End-tidal Carbon Dioxide Trajectory-based Prognostication of Out-of-hospital Cardiac Arrest

Background During cardiopulmonary resuscitation (CPR), end-tidal carbon dioxide (EtCO2) is primarily determined by pulmonary blood flow, thereby reflecting the blood flow generated by CPR. We aimed to develop an EtCO2 trajectory-based prediction model for prognostication at specific time points during CPR in patients with out-of-hospital cardiac arrest (OHCA). Methods We screened patients receiving CPR between 2015–2021 from a prospectively collected database of a tertiary-care medical center. The primary outcome was survival to hospital discharge. We used group-based trajectory modeling to identify the EtCO2 trajectories. Multivariable logistic regression analysis was used for model development and internally validated using bootstrapping. We assessed performance of the model using the area under the receiver operating characteristic curve (AUC). Results The primary analysis included 542 patients with a median age of 68.0 years. Three distinct EtCO2 trajectories were identified in patients resuscitated for 20 minutes (min): low (average EtCO2 10.0 millimeters of mercury [mm Hg]; intermediate (average EtCO2 26.5 mm Hg); and high (average EtCO2: 51.5 mm Hg). Twenty-min EtCO2 trajectory was fitted as an ordinal variable (low, intermediate, and high) and positively associated with survival (odds ratio 2.25, 95% confidence interval [CI] 1.07–4.74). When the 20-min EtCO2 trajectory was combined with other variables, including arrest location and arrest rhythms, the AUC of the 20-min prediction model for survival was 0.89 (95% CI 0.86–0.92). All predictors in the 20-min model remained statistically significant after bootstrapping. Conclusion Time-specific EtCO2 trajectory was a significant predictor of OHCA outcomes, which could be combined with other baseline variables for intra-arrest prognostication. For this purpose, the 20-min survival model achieved excellent discriminative performance in predicting survival to hospital discharge.


INTRODUCTION
The annual incidence of out-of-hospital cardiac arrest (OHCA) is estimated to be 28-44 cases per 100,000 population worldwide. 1The estimated proportion of survival to discharge in OHCA was 7.6% in Europe, 6.8% in North America, 3.0% in Asia, and 9.7% in Australia. 1 High-quality cardiopulmonary resuscitation (CPR) is critical in improving OHCA outcomes. 2,3Capnography is recommended to monitor CPR quality in real time and adjust chest compression quality accordingly. 2,3During CPR, end-tidal carbon dioxide (EtCO 2 ) is primarily determined by pulmonary blood flow, thereby reflecting the blood flow generated by CPR. 4,5he 2020 International Liaison Committee on Resuscitation (ILCOR) consensus 6,7 recommended that EtCO 2 ≥20 millimeters of mercury (mm Hg) measured after 20 minutes (min) of CPR may predict survival to discharge.Nonetheless, this weak recommendation was supported by only moderate-quality evidence.A 2018 ILCOR systematic review noticed that the measurement time points of EtCO 2 were very heterogeneous across different studies. 8Accordingly, ILCOR 6,7 suggested that instead of single EtCO 2 values, the EtCO 2 trend should be further explored in future studies for its prognostic performance.
The previous study noted that EtCO 2 trajectory during CPR was associated with OHCA outcomes. 9However, the predictive ability of EtCO 2 trajectory at a specific timing was not explored in the previous study. 9Whether EtCO 2 can be combined with other metrics for intra-arrest prognostication was considered a critical knowledge gap by the 2020 American Heart Association (AHA) guidelines. 2 In our recent study, 10 we incorporated the minimum EtCO 2 value into the return of spontaneous circulation after cardiac arrest (RACA) score and improved the performance of RACA score in predicting ROSC, suggesting that EtCO 2 could potentially help intra-arrest prognostication.
In the current study, we further developed models that could predict survival at hospital discharge.Instead of a single EtCO 2 value, 10 we attempted to combine EtCO 2 trajectory and other predictors in deriving prediction models.Moreover, these models were developed using time-specific windows to prognosticate patient outcomes during resuscitation, including 10-and 20 min 6,7 after initiation of CPR.

