Background: Countershock outcome prediction using ventricular fibrillation (VF) feature analysis needs undisturbed electrocardiogram (ECG) signals and therefore requires interruption of cardiopulmonary resuscitation (CPR). Features that originate from higher frequency bands of the VF power spectrum may be less affected by CPR artefacts and as such reduce cumulative hands-off intervals.
Materials and methods: From 192 patients with in-hospital and out-of-hospital cardiac arrest, four countershock outcome prediction features (peak-peak amplitude, mean slope, median slope, power spectrum analysis) were analysed in 550 short time ECG records, each including a CPR corrupted and a subsequent undisturbed sequence. ECG features calculated from the main frequency band (0-26Hz) and from bandpass-filtered subbands (>10-26Hz) were compared using the similarity level method and differences in shock advice numbers.
Results: The feature similarity between ECG periods with and without CPR artefacts was higher in bandpass-filtered (Sim=0.79, 0.8, 0.78, 0.66) than in unfiltered ECG traces (Sim=0.58, 0.69, 0.68, 0.47). For the features evaluated, the difference in number of shock advices between subsequent traces with and without CPR artefact was significantly reduced using VF analysis from higher frequency bands.
Conclusion: The accuracy of shock outcome prediction during CPR could be increased by using filtered ECG features from higher ECG subbands instead of features derived from the main ECG spectrum.