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

Figure A1. Entropy values of the original VF signals (baseline), and the difference in entropy with respect to baseline for the 100 surrogate signals created using the amplitude adjusted fourier transform (AAFT) method. The red line shows the mean trends of the difference in entropy with respect to baseline of the surrogate data. The entropy values are significantly higher for the surrogate test signals for FuzzyEn (top) and SampEn (bottom) computed for the optimal values to discriminate recovery of QRS (left) or survival of the patient (right).

Figure A2. Values of SampEn(m,r,τ,N) and FuzzyEn(m,2,r,τ,N) and their predictive powers for signals sampled at fs=250 Hz. All calculations were done for the optimal points for prediction of survival.

Figure A3. Values of SampEn(m,r,τ,N) and FuzzyEn(m,2,r,τ,N) and their predictive powers for signals sampled at fs=60 Hz. All calculations were done for the optimal points for prediction of survival.

Figure A4. Values of the entropies and their predictive power as a function of the sampling frequency. All values were computed for the optimal (m, r) pairs reported in Table 1.

Figure A5. Relation between FuzzyEn and the typical VF waveform measures for which associated ATP level values and CPP values exist. As shown in the figure there is a strong positive association between AMSA and FuzzyEn for succesful and unsuccessful shocks (total R=0.96), and a strong negative association between ScE and FuzzyEn (total R=−0.87).

Figure 1. Example of the 2-min time period around a shock labeled as successful using the electrocardiographic criterion. An electrocardiogram (ECG) (top) was used to assess the rhythm and annotate the post-shock rhythm. The impedance (middle) shows chest compression activity and was used to determine the pre-shock pause interval. The ECG around the shock (bottom) shows how the shock restored a rhythm with sustained QRS complexes, indicated by arrows.

Figure 2. Analysis of the optimal (m,r) values for fuzzy entropy (FuzzyEn) (top) and sample entropy (SampEn) (bottom) using a 5 s analysis interval. Both success criteria were analyzed separately, and the optimal ranges to predict shock success were derived from the receiver operating characteristics (ROC) curve analyses (right) and are summarized in Table 1.

Figure 3. ROC curves for AMSA compared to entropy measures for both success criteria.

Figure 4. Values of FuzzyEn (top) and SampEn (bottom) computed using different window lengths (wl). The length of the interval in samples was Nw=fs·wl=60· wl for each case. The AUC values for each segment length and the two outcome criteria are shown in the rightmost graphs. FuzzyEn and SampEn were computed using the optimal (m,r) pairs obtained for 5-s segments, as reported in Table 1.

Figure 5. Values of AMSA and its predictive power for different window lengths (wl), and the two outcome criteria.

Figure 6. Evolution of entropies during the pre-shock pause, disaggregated for the different outcomes. t=0 s corresponds to the start of the pre-shock pause. In all cases, entropies decreased (almost) linearly as the interval without chest compression therapy increased. Entropies were computed using 3 s intervals, and values were computed every 0.5 s during the pause.