MATERIALS AND METHODS
This observational study was a secondary analysis of a prospectively collected OHCA database registered in the emergency department (ED) of National Taiwan University Hospital (NTUH).The institutional review board approved this study (reference number: 201906082RINB) and waived the requirement for informed consent.The study was performed according to the recommendations from Worster et al 11 regarding health record review studies in emergency medicine research with all elements followed.The results are reported according to the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement. 12

Study Setting
The NTUH is a tertiary-care medical center with 2,600 beds, including 220 beds in intensive care units.Approximately 100,000 patients visit NTUH ED annually.Patients with OHCA are transported directly to the resuscitation bay of the critical care area in the ED for CPR, which is delivered according to resuscitation guidelines. 2,3lso, since 2013 ED staff have been trained with the A-C-L-S (airway-circulation-leadership-support) teamwork model 9,13,14 to streamline the resuscitation process via both strengthened technical and non-technical skills. 15,16Any intervention, such as tracheal intubation performed during CPR, are timestamped by nurses with a specially designed mobile application.The EtCO 2 is recorded every two min right before pulse check.The EtCO 2 is monitored with devices attached to the advanced airways, including supraglottic airways and endotracheal tubes.For patients with OHCA who never achieve return of spontaneous circulation (ROSC), CPR is usually performed for at least 30 min in the ED, except for those with a documented do-not-resuscitate (DNR) order.

What do we already know about this issue?
The end-tidal carbon dioxide (EtCO 2 ) level during cardiopulmonary resuscitation (CPR) is associated with outcomes following out-ofhospital cardiac arrest (OHCA).
What was the research question?Could EtCO 2 trajectories during CPR be combined with baseline variables to predict outcomes of OHCA?
What was the major finding of the study?
The area under the curve of the EtCO 2 -based model for survival was 0.89 (95% confidence interval 0.86-0.92).

How does this improve population health?
An EtCO 2 trajectory-based prediction model may help emergency medical services to predict OHCA outcomes and facilitate allocation of medical resources.

Study Population
Patients with OHCA sent to the NTUH ED between January 1, 2015-December 31, 2021 were screened.The inclusion criteria for the study were as follows: 1) nontraumatic arrest; 2) absence of ROSC before ED arrival; (3) absence of documented DNR order before CPR; 4) age ≥18 years; and 5) insertion of advanced airways during CPR.Based on the CPR duration, the included patients would be further selected for primary and secondary analyses.If the included patients received CPR ≥20 min and had EtCO 2 measurements ≥3 times within 20 min of CPR, they would be selected into the 20-min group for the primary analysis.Similarly, if the included patients received CPR ≥10 min and had EtCO 2 measurements ≥3 times within 10 min of CPR, they would be selected into the 10-min group for secondary analysis.

Data Collection, Variable Definitions, and Outcome Measures
In the NTUH database, OHCA events were recorded based on the Utstein template. 17Data requested for analysis included age, gender, variables derived from the Utstein template, advanced airway insertion timing, EtCO 2 values with measurement timing, and outcomes.For ED resuscitation, the time point of the initial chest compression delivered in the ED was set as time zero for reference.Time to advanced airway use was defined as the interval between time zero and time for completing advanced airway insertion.If advanced airway devices were inserted before ED arrival, the time to advanced airway was recorded as zero.Duration of CPR in the ED referred to the time interval between time zero and the end of resuscitation, either due to ROSC or death.Time-specific EtCO 2 referred to the EtCO 2 level measured after the specific time elapsing following time zero.
The primary outcome was survival status at the time of hospital discharge.The secondary outcome was ROSC, defined as a palpable pulse for 20 seconds. 18Data abstraction for the current analysis was performed by trained researchers who were blinded to the study hypothesis.

