Logarithm of the absolute correlations of the ECG waveform estimates duration of ventricular fibrillation and predicts successful defibrillation

Abstract

Background: Measures of the ventricular fibrillation (VF) waveform may enable better allocation of cardiac arrest treatment by discriminating which patients should receive immediate defibrillation versus alternate therapies such as CPR. We derive a new measure based on the 'roughness' of the VF waveform, the Logarithm of the Absolute Correlations (LAC), and assess and contrast how well the LAC and the previously published scaling exponent (ScE) predict the duration of VF and the likelihood of return of spontaneous circulation (ROSC) under both optimal experimental and commercial-defibrillator sampling conditions.

Methods and results: We derived the LAC and ScE from two different populations--an animal study of 44 swine and a retrospective human sample of 158 out-of-hospital VF arrests treated with a commercial defibrillator. In the animal study, the LAC and ScE were calculated on 5s epochs of VF recorded at 1000 samples/s and then down sampled to 125 samples/s. In the human study, the LAC and ScE were calculated using 6s epochs recorded at 200 samples/s that occurred immediately prior to the initial shock. We compared the LAC and ScE measures using the Spearman correlation coefficients (CC) and areas under the receiver operating characteristic curve (AUC).

Results: In the animal study, the LAC and ScE were highly correlated at 1000 sample/s (CC=0.93) but not at 125 samples/s (CC=-0.06). These correlations were reflected in how well the measures discriminated VF of < or =5 versus >5 min: AUC at 1000 samples/s was similar for LAC compared to ScE (0.71 versus 0.76). However AUC at 125 samples was greater for LAC compared to ScE (0.75 versus 0.62). In the human study, the LAC measure was a better predictor of ROSC following initial defibrillation as reflected by an AUC of 0.77 for LAC compared to 0.57 for ScE.

Conclusions: The LAC is an improvement over the ScE because the LAC retains its prognostic characteristics at lower ECG sampling rates typical of current clinical defibrillators. Hence, the LAC may have a role in better allocating treatment in resuscitation of VF cardiac arrest.

Fig. 1

Comparison of the plot of the autocorrelation coefficients versus the lag (Correlogram) for three waveforms; a 5 Hz sine wave, a wave of high amplitude noise and a wave that is the combination of 5 Hz sine wave plus noise. The ability of the correlogram to accentuate periodicity is demonstrated in the comparison of the bottom traces.

Fig. 2

Fig. (2A). 5 second ECG recordings from swine in early VF (at 30 seconds VF duration) and late VF (at 6.5 minutes VF duration). (2B): Autocorrelation of the early and late VF recordings at lags from −500 to 500 samples demonstrating positive and negative deviations from baseline. (2C): The absolute values of the autocorrelations in 2B demonstrating that the area under the curves is now positive at all lags for both time periods. The area under the curve is clearly greater in early VF.

Fig. 3

Fig. (3A). The LAC calculated on the original recordings from 44 swine recorded at 1000 samples/sec up to a maximum VF duration of 13.5 minutes. (3B): The ScE is shown for the same data and The LAC from Fig. 3A has been transformed to produce the LACadjusted as described in the text to facilitate comparison. (3C): The ScE is calculated for swine VF data at 1000 samples/sec and again on the same data down sampled to 125 samples/sec. (3D): The LAC and LACadjusted are shown for the swine VF data at 1000 samples/sec and on the same data down sampled to 125 samples/sec.

Fig. 4

Fig. (4A). A scatter plot of the LACadjusted versus the ScE from swine VF data recorded at1000 samples/sec versus is shown with trend line. The Spearman correlation coefficient (CC) is 0.93. (4B): A scatter plot of the LACadjusted versus the ScE from swine data down sampled to 125 samples/second showing a Spearman correlation coefficient of −0.06. (4C): Scatter plot of ScE for swine data at 1000 samples/second versus the same data down sampled to 125 samples/second with Spearman correlation of −0.05. (4D): The scatter plot of the LACadjusted from data recorded at 1000 samples/second versus the LACadjusted from data down sampled to 125samples/second demonstrating a Spearman correlation coefficient of 0.90.

Fig. 5

Fig. (5A). The ROC curves for the LACand ScE with positive result defined as VF less than 5 minutes in duration for swine VF recorded at 1000 samples/second. Area under the curve is 0.71 for the LAC and 0.76 for the ScE. (5B): The ROC curves for the LAC and ScE for swine VF at 125 samples/second. Area under the curve is 0.75 for the LAC and 0.62 for the ScE.

Fig. 6

Fig. (6A). Results for human data: mean values of the ScE for the initial shock of VF by AED. The ROSC group has mean of 1.093 (SD +/− 0.083, n=36) and the No ROSC group has a mean of 1.128 (SD +/−0.1540, n=122, p=0.19). (6B): LACadjusted for the initial shock for VF by AED. The ROSC group has mean= 1.196 (SD +/−.071, n=36) and the No ROSC group has mean 1.273 (SD +/−.092, n=122, p < 0.0001).

Fig. 7

Fig. (7). ROC curves for the LAC for human data for first shock of VF with ROSC as the positive response has an AUC of 0.77 (ROSC; n=36, NO ROSC; n=122). The ROC curve for the ScE is also shown for comparison with AUC of 0.57.