Analysis of the Ventricular Fibrillation Electrocardiogram During Cardiopulmonary Resuscitation to Predict Outcome of Out-of-Hospital Cardiac Arrest

Abstract

Out-of-hospital ventricular fibrillation (VF) cardiac arrest results in approximately 50,000 deaths per year in the United States. Treatment includes defibrillation shock supported by cardiopulmonary resuscitation (CPR). Current defibrillator-guided resuscitation protocol follows a fixed sequence. This sequence consists of CPR interrupted at 2-minute intervals to allow rhythm classification and defibrillation when VF is detected. However, defibrillation attempts frequently fail and overall survival from VF arrest is low. Studies have demonstrated that outcomes might be improved by (1) minimizing interruptions in CPR and (2) delaying shock to allow an interim of CPR and medications when patient myocardial status is poor and likelihood of defibrillation success is low. Quantitative measures of the VF electrocardiogram (ECG) have therefore been proposed to estimate the likelihood of defibrillation success and serve as a surrogate for a patient’s myocardial status. These measures have potential to improve survival by guiding the administration of shock, CPR, and medications. However, clinical application of VF waveform measures is challenged by their modest prognostic performance and the requirement to pause CPR for analysis (as interrupting CPR reduces patient survival). In these works, we therefore sought to determine whether the performance of waveform measures could be improved and whether measures could be calculated without interrupting CPR. Using retrospective analysis of out-of-hospital cardiac arrests, we investigated the potential for VF waveform measure calculation during short incidental pauses in CPR (Chapter 2) and during continuous uninterrupted CPR (Chapter 3, Chapter 4). We also sought to determine whether a novel prognostic algorithm could improve performance during uninterrupted CPR versus existing measures (Chapter 5, Chapter 6). In these investigations, we observed that waveform measures can indeed be calculated during incidental pauses in compressions a fraction of a second in length. We confirmed that measures are compromised when applied during uninterrupted CPR, but that their performance in presence of CPR artifact can be improved by incorporating ECG rhythm history, modifying parameters to mitigate chest compression artifact, and using machine learning to combine multiple measures. Furthermore, we observed that a novel algorithm specifically designed to predict defibrillation outcomes during CPR can further-improve prognostic performance by incorporating automatic CPR detection, adaptive filtering, patient characteristics, and features of the ECG designed to reduce the effects of CPR artifact. These results suggest potential for continuous real-time monitoring of patient status during resuscitation to guide administration of therapy and improve survival from out-of-hospital VF arrest.