Abstract Sun508: Chest Compression Quality Phenotypes and ROSC in Out-of-Hospital Cardiac Arrest

Abstract

Background: High-quality chest compressions are associated with improved outcomes from out-of-hospital cardiac arrest (OHCA). However, minute-to-minute chest compression patterns may vary across patients. Using artificial intelligence (AI), we aimed to characterize novel chest compression quality patterns and their associations with return of spontaneous circulation (ROSC). Method: We analyzed cardiopulmonary resuscitation (CPR) process files collected from adult OHCA enrolled in the Pragmatic Airway Resuscitation Trial (PART). We used validated automated signal processing techniques to calculate chest compression rates (CCR) in one-minute epochs. We categorized each minute into four CCR categories: 1) low (<95 compressions per minute, cpm), 2) acceptable (95-125 cpm), 3) high (>125 cpm) and 4) chest compression pauses or missing (0 cpm). For each patient, we computed the proportion of time spent in each category up to the first 40 minutes. We applied AI K-means clustering with Elbow and Silhouette methods to identify distinct CCR pattern phenotypes. We visualized the CCR patterns using heatmaps. We assessed the association of CCR phenotype with ROSC using logistic regression. Results: Of 3004 patients, we included 2231 patients with recorded chest compressions. Among these, 559 (25.1%) achieved ROSC on ED arrival. AI identified different CCR patterns between patients: phenotype 1, n = 378 (mostly acceptable CCR: 61.4%); phenotype 2, n = 110 (CCR split between low: 38% and acceptable: 51.2%); phenotype 3, n = 193 (mostly high CCR: 71.6%); phenotype 4, n = 1,341 (predominantly acceptable CCR: 90.8%); phenotype 5, n= 206 (Missing/Pauses: 41.7%) (Figure 1). Excluding phenotype 5, CCR phenotype was not associated with ROSC; phenotype 1, reference; phenotype 2, OR: 0.79 (95% CI: 0.46–1.34); phenotype 3, OR: 1.17 (95% CI: 0.78–1.75); phenotype 4, OR: 0.99 (95% CI: 0.75–1.30). Conclusion: In this series, AI analysis revealed 5 distinct CCR quality phenotypes. CCR phenotype was not associated with ROSC. AI analysis may provide an important tool for characterizing resuscitation quality.