May 2003, Pages 153-159

https://doi.org/10.1016/S0300-9572(03)00030-3 Get rights and content

Abstract

Background: In some patients with ventricular fibrillation (VF) there may be a better chance of successful defibrillation after a period of chest compression and ventilation before the defibrillation attempt. It is therefore important to know whether a defibrillation attempt will be successful. The predictive power of a model developed by ‘genetic’ programming (GP) to predict defibrillation success was studied. Methods and Results: 203 defibrillations were administered in 47 patients with out-of-hospital cardiac arrest due to a cardiac cause. Maximal amplitude, a total energy of power spectral density, and the Hurst exponent of the VF electrocardiogram (ECG) signal were included in the model developed by GP. Positive and negative likelihood ratios of the model for testing data were 35.5 and 0.00, respectively. Using a model developed by GP on the complete database, 120 of the 124 unsuccessful defibrillations would have been avoided, whereas all of the 79 successful defibrillations would have been administered. Conclusion: The VF ECG contains information predictive of defibrillation success. The model developed by GP, including data from the time-domain, frequency-domain and nonlinear dynamics, could reduce the incidence of unsuccessful defibrillations.

Sumàrio

Contexto: Em alguns doentes com fibrilhação ventricular (VF) pode haver mais hipóteses de sucesso se antes da tentativa de desfibrilhação se fizer um perı́odo de compressões torácicas e ventilação. É, portanto, importante saber se a tentativa de desfibrilhação terá sucesso. Estudou-se o poder predictivo para desfibrilhação com sucesso de um modelo desenvolvido por programação “genética” (PG). Métodos e Resultados: Foram feitas 203 desfibrilhações em 47 doentes com paragem cardı́aca pré-hospitalar de causa cardı́aca. A amplitude máxima, uma energia total de densidade de poder espectral, e o expoente Hurst do sinal de VF do electrocardiograma (ECG) foram incluı́dos no modelo desenvolvido por PG. A probabilidade de ocorrência positiva e negativa do modelo para os dados testados foram 35,5 e 0,00, respectivamente. Utilizando um modelo desenvolvido por PG na base de dados completa, poder-se-iam ter dispensado 120 das 124 desfibrilhações, enquanto todas as 79 desfibrilhações com sucesso teriam sido administradas. Conclusão: O ECG de VF contém informação predictiva do sucesso da desfibrilhação. Um modelo desenvolvido por PG, incluindo dados a partir do domı́nio-tempo, domı́nio-frequência e dinâmicas não lineares, pode reduzir a incidência de desfibrilhações sem sucesso.

Resumen

Antecedentes: En algunos pacientes con fibrilación ventricular (VF) puede haber una mejor posibilidad de desfibrilación exitosa después de un perı́odo de masaje cardiaco y ventilación antes de intentar desfibrilar. Por lo tanto es importante saber si acaso será exitoso un intento de desfibrilación. Se estudió el poder de predicción de un modelo desarrollado por programación ‘genética’(GP) para predecir el éxito de una desfibrilación. Métodos y Resultados: Se administraron 203 desfibrilaciones en 47 pacientes con paro cardiaco extrahospitalario de causa cardiaca. En el modelo de GP se incluyeron la amplitud máxima, la energı́a total de densidad de poder espectral, y el exponente Hurst de la señal electrocardiográfica (ECG) en la VF. Las relaciones de probabilidades positivas y negativas del modelo para examinar los datos fueron 35.5 y 0.00, respectivamente. Usando el modelo GP en la base de datos completa, 120 de 124 desfibrilaciones no exitosa habrı́a sido evitada, mientras que todas las 79 desfibrilaciones exitosas habrı́an sido administradas. Conclusión: El ECG de la VF contiene información que permite predecir el éxito de la desfibrilación. El modelo desarrollado por GP, incluyendo datos de tiempo y de frecuencia y dinámica no linear, podrı́a reducir la incidencia de desfibrilaciones no exitosas.

Introduction

The current guidelines of advance life support prescribe immediate defibrillation after the onset of ventricular fibrillation (VF). With increasing duration of VF, the success of electrical defibrillation decreases [1]. However, in some patients with VF there might be a better chance of return of spontaneous circulation after a period of chest compression and ventilation before the defibrillation attempt [2]. Futile defibrillation attempts are in themselves detrimental, because tissue damage and post-resuscitation myocardial dysfunction may be caused by the shock itself and by the lack of tissue perfusion from chest compression during the shock period (analysis, charging, defibrillation and outcome evaluation) [3]. It is therefore important to know whether a defibrillation attempt will be successful.

