Prediction of Countershock Success in Patients Using the Autoregressive Spectral Estimation

Abstract

Ventricular fibrillation (VF) is a life-threatening cardiac arrhythmia and within of minutes of its occurrence, optimal timing of countershock therapy is highly warranted to improve the chance of survival. This study was designed to investigate whether the autoregressive (AR) estimation technique was capable to reliably predict countershock success in VF cardiac arrest patients. ECG data of 1077 countershocks applied to 197 cardiac arrest patients with out-of-hospital and in-hospital cardiac arrest between March 2002 and July 2004 were retrospectively analyzed. The ECG from the 2.5 s interval of the precountershock VF ECG was used for computing the AR based features Spectral Pole Power (SPP) and Spectral Pole Power with Dominant Frequency weighing (SPPDF) and Centroid Frequency (CF) and Amplitude Spectrum Area (AMSA) based on Fast Fourier Transformation (FFT). With ROC AUC values up to 84.1% and diagnostic odds ratio up to 19.12 AR based features SPP and SPPDF have better prediction power than the FFT based features CF (80.5%; 6.56) and AMSA (82.1%; 8.79). AR estimation based features are promising alternatives to FFT based features for countershock outcome when analyzing human data.