Statistical Analysis
In the primary analysis, we used the 20-min group to build models for predicting survival (20-min survival model) and ROSC (20-min ROSC model).In the secondary analysis, similar procedures were applied to develop the 10-min survival model and 10-min ROSC model.We first performed group-based trajectory modeling (GBTM) to identify trajectory groups based on the EtCO 2 level.The GBTM is an explanatory modeling technique to identify hidden groups of individuals with similar trajectories for a particular variable of interest. 19The GBTM performs better when longitudinal data is measured at least three times.
For descriptive statistics, categorical variables are presented as proportions, and continuous variables are presented as medians with interquartile ranges.We examined categorical variables using the chi-squared test, whereas continuous variables were compared using the Kruskal-Wallis test or Mann-Whitney test, as appropriate.We used multivariable logistic regression analyses to develop the prediction models.All available variables, including basic demographics, peri-CPR events, and EtCO 2 trajectory were accounted for in the regression model via a stepwise, variable selection procedure.The EtCO 2 trajectory would be tested as ordinal or categorical variables in the modelbuilding process.We used generalized additive models (GAM) 20 to identify the appropriate cutoff point(s) for dichotomization.The discriminative performance and calibration of the prediction model were assessed by area under the receiver operating characteristic curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test, respectively.We internally validated the prediction model using the bootstrapping procedure with 1,000 repetitions to examine the robustness of the effect estimate of each variable in the prediction model.
We performed GBTM and bootstrapping using the traj package and bootstrap procedure of Stata software (StataCorp LLC, College Station, TX), respectively.We used the R 4.1.1software (R Foundation for Statistical Computing, Vienna, Austria) for other analyses.A twotailed P-value <0.05 was considered statistically significant.

RESULTS
The patient selection procedure resulted in 542 and 532 patients in the 20-min and 10-min groups, respectively (Figure 1).The two groups were not mutually exclusive.Because not all patients in the 20-min group had EtCO 2 measurements ≥3 times within 10 mins, the 20-min group patients may not have been necessarily included in the 10-min group.Also, because some of the patients in the 10-min group would achieve ROSC within 20 min of CPR, the 10-min group patients would not necessarily have been included in the 20-min group.Therefore, there was an overlap of 385 patients between the 20-min and 10-min groups who met the selection criteria for both groups.
In the primary analysis, we identified and named three EtCO 2 trajectories as low, intermediate, and high trajectories according to their respective average EtCO 2 levels (Figure 2).The characteristics of the 20-min group and comparisons between these EtCO 2 trajectories are presented in Table 1.The median CPR duration in the ED was 31.0 minutes, and the median number of EtCO 2 measurements was eight.A total of 25 (4.6%) patients survived at hospital discharge.There seems to be an increasing trend of survival from low to high EtCO 2 trajectory.The comparisons between patients stratified by survival are shown in Supplemental Table 1.During the model development, the 20-min EtCO 2 trajectory was fitted as an ordinal variable by the logistic regression analysis and positively associated with survival (odds ratio [OR] 2.25, 95% confidence interval [CI] 1.07-4.74)and ROSC (OR 2.46, 95% CI 1.78-3.41)(Table 2).In other words, compared with the low EtCO 2 trajectory, the intermediate trajectory had 2.25 times higher odds of survival to hospital discharge.Similarly, compared with the intermediate trajectory, the high EtCO 2 trajectory also had 2.25 times higher odds of survival.When the 20-min EtCO 2 trajectory was combined with other variables, the AUCs of the 20-min survival and ROSC models were 0.89 (95% CI 0.86-0.92)and 0.78 (95% CI 0.74-0.81),respectively.
Similarly, in the secondary analysis we identified three EtCO 2 trajectories (Figure 2 and Table 3).The median CPR duration in the ED was 30.0 min, and the median number of EtCO 2 measurements was four.A total of 34 (6.4%) patients survived at hospital discharge.Significant survival differences were noted among the three EtCO 2 trajectories; nonetheless, the survival of intermediate and high EtCO 2 trajectories was similar.The survival-stratified comparisons are shown in Supplemental Table 2.During the model-fitting process, the 10-min EtCO 2 trajectory was fitted as a categorical variable.As shown in Table 4, compared with the 10-min low EtCO 2 trajectory, the 10-min intermediate or high EtCO 2 trajectory was significantly associated with    For the 20-and 10-min models, all the predictors remained significantly associated with outcomes after the bootstrapping procedure, indicating the robustness of these models (Supplemental Table 3).

