Abstract 219: Prediction of Shock-Refractory Ventricular Fibrillation Amidst Continuous Chest Compressions During Out-of-Hospital Cardiac Arrest Resuscitation

Background: Shock-refractory (SR) ventricular fibrillation (VF) patients (requiring ≥ 3 shocks) have relatively poor outcomes. Prediction of SR-VF may enable preemptive interventions targeted to improve SR-VF outcomes, such as earlier antiarrhythmics, alternative shock delivery strategies, or early transport for hospital-based interventions. We recently demonstrated prediction of SR-VF in the context of current practice whereby chest compressions (CCs) are interrupted for electrocardiogram (ECG) analysis prior to the initial shock. However, since pausing CCs is detrimental to outcome, ideally next-generation defibrillator algorithms will analyze during uninterrupted CCs, precluding CC interruption for SR-VF prediction.

Aim: We designed a machine learning algorithm to predict SR-VF based on ECGs surrounding the first shock during CCs.

Methods: We performed a retrospective study of adult out-of-hospital VF cardiac arrests treated by paramedics from 2008-2020 in King County WA. From each patient we collected a pair of 3-s VF ECG segments during uninterrupted CCs prior to and 1 minute following the first shock. Patients were randomized into 80%/20% groups for training/test. A random forest classifier was trained to predict SR-VF using dimensionally-reduced ECG scalograms. Performance was assessed by area under the receiver operating characteristic curve (AUC) and sensitivity for >90% specificity.

Results: Of 940 included patients, 584 (62%) had SR-VF, median age was 62, 206 (22%) were female, and 429 (46%) survived to hospital discharge. AUC for predicting SR-VF was 0.80 (95% CI: 0.72-0.85) in 188 test patients (Figure), with specificity of 92% and sensitivity of 52%.

Conclusion: Automated SR-VF prediction during ongoing CCs is possible using the ECG surrounding the initial shock. Early identification of SR-VF during CCs could prompt preemptive, patient-specific treatments without compromising CPR, potentially improving resuscitation outcome.