Abstract 340: An Algorithm that Analyzes During Chest Compressions Exceeds Sensitivity and Specificity Recommendations
Background: Pauses in chest compressions during cardiopulmonary resuscitation (CPR) reduce survival from sudden cardiac arrest. Analyzing ECG during compressions (ADC) instead of pausing for analysis may improve outcomes. We assembled large development and test sets of clinical data and evaluated the performance of an ADC algorithm that classifies rhythms into categories of “Shock,” “No Shock,” and “Pause CPR for Analysis.”
Methods: Defibrillator ECG and impedance recordings from 1097 (585 development, 565 test) patients with cardiac arrest were retrospectively gathered from 7 emergency medical systems. To compare against American Heart Association (AHA) recommendations, CPR-artifacted segments with coarse VF (> 0.2 mV peak-peak) or non-shockable rhythms were paired with adjacent segments free of CPR artifact, enabling annotators to accurately identify the rhythm. One reviewer annotated and another verified rhythm designations, resulting in a development set of 2811 (716 coarse VF + 2095 non-shockable) CPR-artifacted segments and an independent test set of 2758 (707 coarse VF + 2051 non-shockable) segments. “Pct Pauses” was defined as the number of segments classified “Pause CPR for Analysis” divided by the total number of segments. Sensitivity, specificity, positive predictivity, and negative predictivity were calculated for segments not designated for a pause (i.e. analyzed during CPR). Results were compared to AHA recommended sensitivity and specificity for artifact-free ECG.
Results: See table.
Conclusions: ADC algorithm performance, evaluated on development and test sets of ECG with CPR artifact, exceeded the AHA recommendation for performance on ECG without CPR artifact. Incorporation of this algorithm into an AED may eliminate about 80% of analysis pauses without compromising analysis accuracy and may consequently improve the likelihood of resuscitation.
- © 2012 by American Heart Association, Inc.