DISCUSSION Main Findings
By using a prospectively collected database, we identified that the time-specific EtCO 2 trajectory was a significant intra-arrest outcome predictor.Time-specific EtCO 2 trajectory could be combined with other predictors to assist in intra-arrest prognostication at different time points during CPR.Among all the prediction models, the 20-min EtCO 2 trajectory-based survival model achieved the highest discriminative performance (AUC 0.89).

Comparison with Previous Studies
For outcome prediction in OHCA, most models were developed for patients who had already achieved ROSC. 21here were few, if any, models available for patients who were still undergoing CPR.3][24] All the predictors included in the RACA score were baseline variables, such as arrest location and arrest rhythms, which did not consider the treatment effects of CPR.Nonetheless, it was possible that even though the RACA score-predicted ROSC probabilities were similar, the actual outcomes may differ because of different CPR qualities and durations delivered by rescuers.To make individualized intraarrest prognostication, variables specific to the patient and resuscitation process, such as EtCO 2 , may be necessary,.
The 2018 ILCOR systematic review 8 indicated that EtCO 2 was associated with ROSC probability.Nonetheless, the optimal parameter of EtCO 2 for prognostication is still debated. 8For example, despite its convenience in statistical analysis, average EtCO 2 could not differentiate between different EtCO 2 trajectories.Ascending and descending EtCO 2 trajectories may have similar average EtCO 2 , but their prognoses may be very different. 25,26Moreover, the term "initial" EtCO 2 may not accurately reflect the EtCO 2 level during the early phase of CPR, as the endotracheal tube could potentially be introduced later during the resuscitation.It was reported that the specificity of EtCO 2 in predicting ROSC would increase progressively from 50% at 0 min to 60%, 98%, and 100% at 10, 15, and 20 min, respectively. 27herefore, for EtCO 2 to be a valid predictor, the timing of prognostication should be specified, and its trend during CPR, instead of a single value, should be adopted.

Interpretation of Current Analysis
The 2020 ILCOR consensus 6,7 recommends that EtCO 2 measured after 20 min of CPR may be a predictor of survival to discharge.Rosman et al 28 indicated that when higher EtCO 2 levels were reached beyond 20 min of CPR they may not lead to ROSC.Progressively worsening ischemia may cause refractoriness to CPR during the metabolic phase of cardiac arrest, 29 and EtCO 2 trajectories beyond 20 min may not be prognostic of outcomes.Therefore, CPR for 20 min was used to select the 20-min cohort and identify the 20-min EtCO 2 trajectory.The advantage of employing GBTM was that it offered an efficient method to unravel the hidden trajectories that may not be readily recognizable from the   1), which may also explain better outcomes in the latter.
In the 20-min survival model, the multivariable logistical regression analysis indicated that the 20-min EtCO 2 trajectory was positively associated with survival, demonstrating the trend    1).Therefore, instead of the arrest etiology, the CPR quality may account for the positive association between 20-min EtCO 2 trajectory and survival.Whether EtCO 2 , along with other factors, can be used for intra-arrest prognostication was listed by AHA guidelines 2 as an important knowledge gap.In the 20-min survival model, besides EtCO 2 trajectory, other baseline variables, including arrest at home, prehospital defibrillation by EMS, and initial shockable rhythms on ED arrival, were also selected as significant predictors.These baseline variables had been well-validated for their predictive performance in previous studies. 18The 20-min survival model achieved excellent discriminative performance and may first answer the question presented by the AHA. 2 Moreover, we further tested whether the 20-min EtCO 2 trajectory could facilitate predicting ROSC.However, the AUC of the 20-min ROSC model was 0.78, lower than that of the 20-min survival model.In our study, ROSC was defined as a palpable pulse for 20 seconds, as used by RACA score. 18The swift nature of this secondary outcome may render it difficult to be predicted, even though the 20-min ROSC model included more variables than the 20-min survival model.
Finally, we developed the 10-min prediction models to explore whether outcomes could be predicted at an earlier time point during CPR.Nevertheless, the AUCs of both 10-min models were respectively lower than their counterparts of 20-min models.As shown in Figure 2, the 10-min EtCO 2 trajectory was slightly different from the 20-min EtCO 2 trajectory in the trend pattern.For example, the high EtCO 2 trajectory continued to rise within 10 min; it was only evident later in the 20-min window that the trajectory had plateaued.Taken together, these time-specific models varied over time in terms of trajectory shapes and model performance.Earlier trajectories may still be evolving with moderate model performance, while late trajectories may have improved model performance at the cost of more medical recourses consumed.Our data suggested that 20 min after CPR may be the earliest point in time with excellent model performance to predict distant, clinically important outcomes, such as survival to hospital discharge.