The electrocardiogram (ECG) is a part of the current practice of advanced life support. Different algorithms and variables to analyze VF ECG signals and to predict success of defibrillation, including amplitude, frequency, bispectral analysis, amplitude spectrum area, wavelets, nonlinear dynamics, N(α) histograms and combinations of several of these parameters, have been studied [4].

There are essentially two kinds of predictive models: deterministic (mathematical models, empirical models, computer simulation models) and non-deterministic (models developed by genetic methods, neural network models, models based on chaos theory and soft logic theory), and each has its advantages and disadvantages [5]. In general, when deterministic modeling is used the models obtained are the result of strict mathematical rules or they are set in advance. In that case, the goal is merely to discover a set of numerical coefficients for a model whose form has been pre-specified. However, nowadays more and more processes and systems are modeled and optimized using non-deterministic approaches. This is due to the high degree of complexity of the systems, and consequently, the inability to study them successfully with conventional methods only. In non-deterministic modeling of systems, there are no precise, strict mathematical rules. For example, in ‘genetic’ programming (GP), no assumptions about the form, size and complexity of models are made in advance. They are left to stochastic, self-organized, intelligent and non-centralized evolutionary processes [6].

The aim of the study was to develop a model by GP to predict success of defibrillation in patients with out-of-hospital cardiac arrest using features of VF ECG form time-domain, frequency-domain and nonlinear dynamics, which were shown previously to contain information to predict successful defibrillation.

Section snippets

Patients

Two hundred and three ECG recordings in 47 patients with out-of-hospital cardiac arrest occurring between March 1998 and December 2000 were evaluated retrospectively. In all patients, cardiac aetiology was the immediate cause of cardiac arrest and VF was identified as the initial ECG rhythm. Cardiopulmonary resuscitation and defibrillation therapy were performed in the out-of-hospital setting by the staff of the medical emergency service and all patients were treated in accordance with 1998

Results

Two hundred and three (mean 4.3, median 4, range 1–16) defibrillations were administered in 47 (13 (27.6%) female) with out-of-hospital cardiac arrest due to VF. Mean age of the patients was 64.5±10.5 years. Seventy-nine (38.9%) defibrillations were successful. Characteristics of VF signal are shown in Table 1.

At the beginning the predictive model developed by GP included all four variables: TD, A, TE and H. TD was excluded during evolution of the model. The equation is presented in Fig. 2. The

Discussion

In this study of 203 defibrillations, the prediction power of a model developed by GP to predict successful defibrillation was studied. High positive and low negative likelihood ratios of the GP model indicate that the model may have a very substantial impact on clinical decision-making through a meaningful assessment of whether to administer defibrillation or not [14].

In the model developed by GP four variables from time-domain (TD and A), frequency-domain (TE) and nonlinear dynamics (H) were

References (21)

E.J. Gallagher
Clinical utility of likelihood ratios
Ann. Emerg. Med.
(1998)
H.U. Strohmenger et al.
Analysis of ventricular fibrillation ECG signal amplitude and frequency parameters as predictors of countershock success in humans
Chest
(1997)
C.G. Brown et al.
Median frequency—a new parameter for predicting defibrillation success rate
Ann. Emerg. Med.
(1991)
C.G. Brown et al.
Physiologic measurement of ventricular fibrillation ECG signal: estimating the duration of ventricular fibrillation
Ann. Emerg. Med.
(1993)
L.D. Sherman et al.
Ventricular fibrillation exhibits dynamical properties and self-similarity
Resuscitation
(2000)
W.D. Weaver et al.
Amplitude of ventricular fibrillation waveform and outcome after cardiac arrest
Ann. Inter. Med.
(1985)
L.A. Cobb et al.
Influence of cardiopulmonary resuscitation prior to defibrillation in patients with out-of-hospital ventricular fibrillation
J. Am. Med. Assoc.
(1999)
J. Xie et al.
High-energy defibrillation increases the severity of postresuscitation myocardial dysfunction
Circulation
(1997)
A. Amann et al.
Algorithms to analyze ventricular fibrillation signals
Curr. Opin. Crit. Care
(2001)
P. Dobnik-Dubovski et al.
Using genetic programming to predict the macroporosity of woven cotton fabrics
Textil. Res. J.
(2002)

There are more references available in the full text version of this article.

