Automatic ROSC detection during cardiopulmonary resuscitation using accelerometry of feedback devices

Research question Manual pulse check leaves the identification of a return of spontaneous circulation (ROSC) during cardiac arrest treatment to be challenging due to its lack of reliability. Thus, we developed an algorithm which reliably determines circulation state and detects a ROSC during cardiopulmonary resuscitation (CPR) based on electrocardiogram and accelerometry data from real-world defibrillator records. Methodology 480 cases were obtained from the German Resuscitation Registry, and annotated by physicians using newly developed Python scripts. A recently published algorithm [1][2] was used to flag intervals with chest compressions absent. Those intervals were fragmented into 4-second long snippets containing electrocardiogram (ECG) and accelerometry and a label (’spontaneous circulation’ or ’cardiac arrest), which were further preprocessed to overcome shortcomings of retrospective annotation and corrupted data. From these snippets we computed 14 features as an input for a machine learning algorithm, which we trained on a subset of the data set. The performance of the algorithm to detect the circulation state was assessed on a separate subset unused for training. Results The algorithm exhibits a sensitivity of 93.0% and a specificity of 95.7% for detecting the circulation state during CPR. Using only ECG data leads to an inferior performance compared to additionally employing accelerometry data for classification. InterpretationAccelerometry data improves classification performance significantly compared to using only ECG data. The proposed algorithm, if implemented into defibrillators by a firmware update, may aid clinicians in ROSC detection and subsequent decision making. Another potential application is the retrospective analysis of defibrillator data for clinical quality management and research, by simplifying the laborious manual annotation. Further improvements to the algorithm (including better data quality) could additionally enhance its performance, although further clinical validation of the algorithm is needed. References [1] Simon Orlob, Wolfgang J. Kern, Birgitt Alpers et al. Chest compression fraction calculation: A new, automated, robust method to identify periods of chest compressions from defibrillator data – tested in zoll x series. Resuscitation, 2022. doi: 10.1016/j.resuscitation.2021.12.028 [2] Wolfgang J. Kern, Simon Orlob, Birgitt Alpers et al. A sliding-window based algorithm to determine the presence of chest compressions from acceleration data. Data in Brief, 2022, S. 107973. doi: 10.1016/j.dib.2022.107973