Cycle-to-Cycle Prediction of Return of Spontaneous Circulation Using End-Tidal Carbon Dioxide Dynamics in Out-of-Hospital Cardiac Arrest.

Resuscitation · 2026-11-12 · Method & validation

Abstract

BACKGROUND: Prognostication in out-of-hospital cardiac arrest (OHCA) commonly relies on static pre-resuscitation variables, although resuscitation is a rapidly evolving physiologic process. End-tidal carbon dioxide (EtCO2) is continuously available during cardiopulmonary resuscitation, but its sequential information has not been fully used for real-time prediction. We aim to develop and internally validate a cycle-to-cycle time-updated prediction model using sequential EtCO2 values to estimate the probability of return of spontaneous circulation (ROSC) in the subsequent 2-minute resuscitation cycle. METHODS: We conducted a multicenter retrospective study of adult OHCA patients from the NTUH-HYCARD database between 2016 and 2023. Event-level EtCO2 sequences were extracted from resuscitation records and standardized into 2-minute intervals for up to 30 minutes, consistent with Advanced Cardiac Life Support reassessment cycles. At each prediction time point, the model used EtCO2 values available from event onset to the current time point, without using future values. A two-layer long short-term memory model estimated the probability of ROSC in the subsequent cycle. Performance was evaluated using 10-fold cross-validation at two levels: the event level (one prediction per event) and the cycle level (one prediction per 2-minute cycle). RESULTS: The final analysis included 1,651 event-level EtCO2 sequences from 1,480 patients; 583 events ended with ROSC. At the event level, the model achieved an area under the receiver operating characteristic curve (AUROC) of 0.876 (95% CI, 0.863-0.890) and a Brier score of 0.140 (95% CI, 0.131-0.149). When applied to every 2-minute cycle (17,029 prediction points; ROSC prevalence, 3.4%), the AUROC was 0.685 (95% CI, 0.664-0.706), and the model systematically overestimated ROSC probability (Brier score, 0.285; 95% CI, 0.268-0.303). CONCLUSION: Sequential EtCO2 dynamics showed good event-level discrimination for predicting ROSC in the subsequent cycle. When applied at each cycle, the model retained modest discrimination but overestimated ROSC probability, indicating that recalibration is required before cycle-to-cycle use.

Abstract © the authors or publisher, via PubMed. Reuse and removal

Tags

Capnography · Mortality & survival

Related articles