Future Applications
For OHCA patients transported to the ED for continuous CPR, emergency clinicians are faced with the problem of balancing the probability of a favorable outcome with the utilization of current and future resources when making important decisions, such as termination of resuscitation or implementation of invasive extracorporeal CPR. 34Most of these advanced interventions are reserved for patients receiving CPR within a certain duration. 34espite the fact that CPR duration is known to be inversely associated with favorable outcomes, 35 it may not be the sole prognostic factor.Quality CPR may facilitate maintaining patients' potential for favorable outcomes and lengthen the time window for advanced interventions to be implemented.Our prediction models demonstrated that time-specific EtCO 2 trajectory, taking into account both the CPR duration and quality, could be a significant intra-arrest prognostic factor.In the future, time-specific EtCO 2 may be transmitted instantaneously from EtCO 2 monitors to mobile devices with the assistance of advanced information and communication technology.The predicted outcomes could be updated instantaneously minute by minute for each individual patient and may not be restricted to a certain time point during CPR, such as 20 min or 10 min, as used in our study.

LIMITATIONS
First, while we had internally validated the prediction models by using the bootstrap method, further external validation in other datasets should be performed.Second, the analyzed EtCO 2 dataset was derived from a prospectively collected database of a single ED with a specialized training model for CPR.Further studies are needed to investigate whether these models could be generalized to other EDs or prehospital resuscitation.

CONCLUSION
Time-specific EtCO 2 trajectory was a significant predictor of OHCA outcomes, which could be combined with other baseline variables for intra-arrest prognostication.For this purpose, the 20-min survival model achieved the highest discriminative performance in predicting survival to hospital discharge.

Table 1 .
Characteristics of patients included in the twenty-minute group stratified by end-tidal carbon dioxide trajectory group.

Table 3 .
Characteristics of patients included in the ten-min group stratified by end-tidal carbon dioxide trajectory group.

Table 4 .
33, emergency department; EMS, emergency medical service; mm HG, millimeters of mercury; ETT, endotracheal tube; ROSC, return of spontaneous circulation; SGA, supraglottic airway.Multivariable logistic regression analysis for ten-minute group to build end-tidal carbon dioxide trajectory-based prediction models.Ten-min survival model: goodness-of-fit assessment: n = 532, adjusted generalized R 2 = 0.14, estimated area under the receiver operating characteristic curve = 0.76 (95% confidence interval: 0.72-0.79),andHosmerandLemeshowgoodness-of-fitchi-squaredtestP= 0.79; ten-min ROSC model: goodness-of-fit assessment: n = 532, adjusted generalized R 2 = 0.25, estimated area under the receiver operating characteristic curve = 0.75 (95% confidence interval: 0.71-0.79),andHosmerandLemeshowgoodness-of-fitchi-squaredtestP= 0.65.CPR, cardiopulmonary resuscitation; ED, emergency department; EMS, emergency medical service; ROSC, return of spontaneous circulation. of a higher EtCO 2 trajectory with increased survival.Studies revealed that for every 10 mm increase in chest compression depth, EtCO 2 would increase by 1.4 mm Hg 30 or 4.0%.31HigherEtCO 2 trajectory may suggest better CPR quality, which may explain the positive association between EtCO 2 trajectory and chances of survival.In contrast, arrest etiology may also be a confounding factor in explaining the associations between favorable outcomes and intermediate or high EtCO 2 trajectory.Studies have shown that patients with asphyxial arrest32or suspected respiratory etiology 33 may have higher EtCO 2 levels than those with initial shockable rhythms32or suspected cardiac etiology,33respectively. Nonetheless, in our cohort, patients of intermediate or high EtCO 2 trajectory had higher proportions of prehospital defibrillation by emergency medical services (EMS) (Table