Cited by (37)

European Resuscitation Council Guidelines 2021: Adult advanced life support
2021, Resuscitation
ECG waveform dataset for predicting defibrillation outcome in out-of-hospital cardiac arrested patients
2021, Data in Brief
The provided database of 260 ECG signals was collected from patients with out-of-hospital cardiac arrest while treated by the emergency medical services. Each ECG signal contains a 9 second waveform showing ventricular fibrillation, followed by 1 min of post-shock waveform. Patients’ ECGs are made available in multiple formats. All ECGs recorded during the prehospital treatment are provided in PFD files, after being anonymized, printed in paper, and scanned. For each ECG, the dataset also includes the whole digitized waveform (9 s pre- and 1 min post-shock each) and numerous features in temporal and frequency domain extracted from the 9 s episode immediately prior to the first defibrillation shock. Based on the shock outcome, each ECG file has been annotated by three expert cardiologists, - using majority decision -, as successful (56 cases), unsuccessful (195 cases), or indeterminable (9 cases). The code for preprocessing, for feature extraction, and for limiting the investigation to different temporal intervals before the shock is also provided. These data could be reused to design algorithms to predict shock outcome based on ventricular fibrillation analysis, with the goal to optimize the defibrillation strategy (immediate defibrillation versus cardiopulmonary resuscitation and/or drug administration) for enhancing resuscitation.
Predicting defibrillation success in out-of-hospital cardiac arrested patients: Moving beyond feature design
2020, Artificial Intelligence in Medicine
Citation Excerpt :
However, by choosing the decision as in this study, this model would not be considered safe. The only model that outperforms the CNN model was suggested by Podbregar et al. [35]. Nonetheless, as we stated above there is a possibility of data leaking related to this study, and possibly unrealistically high performance.
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.
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.
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.
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.
European Resuscitation Council Guidelines for Resuscitation 2015. Section 3. Adult advanced life support.
2015, Resuscitation
Citation Excerpt :
Auto-PEEP (gas trapping) may be particularly high in asthmatics and may necessitate higher than usual energy levels for defibrillation.446 It is possible to predict, with varying reliability, the success of defibrillation from the fibrillation waveform.342,343,447–467 If optimal defibrillation waveforms and the optimal timing of shock delivery can be determined in prospective studies, it should be possible to prevent the delivery of unsuccessful high energy shocks and minimise myocardial injury.
Optimizing the timing of defibrillation: The role of ventricular fibrillation waveform analysis during cardiopulmonary resuscitation
2012, Critical Care Clinics
Citation Excerpt :
In addition, cardiomyopathy, autonomic dysfunction, and differences in drug therapy make it probable that VF waveform analysis will never demonstrate perfect predictive ability. Because different measurements extract slightly different information from the VF waveform, it is likely that combinations of these measurements will provide superior discriminative ability.32,65,66 Neurauter and coworkers35 analyzed 770 ECG recordings of countershock parameters from 197 patients with CA.
Development and validation of clinical prediction models: Marginal differences between logistic regression, penalized maximum likelihood estimation, and genetic programming
2012, Journal of Clinical Epidemiology
Many prediction models are developed by multivariable logistic regression. However, there are several alternative methods to develop prediction models. We compared the accuracy of a model that predicts the presence of deep venous thrombosis (DVT) when developed by four different methods.
We used the data of 2,086 primary care patients suspected of DVT, which included 21 candidate predictors. The cohort was split into a derivation set (1,668 patients, 329 with DVT) and a validation set (418 patients, 86 with DVT). Also, 100 cross-validations were conducted in the full cohort. The models were developed by logistic regression, logistic regression with shrinkage by bootstrapping techniques, logistic regression with shrinkage by penalized maximum likelihood estimation, and genetic programming. The accuracy of the models was tested by assessing discrimination and calibration.
There were only marginal differences in the discrimination and calibration of the models in the validation set and cross-validations.
The accuracy measures of the models developed by the four different methods were only slightly different, and the 95% confidence intervals were mostly overlapped. We have shown that models with good predictive accuracy are most likely developed by sensible modeling strategies rather than by complex development methods.

View all citing articles on Scopus

View full text

Predicting defibrillation success by ‘genetic’ programming in patients with out-of-hospital cardiac arrest

Abstract

Sumàrio

Resumen

Introduction

Section snippets

Patients

Results

Discussion

Ann. Emerg. Med.

Chest

Ann. Emerg. Med.

Ann. Emerg. Med.

Resuscitation

Amplitude of ventricular fibrillation waveform and outcome after cardiac arrest

Ann. Inter. Med.

Influence of cardiopulmonary resuscitation prior to defibrillation in patients with out-of-hospital ventricular fibrillation

J. Am. Med. Assoc.

High-energy defibrillation increases the severity of postresuscitation myocardial dysfunction

Circulation

Algorithms to analyze ventricular fibrillation signals

Curr. Opin. Crit. Care

Using genetic programming to predict the macroporosity of woven cotton fabrics

Textil. Res. J.