Abstract 331: Novel Electrocardiogram Features and Dichotomous Variables to Predict Resuscitation Outcomes During Chest Compressions

• Published in: Circulation (New York, N.Y.) Vol. 140; no. Suppl_2
• Main Authors: Coult, Jason, Blackwood, Jennifer E, Rea, Thomas, Kudenchuk, Peter, Kwok, Heemun
• Format: Journal Article
• Language: English
• Published: 19.11.2019
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.140.suppl_2.331

Abstract only Objective: Analysis of the ventricular fibrillation (VF) ECG to predict patient outcomes can potentially improve defibrillation timing and monitor prognosis during resuscitation. However, accurate ECG analysis requires pausing CPR. We evaluated strategies to predict patient outcomes during CPR using novel ECG features, adaptive filtering, and dichotomous variables. Methods: We collected a 5-s VF ECG segment during chest compressions prior to 2137 shocks in 942 out-of-hospital VF arrests. Using training segments from 376 patients (40%), we designed ten features of the VF ECG to characterize amplitude, time-frequency energy, and complexity. Features were calculated with and without adaptive filtering. We used support vector machine and logistic models to combine ECG features with dichotomous variables presumed available during treatment: age<60, sex, and prior shock success. Using segments from 566 patients (60%) for validation, we evaluated performance of individual variables and the combined model. For comparison, we also computed the amplitude spectrum area (AMSA) ECG metric. Performance was characterized by area under the receiver operating characteristic curve (AUC) to predict return of organized rhythm after shock and survival with Cerebral Performance Category of 1-2. Results: Of individual ECG variables, filtering improved amplitude features, and the best performance was observed using high-frequency energy features above CPR artifact frequencies (Table 1). Of dichotomous variables, age<60 predicted survival but not return of rhythm, and prior shock success was more indicative of return of rhythm than survival. The combined model had improved performance versus AMSA for predicting return of rhythm (AUC=0.74 vs. 0.66, p<0.001) and survival (AUC=0.75 vs. 0.70, p<0.001). Conclusion: Combining ECG characteristics, adaptive filtering, and dichotomous variables may improve prediction of defibrillation outcomes during uninterrupted CPR.